xref: /freebsd/sys/contrib/openzfs/cmd/ztest.c (revision a4adfaf712694ce7923d5309cf87d0cd2a598953)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Joyent, Inc.
28  * Copyright (c) 2017, Intel Corporation.
29  */
30 
31 /*
32  * The objective of this program is to provide a DMU/ZAP/SPA stress test
33  * that runs entirely in userland, is easy to use, and easy to extend.
34  *
35  * The overall design of the ztest program is as follows:
36  *
37  * (1) For each major functional area (e.g. adding vdevs to a pool,
38  *     creating and destroying datasets, reading and writing objects, etc)
39  *     we have a simple routine to test that functionality.  These
40  *     individual routines do not have to do anything "stressful".
41  *
42  * (2) We turn these simple functionality tests into a stress test by
43  *     running them all in parallel, with as many threads as desired,
44  *     and spread across as many datasets, objects, and vdevs as desired.
45  *
46  * (3) While all this is happening, we inject faults into the pool to
47  *     verify that self-healing data really works.
48  *
49  * (4) Every time we open a dataset, we change its checksum and compression
50  *     functions.  Thus even individual objects vary from block to block
51  *     in which checksum they use and whether they're compressed.
52  *
53  * (5) To verify that we never lose on-disk consistency after a crash,
54  *     we run the entire test in a child of the main process.
55  *     At random times, the child self-immolates with a SIGKILL.
56  *     This is the software equivalent of pulling the power cord.
57  *     The parent then runs the test again, using the existing
58  *     storage pool, as many times as desired. If backwards compatibility
59  *     testing is enabled ztest will sometimes run the "older" version
60  *     of ztest after a SIGKILL.
61  *
62  * (6) To verify that we don't have future leaks or temporal incursions,
63  *     many of the functional tests record the transaction group number
64  *     as part of their data.  When reading old data, they verify that
65  *     the transaction group number is less than the current, open txg.
66  *     If you add a new test, please do this if applicable.
67  *
68  * (7) Threads are created with a reduced stack size, for sanity checking.
69  *     Therefore, it's important not to allocate huge buffers on the stack.
70  *
71  * When run with no arguments, ztest runs for about five minutes and
72  * produces no output if successful.  To get a little bit of information,
73  * specify -V.  To get more information, specify -VV, and so on.
74  *
75  * To turn this into an overnight stress test, use -T to specify run time.
76  *
77  * You can ask more vdevs [-v], datasets [-d], or threads [-t]
78  * to increase the pool capacity, fanout, and overall stress level.
79  *
80  * Use the -k option to set the desired frequency of kills.
81  *
82  * When ztest invokes itself it passes all relevant information through a
83  * temporary file which is mmap-ed in the child process. This allows shared
84  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
85  * stored at offset 0 of this file and contains information on the size and
86  * number of shared structures in the file. The information stored in this file
87  * must remain backwards compatible with older versions of ztest so that
88  * ztest can invoke them during backwards compatibility testing (-B).
89  */
90 
91 #include <sys/zfs_context.h>
92 #include <sys/spa.h>
93 #include <sys/dmu.h>
94 #include <sys/txg.h>
95 #include <sys/dbuf.h>
96 #include <sys/zap.h>
97 #include <sys/dmu_objset.h>
98 #include <sys/poll.h>
99 #include <sys/stat.h>
100 #include <sys/time.h>
101 #include <sys/wait.h>
102 #include <sys/mman.h>
103 #include <sys/resource.h>
104 #include <sys/zio.h>
105 #include <sys/zil.h>
106 #include <sys/zil_impl.h>
107 #include <sys/vdev_draid.h>
108 #include <sys/vdev_impl.h>
109 #include <sys/vdev_file.h>
110 #include <sys/vdev_initialize.h>
111 #include <sys/vdev_raidz.h>
112 #include <sys/vdev_trim.h>
113 #include <sys/spa_impl.h>
114 #include <sys/metaslab_impl.h>
115 #include <sys/dsl_prop.h>
116 #include <sys/dsl_dataset.h>
117 #include <sys/dsl_destroy.h>
118 #include <sys/dsl_scan.h>
119 #include <sys/zio_checksum.h>
120 #include <sys/zfs_refcount.h>
121 #include <sys/zfeature.h>
122 #include <sys/dsl_userhold.h>
123 #include <sys/abd.h>
124 #include <sys/blake3.h>
125 #include <stdio.h>
126 #include <stdlib.h>
127 #include <unistd.h>
128 #include <getopt.h>
129 #include <signal.h>
130 #include <umem.h>
131 #include <ctype.h>
132 #include <math.h>
133 #include <sys/fs/zfs.h>
134 #include <zfs_fletcher.h>
135 #include <libnvpair.h>
136 #include <libzutil.h>
137 #include <sys/crypto/icp.h>
138 #if (__GLIBC__ && !__UCLIBC__)
139 #include <execinfo.h> /* for backtrace() */
140 #endif
141 
142 static int ztest_fd_data = -1;
143 static int ztest_fd_rand = -1;
144 
145 typedef struct ztest_shared_hdr {
146 	uint64_t	zh_hdr_size;
147 	uint64_t	zh_opts_size;
148 	uint64_t	zh_size;
149 	uint64_t	zh_stats_size;
150 	uint64_t	zh_stats_count;
151 	uint64_t	zh_ds_size;
152 	uint64_t	zh_ds_count;
153 } ztest_shared_hdr_t;
154 
155 static ztest_shared_hdr_t *ztest_shared_hdr;
156 
157 enum ztest_class_state {
158 	ZTEST_VDEV_CLASS_OFF,
159 	ZTEST_VDEV_CLASS_ON,
160 	ZTEST_VDEV_CLASS_RND
161 };
162 
163 #define	ZO_GVARS_MAX_ARGLEN	((size_t)64)
164 #define	ZO_GVARS_MAX_COUNT	((size_t)10)
165 
166 typedef struct ztest_shared_opts {
167 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
168 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
169 	char zo_alt_ztest[MAXNAMELEN];
170 	char zo_alt_libpath[MAXNAMELEN];
171 	uint64_t zo_vdevs;
172 	uint64_t zo_vdevtime;
173 	size_t zo_vdev_size;
174 	int zo_ashift;
175 	int zo_mirrors;
176 	int zo_raid_children;
177 	int zo_raid_parity;
178 	char zo_raid_type[8];
179 	int zo_draid_data;
180 	int zo_draid_spares;
181 	int zo_datasets;
182 	int zo_threads;
183 	uint64_t zo_passtime;
184 	uint64_t zo_killrate;
185 	int zo_verbose;
186 	int zo_init;
187 	uint64_t zo_time;
188 	uint64_t zo_maxloops;
189 	uint64_t zo_metaslab_force_ganging;
190 	int zo_mmp_test;
191 	int zo_special_vdevs;
192 	int zo_dump_dbgmsg;
193 	int zo_gvars_count;
194 	char zo_gvars[ZO_GVARS_MAX_COUNT][ZO_GVARS_MAX_ARGLEN];
195 } ztest_shared_opts_t;
196 
197 /* Default values for command line options. */
198 #define	DEFAULT_POOL "ztest"
199 #define	DEFAULT_VDEV_DIR "/tmp"
200 #define	DEFAULT_VDEV_COUNT 5
201 #define	DEFAULT_VDEV_SIZE (SPA_MINDEVSIZE * 4)	/* 256m default size */
202 #define	DEFAULT_VDEV_SIZE_STR "256M"
203 #define	DEFAULT_ASHIFT SPA_MINBLOCKSHIFT
204 #define	DEFAULT_MIRRORS 2
205 #define	DEFAULT_RAID_CHILDREN 4
206 #define	DEFAULT_RAID_PARITY 1
207 #define	DEFAULT_DRAID_DATA 4
208 #define	DEFAULT_DRAID_SPARES 1
209 #define	DEFAULT_DATASETS_COUNT 7
210 #define	DEFAULT_THREADS 23
211 #define	DEFAULT_RUN_TIME 300 /* 300 seconds */
212 #define	DEFAULT_RUN_TIME_STR "300 sec"
213 #define	DEFAULT_PASS_TIME 60 /* 60 seconds */
214 #define	DEFAULT_PASS_TIME_STR "60 sec"
215 #define	DEFAULT_KILL_RATE 70 /* 70% kill rate */
216 #define	DEFAULT_KILLRATE_STR "70%"
217 #define	DEFAULT_INITS 1
218 #define	DEFAULT_MAX_LOOPS 50 /* 5 minutes */
219 #define	DEFAULT_FORCE_GANGING (64 << 10)
220 #define	DEFAULT_FORCE_GANGING_STR "64K"
221 
222 /* Simplifying assumption: -1 is not a valid default. */
223 #define	NO_DEFAULT -1
224 
225 static const ztest_shared_opts_t ztest_opts_defaults = {
226 	.zo_pool = DEFAULT_POOL,
227 	.zo_dir = DEFAULT_VDEV_DIR,
228 	.zo_alt_ztest = { '\0' },
229 	.zo_alt_libpath = { '\0' },
230 	.zo_vdevs = DEFAULT_VDEV_COUNT,
231 	.zo_ashift = DEFAULT_ASHIFT,
232 	.zo_mirrors = DEFAULT_MIRRORS,
233 	.zo_raid_children = DEFAULT_RAID_CHILDREN,
234 	.zo_raid_parity = DEFAULT_RAID_PARITY,
235 	.zo_raid_type = VDEV_TYPE_RAIDZ,
236 	.zo_vdev_size = DEFAULT_VDEV_SIZE,
237 	.zo_draid_data = DEFAULT_DRAID_DATA,	/* data drives */
238 	.zo_draid_spares = DEFAULT_DRAID_SPARES, /* distributed spares */
239 	.zo_datasets = DEFAULT_DATASETS_COUNT,
240 	.zo_threads = DEFAULT_THREADS,
241 	.zo_passtime = DEFAULT_PASS_TIME,
242 	.zo_killrate = DEFAULT_KILL_RATE,
243 	.zo_verbose = 0,
244 	.zo_mmp_test = 0,
245 	.zo_init = DEFAULT_INITS,
246 	.zo_time = DEFAULT_RUN_TIME,
247 	.zo_maxloops = DEFAULT_MAX_LOOPS, /* max loops during spa_freeze() */
248 	.zo_metaslab_force_ganging = DEFAULT_FORCE_GANGING,
249 	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
250 	.zo_gvars_count = 0,
251 };
252 
253 extern uint64_t metaslab_force_ganging;
254 extern uint64_t metaslab_df_alloc_threshold;
255 extern unsigned long zfs_deadman_synctime_ms;
256 extern int metaslab_preload_limit;
257 extern int zfs_compressed_arc_enabled;
258 extern int zfs_abd_scatter_enabled;
259 extern int dmu_object_alloc_chunk_shift;
260 extern boolean_t zfs_force_some_double_word_sm_entries;
261 extern unsigned long zio_decompress_fail_fraction;
262 extern unsigned long zfs_reconstruct_indirect_damage_fraction;
263 
264 
265 static ztest_shared_opts_t *ztest_shared_opts;
266 static ztest_shared_opts_t ztest_opts;
267 static char *ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
268 
269 typedef struct ztest_shared_ds {
270 	uint64_t	zd_seq;
271 } ztest_shared_ds_t;
272 
273 static ztest_shared_ds_t *ztest_shared_ds;
274 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
275 
276 #define	BT_MAGIC	0x123456789abcdefULL
277 #define	MAXFAULTS(zs) \
278 	(MAX((zs)->zs_mirrors, 1) * (ztest_opts.zo_raid_parity + 1) - 1)
279 
280 enum ztest_io_type {
281 	ZTEST_IO_WRITE_TAG,
282 	ZTEST_IO_WRITE_PATTERN,
283 	ZTEST_IO_WRITE_ZEROES,
284 	ZTEST_IO_TRUNCATE,
285 	ZTEST_IO_SETATTR,
286 	ZTEST_IO_REWRITE,
287 	ZTEST_IO_TYPES
288 };
289 
290 typedef struct ztest_block_tag {
291 	uint64_t	bt_magic;
292 	uint64_t	bt_objset;
293 	uint64_t	bt_object;
294 	uint64_t	bt_dnodesize;
295 	uint64_t	bt_offset;
296 	uint64_t	bt_gen;
297 	uint64_t	bt_txg;
298 	uint64_t	bt_crtxg;
299 } ztest_block_tag_t;
300 
301 typedef struct bufwad {
302 	uint64_t	bw_index;
303 	uint64_t	bw_txg;
304 	uint64_t	bw_data;
305 } bufwad_t;
306 
307 /*
308  * It would be better to use a rangelock_t per object.  Unfortunately
309  * the rangelock_t is not a drop-in replacement for rl_t, because we
310  * still need to map from object ID to rangelock_t.
311  */
312 typedef enum {
313 	RL_READER,
314 	RL_WRITER,
315 	RL_APPEND
316 } rl_type_t;
317 
318 typedef struct rll {
319 	void		*rll_writer;
320 	int		rll_readers;
321 	kmutex_t	rll_lock;
322 	kcondvar_t	rll_cv;
323 } rll_t;
324 
325 typedef struct rl {
326 	uint64_t	rl_object;
327 	uint64_t	rl_offset;
328 	uint64_t	rl_size;
329 	rll_t		*rl_lock;
330 } rl_t;
331 
332 #define	ZTEST_RANGE_LOCKS	64
333 #define	ZTEST_OBJECT_LOCKS	64
334 
335 /*
336  * Object descriptor.  Used as a template for object lookup/create/remove.
337  */
338 typedef struct ztest_od {
339 	uint64_t	od_dir;
340 	uint64_t	od_object;
341 	dmu_object_type_t od_type;
342 	dmu_object_type_t od_crtype;
343 	uint64_t	od_blocksize;
344 	uint64_t	od_crblocksize;
345 	uint64_t	od_crdnodesize;
346 	uint64_t	od_gen;
347 	uint64_t	od_crgen;
348 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
349 } ztest_od_t;
350 
351 /*
352  * Per-dataset state.
353  */
354 typedef struct ztest_ds {
355 	ztest_shared_ds_t *zd_shared;
356 	objset_t	*zd_os;
357 	pthread_rwlock_t zd_zilog_lock;
358 	zilog_t		*zd_zilog;
359 	ztest_od_t	*zd_od;		/* debugging aid */
360 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
361 	kmutex_t	zd_dirobj_lock;
362 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
363 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
364 } ztest_ds_t;
365 
366 /*
367  * Per-iteration state.
368  */
369 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
370 
371 typedef struct ztest_info {
372 	ztest_func_t	*zi_func;	/* test function */
373 	uint64_t	zi_iters;	/* iterations per execution */
374 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
375 	const char	*zi_funcname;	/* name of test function */
376 } ztest_info_t;
377 
378 typedef struct ztest_shared_callstate {
379 	uint64_t	zc_count;	/* per-pass count */
380 	uint64_t	zc_time;	/* per-pass time */
381 	uint64_t	zc_next;	/* next time to call this function */
382 } ztest_shared_callstate_t;
383 
384 static ztest_shared_callstate_t *ztest_shared_callstate;
385 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
386 
387 ztest_func_t ztest_dmu_read_write;
388 ztest_func_t ztest_dmu_write_parallel;
389 ztest_func_t ztest_dmu_object_alloc_free;
390 ztest_func_t ztest_dmu_object_next_chunk;
391 ztest_func_t ztest_dmu_commit_callbacks;
392 ztest_func_t ztest_zap;
393 ztest_func_t ztest_zap_parallel;
394 ztest_func_t ztest_zil_commit;
395 ztest_func_t ztest_zil_remount;
396 ztest_func_t ztest_dmu_read_write_zcopy;
397 ztest_func_t ztest_dmu_objset_create_destroy;
398 ztest_func_t ztest_dmu_prealloc;
399 ztest_func_t ztest_fzap;
400 ztest_func_t ztest_dmu_snapshot_create_destroy;
401 ztest_func_t ztest_dsl_prop_get_set;
402 ztest_func_t ztest_spa_prop_get_set;
403 ztest_func_t ztest_spa_create_destroy;
404 ztest_func_t ztest_fault_inject;
405 ztest_func_t ztest_dmu_snapshot_hold;
406 ztest_func_t ztest_mmp_enable_disable;
407 ztest_func_t ztest_scrub;
408 ztest_func_t ztest_dsl_dataset_promote_busy;
409 ztest_func_t ztest_vdev_attach_detach;
410 ztest_func_t ztest_vdev_LUN_growth;
411 ztest_func_t ztest_vdev_add_remove;
412 ztest_func_t ztest_vdev_class_add;
413 ztest_func_t ztest_vdev_aux_add_remove;
414 ztest_func_t ztest_split_pool;
415 ztest_func_t ztest_reguid;
416 ztest_func_t ztest_spa_upgrade;
417 ztest_func_t ztest_device_removal;
418 ztest_func_t ztest_spa_checkpoint_create_discard;
419 ztest_func_t ztest_initialize;
420 ztest_func_t ztest_trim;
421 ztest_func_t ztest_blake3;
422 ztest_func_t ztest_fletcher;
423 ztest_func_t ztest_fletcher_incr;
424 ztest_func_t ztest_verify_dnode_bt;
425 
426 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
427 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
428 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
429 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
430 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
431 
432 #define	ZTI_INIT(func, iters, interval) \
433 	{   .zi_func = (func), \
434 	    .zi_iters = (iters), \
435 	    .zi_interval = (interval), \
436 	    .zi_funcname = # func }
437 
438 ztest_info_t ztest_info[] = {
439 	ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
440 	ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
441 	ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
442 	ZTI_INIT(ztest_dmu_object_next_chunk, 1, &zopt_sometimes),
443 	ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
444 	ZTI_INIT(ztest_zap, 30, &zopt_always),
445 	ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
446 	ZTI_INIT(ztest_split_pool, 1, &zopt_always),
447 	ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
448 	ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
449 	ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
450 	ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
451 	ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
452 	ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
453 #if 0
454 	ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
455 #endif
456 	ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
457 	ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
458 	ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
459 	ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
460 	ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
461 	ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
462 	ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
463 	ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
464 	ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
465 	ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
466 	ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
467 	ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
468 	ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
469 	ZTI_INIT(ztest_vdev_class_add, 1, &ztest_opts.zo_vdevtime),
470 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
471 	ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
472 	ZTI_INIT(ztest_spa_checkpoint_create_discard, 1, &zopt_rarely),
473 	ZTI_INIT(ztest_initialize, 1, &zopt_sometimes),
474 	ZTI_INIT(ztest_trim, 1, &zopt_sometimes),
475 	ZTI_INIT(ztest_blake3, 1, &zopt_rarely),
476 	ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
477 	ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
478 	ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
479 };
480 
481 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
482 
483 /*
484  * The following struct is used to hold a list of uncalled commit callbacks.
485  * The callbacks are ordered by txg number.
486  */
487 typedef struct ztest_cb_list {
488 	kmutex_t	zcl_callbacks_lock;
489 	list_t		zcl_callbacks;
490 } ztest_cb_list_t;
491 
492 /*
493  * Stuff we need to share writably between parent and child.
494  */
495 typedef struct ztest_shared {
496 	boolean_t	zs_do_init;
497 	hrtime_t	zs_proc_start;
498 	hrtime_t	zs_proc_stop;
499 	hrtime_t	zs_thread_start;
500 	hrtime_t	zs_thread_stop;
501 	hrtime_t	zs_thread_kill;
502 	uint64_t	zs_enospc_count;
503 	uint64_t	zs_vdev_next_leaf;
504 	uint64_t	zs_vdev_aux;
505 	uint64_t	zs_alloc;
506 	uint64_t	zs_space;
507 	uint64_t	zs_splits;
508 	uint64_t	zs_mirrors;
509 	uint64_t	zs_metaslab_sz;
510 	uint64_t	zs_metaslab_df_alloc_threshold;
511 	uint64_t	zs_guid;
512 } ztest_shared_t;
513 
514 #define	ID_PARALLEL	-1ULL
515 
516 static char ztest_dev_template[] = "%s/%s.%llua";
517 static char ztest_aux_template[] = "%s/%s.%s.%llu";
518 ztest_shared_t *ztest_shared;
519 
520 static spa_t *ztest_spa = NULL;
521 static ztest_ds_t *ztest_ds;
522 
523 static kmutex_t ztest_vdev_lock;
524 static boolean_t ztest_device_removal_active = B_FALSE;
525 static boolean_t ztest_pool_scrubbed = B_FALSE;
526 static kmutex_t ztest_checkpoint_lock;
527 
528 /*
529  * The ztest_name_lock protects the pool and dataset namespace used by
530  * the individual tests. To modify the namespace, consumers must grab
531  * this lock as writer. Grabbing the lock as reader will ensure that the
532  * namespace does not change while the lock is held.
533  */
534 static pthread_rwlock_t ztest_name_lock;
535 
536 static boolean_t ztest_dump_core = B_TRUE;
537 static boolean_t ztest_exiting;
538 
539 /* Global commit callback list */
540 static ztest_cb_list_t zcl;
541 /* Commit cb delay */
542 static uint64_t zc_min_txg_delay = UINT64_MAX;
543 static int zc_cb_counter = 0;
544 
545 /*
546  * Minimum number of commit callbacks that need to be registered for us to check
547  * whether the minimum txg delay is acceptable.
548  */
549 #define	ZTEST_COMMIT_CB_MIN_REG	100
550 
551 /*
552  * If a number of txgs equal to this threshold have been created after a commit
553  * callback has been registered but not called, then we assume there is an
554  * implementation bug.
555  */
556 #define	ZTEST_COMMIT_CB_THRESH	(TXG_CONCURRENT_STATES + 1000)
557 
558 enum ztest_object {
559 	ZTEST_META_DNODE = 0,
560 	ZTEST_DIROBJ,
561 	ZTEST_OBJECTS
562 };
563 
564 static __attribute__((noreturn)) void usage(boolean_t requested);
565 static int ztest_scrub_impl(spa_t *spa);
566 
567 /*
568  * These libumem hooks provide a reasonable set of defaults for the allocator's
569  * debugging facilities.
570  */
571 const char *
572 _umem_debug_init(void)
573 {
574 	return ("default,verbose"); /* $UMEM_DEBUG setting */
575 }
576 
577 const char *
578 _umem_logging_init(void)
579 {
580 	return ("fail,contents"); /* $UMEM_LOGGING setting */
581 }
582 
583 static void
584 dump_debug_buffer(void)
585 {
586 	ssize_t ret __attribute__((unused));
587 
588 	if (!ztest_opts.zo_dump_dbgmsg)
589 		return;
590 
591 	/*
592 	 * We use write() instead of printf() so that this function
593 	 * is safe to call from a signal handler.
594 	 */
595 	ret = write(STDOUT_FILENO, "\n", 1);
596 	zfs_dbgmsg_print("ztest");
597 }
598 
599 #define	BACKTRACE_SZ	100
600 
601 static void sig_handler(int signo)
602 {
603 	struct sigaction action;
604 #if (__GLIBC__ && !__UCLIBC__) /* backtrace() is a GNU extension */
605 	int nptrs;
606 	void *buffer[BACKTRACE_SZ];
607 
608 	nptrs = backtrace(buffer, BACKTRACE_SZ);
609 	backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
610 #endif
611 	dump_debug_buffer();
612 
613 	/*
614 	 * Restore default action and re-raise signal so SIGSEGV and
615 	 * SIGABRT can trigger a core dump.
616 	 */
617 	action.sa_handler = SIG_DFL;
618 	sigemptyset(&action.sa_mask);
619 	action.sa_flags = 0;
620 	(void) sigaction(signo, &action, NULL);
621 	raise(signo);
622 }
623 
624 #define	FATAL_MSG_SZ	1024
625 
626 char *fatal_msg;
627 
628 static __attribute__((format(printf, 2, 3))) __attribute__((noreturn)) void
629 fatal(int do_perror, char *message, ...)
630 {
631 	va_list args;
632 	int save_errno = errno;
633 	char *buf;
634 
635 	(void) fflush(stdout);
636 	buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
637 	if (buf == NULL)
638 		goto out;
639 
640 	va_start(args, message);
641 	(void) sprintf(buf, "ztest: ");
642 	/* LINTED */
643 	(void) vsprintf(buf + strlen(buf), message, args);
644 	va_end(args);
645 	if (do_perror) {
646 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
647 		    ": %s", strerror(save_errno));
648 	}
649 	(void) fprintf(stderr, "%s\n", buf);
650 	fatal_msg = buf;			/* to ease debugging */
651 
652 out:
653 	if (ztest_dump_core)
654 		abort();
655 	else
656 		dump_debug_buffer();
657 
658 	exit(3);
659 }
660 
661 static int
662 str2shift(const char *buf)
663 {
664 	const char *ends = "BKMGTPEZ";
665 	int i;
666 
667 	if (buf[0] == '\0')
668 		return (0);
669 	for (i = 0; i < strlen(ends); i++) {
670 		if (toupper(buf[0]) == ends[i])
671 			break;
672 	}
673 	if (i == strlen(ends)) {
674 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
675 		    buf);
676 		usage(B_FALSE);
677 	}
678 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
679 		return (10*i);
680 	}
681 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
682 	usage(B_FALSE);
683 }
684 
685 static uint64_t
686 nicenumtoull(const char *buf)
687 {
688 	char *end;
689 	uint64_t val;
690 
691 	val = strtoull(buf, &end, 0);
692 	if (end == buf) {
693 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
694 		usage(B_FALSE);
695 	} else if (end[0] == '.') {
696 		double fval = strtod(buf, &end);
697 		fval *= pow(2, str2shift(end));
698 		/*
699 		 * UINT64_MAX is not exactly representable as a double.
700 		 * The closest representation is UINT64_MAX + 1, so we
701 		 * use a >= comparison instead of > for the bounds check.
702 		 */
703 		if (fval >= (double)UINT64_MAX) {
704 			(void) fprintf(stderr, "ztest: value too large: %s\n",
705 			    buf);
706 			usage(B_FALSE);
707 		}
708 		val = (uint64_t)fval;
709 	} else {
710 		int shift = str2shift(end);
711 		if (shift >= 64 || (val << shift) >> shift != val) {
712 			(void) fprintf(stderr, "ztest: value too large: %s\n",
713 			    buf);
714 			usage(B_FALSE);
715 		}
716 		val <<= shift;
717 	}
718 	return (val);
719 }
720 
721 typedef struct ztest_option {
722 	const char	short_opt;
723 	const char	*long_opt;
724 	const char	*long_opt_param;
725 	const char	*comment;
726 	unsigned int	default_int;
727 	char		*default_str;
728 } ztest_option_t;
729 
730 /*
731  * The following option_table is used for generating the usage info as well as
732  * the long and short option information for calling getopt_long().
733  */
734 static ztest_option_t option_table[] = {
735 	{ 'v',	"vdevs", "INTEGER", "Number of vdevs", DEFAULT_VDEV_COUNT,
736 	    NULL},
737 	{ 's',	"vdev-size", "INTEGER", "Size of each vdev",
738 	    NO_DEFAULT, DEFAULT_VDEV_SIZE_STR},
739 	{ 'a',	"alignment-shift", "INTEGER",
740 	    "Alignment shift; use 0 for random", DEFAULT_ASHIFT, NULL},
741 	{ 'm',	"mirror-copies", "INTEGER", "Number of mirror copies",
742 	    DEFAULT_MIRRORS, NULL},
743 	{ 'r',	"raid-disks", "INTEGER", "Number of raidz/draid disks",
744 	    DEFAULT_RAID_CHILDREN, NULL},
745 	{ 'R',	"raid-parity", "INTEGER", "Raid parity",
746 	    DEFAULT_RAID_PARITY, NULL},
747 	{ 'K',	"raid-kind", "raidz|draid|random", "Raid kind",
748 	    NO_DEFAULT, "random"},
749 	{ 'D',	"draid-data", "INTEGER", "Number of draid data drives",
750 	    DEFAULT_DRAID_DATA, NULL},
751 	{ 'S',	"draid-spares", "INTEGER", "Number of draid spares",
752 	    DEFAULT_DRAID_SPARES, NULL},
753 	{ 'd',	"datasets", "INTEGER", "Number of datasets",
754 	    DEFAULT_DATASETS_COUNT, NULL},
755 	{ 't',	"threads", "INTEGER", "Number of ztest threads",
756 	    DEFAULT_THREADS, NULL},
757 	{ 'g',	"gang-block-threshold", "INTEGER",
758 	    "Metaslab gang block threshold",
759 	    NO_DEFAULT, DEFAULT_FORCE_GANGING_STR},
760 	{ 'i',	"init-count", "INTEGER", "Number of times to initialize pool",
761 	    DEFAULT_INITS, NULL},
762 	{ 'k',	"kill-percentage", "INTEGER", "Kill percentage",
763 	    NO_DEFAULT, DEFAULT_KILLRATE_STR},
764 	{ 'p',	"pool-name", "STRING", "Pool name",
765 	    NO_DEFAULT, DEFAULT_POOL},
766 	{ 'f',	"vdev-file-directory", "PATH", "File directory for vdev files",
767 	    NO_DEFAULT, DEFAULT_VDEV_DIR},
768 	{ 'M',	"multi-host", NULL,
769 	    "Multi-host; simulate pool imported on remote host",
770 	    NO_DEFAULT, NULL},
771 	{ 'E',	"use-existing-pool", NULL,
772 	    "Use existing pool instead of creating new one", NO_DEFAULT, NULL},
773 	{ 'T',	"run-time", "INTEGER", "Total run time",
774 	    NO_DEFAULT, DEFAULT_RUN_TIME_STR},
775 	{ 'P',	"pass-time", "INTEGER", "Time per pass",
776 	    NO_DEFAULT, DEFAULT_PASS_TIME_STR},
777 	{ 'F',	"freeze-loops", "INTEGER", "Max loops in spa_freeze()",
778 	    DEFAULT_MAX_LOOPS, NULL},
779 	{ 'B',	"alt-ztest", "PATH", "Alternate ztest path",
780 	    NO_DEFAULT, NULL},
781 	{ 'C',	"vdev-class-state", "on|off|random", "vdev class state",
782 	    NO_DEFAULT, "random"},
783 	{ 'o',	"option", "\"OPTION=INTEGER\"",
784 	    "Set global variable to an unsigned 32-bit integer value",
785 	    NO_DEFAULT, NULL},
786 	{ 'G',	"dump-debug-msg", NULL,
787 	    "Dump zfs_dbgmsg buffer before exiting due to an error",
788 	    NO_DEFAULT, NULL},
789 	{ 'V',	"verbose", NULL,
790 	    "Verbose (use multiple times for ever more verbosity)",
791 	    NO_DEFAULT, NULL},
792 	{ 'h',	"help",	NULL, "Show this help",
793 	    NO_DEFAULT, NULL},
794 	{0, 0, 0, 0, 0, 0}
795 };
796 
797 static struct option *long_opts = NULL;
798 static char *short_opts = NULL;
799 
800 static void
801 init_options(void)
802 {
803 	ASSERT3P(long_opts, ==, NULL);
804 	ASSERT3P(short_opts, ==, NULL);
805 
806 	int count = sizeof (option_table) / sizeof (option_table[0]);
807 	long_opts = umem_alloc(sizeof (struct option) * count, UMEM_NOFAIL);
808 
809 	short_opts = umem_alloc(sizeof (char) * 2 * count, UMEM_NOFAIL);
810 	int short_opt_index = 0;
811 
812 	for (int i = 0; i < count; i++) {
813 		long_opts[i].val = option_table[i].short_opt;
814 		long_opts[i].name = option_table[i].long_opt;
815 		long_opts[i].has_arg = option_table[i].long_opt_param != NULL
816 		    ? required_argument : no_argument;
817 		long_opts[i].flag = NULL;
818 		short_opts[short_opt_index++] = option_table[i].short_opt;
819 		if (option_table[i].long_opt_param != NULL) {
820 			short_opts[short_opt_index++] = ':';
821 		}
822 	}
823 }
824 
825 static void
826 fini_options(void)
827 {
828 	int count = sizeof (option_table) / sizeof (option_table[0]);
829 
830 	umem_free(long_opts, sizeof (struct option) * count);
831 	umem_free(short_opts, sizeof (char) * 2 * count);
832 
833 	long_opts = NULL;
834 	short_opts = NULL;
835 }
836 
837 static __attribute__((noreturn)) void
838 usage(boolean_t requested)
839 {
840 	char option[80];
841 	FILE *fp = requested ? stdout : stderr;
842 
843 	(void) fprintf(fp, "Usage: %s [OPTIONS...]\n", DEFAULT_POOL);
844 	for (int i = 0; option_table[i].short_opt != 0; i++) {
845 		if (option_table[i].long_opt_param != NULL) {
846 			(void) sprintf(option, "  -%c --%s=%s",
847 			    option_table[i].short_opt,
848 			    option_table[i].long_opt,
849 			    option_table[i].long_opt_param);
850 		} else {
851 			(void) sprintf(option, "  -%c --%s",
852 			    option_table[i].short_opt,
853 			    option_table[i].long_opt);
854 		}
855 		(void) fprintf(fp, "  %-40s%s", option,
856 		    option_table[i].comment);
857 
858 		if (option_table[i].long_opt_param != NULL) {
859 			if (option_table[i].default_str != NULL) {
860 				(void) fprintf(fp, " (default: %s)",
861 				    option_table[i].default_str);
862 			} else if (option_table[i].default_int != NO_DEFAULT) {
863 				(void) fprintf(fp, " (default: %u)",
864 				    option_table[i].default_int);
865 			}
866 		}
867 		(void) fprintf(fp, "\n");
868 	}
869 	exit(requested ? 0 : 1);
870 }
871 
872 static uint64_t
873 ztest_random(uint64_t range)
874 {
875 	uint64_t r;
876 
877 	ASSERT3S(ztest_fd_rand, >=, 0);
878 
879 	if (range == 0)
880 		return (0);
881 
882 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
883 		fatal(B_TRUE, "short read from /dev/urandom");
884 
885 	return (r % range);
886 }
887 
888 static void
889 ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
890 {
891 	char name[32];
892 	char *value;
893 	int state = ZTEST_VDEV_CLASS_RND;
894 
895 	(void) strlcpy(name, input, sizeof (name));
896 
897 	value = strchr(name, '=');
898 	if (value == NULL) {
899 		(void) fprintf(stderr, "missing value in property=value "
900 		    "'-C' argument (%s)\n", input);
901 		usage(B_FALSE);
902 	}
903 	*(value) = '\0';
904 	value++;
905 
906 	if (strcmp(value, "on") == 0) {
907 		state = ZTEST_VDEV_CLASS_ON;
908 	} else if (strcmp(value, "off") == 0) {
909 		state = ZTEST_VDEV_CLASS_OFF;
910 	} else if (strcmp(value, "random") == 0) {
911 		state = ZTEST_VDEV_CLASS_RND;
912 	} else {
913 		(void) fprintf(stderr, "invalid property value '%s'\n", value);
914 		usage(B_FALSE);
915 	}
916 
917 	if (strcmp(name, "special") == 0) {
918 		zo->zo_special_vdevs = state;
919 	} else {
920 		(void) fprintf(stderr, "invalid property name '%s'\n", name);
921 		usage(B_FALSE);
922 	}
923 	if (zo->zo_verbose >= 3)
924 		(void) printf("%s vdev state is '%s'\n", name, value);
925 }
926 
927 static void
928 process_options(int argc, char **argv)
929 {
930 	char *path;
931 	ztest_shared_opts_t *zo = &ztest_opts;
932 
933 	int opt;
934 	uint64_t value;
935 	const char *raid_kind = "random";
936 
937 	memcpy(zo, &ztest_opts_defaults, sizeof (*zo));
938 
939 	init_options();
940 
941 	while ((opt = getopt_long(argc, argv, short_opts, long_opts,
942 	    NULL)) != EOF) {
943 		value = 0;
944 		switch (opt) {
945 		case 'v':
946 		case 's':
947 		case 'a':
948 		case 'm':
949 		case 'r':
950 		case 'R':
951 		case 'D':
952 		case 'S':
953 		case 'd':
954 		case 't':
955 		case 'g':
956 		case 'i':
957 		case 'k':
958 		case 'T':
959 		case 'P':
960 		case 'F':
961 			value = nicenumtoull(optarg);
962 		}
963 		switch (opt) {
964 		case 'v':
965 			zo->zo_vdevs = value;
966 			break;
967 		case 's':
968 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
969 			break;
970 		case 'a':
971 			zo->zo_ashift = value;
972 			break;
973 		case 'm':
974 			zo->zo_mirrors = value;
975 			break;
976 		case 'r':
977 			zo->zo_raid_children = MAX(1, value);
978 			break;
979 		case 'R':
980 			zo->zo_raid_parity = MIN(MAX(value, 1), 3);
981 			break;
982 		case 'K':
983 			raid_kind = optarg;
984 			break;
985 		case 'D':
986 			zo->zo_draid_data = MAX(1, value);
987 			break;
988 		case 'S':
989 			zo->zo_draid_spares = MAX(1, value);
990 			break;
991 		case 'd':
992 			zo->zo_datasets = MAX(1, value);
993 			break;
994 		case 't':
995 			zo->zo_threads = MAX(1, value);
996 			break;
997 		case 'g':
998 			zo->zo_metaslab_force_ganging =
999 			    MAX(SPA_MINBLOCKSIZE << 1, value);
1000 			break;
1001 		case 'i':
1002 			zo->zo_init = value;
1003 			break;
1004 		case 'k':
1005 			zo->zo_killrate = value;
1006 			break;
1007 		case 'p':
1008 			(void) strlcpy(zo->zo_pool, optarg,
1009 			    sizeof (zo->zo_pool));
1010 			break;
1011 		case 'f':
1012 			path = realpath(optarg, NULL);
1013 			if (path == NULL) {
1014 				(void) fprintf(stderr, "error: %s: %s\n",
1015 				    optarg, strerror(errno));
1016 				usage(B_FALSE);
1017 			} else {
1018 				(void) strlcpy(zo->zo_dir, path,
1019 				    sizeof (zo->zo_dir));
1020 				free(path);
1021 			}
1022 			break;
1023 		case 'M':
1024 			zo->zo_mmp_test = 1;
1025 			break;
1026 		case 'V':
1027 			zo->zo_verbose++;
1028 			break;
1029 		case 'E':
1030 			zo->zo_init = 0;
1031 			break;
1032 		case 'T':
1033 			zo->zo_time = value;
1034 			break;
1035 		case 'P':
1036 			zo->zo_passtime = MAX(1, value);
1037 			break;
1038 		case 'F':
1039 			zo->zo_maxloops = MAX(1, value);
1040 			break;
1041 		case 'B':
1042 			(void) strlcpy(zo->zo_alt_ztest, optarg,
1043 			    sizeof (zo->zo_alt_ztest));
1044 			break;
1045 		case 'C':
1046 			ztest_parse_name_value(optarg, zo);
1047 			break;
1048 		case 'o':
1049 			if (zo->zo_gvars_count >= ZO_GVARS_MAX_COUNT) {
1050 				(void) fprintf(stderr,
1051 				    "max global var count (%zu) exceeded\n",
1052 				    ZO_GVARS_MAX_COUNT);
1053 				usage(B_FALSE);
1054 			}
1055 			char *v = zo->zo_gvars[zo->zo_gvars_count];
1056 			if (strlcpy(v, optarg, ZO_GVARS_MAX_ARGLEN) >=
1057 			    ZO_GVARS_MAX_ARGLEN) {
1058 				(void) fprintf(stderr,
1059 				    "global var option '%s' is too long\n",
1060 				    optarg);
1061 				usage(B_FALSE);
1062 			}
1063 			zo->zo_gvars_count++;
1064 			break;
1065 		case 'G':
1066 			zo->zo_dump_dbgmsg = 1;
1067 			break;
1068 		case 'h':
1069 			usage(B_TRUE);
1070 			break;
1071 		case '?':
1072 		default:
1073 			usage(B_FALSE);
1074 			break;
1075 		}
1076 	}
1077 
1078 	fini_options();
1079 
1080 	/* When raid choice is 'random' add a draid pool 50% of the time */
1081 	if (strcmp(raid_kind, "random") == 0) {
1082 		raid_kind = (ztest_random(2) == 0) ? "draid" : "raidz";
1083 
1084 		if (ztest_opts.zo_verbose >= 3)
1085 			(void) printf("choosing RAID type '%s'\n", raid_kind);
1086 	}
1087 
1088 	if (strcmp(raid_kind, "draid") == 0) {
1089 		uint64_t min_devsize;
1090 
1091 		/* With fewer disk use 256M, otherwise 128M is OK */
1092 		min_devsize = (ztest_opts.zo_raid_children < 16) ?
1093 		    (256ULL << 20) : (128ULL << 20);
1094 
1095 		/* No top-level mirrors with dRAID for now */
1096 		zo->zo_mirrors = 0;
1097 
1098 		/* Use more appropriate defaults for dRAID */
1099 		if (zo->zo_vdevs == ztest_opts_defaults.zo_vdevs)
1100 			zo->zo_vdevs = 1;
1101 		if (zo->zo_raid_children ==
1102 		    ztest_opts_defaults.zo_raid_children)
1103 			zo->zo_raid_children = 16;
1104 		if (zo->zo_ashift < 12)
1105 			zo->zo_ashift = 12;
1106 		if (zo->zo_vdev_size < min_devsize)
1107 			zo->zo_vdev_size = min_devsize;
1108 
1109 		if (zo->zo_draid_data + zo->zo_raid_parity >
1110 		    zo->zo_raid_children - zo->zo_draid_spares) {
1111 			(void) fprintf(stderr, "error: too few draid "
1112 			    "children (%d) for stripe width (%d)\n",
1113 			    zo->zo_raid_children,
1114 			    zo->zo_draid_data + zo->zo_raid_parity);
1115 			usage(B_FALSE);
1116 		}
1117 
1118 		(void) strlcpy(zo->zo_raid_type, VDEV_TYPE_DRAID,
1119 		    sizeof (zo->zo_raid_type));
1120 
1121 	} else /* using raidz */ {
1122 		ASSERT0(strcmp(raid_kind, "raidz"));
1123 
1124 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
1125 		    zo->zo_raid_children - 1);
1126 	}
1127 
1128 	zo->zo_vdevtime =
1129 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
1130 	    UINT64_MAX >> 2);
1131 
1132 	if (*zo->zo_alt_ztest) {
1133 		const char *invalid_what = "ztest";
1134 		char *val = zo->zo_alt_ztest;
1135 		if (0 != access(val, X_OK) ||
1136 		    (strrchr(val, '/') == NULL && (errno = EINVAL)))
1137 			goto invalid;
1138 
1139 		int dirlen = strrchr(val, '/') - val;
1140 		strncpy(zo->zo_alt_libpath, val, dirlen);
1141 		invalid_what = "library path", val = zo->zo_alt_libpath;
1142 		if (strrchr(val, '/') == NULL && (errno = EINVAL))
1143 			goto invalid;
1144 		*strrchr(val, '/') = '\0';
1145 		strlcat(val, "/lib", sizeof (zo->zo_alt_libpath));
1146 
1147 		if (0 != access(zo->zo_alt_libpath, X_OK))
1148 			goto invalid;
1149 		return;
1150 
1151 invalid:
1152 		ztest_dump_core = B_FALSE;
1153 		fatal(B_TRUE, "invalid alternate %s %s", invalid_what, val);
1154 	}
1155 }
1156 
1157 static void
1158 ztest_kill(ztest_shared_t *zs)
1159 {
1160 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
1161 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
1162 
1163 	/*
1164 	 * Before we kill ourselves, make sure that the config is updated.
1165 	 * See comment above spa_write_cachefile().
1166 	 */
1167 	mutex_enter(&spa_namespace_lock);
1168 	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
1169 	mutex_exit(&spa_namespace_lock);
1170 
1171 	(void) raise(SIGKILL);
1172 }
1173 
1174 static void
1175 ztest_record_enospc(const char *s)
1176 {
1177 	(void) s;
1178 	ztest_shared->zs_enospc_count++;
1179 }
1180 
1181 static uint64_t
1182 ztest_get_ashift(void)
1183 {
1184 	if (ztest_opts.zo_ashift == 0)
1185 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
1186 	return (ztest_opts.zo_ashift);
1187 }
1188 
1189 static boolean_t
1190 ztest_is_draid_spare(const char *name)
1191 {
1192 	uint64_t spare_id = 0, parity = 0, vdev_id = 0;
1193 
1194 	if (sscanf(name, VDEV_TYPE_DRAID "%"PRIu64"-%"PRIu64"-%"PRIu64"",
1195 	    &parity, &vdev_id, &spare_id) == 3) {
1196 		return (B_TRUE);
1197 	}
1198 
1199 	return (B_FALSE);
1200 }
1201 
1202 static nvlist_t *
1203 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
1204 {
1205 	char *pathbuf;
1206 	uint64_t vdev;
1207 	nvlist_t *file;
1208 	boolean_t draid_spare = B_FALSE;
1209 
1210 	pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1211 
1212 	if (ashift == 0)
1213 		ashift = ztest_get_ashift();
1214 
1215 	if (path == NULL) {
1216 		path = pathbuf;
1217 
1218 		if (aux != NULL) {
1219 			vdev = ztest_shared->zs_vdev_aux;
1220 			(void) snprintf(path, MAXPATHLEN,
1221 			    ztest_aux_template, ztest_opts.zo_dir,
1222 			    pool == NULL ? ztest_opts.zo_pool : pool,
1223 			    aux, vdev);
1224 		} else {
1225 			vdev = ztest_shared->zs_vdev_next_leaf++;
1226 			(void) snprintf(path, MAXPATHLEN,
1227 			    ztest_dev_template, ztest_opts.zo_dir,
1228 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
1229 		}
1230 	} else {
1231 		draid_spare = ztest_is_draid_spare(path);
1232 	}
1233 
1234 	if (size != 0 && !draid_spare) {
1235 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
1236 		if (fd == -1)
1237 			fatal(B_TRUE, "can't open %s", path);
1238 		if (ftruncate(fd, size) != 0)
1239 			fatal(B_TRUE, "can't ftruncate %s", path);
1240 		(void) close(fd);
1241 	}
1242 
1243 	file = fnvlist_alloc();
1244 	fnvlist_add_string(file, ZPOOL_CONFIG_TYPE,
1245 	    draid_spare ? VDEV_TYPE_DRAID_SPARE : VDEV_TYPE_FILE);
1246 	fnvlist_add_string(file, ZPOOL_CONFIG_PATH, path);
1247 	fnvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift);
1248 	umem_free(pathbuf, MAXPATHLEN);
1249 
1250 	return (file);
1251 }
1252 
1253 static nvlist_t *
1254 make_vdev_raid(char *path, char *aux, char *pool, size_t size,
1255     uint64_t ashift, int r)
1256 {
1257 	nvlist_t *raid, **child;
1258 	int c;
1259 
1260 	if (r < 2)
1261 		return (make_vdev_file(path, aux, pool, size, ashift));
1262 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
1263 
1264 	for (c = 0; c < r; c++)
1265 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
1266 
1267 	raid = fnvlist_alloc();
1268 	fnvlist_add_string(raid, ZPOOL_CONFIG_TYPE,
1269 	    ztest_opts.zo_raid_type);
1270 	fnvlist_add_uint64(raid, ZPOOL_CONFIG_NPARITY,
1271 	    ztest_opts.zo_raid_parity);
1272 	fnvlist_add_nvlist_array(raid, ZPOOL_CONFIG_CHILDREN,
1273 	    (const nvlist_t **)child, r);
1274 
1275 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0) {
1276 		uint64_t ndata = ztest_opts.zo_draid_data;
1277 		uint64_t nparity = ztest_opts.zo_raid_parity;
1278 		uint64_t nspares = ztest_opts.zo_draid_spares;
1279 		uint64_t children = ztest_opts.zo_raid_children;
1280 		uint64_t ngroups = 1;
1281 
1282 		/*
1283 		 * Calculate the minimum number of groups required to fill a
1284 		 * slice. This is the LCM of the stripe width (data + parity)
1285 		 * and the number of data drives (children - spares).
1286 		 */
1287 		while (ngroups * (ndata + nparity) % (children - nspares) != 0)
1288 			ngroups++;
1289 
1290 		/* Store the basic dRAID configuration. */
1291 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NDATA, ndata);
1292 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NSPARES, nspares);
1293 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups);
1294 	}
1295 
1296 	for (c = 0; c < r; c++)
1297 		fnvlist_free(child[c]);
1298 
1299 	umem_free(child, r * sizeof (nvlist_t *));
1300 
1301 	return (raid);
1302 }
1303 
1304 static nvlist_t *
1305 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
1306     uint64_t ashift, int r, int m)
1307 {
1308 	nvlist_t *mirror, **child;
1309 	int c;
1310 
1311 	if (m < 1)
1312 		return (make_vdev_raid(path, aux, pool, size, ashift, r));
1313 
1314 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1315 
1316 	for (c = 0; c < m; c++)
1317 		child[c] = make_vdev_raid(path, aux, pool, size, ashift, r);
1318 
1319 	mirror = fnvlist_alloc();
1320 	fnvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, VDEV_TYPE_MIRROR);
1321 	fnvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1322 	    (const nvlist_t **)child, m);
1323 
1324 	for (c = 0; c < m; c++)
1325 		fnvlist_free(child[c]);
1326 
1327 	umem_free(child, m * sizeof (nvlist_t *));
1328 
1329 	return (mirror);
1330 }
1331 
1332 static nvlist_t *
1333 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
1334     const char *class, int r, int m, int t)
1335 {
1336 	nvlist_t *root, **child;
1337 	int c;
1338 	boolean_t log;
1339 
1340 	ASSERT3S(t, >, 0);
1341 
1342 	log = (class != NULL && strcmp(class, "log") == 0);
1343 
1344 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1345 
1346 	for (c = 0; c < t; c++) {
1347 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1348 		    r, m);
1349 		fnvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, log);
1350 
1351 		if (class != NULL && class[0] != '\0') {
1352 			ASSERT(m > 1 || log);   /* expecting a mirror */
1353 			fnvlist_add_string(child[c],
1354 			    ZPOOL_CONFIG_ALLOCATION_BIAS, class);
1355 		}
1356 	}
1357 
1358 	root = fnvlist_alloc();
1359 	fnvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
1360 	fnvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1361 	    (const nvlist_t **)child, t);
1362 
1363 	for (c = 0; c < t; c++)
1364 		fnvlist_free(child[c]);
1365 
1366 	umem_free(child, t * sizeof (nvlist_t *));
1367 
1368 	return (root);
1369 }
1370 
1371 /*
1372  * Find a random spa version. Returns back a random spa version in the
1373  * range [initial_version, SPA_VERSION_FEATURES].
1374  */
1375 static uint64_t
1376 ztest_random_spa_version(uint64_t initial_version)
1377 {
1378 	uint64_t version = initial_version;
1379 
1380 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1381 		version = version +
1382 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1383 	}
1384 
1385 	if (version > SPA_VERSION_BEFORE_FEATURES)
1386 		version = SPA_VERSION_FEATURES;
1387 
1388 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1389 	return (version);
1390 }
1391 
1392 static int
1393 ztest_random_blocksize(void)
1394 {
1395 	ASSERT3U(ztest_spa->spa_max_ashift, !=, 0);
1396 
1397 	/*
1398 	 * Choose a block size >= the ashift.
1399 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1400 	 */
1401 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1402 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1403 		maxbs = 20;
1404 	uint64_t block_shift =
1405 	    ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1406 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1407 }
1408 
1409 static int
1410 ztest_random_dnodesize(void)
1411 {
1412 	int slots;
1413 	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1414 
1415 	if (max_slots == DNODE_MIN_SLOTS)
1416 		return (DNODE_MIN_SIZE);
1417 
1418 	/*
1419 	 * Weight the random distribution more heavily toward smaller
1420 	 * dnode sizes since that is more likely to reflect real-world
1421 	 * usage.
1422 	 */
1423 	ASSERT3U(max_slots, >, 4);
1424 	switch (ztest_random(10)) {
1425 	case 0:
1426 		slots = 5 + ztest_random(max_slots - 4);
1427 		break;
1428 	case 1 ... 4:
1429 		slots = 2 + ztest_random(3);
1430 		break;
1431 	default:
1432 		slots = 1;
1433 		break;
1434 	}
1435 
1436 	return (slots << DNODE_SHIFT);
1437 }
1438 
1439 static int
1440 ztest_random_ibshift(void)
1441 {
1442 	return (DN_MIN_INDBLKSHIFT +
1443 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1444 }
1445 
1446 static uint64_t
1447 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1448 {
1449 	uint64_t top;
1450 	vdev_t *rvd = spa->spa_root_vdev;
1451 	vdev_t *tvd;
1452 
1453 	ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
1454 
1455 	do {
1456 		top = ztest_random(rvd->vdev_children);
1457 		tvd = rvd->vdev_child[top];
1458 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1459 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1460 
1461 	return (top);
1462 }
1463 
1464 static uint64_t
1465 ztest_random_dsl_prop(zfs_prop_t prop)
1466 {
1467 	uint64_t value;
1468 
1469 	do {
1470 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1471 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1472 
1473 	return (value);
1474 }
1475 
1476 static int
1477 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1478     boolean_t inherit)
1479 {
1480 	const char *propname = zfs_prop_to_name(prop);
1481 	const char *valname;
1482 	char *setpoint;
1483 	uint64_t curval;
1484 	int error;
1485 
1486 	error = dsl_prop_set_int(osname, propname,
1487 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1488 
1489 	if (error == ENOSPC) {
1490 		ztest_record_enospc(FTAG);
1491 		return (error);
1492 	}
1493 	ASSERT0(error);
1494 
1495 	setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1496 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1497 
1498 	if (ztest_opts.zo_verbose >= 6) {
1499 		int err;
1500 
1501 		err = zfs_prop_index_to_string(prop, curval, &valname);
1502 		if (err)
1503 			(void) printf("%s %s = %llu at '%s'\n", osname,
1504 			    propname, (unsigned long long)curval, setpoint);
1505 		else
1506 			(void) printf("%s %s = %s at '%s'\n",
1507 			    osname, propname, valname, setpoint);
1508 	}
1509 	umem_free(setpoint, MAXPATHLEN);
1510 
1511 	return (error);
1512 }
1513 
1514 static int
1515 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1516 {
1517 	spa_t *spa = ztest_spa;
1518 	nvlist_t *props = NULL;
1519 	int error;
1520 
1521 	props = fnvlist_alloc();
1522 	fnvlist_add_uint64(props, zpool_prop_to_name(prop), value);
1523 
1524 	error = spa_prop_set(spa, props);
1525 
1526 	fnvlist_free(props);
1527 
1528 	if (error == ENOSPC) {
1529 		ztest_record_enospc(FTAG);
1530 		return (error);
1531 	}
1532 	ASSERT0(error);
1533 
1534 	return (error);
1535 }
1536 
1537 static int
1538 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1539     boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
1540 {
1541 	int err;
1542 	char *cp = NULL;
1543 	char ddname[ZFS_MAX_DATASET_NAME_LEN];
1544 
1545 	strcpy(ddname, name);
1546 	cp = strchr(ddname, '@');
1547 	if (cp != NULL)
1548 		*cp = '\0';
1549 
1550 	err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1551 	while (decrypt && err == EACCES) {
1552 		dsl_crypto_params_t *dcp;
1553 		nvlist_t *crypto_args = fnvlist_alloc();
1554 
1555 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
1556 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1557 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1558 		    crypto_args, &dcp));
1559 		err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1560 		/*
1561 		 * Note: if there was an error loading, the wkey was not
1562 		 * consumed, and needs to be freed.
1563 		 */
1564 		dsl_crypto_params_free(dcp, (err != 0));
1565 		fnvlist_free(crypto_args);
1566 
1567 		if (err == EINVAL) {
1568 			/*
1569 			 * We couldn't load a key for this dataset so try
1570 			 * the parent. This loop will eventually hit the
1571 			 * encryption root since ztest only makes clones
1572 			 * as children of their origin datasets.
1573 			 */
1574 			cp = strrchr(ddname, '/');
1575 			if (cp == NULL)
1576 				return (err);
1577 
1578 			*cp = '\0';
1579 			err = EACCES;
1580 			continue;
1581 		} else if (err != 0) {
1582 			break;
1583 		}
1584 
1585 		err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1586 		break;
1587 	}
1588 
1589 	return (err);
1590 }
1591 
1592 static void
1593 ztest_rll_init(rll_t *rll)
1594 {
1595 	rll->rll_writer = NULL;
1596 	rll->rll_readers = 0;
1597 	mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1598 	cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1599 }
1600 
1601 static void
1602 ztest_rll_destroy(rll_t *rll)
1603 {
1604 	ASSERT3P(rll->rll_writer, ==, NULL);
1605 	ASSERT0(rll->rll_readers);
1606 	mutex_destroy(&rll->rll_lock);
1607 	cv_destroy(&rll->rll_cv);
1608 }
1609 
1610 static void
1611 ztest_rll_lock(rll_t *rll, rl_type_t type)
1612 {
1613 	mutex_enter(&rll->rll_lock);
1614 
1615 	if (type == RL_READER) {
1616 		while (rll->rll_writer != NULL)
1617 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1618 		rll->rll_readers++;
1619 	} else {
1620 		while (rll->rll_writer != NULL || rll->rll_readers)
1621 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1622 		rll->rll_writer = curthread;
1623 	}
1624 
1625 	mutex_exit(&rll->rll_lock);
1626 }
1627 
1628 static void
1629 ztest_rll_unlock(rll_t *rll)
1630 {
1631 	mutex_enter(&rll->rll_lock);
1632 
1633 	if (rll->rll_writer) {
1634 		ASSERT0(rll->rll_readers);
1635 		rll->rll_writer = NULL;
1636 	} else {
1637 		ASSERT3S(rll->rll_readers, >, 0);
1638 		ASSERT3P(rll->rll_writer, ==, NULL);
1639 		rll->rll_readers--;
1640 	}
1641 
1642 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1643 		cv_broadcast(&rll->rll_cv);
1644 
1645 	mutex_exit(&rll->rll_lock);
1646 }
1647 
1648 static void
1649 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1650 {
1651 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1652 
1653 	ztest_rll_lock(rll, type);
1654 }
1655 
1656 static void
1657 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1658 {
1659 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1660 
1661 	ztest_rll_unlock(rll);
1662 }
1663 
1664 static rl_t *
1665 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1666     uint64_t size, rl_type_t type)
1667 {
1668 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1669 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1670 	rl_t *rl;
1671 
1672 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1673 	rl->rl_object = object;
1674 	rl->rl_offset = offset;
1675 	rl->rl_size = size;
1676 	rl->rl_lock = rll;
1677 
1678 	ztest_rll_lock(rll, type);
1679 
1680 	return (rl);
1681 }
1682 
1683 static void
1684 ztest_range_unlock(rl_t *rl)
1685 {
1686 	rll_t *rll = rl->rl_lock;
1687 
1688 	ztest_rll_unlock(rll);
1689 
1690 	umem_free(rl, sizeof (*rl));
1691 }
1692 
1693 static void
1694 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1695 {
1696 	zd->zd_os = os;
1697 	zd->zd_zilog = dmu_objset_zil(os);
1698 	zd->zd_shared = szd;
1699 	dmu_objset_name(os, zd->zd_name);
1700 	int l;
1701 
1702 	if (zd->zd_shared != NULL)
1703 		zd->zd_shared->zd_seq = 0;
1704 
1705 	VERIFY0(pthread_rwlock_init(&zd->zd_zilog_lock, NULL));
1706 	mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1707 
1708 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1709 		ztest_rll_init(&zd->zd_object_lock[l]);
1710 
1711 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1712 		ztest_rll_init(&zd->zd_range_lock[l]);
1713 }
1714 
1715 static void
1716 ztest_zd_fini(ztest_ds_t *zd)
1717 {
1718 	int l;
1719 
1720 	mutex_destroy(&zd->zd_dirobj_lock);
1721 	(void) pthread_rwlock_destroy(&zd->zd_zilog_lock);
1722 
1723 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1724 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1725 
1726 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1727 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1728 }
1729 
1730 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1731 
1732 static uint64_t
1733 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1734 {
1735 	uint64_t txg;
1736 	int error;
1737 
1738 	/*
1739 	 * Attempt to assign tx to some transaction group.
1740 	 */
1741 	error = dmu_tx_assign(tx, txg_how);
1742 	if (error) {
1743 		if (error == ERESTART) {
1744 			ASSERT3U(txg_how, ==, TXG_NOWAIT);
1745 			dmu_tx_wait(tx);
1746 		} else {
1747 			ASSERT3U(error, ==, ENOSPC);
1748 			ztest_record_enospc(tag);
1749 		}
1750 		dmu_tx_abort(tx);
1751 		return (0);
1752 	}
1753 	txg = dmu_tx_get_txg(tx);
1754 	ASSERT3U(txg, !=, 0);
1755 	return (txg);
1756 }
1757 
1758 static void
1759 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1760     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1761     uint64_t crtxg)
1762 {
1763 	bt->bt_magic = BT_MAGIC;
1764 	bt->bt_objset = dmu_objset_id(os);
1765 	bt->bt_object = object;
1766 	bt->bt_dnodesize = dnodesize;
1767 	bt->bt_offset = offset;
1768 	bt->bt_gen = gen;
1769 	bt->bt_txg = txg;
1770 	bt->bt_crtxg = crtxg;
1771 }
1772 
1773 static void
1774 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1775     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1776     uint64_t crtxg)
1777 {
1778 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1779 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1780 	ASSERT3U(bt->bt_object, ==, object);
1781 	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1782 	ASSERT3U(bt->bt_offset, ==, offset);
1783 	ASSERT3U(bt->bt_gen, <=, gen);
1784 	ASSERT3U(bt->bt_txg, <=, txg);
1785 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1786 }
1787 
1788 static ztest_block_tag_t *
1789 ztest_bt_bonus(dmu_buf_t *db)
1790 {
1791 	dmu_object_info_t doi;
1792 	ztest_block_tag_t *bt;
1793 
1794 	dmu_object_info_from_db(db, &doi);
1795 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1796 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1797 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1798 
1799 	return (bt);
1800 }
1801 
1802 /*
1803  * Generate a token to fill up unused bonus buffer space.  Try to make
1804  * it unique to the object, generation, and offset to verify that data
1805  * is not getting overwritten by data from other dnodes.
1806  */
1807 #define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1808 	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1809 
1810 /*
1811  * Fill up the unused bonus buffer region before the block tag with a
1812  * verifiable pattern. Filling the whole bonus area with non-zero data
1813  * helps ensure that all dnode traversal code properly skips the
1814  * interior regions of large dnodes.
1815  */
1816 static void
1817 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1818     objset_t *os, uint64_t gen)
1819 {
1820 	uint64_t *bonusp;
1821 
1822 	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1823 
1824 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1825 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1826 		    gen, bonusp - (uint64_t *)db->db_data);
1827 		*bonusp = token;
1828 	}
1829 }
1830 
1831 /*
1832  * Verify that the unused area of a bonus buffer is filled with the
1833  * expected tokens.
1834  */
1835 static void
1836 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1837     objset_t *os, uint64_t gen)
1838 {
1839 	uint64_t *bonusp;
1840 
1841 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1842 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1843 		    gen, bonusp - (uint64_t *)db->db_data);
1844 		VERIFY3U(*bonusp, ==, token);
1845 	}
1846 }
1847 
1848 /*
1849  * ZIL logging ops
1850  */
1851 
1852 #define	lrz_type	lr_mode
1853 #define	lrz_blocksize	lr_uid
1854 #define	lrz_ibshift	lr_gid
1855 #define	lrz_bonustype	lr_rdev
1856 #define	lrz_dnodesize	lr_crtime[1]
1857 
1858 static void
1859 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1860 {
1861 	char *name = (void *)(lr + 1);		/* name follows lr */
1862 	size_t namesize = strlen(name) + 1;
1863 	itx_t *itx;
1864 
1865 	if (zil_replaying(zd->zd_zilog, tx))
1866 		return;
1867 
1868 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1869 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1870 	    sizeof (*lr) + namesize - sizeof (lr_t));
1871 
1872 	zil_itx_assign(zd->zd_zilog, itx, tx);
1873 }
1874 
1875 static void
1876 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1877 {
1878 	char *name = (void *)(lr + 1);		/* name follows lr */
1879 	size_t namesize = strlen(name) + 1;
1880 	itx_t *itx;
1881 
1882 	if (zil_replaying(zd->zd_zilog, tx))
1883 		return;
1884 
1885 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1886 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1887 	    sizeof (*lr) + namesize - sizeof (lr_t));
1888 
1889 	itx->itx_oid = object;
1890 	zil_itx_assign(zd->zd_zilog, itx, tx);
1891 }
1892 
1893 static void
1894 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1895 {
1896 	itx_t *itx;
1897 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1898 
1899 	if (zil_replaying(zd->zd_zilog, tx))
1900 		return;
1901 
1902 	if (lr->lr_length > zil_max_log_data(zd->zd_zilog))
1903 		write_state = WR_INDIRECT;
1904 
1905 	itx = zil_itx_create(TX_WRITE,
1906 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1907 
1908 	if (write_state == WR_COPIED &&
1909 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1910 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1911 		zil_itx_destroy(itx);
1912 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1913 		write_state = WR_NEED_COPY;
1914 	}
1915 	itx->itx_private = zd;
1916 	itx->itx_wr_state = write_state;
1917 	itx->itx_sync = (ztest_random(8) == 0);
1918 
1919 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1920 	    sizeof (*lr) - sizeof (lr_t));
1921 
1922 	zil_itx_assign(zd->zd_zilog, itx, tx);
1923 }
1924 
1925 static void
1926 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1927 {
1928 	itx_t *itx;
1929 
1930 	if (zil_replaying(zd->zd_zilog, tx))
1931 		return;
1932 
1933 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1934 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1935 	    sizeof (*lr) - sizeof (lr_t));
1936 
1937 	itx->itx_sync = B_FALSE;
1938 	zil_itx_assign(zd->zd_zilog, itx, tx);
1939 }
1940 
1941 static void
1942 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1943 {
1944 	itx_t *itx;
1945 
1946 	if (zil_replaying(zd->zd_zilog, tx))
1947 		return;
1948 
1949 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1950 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1951 	    sizeof (*lr) - sizeof (lr_t));
1952 
1953 	itx->itx_sync = B_FALSE;
1954 	zil_itx_assign(zd->zd_zilog, itx, tx);
1955 }
1956 
1957 /*
1958  * ZIL replay ops
1959  */
1960 static int
1961 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1962 {
1963 	ztest_ds_t *zd = arg1;
1964 	lr_create_t *lr = arg2;
1965 	char *name = (void *)(lr + 1);		/* name follows lr */
1966 	objset_t *os = zd->zd_os;
1967 	ztest_block_tag_t *bbt;
1968 	dmu_buf_t *db;
1969 	dmu_tx_t *tx;
1970 	uint64_t txg;
1971 	int error = 0;
1972 	int bonuslen;
1973 
1974 	if (byteswap)
1975 		byteswap_uint64_array(lr, sizeof (*lr));
1976 
1977 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
1978 	ASSERT3S(name[0], !=, '\0');
1979 
1980 	tx = dmu_tx_create(os);
1981 
1982 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1983 
1984 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1985 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1986 	} else {
1987 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1988 	}
1989 
1990 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1991 	if (txg == 0)
1992 		return (ENOSPC);
1993 
1994 	ASSERT3U(dmu_objset_zil(os)->zl_replay, ==, !!lr->lr_foid);
1995 	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1996 
1997 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1998 		if (lr->lr_foid == 0) {
1999 			lr->lr_foid = zap_create_dnsize(os,
2000 			    lr->lrz_type, lr->lrz_bonustype,
2001 			    bonuslen, lr->lrz_dnodesize, tx);
2002 		} else {
2003 			error = zap_create_claim_dnsize(os, lr->lr_foid,
2004 			    lr->lrz_type, lr->lrz_bonustype,
2005 			    bonuslen, lr->lrz_dnodesize, tx);
2006 		}
2007 	} else {
2008 		if (lr->lr_foid == 0) {
2009 			lr->lr_foid = dmu_object_alloc_dnsize(os,
2010 			    lr->lrz_type, 0, lr->lrz_bonustype,
2011 			    bonuslen, lr->lrz_dnodesize, tx);
2012 		} else {
2013 			error = dmu_object_claim_dnsize(os, lr->lr_foid,
2014 			    lr->lrz_type, 0, lr->lrz_bonustype,
2015 			    bonuslen, lr->lrz_dnodesize, tx);
2016 		}
2017 	}
2018 
2019 	if (error) {
2020 		ASSERT3U(error, ==, EEXIST);
2021 		ASSERT(zd->zd_zilog->zl_replay);
2022 		dmu_tx_commit(tx);
2023 		return (error);
2024 	}
2025 
2026 	ASSERT3U(lr->lr_foid, !=, 0);
2027 
2028 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
2029 		VERIFY0(dmu_object_set_blocksize(os, lr->lr_foid,
2030 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
2031 
2032 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2033 	bbt = ztest_bt_bonus(db);
2034 	dmu_buf_will_dirty(db, tx);
2035 	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
2036 	    lr->lr_gen, txg, txg);
2037 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
2038 	dmu_buf_rele(db, FTAG);
2039 
2040 	VERIFY0(zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
2041 	    &lr->lr_foid, tx));
2042 
2043 	(void) ztest_log_create(zd, tx, lr);
2044 
2045 	dmu_tx_commit(tx);
2046 
2047 	return (0);
2048 }
2049 
2050 static int
2051 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
2052 {
2053 	ztest_ds_t *zd = arg1;
2054 	lr_remove_t *lr = arg2;
2055 	char *name = (void *)(lr + 1);		/* name follows lr */
2056 	objset_t *os = zd->zd_os;
2057 	dmu_object_info_t doi;
2058 	dmu_tx_t *tx;
2059 	uint64_t object, txg;
2060 
2061 	if (byteswap)
2062 		byteswap_uint64_array(lr, sizeof (*lr));
2063 
2064 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
2065 	ASSERT3S(name[0], !=, '\0');
2066 
2067 	VERIFY0(
2068 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
2069 	ASSERT3U(object, !=, 0);
2070 
2071 	ztest_object_lock(zd, object, RL_WRITER);
2072 
2073 	VERIFY0(dmu_object_info(os, object, &doi));
2074 
2075 	tx = dmu_tx_create(os);
2076 
2077 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
2078 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2079 
2080 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2081 	if (txg == 0) {
2082 		ztest_object_unlock(zd, object);
2083 		return (ENOSPC);
2084 	}
2085 
2086 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
2087 		VERIFY0(zap_destroy(os, object, tx));
2088 	} else {
2089 		VERIFY0(dmu_object_free(os, object, tx));
2090 	}
2091 
2092 	VERIFY0(zap_remove(os, lr->lr_doid, name, tx));
2093 
2094 	(void) ztest_log_remove(zd, tx, lr, object);
2095 
2096 	dmu_tx_commit(tx);
2097 
2098 	ztest_object_unlock(zd, object);
2099 
2100 	return (0);
2101 }
2102 
2103 static int
2104 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
2105 {
2106 	ztest_ds_t *zd = arg1;
2107 	lr_write_t *lr = arg2;
2108 	objset_t *os = zd->zd_os;
2109 	void *data = lr + 1;			/* data follows lr */
2110 	uint64_t offset, length;
2111 	ztest_block_tag_t *bt = data;
2112 	ztest_block_tag_t *bbt;
2113 	uint64_t gen, txg, lrtxg, crtxg;
2114 	dmu_object_info_t doi;
2115 	dmu_tx_t *tx;
2116 	dmu_buf_t *db;
2117 	arc_buf_t *abuf = NULL;
2118 	rl_t *rl;
2119 
2120 	if (byteswap)
2121 		byteswap_uint64_array(lr, sizeof (*lr));
2122 
2123 	offset = lr->lr_offset;
2124 	length = lr->lr_length;
2125 
2126 	/* If it's a dmu_sync() block, write the whole block */
2127 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
2128 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
2129 		if (length < blocksize) {
2130 			offset -= offset % blocksize;
2131 			length = blocksize;
2132 		}
2133 	}
2134 
2135 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
2136 		byteswap_uint64_array(bt, sizeof (*bt));
2137 
2138 	if (bt->bt_magic != BT_MAGIC)
2139 		bt = NULL;
2140 
2141 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
2142 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
2143 
2144 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2145 
2146 	dmu_object_info_from_db(db, &doi);
2147 
2148 	bbt = ztest_bt_bonus(db);
2149 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2150 	gen = bbt->bt_gen;
2151 	crtxg = bbt->bt_crtxg;
2152 	lrtxg = lr->lr_common.lrc_txg;
2153 
2154 	tx = dmu_tx_create(os);
2155 
2156 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
2157 
2158 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
2159 	    P2PHASE(offset, length) == 0)
2160 		abuf = dmu_request_arcbuf(db, length);
2161 
2162 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2163 	if (txg == 0) {
2164 		if (abuf != NULL)
2165 			dmu_return_arcbuf(abuf);
2166 		dmu_buf_rele(db, FTAG);
2167 		ztest_range_unlock(rl);
2168 		ztest_object_unlock(zd, lr->lr_foid);
2169 		return (ENOSPC);
2170 	}
2171 
2172 	if (bt != NULL) {
2173 		/*
2174 		 * Usually, verify the old data before writing new data --
2175 		 * but not always, because we also want to verify correct
2176 		 * behavior when the data was not recently read into cache.
2177 		 */
2178 		ASSERT0(offset % doi.doi_data_block_size);
2179 		if (ztest_random(4) != 0) {
2180 			int prefetch = ztest_random(2) ?
2181 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
2182 			ztest_block_tag_t rbt;
2183 
2184 			VERIFY(dmu_read(os, lr->lr_foid, offset,
2185 			    sizeof (rbt), &rbt, prefetch) == 0);
2186 			if (rbt.bt_magic == BT_MAGIC) {
2187 				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
2188 				    offset, gen, txg, crtxg);
2189 			}
2190 		}
2191 
2192 		/*
2193 		 * Writes can appear to be newer than the bonus buffer because
2194 		 * the ztest_get_data() callback does a dmu_read() of the
2195 		 * open-context data, which may be different than the data
2196 		 * as it was when the write was generated.
2197 		 */
2198 		if (zd->zd_zilog->zl_replay) {
2199 			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
2200 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
2201 			    bt->bt_crtxg);
2202 		}
2203 
2204 		/*
2205 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
2206 		 * so that all of the usual ASSERTs will work.
2207 		 */
2208 		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
2209 		    crtxg);
2210 	}
2211 
2212 	if (abuf == NULL) {
2213 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
2214 	} else {
2215 		memcpy(abuf->b_data, data, length);
2216 		dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx);
2217 	}
2218 
2219 	(void) ztest_log_write(zd, tx, lr);
2220 
2221 	dmu_buf_rele(db, FTAG);
2222 
2223 	dmu_tx_commit(tx);
2224 
2225 	ztest_range_unlock(rl);
2226 	ztest_object_unlock(zd, lr->lr_foid);
2227 
2228 	return (0);
2229 }
2230 
2231 static int
2232 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
2233 {
2234 	ztest_ds_t *zd = arg1;
2235 	lr_truncate_t *lr = arg2;
2236 	objset_t *os = zd->zd_os;
2237 	dmu_tx_t *tx;
2238 	uint64_t txg;
2239 	rl_t *rl;
2240 
2241 	if (byteswap)
2242 		byteswap_uint64_array(lr, sizeof (*lr));
2243 
2244 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
2245 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
2246 	    RL_WRITER);
2247 
2248 	tx = dmu_tx_create(os);
2249 
2250 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
2251 
2252 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2253 	if (txg == 0) {
2254 		ztest_range_unlock(rl);
2255 		ztest_object_unlock(zd, lr->lr_foid);
2256 		return (ENOSPC);
2257 	}
2258 
2259 	VERIFY0(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2260 	    lr->lr_length, tx));
2261 
2262 	(void) ztest_log_truncate(zd, tx, lr);
2263 
2264 	dmu_tx_commit(tx);
2265 
2266 	ztest_range_unlock(rl);
2267 	ztest_object_unlock(zd, lr->lr_foid);
2268 
2269 	return (0);
2270 }
2271 
2272 static int
2273 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
2274 {
2275 	ztest_ds_t *zd = arg1;
2276 	lr_setattr_t *lr = arg2;
2277 	objset_t *os = zd->zd_os;
2278 	dmu_tx_t *tx;
2279 	dmu_buf_t *db;
2280 	ztest_block_tag_t *bbt;
2281 	uint64_t txg, lrtxg, crtxg, dnodesize;
2282 
2283 	if (byteswap)
2284 		byteswap_uint64_array(lr, sizeof (*lr));
2285 
2286 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
2287 
2288 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2289 
2290 	tx = dmu_tx_create(os);
2291 	dmu_tx_hold_bonus(tx, lr->lr_foid);
2292 
2293 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2294 	if (txg == 0) {
2295 		dmu_buf_rele(db, FTAG);
2296 		ztest_object_unlock(zd, lr->lr_foid);
2297 		return (ENOSPC);
2298 	}
2299 
2300 	bbt = ztest_bt_bonus(db);
2301 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2302 	crtxg = bbt->bt_crtxg;
2303 	lrtxg = lr->lr_common.lrc_txg;
2304 	dnodesize = bbt->bt_dnodesize;
2305 
2306 	if (zd->zd_zilog->zl_replay) {
2307 		ASSERT3U(lr->lr_size, !=, 0);
2308 		ASSERT3U(lr->lr_mode, !=, 0);
2309 		ASSERT3U(lrtxg, !=, 0);
2310 	} else {
2311 		/*
2312 		 * Randomly change the size and increment the generation.
2313 		 */
2314 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2315 		    sizeof (*bbt);
2316 		lr->lr_mode = bbt->bt_gen + 1;
2317 		ASSERT0(lrtxg);
2318 	}
2319 
2320 	/*
2321 	 * Verify that the current bonus buffer is not newer than our txg.
2322 	 */
2323 	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2324 	    MAX(txg, lrtxg), crtxg);
2325 
2326 	dmu_buf_will_dirty(db, tx);
2327 
2328 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2329 	ASSERT3U(lr->lr_size, <=, db->db_size);
2330 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2331 	bbt = ztest_bt_bonus(db);
2332 
2333 	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2334 	    txg, crtxg);
2335 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2336 	dmu_buf_rele(db, FTAG);
2337 
2338 	(void) ztest_log_setattr(zd, tx, lr);
2339 
2340 	dmu_tx_commit(tx);
2341 
2342 	ztest_object_unlock(zd, lr->lr_foid);
2343 
2344 	return (0);
2345 }
2346 
2347 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2348 	NULL,			/* 0 no such transaction type */
2349 	ztest_replay_create,	/* TX_CREATE */
2350 	NULL,			/* TX_MKDIR */
2351 	NULL,			/* TX_MKXATTR */
2352 	NULL,			/* TX_SYMLINK */
2353 	ztest_replay_remove,	/* TX_REMOVE */
2354 	NULL,			/* TX_RMDIR */
2355 	NULL,			/* TX_LINK */
2356 	NULL,			/* TX_RENAME */
2357 	ztest_replay_write,	/* TX_WRITE */
2358 	ztest_replay_truncate,	/* TX_TRUNCATE */
2359 	ztest_replay_setattr,	/* TX_SETATTR */
2360 	NULL,			/* TX_ACL */
2361 	NULL,			/* TX_CREATE_ACL */
2362 	NULL,			/* TX_CREATE_ATTR */
2363 	NULL,			/* TX_CREATE_ACL_ATTR */
2364 	NULL,			/* TX_MKDIR_ACL */
2365 	NULL,			/* TX_MKDIR_ATTR */
2366 	NULL,			/* TX_MKDIR_ACL_ATTR */
2367 	NULL,			/* TX_WRITE2 */
2368 	NULL,			/* TX_SETSAXATTR */
2369 };
2370 
2371 /*
2372  * ZIL get_data callbacks
2373  */
2374 
2375 static void
2376 ztest_get_done(zgd_t *zgd, int error)
2377 {
2378 	(void) error;
2379 	ztest_ds_t *zd = zgd->zgd_private;
2380 	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
2381 
2382 	if (zgd->zgd_db)
2383 		dmu_buf_rele(zgd->zgd_db, zgd);
2384 
2385 	ztest_range_unlock((rl_t *)zgd->zgd_lr);
2386 	ztest_object_unlock(zd, object);
2387 
2388 	umem_free(zgd, sizeof (*zgd));
2389 }
2390 
2391 static int
2392 ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
2393     struct lwb *lwb, zio_t *zio)
2394 {
2395 	(void) arg2;
2396 	ztest_ds_t *zd = arg;
2397 	objset_t *os = zd->zd_os;
2398 	uint64_t object = lr->lr_foid;
2399 	uint64_t offset = lr->lr_offset;
2400 	uint64_t size = lr->lr_length;
2401 	uint64_t txg = lr->lr_common.lrc_txg;
2402 	uint64_t crtxg;
2403 	dmu_object_info_t doi;
2404 	dmu_buf_t *db;
2405 	zgd_t *zgd;
2406 	int error;
2407 
2408 	ASSERT3P(lwb, !=, NULL);
2409 	ASSERT3P(zio, !=, NULL);
2410 	ASSERT3U(size, !=, 0);
2411 
2412 	ztest_object_lock(zd, object, RL_READER);
2413 	error = dmu_bonus_hold(os, object, FTAG, &db);
2414 	if (error) {
2415 		ztest_object_unlock(zd, object);
2416 		return (error);
2417 	}
2418 
2419 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
2420 
2421 	if (crtxg == 0 || crtxg > txg) {
2422 		dmu_buf_rele(db, FTAG);
2423 		ztest_object_unlock(zd, object);
2424 		return (ENOENT);
2425 	}
2426 
2427 	dmu_object_info_from_db(db, &doi);
2428 	dmu_buf_rele(db, FTAG);
2429 	db = NULL;
2430 
2431 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2432 	zgd->zgd_lwb = lwb;
2433 	zgd->zgd_private = zd;
2434 
2435 	if (buf != NULL) {	/* immediate write */
2436 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2437 		    object, offset, size, RL_READER);
2438 
2439 		error = dmu_read(os, object, offset, size, buf,
2440 		    DMU_READ_NO_PREFETCH);
2441 		ASSERT0(error);
2442 	} else {
2443 		size = doi.doi_data_block_size;
2444 		if (ISP2(size)) {
2445 			offset = P2ALIGN(offset, size);
2446 		} else {
2447 			ASSERT3U(offset, <, size);
2448 			offset = 0;
2449 		}
2450 
2451 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2452 		    object, offset, size, RL_READER);
2453 
2454 		error = dmu_buf_hold(os, object, offset, zgd, &db,
2455 		    DMU_READ_NO_PREFETCH);
2456 
2457 		if (error == 0) {
2458 			blkptr_t *bp = &lr->lr_blkptr;
2459 
2460 			zgd->zgd_db = db;
2461 			zgd->zgd_bp = bp;
2462 
2463 			ASSERT3U(db->db_offset, ==, offset);
2464 			ASSERT3U(db->db_size, ==, size);
2465 
2466 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2467 			    ztest_get_done, zgd);
2468 
2469 			if (error == 0)
2470 				return (0);
2471 		}
2472 	}
2473 
2474 	ztest_get_done(zgd, error);
2475 
2476 	return (error);
2477 }
2478 
2479 static void *
2480 ztest_lr_alloc(size_t lrsize, char *name)
2481 {
2482 	char *lr;
2483 	size_t namesize = name ? strlen(name) + 1 : 0;
2484 
2485 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2486 
2487 	if (name)
2488 		memcpy(lr + lrsize, name, namesize);
2489 
2490 	return (lr);
2491 }
2492 
2493 static void
2494 ztest_lr_free(void *lr, size_t lrsize, char *name)
2495 {
2496 	size_t namesize = name ? strlen(name) + 1 : 0;
2497 
2498 	umem_free(lr, lrsize + namesize);
2499 }
2500 
2501 /*
2502  * Lookup a bunch of objects.  Returns the number of objects not found.
2503  */
2504 static int
2505 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2506 {
2507 	int missing = 0;
2508 	int error;
2509 	int i;
2510 
2511 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2512 
2513 	for (i = 0; i < count; i++, od++) {
2514 		od->od_object = 0;
2515 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2516 		    sizeof (uint64_t), 1, &od->od_object);
2517 		if (error) {
2518 			ASSERT3S(error, ==, ENOENT);
2519 			ASSERT0(od->od_object);
2520 			missing++;
2521 		} else {
2522 			dmu_buf_t *db;
2523 			ztest_block_tag_t *bbt;
2524 			dmu_object_info_t doi;
2525 
2526 			ASSERT3U(od->od_object, !=, 0);
2527 			ASSERT0(missing);	/* there should be no gaps */
2528 
2529 			ztest_object_lock(zd, od->od_object, RL_READER);
2530 			VERIFY0(dmu_bonus_hold(zd->zd_os, od->od_object,
2531 			    FTAG, &db));
2532 			dmu_object_info_from_db(db, &doi);
2533 			bbt = ztest_bt_bonus(db);
2534 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2535 			od->od_type = doi.doi_type;
2536 			od->od_blocksize = doi.doi_data_block_size;
2537 			od->od_gen = bbt->bt_gen;
2538 			dmu_buf_rele(db, FTAG);
2539 			ztest_object_unlock(zd, od->od_object);
2540 		}
2541 	}
2542 
2543 	return (missing);
2544 }
2545 
2546 static int
2547 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2548 {
2549 	int missing = 0;
2550 	int i;
2551 
2552 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2553 
2554 	for (i = 0; i < count; i++, od++) {
2555 		if (missing) {
2556 			od->od_object = 0;
2557 			missing++;
2558 			continue;
2559 		}
2560 
2561 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2562 
2563 		lr->lr_doid = od->od_dir;
2564 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2565 		lr->lrz_type = od->od_crtype;
2566 		lr->lrz_blocksize = od->od_crblocksize;
2567 		lr->lrz_ibshift = ztest_random_ibshift();
2568 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2569 		lr->lrz_dnodesize = od->od_crdnodesize;
2570 		lr->lr_gen = od->od_crgen;
2571 		lr->lr_crtime[0] = time(NULL);
2572 
2573 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2574 			ASSERT0(missing);
2575 			od->od_object = 0;
2576 			missing++;
2577 		} else {
2578 			od->od_object = lr->lr_foid;
2579 			od->od_type = od->od_crtype;
2580 			od->od_blocksize = od->od_crblocksize;
2581 			od->od_gen = od->od_crgen;
2582 			ASSERT3U(od->od_object, !=, 0);
2583 		}
2584 
2585 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2586 	}
2587 
2588 	return (missing);
2589 }
2590 
2591 static int
2592 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2593 {
2594 	int missing = 0;
2595 	int error;
2596 	int i;
2597 
2598 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2599 
2600 	od += count - 1;
2601 
2602 	for (i = count - 1; i >= 0; i--, od--) {
2603 		if (missing) {
2604 			missing++;
2605 			continue;
2606 		}
2607 
2608 		/*
2609 		 * No object was found.
2610 		 */
2611 		if (od->od_object == 0)
2612 			continue;
2613 
2614 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2615 
2616 		lr->lr_doid = od->od_dir;
2617 
2618 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2619 			ASSERT3U(error, ==, ENOSPC);
2620 			missing++;
2621 		} else {
2622 			od->od_object = 0;
2623 		}
2624 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2625 	}
2626 
2627 	return (missing);
2628 }
2629 
2630 static int
2631 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2632     void *data)
2633 {
2634 	lr_write_t *lr;
2635 	int error;
2636 
2637 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2638 
2639 	lr->lr_foid = object;
2640 	lr->lr_offset = offset;
2641 	lr->lr_length = size;
2642 	lr->lr_blkoff = 0;
2643 	BP_ZERO(&lr->lr_blkptr);
2644 
2645 	memcpy(lr + 1, data, size);
2646 
2647 	error = ztest_replay_write(zd, lr, B_FALSE);
2648 
2649 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2650 
2651 	return (error);
2652 }
2653 
2654 static int
2655 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2656 {
2657 	lr_truncate_t *lr;
2658 	int error;
2659 
2660 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2661 
2662 	lr->lr_foid = object;
2663 	lr->lr_offset = offset;
2664 	lr->lr_length = size;
2665 
2666 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2667 
2668 	ztest_lr_free(lr, sizeof (*lr), NULL);
2669 
2670 	return (error);
2671 }
2672 
2673 static int
2674 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2675 {
2676 	lr_setattr_t *lr;
2677 	int error;
2678 
2679 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2680 
2681 	lr->lr_foid = object;
2682 	lr->lr_size = 0;
2683 	lr->lr_mode = 0;
2684 
2685 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2686 
2687 	ztest_lr_free(lr, sizeof (*lr), NULL);
2688 
2689 	return (error);
2690 }
2691 
2692 static void
2693 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2694 {
2695 	objset_t *os = zd->zd_os;
2696 	dmu_tx_t *tx;
2697 	uint64_t txg;
2698 	rl_t *rl;
2699 
2700 	txg_wait_synced(dmu_objset_pool(os), 0);
2701 
2702 	ztest_object_lock(zd, object, RL_READER);
2703 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2704 
2705 	tx = dmu_tx_create(os);
2706 
2707 	dmu_tx_hold_write(tx, object, offset, size);
2708 
2709 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2710 
2711 	if (txg != 0) {
2712 		dmu_prealloc(os, object, offset, size, tx);
2713 		dmu_tx_commit(tx);
2714 		txg_wait_synced(dmu_objset_pool(os), txg);
2715 	} else {
2716 		(void) dmu_free_long_range(os, object, offset, size);
2717 	}
2718 
2719 	ztest_range_unlock(rl);
2720 	ztest_object_unlock(zd, object);
2721 }
2722 
2723 static void
2724 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2725 {
2726 	int err;
2727 	ztest_block_tag_t wbt;
2728 	dmu_object_info_t doi;
2729 	enum ztest_io_type io_type;
2730 	uint64_t blocksize;
2731 	void *data;
2732 
2733 	VERIFY0(dmu_object_info(zd->zd_os, object, &doi));
2734 	blocksize = doi.doi_data_block_size;
2735 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2736 
2737 	/*
2738 	 * Pick an i/o type at random, biased toward writing block tags.
2739 	 */
2740 	io_type = ztest_random(ZTEST_IO_TYPES);
2741 	if (ztest_random(2) == 0)
2742 		io_type = ZTEST_IO_WRITE_TAG;
2743 
2744 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2745 
2746 	switch (io_type) {
2747 
2748 	case ZTEST_IO_WRITE_TAG:
2749 		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2750 		    offset, 0, 0, 0);
2751 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2752 		break;
2753 
2754 	case ZTEST_IO_WRITE_PATTERN:
2755 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2756 		if (ztest_random(2) == 0) {
2757 			/*
2758 			 * Induce fletcher2 collisions to ensure that
2759 			 * zio_ddt_collision() detects and resolves them
2760 			 * when using fletcher2-verify for deduplication.
2761 			 */
2762 			((uint64_t *)data)[0] ^= 1ULL << 63;
2763 			((uint64_t *)data)[4] ^= 1ULL << 63;
2764 		}
2765 		(void) ztest_write(zd, object, offset, blocksize, data);
2766 		break;
2767 
2768 	case ZTEST_IO_WRITE_ZEROES:
2769 		memset(data, 0, blocksize);
2770 		(void) ztest_write(zd, object, offset, blocksize, data);
2771 		break;
2772 
2773 	case ZTEST_IO_TRUNCATE:
2774 		(void) ztest_truncate(zd, object, offset, blocksize);
2775 		break;
2776 
2777 	case ZTEST_IO_SETATTR:
2778 		(void) ztest_setattr(zd, object);
2779 		break;
2780 	default:
2781 		break;
2782 
2783 	case ZTEST_IO_REWRITE:
2784 		(void) pthread_rwlock_rdlock(&ztest_name_lock);
2785 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2786 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2787 		    B_FALSE);
2788 		VERIFY(err == 0 || err == ENOSPC);
2789 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2790 		    ZFS_PROP_COMPRESSION,
2791 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2792 		    B_FALSE);
2793 		VERIFY(err == 0 || err == ENOSPC);
2794 		(void) pthread_rwlock_unlock(&ztest_name_lock);
2795 
2796 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2797 		    DMU_READ_NO_PREFETCH));
2798 
2799 		(void) ztest_write(zd, object, offset, blocksize, data);
2800 		break;
2801 	}
2802 
2803 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2804 
2805 	umem_free(data, blocksize);
2806 }
2807 
2808 /*
2809  * Initialize an object description template.
2810  */
2811 static void
2812 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2813     dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2814     uint64_t gen)
2815 {
2816 	od->od_dir = ZTEST_DIROBJ;
2817 	od->od_object = 0;
2818 
2819 	od->od_crtype = type;
2820 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2821 	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2822 	od->od_crgen = gen;
2823 
2824 	od->od_type = DMU_OT_NONE;
2825 	od->od_blocksize = 0;
2826 	od->od_gen = 0;
2827 
2828 	(void) snprintf(od->od_name, sizeof (od->od_name),
2829 	    "%s(%"PRId64")[%"PRIu64"]",
2830 	    tag, id, index);
2831 }
2832 
2833 /*
2834  * Lookup or create the objects for a test using the od template.
2835  * If the objects do not all exist, or if 'remove' is specified,
2836  * remove any existing objects and create new ones.  Otherwise,
2837  * use the existing objects.
2838  */
2839 static int
2840 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2841 {
2842 	int count = size / sizeof (*od);
2843 	int rv = 0;
2844 
2845 	mutex_enter(&zd->zd_dirobj_lock);
2846 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2847 	    (ztest_remove(zd, od, count) != 0 ||
2848 	    ztest_create(zd, od, count) != 0))
2849 		rv = -1;
2850 	zd->zd_od = od;
2851 	mutex_exit(&zd->zd_dirobj_lock);
2852 
2853 	return (rv);
2854 }
2855 
2856 void
2857 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2858 {
2859 	(void) id;
2860 	zilog_t *zilog = zd->zd_zilog;
2861 
2862 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2863 
2864 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2865 
2866 	/*
2867 	 * Remember the committed values in zd, which is in parent/child
2868 	 * shared memory.  If we die, the next iteration of ztest_run()
2869 	 * will verify that the log really does contain this record.
2870 	 */
2871 	mutex_enter(&zilog->zl_lock);
2872 	ASSERT3P(zd->zd_shared, !=, NULL);
2873 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2874 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2875 	mutex_exit(&zilog->zl_lock);
2876 
2877 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2878 }
2879 
2880 /*
2881  * This function is designed to simulate the operations that occur during a
2882  * mount/unmount operation.  We hold the dataset across these operations in an
2883  * attempt to expose any implicit assumptions about ZIL management.
2884  */
2885 void
2886 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2887 {
2888 	(void) id;
2889 	objset_t *os = zd->zd_os;
2890 
2891 	/*
2892 	 * We hold the ztest_vdev_lock so we don't cause problems with
2893 	 * other threads that wish to remove a log device, such as
2894 	 * ztest_device_removal().
2895 	 */
2896 	mutex_enter(&ztest_vdev_lock);
2897 
2898 	/*
2899 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2900 	 * updating the zil (i.e. adding in-memory log records) and the
2901 	 * zd_zilog_lock to block any I/O.
2902 	 */
2903 	mutex_enter(&zd->zd_dirobj_lock);
2904 	(void) pthread_rwlock_wrlock(&zd->zd_zilog_lock);
2905 
2906 	/* zfsvfs_teardown() */
2907 	zil_close(zd->zd_zilog);
2908 
2909 	/* zfsvfs_setup() */
2910 	VERIFY3P(zil_open(os, ztest_get_data), ==, zd->zd_zilog);
2911 	zil_replay(os, zd, ztest_replay_vector);
2912 
2913 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2914 	mutex_exit(&zd->zd_dirobj_lock);
2915 	mutex_exit(&ztest_vdev_lock);
2916 }
2917 
2918 /*
2919  * Verify that we can't destroy an active pool, create an existing pool,
2920  * or create a pool with a bad vdev spec.
2921  */
2922 void
2923 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2924 {
2925 	(void) zd, (void) id;
2926 	ztest_shared_opts_t *zo = &ztest_opts;
2927 	spa_t *spa;
2928 	nvlist_t *nvroot;
2929 
2930 	if (zo->zo_mmp_test)
2931 		return;
2932 
2933 	/*
2934 	 * Attempt to create using a bad file.
2935 	 */
2936 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2937 	VERIFY3U(ENOENT, ==,
2938 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2939 	fnvlist_free(nvroot);
2940 
2941 	/*
2942 	 * Attempt to create using a bad mirror.
2943 	 */
2944 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
2945 	VERIFY3U(ENOENT, ==,
2946 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2947 	fnvlist_free(nvroot);
2948 
2949 	/*
2950 	 * Attempt to create an existing pool.  It shouldn't matter
2951 	 * what's in the nvroot; we should fail with EEXIST.
2952 	 */
2953 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
2954 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2955 	VERIFY3U(EEXIST, ==,
2956 	    spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2957 	fnvlist_free(nvroot);
2958 
2959 	/*
2960 	 * We open a reference to the spa and then we try to export it
2961 	 * expecting one of the following errors:
2962 	 *
2963 	 * EBUSY
2964 	 *	Because of the reference we just opened.
2965 	 *
2966 	 * ZFS_ERR_EXPORT_IN_PROGRESS
2967 	 *	For the case that there is another ztest thread doing
2968 	 *	an export concurrently.
2969 	 */
2970 	VERIFY0(spa_open(zo->zo_pool, &spa, FTAG));
2971 	int error = spa_destroy(zo->zo_pool);
2972 	if (error != EBUSY && error != ZFS_ERR_EXPORT_IN_PROGRESS) {
2973 		fatal(B_FALSE, "spa_destroy(%s) returned unexpected value %d",
2974 		    spa->spa_name, error);
2975 	}
2976 	spa_close(spa, FTAG);
2977 
2978 	(void) pthread_rwlock_unlock(&ztest_name_lock);
2979 }
2980 
2981 /*
2982  * Start and then stop the MMP threads to ensure the startup and shutdown code
2983  * works properly.  Actual protection and property-related code tested via ZTS.
2984  */
2985 void
2986 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2987 {
2988 	(void) zd, (void) id;
2989 	ztest_shared_opts_t *zo = &ztest_opts;
2990 	spa_t *spa = ztest_spa;
2991 
2992 	if (zo->zo_mmp_test)
2993 		return;
2994 
2995 	/*
2996 	 * Since enabling MMP involves setting a property, it could not be done
2997 	 * while the pool is suspended.
2998 	 */
2999 	if (spa_suspended(spa))
3000 		return;
3001 
3002 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3003 	mutex_enter(&spa->spa_props_lock);
3004 
3005 	zfs_multihost_fail_intervals = 0;
3006 
3007 	if (!spa_multihost(spa)) {
3008 		spa->spa_multihost = B_TRUE;
3009 		mmp_thread_start(spa);
3010 	}
3011 
3012 	mutex_exit(&spa->spa_props_lock);
3013 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3014 
3015 	txg_wait_synced(spa_get_dsl(spa), 0);
3016 	mmp_signal_all_threads();
3017 	txg_wait_synced(spa_get_dsl(spa), 0);
3018 
3019 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3020 	mutex_enter(&spa->spa_props_lock);
3021 
3022 	if (spa_multihost(spa)) {
3023 		mmp_thread_stop(spa);
3024 		spa->spa_multihost = B_FALSE;
3025 	}
3026 
3027 	mutex_exit(&spa->spa_props_lock);
3028 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3029 }
3030 
3031 void
3032 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
3033 {
3034 	(void) zd, (void) id;
3035 	spa_t *spa;
3036 	uint64_t initial_version = SPA_VERSION_INITIAL;
3037 	uint64_t version, newversion;
3038 	nvlist_t *nvroot, *props;
3039 	char *name;
3040 
3041 	if (ztest_opts.zo_mmp_test)
3042 		return;
3043 
3044 	/* dRAID added after feature flags, skip upgrade test. */
3045 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0)
3046 		return;
3047 
3048 	mutex_enter(&ztest_vdev_lock);
3049 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
3050 
3051 	/*
3052 	 * Clean up from previous runs.
3053 	 */
3054 	(void) spa_destroy(name);
3055 
3056 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
3057 	    NULL, ztest_opts.zo_raid_children, ztest_opts.zo_mirrors, 1);
3058 
3059 	/*
3060 	 * If we're configuring a RAIDZ device then make sure that the
3061 	 * initial version is capable of supporting that feature.
3062 	 */
3063 	switch (ztest_opts.zo_raid_parity) {
3064 	case 0:
3065 	case 1:
3066 		initial_version = SPA_VERSION_INITIAL;
3067 		break;
3068 	case 2:
3069 		initial_version = SPA_VERSION_RAIDZ2;
3070 		break;
3071 	case 3:
3072 		initial_version = SPA_VERSION_RAIDZ3;
3073 		break;
3074 	}
3075 
3076 	/*
3077 	 * Create a pool with a spa version that can be upgraded. Pick
3078 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
3079 	 */
3080 	do {
3081 		version = ztest_random_spa_version(initial_version);
3082 	} while (version > SPA_VERSION_BEFORE_FEATURES);
3083 
3084 	props = fnvlist_alloc();
3085 	fnvlist_add_uint64(props,
3086 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
3087 	VERIFY0(spa_create(name, nvroot, props, NULL, NULL));
3088 	fnvlist_free(nvroot);
3089 	fnvlist_free(props);
3090 
3091 	VERIFY0(spa_open(name, &spa, FTAG));
3092 	VERIFY3U(spa_version(spa), ==, version);
3093 	newversion = ztest_random_spa_version(version + 1);
3094 
3095 	if (ztest_opts.zo_verbose >= 4) {
3096 		(void) printf("upgrading spa version from "
3097 		    "%"PRIu64" to %"PRIu64"\n",
3098 		    version, newversion);
3099 	}
3100 
3101 	spa_upgrade(spa, newversion);
3102 	VERIFY3U(spa_version(spa), >, version);
3103 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
3104 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
3105 	spa_close(spa, FTAG);
3106 
3107 	kmem_strfree(name);
3108 	mutex_exit(&ztest_vdev_lock);
3109 }
3110 
3111 static void
3112 ztest_spa_checkpoint(spa_t *spa)
3113 {
3114 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3115 
3116 	int error = spa_checkpoint(spa->spa_name);
3117 
3118 	switch (error) {
3119 	case 0:
3120 	case ZFS_ERR_DEVRM_IN_PROGRESS:
3121 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3122 	case ZFS_ERR_CHECKPOINT_EXISTS:
3123 		break;
3124 	case ENOSPC:
3125 		ztest_record_enospc(FTAG);
3126 		break;
3127 	default:
3128 		fatal(B_FALSE, "spa_checkpoint(%s) = %d", spa->spa_name, error);
3129 	}
3130 }
3131 
3132 static void
3133 ztest_spa_discard_checkpoint(spa_t *spa)
3134 {
3135 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3136 
3137 	int error = spa_checkpoint_discard(spa->spa_name);
3138 
3139 	switch (error) {
3140 	case 0:
3141 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3142 	case ZFS_ERR_NO_CHECKPOINT:
3143 		break;
3144 	default:
3145 		fatal(B_FALSE, "spa_discard_checkpoint(%s) = %d",
3146 		    spa->spa_name, error);
3147 	}
3148 
3149 }
3150 
3151 void
3152 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
3153 {
3154 	(void) zd, (void) id;
3155 	spa_t *spa = ztest_spa;
3156 
3157 	mutex_enter(&ztest_checkpoint_lock);
3158 	if (ztest_random(2) == 0) {
3159 		ztest_spa_checkpoint(spa);
3160 	} else {
3161 		ztest_spa_discard_checkpoint(spa);
3162 	}
3163 	mutex_exit(&ztest_checkpoint_lock);
3164 }
3165 
3166 
3167 static vdev_t *
3168 vdev_lookup_by_path(vdev_t *vd, const char *path)
3169 {
3170 	vdev_t *mvd;
3171 	int c;
3172 
3173 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
3174 		return (vd);
3175 
3176 	for (c = 0; c < vd->vdev_children; c++)
3177 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
3178 		    NULL)
3179 			return (mvd);
3180 
3181 	return (NULL);
3182 }
3183 
3184 static int
3185 spa_num_top_vdevs(spa_t *spa)
3186 {
3187 	vdev_t *rvd = spa->spa_root_vdev;
3188 	ASSERT3U(spa_config_held(spa, SCL_VDEV, RW_READER), ==, SCL_VDEV);
3189 	return (rvd->vdev_children);
3190 }
3191 
3192 /*
3193  * Verify that vdev_add() works as expected.
3194  */
3195 void
3196 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
3197 {
3198 	(void) zd, (void) id;
3199 	ztest_shared_t *zs = ztest_shared;
3200 	spa_t *spa = ztest_spa;
3201 	uint64_t leaves;
3202 	uint64_t guid;
3203 	nvlist_t *nvroot;
3204 	int error;
3205 
3206 	if (ztest_opts.zo_mmp_test)
3207 		return;
3208 
3209 	mutex_enter(&ztest_vdev_lock);
3210 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) *
3211 	    ztest_opts.zo_raid_children;
3212 
3213 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3214 
3215 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3216 
3217 	/*
3218 	 * If we have slogs then remove them 1/4 of the time.
3219 	 */
3220 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
3221 		metaslab_group_t *mg;
3222 
3223 		/*
3224 		 * find the first real slog in log allocation class
3225 		 */
3226 		mg =  spa_log_class(spa)->mc_allocator[0].mca_rotor;
3227 		while (!mg->mg_vd->vdev_islog)
3228 			mg = mg->mg_next;
3229 
3230 		guid = mg->mg_vd->vdev_guid;
3231 
3232 		spa_config_exit(spa, SCL_VDEV, FTAG);
3233 
3234 		/*
3235 		 * We have to grab the zs_name_lock as writer to
3236 		 * prevent a race between removing a slog (dmu_objset_find)
3237 		 * and destroying a dataset. Removing the slog will
3238 		 * grab a reference on the dataset which may cause
3239 		 * dsl_destroy_head() to fail with EBUSY thus
3240 		 * leaving the dataset in an inconsistent state.
3241 		 */
3242 		pthread_rwlock_wrlock(&ztest_name_lock);
3243 		error = spa_vdev_remove(spa, guid, B_FALSE);
3244 		pthread_rwlock_unlock(&ztest_name_lock);
3245 
3246 		switch (error) {
3247 		case 0:
3248 		case EEXIST:	/* Generic zil_reset() error */
3249 		case EBUSY:	/* Replay required */
3250 		case EACCES:	/* Crypto key not loaded */
3251 		case ZFS_ERR_CHECKPOINT_EXISTS:
3252 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3253 			break;
3254 		default:
3255 			fatal(B_FALSE, "spa_vdev_remove() = %d", error);
3256 		}
3257 	} else {
3258 		spa_config_exit(spa, SCL_VDEV, FTAG);
3259 
3260 		/*
3261 		 * Make 1/4 of the devices be log devices
3262 		 */
3263 		nvroot = make_vdev_root(NULL, NULL, NULL,
3264 		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
3265 		    "log" : NULL, ztest_opts.zo_raid_children, zs->zs_mirrors,
3266 		    1);
3267 
3268 		error = spa_vdev_add(spa, nvroot);
3269 		fnvlist_free(nvroot);
3270 
3271 		switch (error) {
3272 		case 0:
3273 			break;
3274 		case ENOSPC:
3275 			ztest_record_enospc("spa_vdev_add");
3276 			break;
3277 		default:
3278 			fatal(B_FALSE, "spa_vdev_add() = %d", error);
3279 		}
3280 	}
3281 
3282 	mutex_exit(&ztest_vdev_lock);
3283 }
3284 
3285 void
3286 ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
3287 {
3288 	(void) zd, (void) id;
3289 	ztest_shared_t *zs = ztest_shared;
3290 	spa_t *spa = ztest_spa;
3291 	uint64_t leaves;
3292 	nvlist_t *nvroot;
3293 	const char *class = (ztest_random(2) == 0) ?
3294 	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
3295 	int error;
3296 
3297 	/*
3298 	 * By default add a special vdev 50% of the time
3299 	 */
3300 	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
3301 	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
3302 	    ztest_random(2) == 0)) {
3303 		return;
3304 	}
3305 
3306 	mutex_enter(&ztest_vdev_lock);
3307 
3308 	/* Only test with mirrors */
3309 	if (zs->zs_mirrors < 2) {
3310 		mutex_exit(&ztest_vdev_lock);
3311 		return;
3312 	}
3313 
3314 	/* requires feature@allocation_classes */
3315 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
3316 		mutex_exit(&ztest_vdev_lock);
3317 		return;
3318 	}
3319 
3320 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) *
3321 	    ztest_opts.zo_raid_children;
3322 
3323 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3324 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3325 	spa_config_exit(spa, SCL_VDEV, FTAG);
3326 
3327 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
3328 	    class, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
3329 
3330 	error = spa_vdev_add(spa, nvroot);
3331 	fnvlist_free(nvroot);
3332 
3333 	if (error == ENOSPC)
3334 		ztest_record_enospc("spa_vdev_add");
3335 	else if (error != 0)
3336 		fatal(B_FALSE, "spa_vdev_add() = %d", error);
3337 
3338 	/*
3339 	 * 50% of the time allow small blocks in the special class
3340 	 */
3341 	if (error == 0 &&
3342 	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
3343 		if (ztest_opts.zo_verbose >= 3)
3344 			(void) printf("Enabling special VDEV small blocks\n");
3345 		(void) ztest_dsl_prop_set_uint64(zd->zd_name,
3346 		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
3347 	}
3348 
3349 	mutex_exit(&ztest_vdev_lock);
3350 
3351 	if (ztest_opts.zo_verbose >= 3) {
3352 		metaslab_class_t *mc;
3353 
3354 		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3355 			mc = spa_special_class(spa);
3356 		else
3357 			mc = spa_dedup_class(spa);
3358 		(void) printf("Added a %s mirrored vdev (of %d)\n",
3359 		    class, (int)mc->mc_groups);
3360 	}
3361 }
3362 
3363 /*
3364  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3365  */
3366 void
3367 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
3368 {
3369 	(void) zd, (void) id;
3370 	ztest_shared_t *zs = ztest_shared;
3371 	spa_t *spa = ztest_spa;
3372 	vdev_t *rvd = spa->spa_root_vdev;
3373 	spa_aux_vdev_t *sav;
3374 	char *aux;
3375 	char *path;
3376 	uint64_t guid = 0;
3377 	int error, ignore_err = 0;
3378 
3379 	if (ztest_opts.zo_mmp_test)
3380 		return;
3381 
3382 	path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3383 
3384 	if (ztest_random(2) == 0) {
3385 		sav = &spa->spa_spares;
3386 		aux = ZPOOL_CONFIG_SPARES;
3387 	} else {
3388 		sav = &spa->spa_l2cache;
3389 		aux = ZPOOL_CONFIG_L2CACHE;
3390 	}
3391 
3392 	mutex_enter(&ztest_vdev_lock);
3393 
3394 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3395 
3396 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
3397 		/*
3398 		 * Pick a random device to remove.
3399 		 */
3400 		vdev_t *svd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3401 
3402 		/* dRAID spares cannot be removed; try anyways to see ENOTSUP */
3403 		if (strstr(svd->vdev_path, VDEV_TYPE_DRAID) != NULL)
3404 			ignore_err = ENOTSUP;
3405 
3406 		guid = svd->vdev_guid;
3407 	} else {
3408 		/*
3409 		 * Find an unused device we can add.
3410 		 */
3411 		zs->zs_vdev_aux = 0;
3412 		for (;;) {
3413 			int c;
3414 			(void) snprintf(path, MAXPATHLEN, ztest_aux_template,
3415 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3416 			    zs->zs_vdev_aux);
3417 			for (c = 0; c < sav->sav_count; c++)
3418 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
3419 				    path) == 0)
3420 					break;
3421 			if (c == sav->sav_count &&
3422 			    vdev_lookup_by_path(rvd, path) == NULL)
3423 				break;
3424 			zs->zs_vdev_aux++;
3425 		}
3426 	}
3427 
3428 	spa_config_exit(spa, SCL_VDEV, FTAG);
3429 
3430 	if (guid == 0) {
3431 		/*
3432 		 * Add a new device.
3433 		 */
3434 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3435 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
3436 		error = spa_vdev_add(spa, nvroot);
3437 
3438 		switch (error) {
3439 		case 0:
3440 			break;
3441 		default:
3442 			fatal(B_FALSE, "spa_vdev_add(%p) = %d", nvroot, error);
3443 		}
3444 		fnvlist_free(nvroot);
3445 	} else {
3446 		/*
3447 		 * Remove an existing device.  Sometimes, dirty its
3448 		 * vdev state first to make sure we handle removal
3449 		 * of devices that have pending state changes.
3450 		 */
3451 		if (ztest_random(2) == 0)
3452 			(void) vdev_online(spa, guid, 0, NULL);
3453 
3454 		error = spa_vdev_remove(spa, guid, B_FALSE);
3455 
3456 		switch (error) {
3457 		case 0:
3458 		case EBUSY:
3459 		case ZFS_ERR_CHECKPOINT_EXISTS:
3460 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3461 			break;
3462 		default:
3463 			if (error != ignore_err)
3464 				fatal(B_FALSE,
3465 				    "spa_vdev_remove(%"PRIu64") = %d",
3466 				    guid, error);
3467 		}
3468 	}
3469 
3470 	mutex_exit(&ztest_vdev_lock);
3471 
3472 	umem_free(path, MAXPATHLEN);
3473 }
3474 
3475 /*
3476  * split a pool if it has mirror tlvdevs
3477  */
3478 void
3479 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3480 {
3481 	(void) zd, (void) id;
3482 	ztest_shared_t *zs = ztest_shared;
3483 	spa_t *spa = ztest_spa;
3484 	vdev_t *rvd = spa->spa_root_vdev;
3485 	nvlist_t *tree, **child, *config, *split, **schild;
3486 	uint_t c, children, schildren = 0, lastlogid = 0;
3487 	int error = 0;
3488 
3489 	if (ztest_opts.zo_mmp_test)
3490 		return;
3491 
3492 	mutex_enter(&ztest_vdev_lock);
3493 
3494 	/* ensure we have a usable config; mirrors of raidz aren't supported */
3495 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raid_children > 1) {
3496 		mutex_exit(&ztest_vdev_lock);
3497 		return;
3498 	}
3499 
3500 	/* clean up the old pool, if any */
3501 	(void) spa_destroy("splitp");
3502 
3503 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3504 
3505 	/* generate a config from the existing config */
3506 	mutex_enter(&spa->spa_props_lock);
3507 	tree = fnvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE);
3508 	mutex_exit(&spa->spa_props_lock);
3509 
3510 	VERIFY0(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN,
3511 	    &child, &children));
3512 
3513 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
3514 	for (c = 0; c < children; c++) {
3515 		vdev_t *tvd = rvd->vdev_child[c];
3516 		nvlist_t **mchild;
3517 		uint_t mchildren;
3518 
3519 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3520 			schild[schildren] = fnvlist_alloc();
3521 			fnvlist_add_string(schild[schildren],
3522 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE);
3523 			fnvlist_add_uint64(schild[schildren],
3524 			    ZPOOL_CONFIG_IS_HOLE, 1);
3525 			if (lastlogid == 0)
3526 				lastlogid = schildren;
3527 			++schildren;
3528 			continue;
3529 		}
3530 		lastlogid = 0;
3531 		VERIFY0(nvlist_lookup_nvlist_array(child[c],
3532 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren));
3533 		schild[schildren++] = fnvlist_dup(mchild[0]);
3534 	}
3535 
3536 	/* OK, create a config that can be used to split */
3537 	split = fnvlist_alloc();
3538 	fnvlist_add_string(split, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
3539 	fnvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN,
3540 	    (const nvlist_t **)schild, lastlogid != 0 ? lastlogid : schildren);
3541 
3542 	config = fnvlist_alloc();
3543 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split);
3544 
3545 	for (c = 0; c < schildren; c++)
3546 		fnvlist_free(schild[c]);
3547 	free(schild);
3548 	fnvlist_free(split);
3549 
3550 	spa_config_exit(spa, SCL_VDEV, FTAG);
3551 
3552 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
3553 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3554 	(void) pthread_rwlock_unlock(&ztest_name_lock);
3555 
3556 	fnvlist_free(config);
3557 
3558 	if (error == 0) {
3559 		(void) printf("successful split - results:\n");
3560 		mutex_enter(&spa_namespace_lock);
3561 		show_pool_stats(spa);
3562 		show_pool_stats(spa_lookup("splitp"));
3563 		mutex_exit(&spa_namespace_lock);
3564 		++zs->zs_splits;
3565 		--zs->zs_mirrors;
3566 	}
3567 	mutex_exit(&ztest_vdev_lock);
3568 }
3569 
3570 /*
3571  * Verify that we can attach and detach devices.
3572  */
3573 void
3574 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3575 {
3576 	(void) zd, (void) id;
3577 	ztest_shared_t *zs = ztest_shared;
3578 	spa_t *spa = ztest_spa;
3579 	spa_aux_vdev_t *sav = &spa->spa_spares;
3580 	vdev_t *rvd = spa->spa_root_vdev;
3581 	vdev_t *oldvd, *newvd, *pvd;
3582 	nvlist_t *root;
3583 	uint64_t leaves;
3584 	uint64_t leaf, top;
3585 	uint64_t ashift = ztest_get_ashift();
3586 	uint64_t oldguid, pguid;
3587 	uint64_t oldsize, newsize;
3588 	char *oldpath, *newpath;
3589 	int replacing;
3590 	int oldvd_has_siblings = B_FALSE;
3591 	int newvd_is_spare = B_FALSE;
3592 	int newvd_is_dspare = B_FALSE;
3593 	int oldvd_is_log;
3594 	int error, expected_error;
3595 
3596 	if (ztest_opts.zo_mmp_test)
3597 		return;
3598 
3599 	oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3600 	newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3601 
3602 	mutex_enter(&ztest_vdev_lock);
3603 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raid_children;
3604 
3605 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3606 
3607 	/*
3608 	 * If a vdev is in the process of being removed, its removal may
3609 	 * finish while we are in progress, leading to an unexpected error
3610 	 * value.  Don't bother trying to attach while we are in the middle
3611 	 * of removal.
3612 	 */
3613 	if (ztest_device_removal_active) {
3614 		spa_config_exit(spa, SCL_ALL, FTAG);
3615 		goto out;
3616 	}
3617 
3618 	/*
3619 	 * Decide whether to do an attach or a replace.
3620 	 */
3621 	replacing = ztest_random(2);
3622 
3623 	/*
3624 	 * Pick a random top-level vdev.
3625 	 */
3626 	top = ztest_random_vdev_top(spa, B_TRUE);
3627 
3628 	/*
3629 	 * Pick a random leaf within it.
3630 	 */
3631 	leaf = ztest_random(leaves);
3632 
3633 	/*
3634 	 * Locate this vdev.
3635 	 */
3636 	oldvd = rvd->vdev_child[top];
3637 
3638 	/* pick a child from the mirror */
3639 	if (zs->zs_mirrors >= 1) {
3640 		ASSERT3P(oldvd->vdev_ops, ==, &vdev_mirror_ops);
3641 		ASSERT3U(oldvd->vdev_children, >=, zs->zs_mirrors);
3642 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raid_children];
3643 	}
3644 
3645 	/* pick a child out of the raidz group */
3646 	if (ztest_opts.zo_raid_children > 1) {
3647 		if (strcmp(oldvd->vdev_ops->vdev_op_type, "raidz") == 0)
3648 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_raidz_ops);
3649 		else
3650 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_draid_ops);
3651 		ASSERT3U(oldvd->vdev_children, ==, ztest_opts.zo_raid_children);
3652 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raid_children];
3653 	}
3654 
3655 	/*
3656 	 * If we're already doing an attach or replace, oldvd may be a
3657 	 * mirror vdev -- in which case, pick a random child.
3658 	 */
3659 	while (oldvd->vdev_children != 0) {
3660 		oldvd_has_siblings = B_TRUE;
3661 		ASSERT3U(oldvd->vdev_children, >=, 2);
3662 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3663 	}
3664 
3665 	oldguid = oldvd->vdev_guid;
3666 	oldsize = vdev_get_min_asize(oldvd);
3667 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3668 	(void) strcpy(oldpath, oldvd->vdev_path);
3669 	pvd = oldvd->vdev_parent;
3670 	pguid = pvd->vdev_guid;
3671 
3672 	/*
3673 	 * If oldvd has siblings, then half of the time, detach it.  Prior
3674 	 * to the detach the pool is scrubbed in order to prevent creating
3675 	 * unrepairable blocks as a result of the data corruption injection.
3676 	 */
3677 	if (oldvd_has_siblings && ztest_random(2) == 0) {
3678 		spa_config_exit(spa, SCL_ALL, FTAG);
3679 
3680 		error = ztest_scrub_impl(spa);
3681 		if (error)
3682 			goto out;
3683 
3684 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3685 		if (error != 0 && error != ENODEV && error != EBUSY &&
3686 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3687 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3688 			fatal(B_FALSE, "detach (%s) returned %d",
3689 			    oldpath, error);
3690 		goto out;
3691 	}
3692 
3693 	/*
3694 	 * For the new vdev, choose with equal probability between the two
3695 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3696 	 */
3697 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3698 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3699 		newvd_is_spare = B_TRUE;
3700 
3701 		if (newvd->vdev_ops == &vdev_draid_spare_ops)
3702 			newvd_is_dspare = B_TRUE;
3703 
3704 		(void) strcpy(newpath, newvd->vdev_path);
3705 	} else {
3706 		(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3707 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3708 		    top * leaves + leaf);
3709 		if (ztest_random(2) == 0)
3710 			newpath[strlen(newpath) - 1] = 'b';
3711 		newvd = vdev_lookup_by_path(rvd, newpath);
3712 	}
3713 
3714 	if (newvd) {
3715 		/*
3716 		 * Reopen to ensure the vdev's asize field isn't stale.
3717 		 */
3718 		vdev_reopen(newvd);
3719 		newsize = vdev_get_min_asize(newvd);
3720 	} else {
3721 		/*
3722 		 * Make newsize a little bigger or smaller than oldsize.
3723 		 * If it's smaller, the attach should fail.
3724 		 * If it's larger, and we're doing a replace,
3725 		 * we should get dynamic LUN growth when we're done.
3726 		 */
3727 		newsize = 10 * oldsize / (9 + ztest_random(3));
3728 	}
3729 
3730 	/*
3731 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3732 	 * unless it's a replace; in that case any non-replacing parent is OK.
3733 	 *
3734 	 * If newvd is already part of the pool, it should fail with EBUSY.
3735 	 *
3736 	 * If newvd is too small, it should fail with EOVERFLOW.
3737 	 *
3738 	 * If newvd is a distributed spare and it's being attached to a
3739 	 * dRAID which is not its parent it should fail with EINVAL.
3740 	 */
3741 	if (pvd->vdev_ops != &vdev_mirror_ops &&
3742 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3743 	    pvd->vdev_ops == &vdev_replacing_ops ||
3744 	    pvd->vdev_ops == &vdev_spare_ops))
3745 		expected_error = ENOTSUP;
3746 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
3747 		expected_error = ENOTSUP;
3748 	else if (newvd == oldvd)
3749 		expected_error = replacing ? 0 : EBUSY;
3750 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3751 		expected_error = EBUSY;
3752 	else if (!newvd_is_dspare && newsize < oldsize)
3753 		expected_error = EOVERFLOW;
3754 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3755 		expected_error = EDOM;
3756 	else if (newvd_is_dspare && pvd != vdev_draid_spare_get_parent(newvd))
3757 		expected_error = ENOTSUP;
3758 	else
3759 		expected_error = 0;
3760 
3761 	spa_config_exit(spa, SCL_ALL, FTAG);
3762 
3763 	/*
3764 	 * Build the nvlist describing newpath.
3765 	 */
3766 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3767 	    ashift, NULL, 0, 0, 1);
3768 
3769 	/*
3770 	 * When supported select either a healing or sequential resilver.
3771 	 */
3772 	boolean_t rebuilding = B_FALSE;
3773 	if (pvd->vdev_ops == &vdev_mirror_ops ||
3774 	    pvd->vdev_ops ==  &vdev_root_ops) {
3775 		rebuilding = !!ztest_random(2);
3776 	}
3777 
3778 	error = spa_vdev_attach(spa, oldguid, root, replacing, rebuilding);
3779 
3780 	fnvlist_free(root);
3781 
3782 	/*
3783 	 * If our parent was the replacing vdev, but the replace completed,
3784 	 * then instead of failing with ENOTSUP we may either succeed,
3785 	 * fail with ENODEV, or fail with EOVERFLOW.
3786 	 */
3787 	if (expected_error == ENOTSUP &&
3788 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3789 		expected_error = error;
3790 
3791 	/*
3792 	 * If someone grew the LUN, the replacement may be too small.
3793 	 */
3794 	if (error == EOVERFLOW || error == EBUSY)
3795 		expected_error = error;
3796 
3797 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3798 	    error == ZFS_ERR_DISCARDING_CHECKPOINT ||
3799 	    error == ZFS_ERR_RESILVER_IN_PROGRESS ||
3800 	    error == ZFS_ERR_REBUILD_IN_PROGRESS)
3801 		expected_error = error;
3802 
3803 	if (error != expected_error && expected_error != EBUSY) {
3804 		fatal(B_FALSE, "attach (%s %"PRIu64", %s %"PRIu64", %d) "
3805 		    "returned %d, expected %d",
3806 		    oldpath, oldsize, newpath,
3807 		    newsize, replacing, error, expected_error);
3808 	}
3809 out:
3810 	mutex_exit(&ztest_vdev_lock);
3811 
3812 	umem_free(oldpath, MAXPATHLEN);
3813 	umem_free(newpath, MAXPATHLEN);
3814 }
3815 
3816 void
3817 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3818 {
3819 	(void) zd, (void) id;
3820 	spa_t *spa = ztest_spa;
3821 	vdev_t *vd;
3822 	uint64_t guid;
3823 	int error;
3824 
3825 	mutex_enter(&ztest_vdev_lock);
3826 
3827 	if (ztest_device_removal_active) {
3828 		mutex_exit(&ztest_vdev_lock);
3829 		return;
3830 	}
3831 
3832 	/*
3833 	 * Remove a random top-level vdev and wait for removal to finish.
3834 	 */
3835 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3836 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3837 	guid = vd->vdev_guid;
3838 	spa_config_exit(spa, SCL_VDEV, FTAG);
3839 
3840 	error = spa_vdev_remove(spa, guid, B_FALSE);
3841 	if (error == 0) {
3842 		ztest_device_removal_active = B_TRUE;
3843 		mutex_exit(&ztest_vdev_lock);
3844 
3845 		/*
3846 		 * spa->spa_vdev_removal is created in a sync task that
3847 		 * is initiated via dsl_sync_task_nowait(). Since the
3848 		 * task may not run before spa_vdev_remove() returns, we
3849 		 * must wait at least 1 txg to ensure that the removal
3850 		 * struct has been created.
3851 		 */
3852 		txg_wait_synced(spa_get_dsl(spa), 0);
3853 
3854 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
3855 			txg_wait_synced(spa_get_dsl(spa), 0);
3856 	} else {
3857 		mutex_exit(&ztest_vdev_lock);
3858 		return;
3859 	}
3860 
3861 	/*
3862 	 * The pool needs to be scrubbed after completing device removal.
3863 	 * Failure to do so may result in checksum errors due to the
3864 	 * strategy employed by ztest_fault_inject() when selecting which
3865 	 * offset are redundant and can be damaged.
3866 	 */
3867 	error = spa_scan(spa, POOL_SCAN_SCRUB);
3868 	if (error == 0) {
3869 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3870 			txg_wait_synced(spa_get_dsl(spa), 0);
3871 	}
3872 
3873 	mutex_enter(&ztest_vdev_lock);
3874 	ztest_device_removal_active = B_FALSE;
3875 	mutex_exit(&ztest_vdev_lock);
3876 }
3877 
3878 /*
3879  * Callback function which expands the physical size of the vdev.
3880  */
3881 static vdev_t *
3882 grow_vdev(vdev_t *vd, void *arg)
3883 {
3884 	spa_t *spa __maybe_unused = vd->vdev_spa;
3885 	size_t *newsize = arg;
3886 	size_t fsize;
3887 	int fd;
3888 
3889 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
3890 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3891 
3892 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3893 		return (vd);
3894 
3895 	fsize = lseek(fd, 0, SEEK_END);
3896 	VERIFY0(ftruncate(fd, *newsize));
3897 
3898 	if (ztest_opts.zo_verbose >= 6) {
3899 		(void) printf("%s grew from %lu to %lu bytes\n",
3900 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3901 	}
3902 	(void) close(fd);
3903 	return (NULL);
3904 }
3905 
3906 /*
3907  * Callback function which expands a given vdev by calling vdev_online().
3908  */
3909 static vdev_t *
3910 online_vdev(vdev_t *vd, void *arg)
3911 {
3912 	(void) arg;
3913 	spa_t *spa = vd->vdev_spa;
3914 	vdev_t *tvd = vd->vdev_top;
3915 	uint64_t guid = vd->vdev_guid;
3916 	uint64_t generation = spa->spa_config_generation + 1;
3917 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3918 	int error;
3919 
3920 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
3921 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3922 
3923 	/* Calling vdev_online will initialize the new metaslabs */
3924 	spa_config_exit(spa, SCL_STATE, spa);
3925 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3926 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3927 
3928 	/*
3929 	 * If vdev_online returned an error or the underlying vdev_open
3930 	 * failed then we abort the expand. The only way to know that
3931 	 * vdev_open fails is by checking the returned newstate.
3932 	 */
3933 	if (error || newstate != VDEV_STATE_HEALTHY) {
3934 		if (ztest_opts.zo_verbose >= 5) {
3935 			(void) printf("Unable to expand vdev, state %u, "
3936 			    "error %d\n", newstate, error);
3937 		}
3938 		return (vd);
3939 	}
3940 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3941 
3942 	/*
3943 	 * Since we dropped the lock we need to ensure that we're
3944 	 * still talking to the original vdev. It's possible this
3945 	 * vdev may have been detached/replaced while we were
3946 	 * trying to online it.
3947 	 */
3948 	if (generation != spa->spa_config_generation) {
3949 		if (ztest_opts.zo_verbose >= 5) {
3950 			(void) printf("vdev configuration has changed, "
3951 			    "guid %"PRIu64", state %"PRIu64", "
3952 			    "expected gen %"PRIu64", got gen %"PRIu64"\n",
3953 			    guid,
3954 			    tvd->vdev_state,
3955 			    generation,
3956 			    spa->spa_config_generation);
3957 		}
3958 		return (vd);
3959 	}
3960 	return (NULL);
3961 }
3962 
3963 /*
3964  * Traverse the vdev tree calling the supplied function.
3965  * We continue to walk the tree until we either have walked all
3966  * children or we receive a non-NULL return from the callback.
3967  * If a NULL callback is passed, then we just return back the first
3968  * leaf vdev we encounter.
3969  */
3970 static vdev_t *
3971 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3972 {
3973 	uint_t c;
3974 
3975 	if (vd->vdev_ops->vdev_op_leaf) {
3976 		if (func == NULL)
3977 			return (vd);
3978 		else
3979 			return (func(vd, arg));
3980 	}
3981 
3982 	for (c = 0; c < vd->vdev_children; c++) {
3983 		vdev_t *cvd = vd->vdev_child[c];
3984 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3985 			return (cvd);
3986 	}
3987 	return (NULL);
3988 }
3989 
3990 /*
3991  * Verify that dynamic LUN growth works as expected.
3992  */
3993 void
3994 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3995 {
3996 	(void) zd, (void) id;
3997 	spa_t *spa = ztest_spa;
3998 	vdev_t *vd, *tvd;
3999 	metaslab_class_t *mc;
4000 	metaslab_group_t *mg;
4001 	size_t psize, newsize;
4002 	uint64_t top;
4003 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
4004 
4005 	mutex_enter(&ztest_checkpoint_lock);
4006 	mutex_enter(&ztest_vdev_lock);
4007 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4008 
4009 	/*
4010 	 * If there is a vdev removal in progress, it could complete while
4011 	 * we are running, in which case we would not be able to verify
4012 	 * that the metaslab_class space increased (because it decreases
4013 	 * when the device removal completes).
4014 	 */
4015 	if (ztest_device_removal_active) {
4016 		spa_config_exit(spa, SCL_STATE, spa);
4017 		mutex_exit(&ztest_vdev_lock);
4018 		mutex_exit(&ztest_checkpoint_lock);
4019 		return;
4020 	}
4021 
4022 	top = ztest_random_vdev_top(spa, B_TRUE);
4023 
4024 	tvd = spa->spa_root_vdev->vdev_child[top];
4025 	mg = tvd->vdev_mg;
4026 	mc = mg->mg_class;
4027 	old_ms_count = tvd->vdev_ms_count;
4028 	old_class_space = metaslab_class_get_space(mc);
4029 
4030 	/*
4031 	 * Determine the size of the first leaf vdev associated with
4032 	 * our top-level device.
4033 	 */
4034 	vd = vdev_walk_tree(tvd, NULL, NULL);
4035 	ASSERT3P(vd, !=, NULL);
4036 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4037 
4038 	psize = vd->vdev_psize;
4039 
4040 	/*
4041 	 * We only try to expand the vdev if it's healthy, less than 4x its
4042 	 * original size, and it has a valid psize.
4043 	 */
4044 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
4045 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
4046 		spa_config_exit(spa, SCL_STATE, spa);
4047 		mutex_exit(&ztest_vdev_lock);
4048 		mutex_exit(&ztest_checkpoint_lock);
4049 		return;
4050 	}
4051 	ASSERT3U(psize, >, 0);
4052 	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
4053 	ASSERT3U(newsize, >, psize);
4054 
4055 	if (ztest_opts.zo_verbose >= 6) {
4056 		(void) printf("Expanding LUN %s from %lu to %lu\n",
4057 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
4058 	}
4059 
4060 	/*
4061 	 * Growing the vdev is a two step process:
4062 	 *	1). expand the physical size (i.e. relabel)
4063 	 *	2). online the vdev to create the new metaslabs
4064 	 */
4065 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
4066 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
4067 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
4068 		if (ztest_opts.zo_verbose >= 5) {
4069 			(void) printf("Could not expand LUN because "
4070 			    "the vdev configuration changed.\n");
4071 		}
4072 		spa_config_exit(spa, SCL_STATE, spa);
4073 		mutex_exit(&ztest_vdev_lock);
4074 		mutex_exit(&ztest_checkpoint_lock);
4075 		return;
4076 	}
4077 
4078 	spa_config_exit(spa, SCL_STATE, spa);
4079 
4080 	/*
4081 	 * Expanding the LUN will update the config asynchronously,
4082 	 * thus we must wait for the async thread to complete any
4083 	 * pending tasks before proceeding.
4084 	 */
4085 	for (;;) {
4086 		boolean_t done;
4087 		mutex_enter(&spa->spa_async_lock);
4088 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
4089 		mutex_exit(&spa->spa_async_lock);
4090 		if (done)
4091 			break;
4092 		txg_wait_synced(spa_get_dsl(spa), 0);
4093 		(void) poll(NULL, 0, 100);
4094 	}
4095 
4096 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4097 
4098 	tvd = spa->spa_root_vdev->vdev_child[top];
4099 	new_ms_count = tvd->vdev_ms_count;
4100 	new_class_space = metaslab_class_get_space(mc);
4101 
4102 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
4103 		if (ztest_opts.zo_verbose >= 5) {
4104 			(void) printf("Could not verify LUN expansion due to "
4105 			    "intervening vdev offline or remove.\n");
4106 		}
4107 		spa_config_exit(spa, SCL_STATE, spa);
4108 		mutex_exit(&ztest_vdev_lock);
4109 		mutex_exit(&ztest_checkpoint_lock);
4110 		return;
4111 	}
4112 
4113 	/*
4114 	 * Make sure we were able to grow the vdev.
4115 	 */
4116 	if (new_ms_count <= old_ms_count) {
4117 		fatal(B_FALSE,
4118 		    "LUN expansion failed: ms_count %"PRIu64" < %"PRIu64"\n",
4119 		    old_ms_count, new_ms_count);
4120 	}
4121 
4122 	/*
4123 	 * Make sure we were able to grow the pool.
4124 	 */
4125 	if (new_class_space <= old_class_space) {
4126 		fatal(B_FALSE,
4127 		    "LUN expansion failed: class_space %"PRIu64" < %"PRIu64"\n",
4128 		    old_class_space, new_class_space);
4129 	}
4130 
4131 	if (ztest_opts.zo_verbose >= 5) {
4132 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
4133 
4134 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
4135 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
4136 		(void) printf("%s grew from %s to %s\n",
4137 		    spa->spa_name, oldnumbuf, newnumbuf);
4138 	}
4139 
4140 	spa_config_exit(spa, SCL_STATE, spa);
4141 	mutex_exit(&ztest_vdev_lock);
4142 	mutex_exit(&ztest_checkpoint_lock);
4143 }
4144 
4145 /*
4146  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
4147  */
4148 static void
4149 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
4150 {
4151 	(void) arg, (void) cr;
4152 
4153 	/*
4154 	 * Create the objects common to all ztest datasets.
4155 	 */
4156 	VERIFY0(zap_create_claim(os, ZTEST_DIROBJ,
4157 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx));
4158 }
4159 
4160 static int
4161 ztest_dataset_create(char *dsname)
4162 {
4163 	int err;
4164 	uint64_t rand;
4165 	dsl_crypto_params_t *dcp = NULL;
4166 
4167 	/*
4168 	 * 50% of the time, we create encrypted datasets
4169 	 * using a random cipher suite and a hard-coded
4170 	 * wrapping key.
4171 	 */
4172 	rand = ztest_random(2);
4173 	if (rand != 0) {
4174 		nvlist_t *crypto_args = fnvlist_alloc();
4175 		nvlist_t *props = fnvlist_alloc();
4176 
4177 		/* slight bias towards the default cipher suite */
4178 		rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
4179 		if (rand < ZIO_CRYPT_AES_128_CCM)
4180 			rand = ZIO_CRYPT_ON;
4181 
4182 		fnvlist_add_uint64(props,
4183 		    zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
4184 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
4185 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
4186 
4187 		/*
4188 		 * These parameters aren't really used by the kernel. They
4189 		 * are simply stored so that userspace knows how to load
4190 		 * the wrapping key.
4191 		 */
4192 		fnvlist_add_uint64(props,
4193 		    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
4194 		fnvlist_add_string(props,
4195 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
4196 		fnvlist_add_uint64(props,
4197 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
4198 		fnvlist_add_uint64(props,
4199 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
4200 
4201 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
4202 		    crypto_args, &dcp));
4203 
4204 		/*
4205 		 * Cycle through all available encryption implementations
4206 		 * to verify interoperability.
4207 		 */
4208 		VERIFY0(gcm_impl_set("cycle"));
4209 		VERIFY0(aes_impl_set("cycle"));
4210 
4211 		fnvlist_free(crypto_args);
4212 		fnvlist_free(props);
4213 	}
4214 
4215 	err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
4216 	    ztest_objset_create_cb, NULL);
4217 	dsl_crypto_params_free(dcp, !!err);
4218 
4219 	rand = ztest_random(100);
4220 	if (err || rand < 80)
4221 		return (err);
4222 
4223 	if (ztest_opts.zo_verbose >= 5)
4224 		(void) printf("Setting dataset %s to sync always\n", dsname);
4225 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
4226 	    ZFS_SYNC_ALWAYS, B_FALSE));
4227 }
4228 
4229 static int
4230 ztest_objset_destroy_cb(const char *name, void *arg)
4231 {
4232 	(void) arg;
4233 	objset_t *os;
4234 	dmu_object_info_t doi;
4235 	int error;
4236 
4237 	/*
4238 	 * Verify that the dataset contains a directory object.
4239 	 */
4240 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
4241 	    B_TRUE, FTAG, &os));
4242 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
4243 	if (error != ENOENT) {
4244 		/* We could have crashed in the middle of destroying it */
4245 		ASSERT0(error);
4246 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
4247 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
4248 	}
4249 	dmu_objset_disown(os, B_TRUE, FTAG);
4250 
4251 	/*
4252 	 * Destroy the dataset.
4253 	 */
4254 	if (strchr(name, '@') != NULL) {
4255 		error = dsl_destroy_snapshot(name, B_TRUE);
4256 		if (error != ECHRNG) {
4257 			/*
4258 			 * The program was executed, but encountered a runtime
4259 			 * error, such as insufficient slop, or a hold on the
4260 			 * dataset.
4261 			 */
4262 			ASSERT0(error);
4263 		}
4264 	} else {
4265 		error = dsl_destroy_head(name);
4266 		if (error == ENOSPC) {
4267 			/* There could be checkpoint or insufficient slop */
4268 			ztest_record_enospc(FTAG);
4269 		} else if (error != EBUSY) {
4270 			/* There could be a hold on this dataset */
4271 			ASSERT0(error);
4272 		}
4273 	}
4274 	return (0);
4275 }
4276 
4277 static boolean_t
4278 ztest_snapshot_create(char *osname, uint64_t id)
4279 {
4280 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4281 	int error;
4282 
4283 	(void) snprintf(snapname, sizeof (snapname), "%"PRIu64"", id);
4284 
4285 	error = dmu_objset_snapshot_one(osname, snapname);
4286 	if (error == ENOSPC) {
4287 		ztest_record_enospc(FTAG);
4288 		return (B_FALSE);
4289 	}
4290 	if (error != 0 && error != EEXIST) {
4291 		fatal(B_FALSE, "ztest_snapshot_create(%s@%s) = %d", osname,
4292 		    snapname, error);
4293 	}
4294 	return (B_TRUE);
4295 }
4296 
4297 static boolean_t
4298 ztest_snapshot_destroy(char *osname, uint64_t id)
4299 {
4300 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4301 	int error;
4302 
4303 	(void) snprintf(snapname, sizeof (snapname), "%s@%"PRIu64"",
4304 	    osname, id);
4305 
4306 	error = dsl_destroy_snapshot(snapname, B_FALSE);
4307 	if (error != 0 && error != ENOENT)
4308 		fatal(B_FALSE, "ztest_snapshot_destroy(%s) = %d",
4309 		    snapname, error);
4310 	return (B_TRUE);
4311 }
4312 
4313 void
4314 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
4315 {
4316 	(void) zd;
4317 	ztest_ds_t *zdtmp;
4318 	int iters;
4319 	int error;
4320 	objset_t *os, *os2;
4321 	char name[ZFS_MAX_DATASET_NAME_LEN];
4322 	zilog_t *zilog;
4323 	int i;
4324 
4325 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
4326 
4327 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4328 
4329 	(void) snprintf(name, sizeof (name), "%s/temp_%"PRIu64"",
4330 	    ztest_opts.zo_pool, id);
4331 
4332 	/*
4333 	 * If this dataset exists from a previous run, process its replay log
4334 	 * half of the time.  If we don't replay it, then dsl_destroy_head()
4335 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
4336 	 */
4337 	if (ztest_random(2) == 0 &&
4338 	    ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
4339 	    B_TRUE, FTAG, &os) == 0) {
4340 		ztest_zd_init(zdtmp, NULL, os);
4341 		zil_replay(os, zdtmp, ztest_replay_vector);
4342 		ztest_zd_fini(zdtmp);
4343 		dmu_objset_disown(os, B_TRUE, FTAG);
4344 	}
4345 
4346 	/*
4347 	 * There may be an old instance of the dataset we're about to
4348 	 * create lying around from a previous run.  If so, destroy it
4349 	 * and all of its snapshots.
4350 	 */
4351 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
4352 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
4353 
4354 	/*
4355 	 * Verify that the destroyed dataset is no longer in the namespace.
4356 	 */
4357 	VERIFY3U(ENOENT, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
4358 	    B_TRUE, FTAG, &os));
4359 
4360 	/*
4361 	 * Verify that we can create a new dataset.
4362 	 */
4363 	error = ztest_dataset_create(name);
4364 	if (error) {
4365 		if (error == ENOSPC) {
4366 			ztest_record_enospc(FTAG);
4367 			goto out;
4368 		}
4369 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", name, error);
4370 	}
4371 
4372 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
4373 	    FTAG, &os));
4374 
4375 	ztest_zd_init(zdtmp, NULL, os);
4376 
4377 	/*
4378 	 * Open the intent log for it.
4379 	 */
4380 	zilog = zil_open(os, ztest_get_data);
4381 
4382 	/*
4383 	 * Put some objects in there, do a little I/O to them,
4384 	 * and randomly take a couple of snapshots along the way.
4385 	 */
4386 	iters = ztest_random(5);
4387 	for (i = 0; i < iters; i++) {
4388 		ztest_dmu_object_alloc_free(zdtmp, id);
4389 		if (ztest_random(iters) == 0)
4390 			(void) ztest_snapshot_create(name, i);
4391 	}
4392 
4393 	/*
4394 	 * Verify that we cannot create an existing dataset.
4395 	 */
4396 	VERIFY3U(EEXIST, ==,
4397 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
4398 
4399 	/*
4400 	 * Verify that we can hold an objset that is also owned.
4401 	 */
4402 	VERIFY0(dmu_objset_hold(name, FTAG, &os2));
4403 	dmu_objset_rele(os2, FTAG);
4404 
4405 	/*
4406 	 * Verify that we cannot own an objset that is already owned.
4407 	 */
4408 	VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
4409 	    B_FALSE, B_TRUE, FTAG, &os2));
4410 
4411 	zil_close(zilog);
4412 	dmu_objset_disown(os, B_TRUE, FTAG);
4413 	ztest_zd_fini(zdtmp);
4414 out:
4415 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4416 
4417 	umem_free(zdtmp, sizeof (ztest_ds_t));
4418 }
4419 
4420 /*
4421  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
4422  */
4423 void
4424 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
4425 {
4426 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4427 	(void) ztest_snapshot_destroy(zd->zd_name, id);
4428 	(void) ztest_snapshot_create(zd->zd_name, id);
4429 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4430 }
4431 
4432 /*
4433  * Cleanup non-standard snapshots and clones.
4434  */
4435 static void
4436 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
4437 {
4438 	char *snap1name;
4439 	char *clone1name;
4440 	char *snap2name;
4441 	char *clone2name;
4442 	char *snap3name;
4443 	int error;
4444 
4445 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4446 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4447 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4448 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4449 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4450 
4451 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4452 	    osname, id);
4453 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4454 	    osname, id);
4455 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4456 	    clone1name, id);
4457 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4458 	    osname, id);
4459 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4460 	    clone1name, id);
4461 
4462 	error = dsl_destroy_head(clone2name);
4463 	if (error && error != ENOENT)
4464 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone2name, error);
4465 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
4466 	if (error && error != ENOENT)
4467 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4468 		    snap3name, error);
4469 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
4470 	if (error && error != ENOENT)
4471 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4472 		    snap2name, error);
4473 	error = dsl_destroy_head(clone1name);
4474 	if (error && error != ENOENT)
4475 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone1name, error);
4476 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
4477 	if (error && error != ENOENT)
4478 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4479 		    snap1name, error);
4480 
4481 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4482 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4483 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4484 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4485 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4486 }
4487 
4488 /*
4489  * Verify dsl_dataset_promote handles EBUSY
4490  */
4491 void
4492 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
4493 {
4494 	objset_t *os;
4495 	char *snap1name;
4496 	char *clone1name;
4497 	char *snap2name;
4498 	char *clone2name;
4499 	char *snap3name;
4500 	char *osname = zd->zd_name;
4501 	int error;
4502 
4503 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4504 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4505 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4506 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4507 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4508 
4509 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4510 
4511 	ztest_dsl_dataset_cleanup(osname, id);
4512 
4513 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4514 	    osname, id);
4515 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4516 	    osname, id);
4517 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4518 	    clone1name, id);
4519 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4520 	    osname, id);
4521 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4522 	    clone1name, id);
4523 
4524 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
4525 	if (error && error != EEXIST) {
4526 		if (error == ENOSPC) {
4527 			ztest_record_enospc(FTAG);
4528 			goto out;
4529 		}
4530 		fatal(B_FALSE, "dmu_take_snapshot(%s) = %d", snap1name, error);
4531 	}
4532 
4533 	error = dmu_objset_clone(clone1name, snap1name);
4534 	if (error) {
4535 		if (error == ENOSPC) {
4536 			ztest_record_enospc(FTAG);
4537 			goto out;
4538 		}
4539 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone1name, error);
4540 	}
4541 
4542 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4543 	if (error && error != EEXIST) {
4544 		if (error == ENOSPC) {
4545 			ztest_record_enospc(FTAG);
4546 			goto out;
4547 		}
4548 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap2name, error);
4549 	}
4550 
4551 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4552 	if (error && error != EEXIST) {
4553 		if (error == ENOSPC) {
4554 			ztest_record_enospc(FTAG);
4555 			goto out;
4556 		}
4557 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap3name, error);
4558 	}
4559 
4560 	error = dmu_objset_clone(clone2name, snap3name);
4561 	if (error) {
4562 		if (error == ENOSPC) {
4563 			ztest_record_enospc(FTAG);
4564 			goto out;
4565 		}
4566 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone2name, error);
4567 	}
4568 
4569 	error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
4570 	    FTAG, &os);
4571 	if (error)
4572 		fatal(B_FALSE, "dmu_objset_own(%s) = %d", snap2name, error);
4573 	error = dsl_dataset_promote(clone2name, NULL);
4574 	if (error == ENOSPC) {
4575 		dmu_objset_disown(os, B_TRUE, FTAG);
4576 		ztest_record_enospc(FTAG);
4577 		goto out;
4578 	}
4579 	if (error != EBUSY)
4580 		fatal(B_FALSE, "dsl_dataset_promote(%s), %d, not EBUSY",
4581 		    clone2name, error);
4582 	dmu_objset_disown(os, B_TRUE, FTAG);
4583 
4584 out:
4585 	ztest_dsl_dataset_cleanup(osname, id);
4586 
4587 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4588 
4589 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4590 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4591 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4592 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4593 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4594 }
4595 
4596 #undef OD_ARRAY_SIZE
4597 #define	OD_ARRAY_SIZE	4
4598 
4599 /*
4600  * Verify that dmu_object_{alloc,free} work as expected.
4601  */
4602 void
4603 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4604 {
4605 	ztest_od_t *od;
4606 	int batchsize;
4607 	int size;
4608 	int b;
4609 
4610 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4611 	od = umem_alloc(size, UMEM_NOFAIL);
4612 	batchsize = OD_ARRAY_SIZE;
4613 
4614 	for (b = 0; b < batchsize; b++)
4615 		ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4616 		    0, 0, 0);
4617 
4618 	/*
4619 	 * Destroy the previous batch of objects, create a new batch,
4620 	 * and do some I/O on the new objects.
4621 	 */
4622 	if (ztest_object_init(zd, od, size, B_TRUE) != 0)
4623 		return;
4624 
4625 	while (ztest_random(4 * batchsize) != 0)
4626 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
4627 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4628 
4629 	umem_free(od, size);
4630 }
4631 
4632 /*
4633  * Rewind the global allocator to verify object allocation backfilling.
4634  */
4635 void
4636 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
4637 {
4638 	(void) id;
4639 	objset_t *os = zd->zd_os;
4640 	int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
4641 	uint64_t object;
4642 
4643 	/*
4644 	 * Rewind the global allocator randomly back to a lower object number
4645 	 * to force backfilling and reclamation of recently freed dnodes.
4646 	 */
4647 	mutex_enter(&os->os_obj_lock);
4648 	object = ztest_random(os->os_obj_next_chunk);
4649 	os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
4650 	mutex_exit(&os->os_obj_lock);
4651 }
4652 
4653 #undef OD_ARRAY_SIZE
4654 #define	OD_ARRAY_SIZE	2
4655 
4656 /*
4657  * Verify that dmu_{read,write} work as expected.
4658  */
4659 void
4660 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4661 {
4662 	int size;
4663 	ztest_od_t *od;
4664 
4665 	objset_t *os = zd->zd_os;
4666 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4667 	od = umem_alloc(size, UMEM_NOFAIL);
4668 	dmu_tx_t *tx;
4669 	int freeit, error;
4670 	uint64_t i, n, s, txg;
4671 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4672 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4673 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4674 	uint64_t regions = 997;
4675 	uint64_t stride = 123456789ULL;
4676 	uint64_t width = 40;
4677 	int free_percent = 5;
4678 
4679 	/*
4680 	 * This test uses two objects, packobj and bigobj, that are always
4681 	 * updated together (i.e. in the same tx) so that their contents are
4682 	 * in sync and can be compared.  Their contents relate to each other
4683 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4684 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4685 	 * for any index n, there are three bufwads that should be identical:
4686 	 *
4687 	 *	packobj, at offset n * sizeof (bufwad_t)
4688 	 *	bigobj, at the head of the nth chunk
4689 	 *	bigobj, at the tail of the nth chunk
4690 	 *
4691 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
4692 	 * and it doesn't have any relation to the object blocksize.
4693 	 * The only requirement is that it can hold at least two bufwads.
4694 	 *
4695 	 * Normally, we write the bufwad to each of these locations.
4696 	 * However, free_percent of the time we instead write zeroes to
4697 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
4698 	 * bigobj to packobj, we can verify that the DMU is correctly
4699 	 * tracking which parts of an object are allocated and free,
4700 	 * and that the contents of the allocated blocks are correct.
4701 	 */
4702 
4703 	/*
4704 	 * Read the directory info.  If it's the first time, set things up.
4705 	 */
4706 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
4707 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4708 	    chunksize);
4709 
4710 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4711 		umem_free(od, size);
4712 		return;
4713 	}
4714 
4715 	bigobj = od[0].od_object;
4716 	packobj = od[1].od_object;
4717 	chunksize = od[0].od_gen;
4718 	ASSERT3U(chunksize, ==, od[1].od_gen);
4719 
4720 	/*
4721 	 * Prefetch a random chunk of the big object.
4722 	 * Our aim here is to get some async reads in flight
4723 	 * for blocks that we may free below; the DMU should
4724 	 * handle this race correctly.
4725 	 */
4726 	n = ztest_random(regions) * stride + ztest_random(width);
4727 	s = 1 + ztest_random(2 * width - 1);
4728 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4729 	    ZIO_PRIORITY_SYNC_READ);
4730 
4731 	/*
4732 	 * Pick a random index and compute the offsets into packobj and bigobj.
4733 	 */
4734 	n = ztest_random(regions) * stride + ztest_random(width);
4735 	s = 1 + ztest_random(width - 1);
4736 
4737 	packoff = n * sizeof (bufwad_t);
4738 	packsize = s * sizeof (bufwad_t);
4739 
4740 	bigoff = n * chunksize;
4741 	bigsize = s * chunksize;
4742 
4743 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4744 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4745 
4746 	/*
4747 	 * free_percent of the time, free a range of bigobj rather than
4748 	 * overwriting it.
4749 	 */
4750 	freeit = (ztest_random(100) < free_percent);
4751 
4752 	/*
4753 	 * Read the current contents of our objects.
4754 	 */
4755 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
4756 	    DMU_READ_PREFETCH);
4757 	ASSERT0(error);
4758 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4759 	    DMU_READ_PREFETCH);
4760 	ASSERT0(error);
4761 
4762 	/*
4763 	 * Get a tx for the mods to both packobj and bigobj.
4764 	 */
4765 	tx = dmu_tx_create(os);
4766 
4767 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
4768 
4769 	if (freeit)
4770 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4771 	else
4772 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4773 
4774 	/* This accounts for setting the checksum/compression. */
4775 	dmu_tx_hold_bonus(tx, bigobj);
4776 
4777 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4778 	if (txg == 0) {
4779 		umem_free(packbuf, packsize);
4780 		umem_free(bigbuf, bigsize);
4781 		umem_free(od, size);
4782 		return;
4783 	}
4784 
4785 	enum zio_checksum cksum;
4786 	do {
4787 		cksum = (enum zio_checksum)
4788 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4789 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4790 	dmu_object_set_checksum(os, bigobj, cksum, tx);
4791 
4792 	enum zio_compress comp;
4793 	do {
4794 		comp = (enum zio_compress)
4795 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4796 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4797 	dmu_object_set_compress(os, bigobj, comp, tx);
4798 
4799 	/*
4800 	 * For each index from n to n + s, verify that the existing bufwad
4801 	 * in packobj matches the bufwads at the head and tail of the
4802 	 * corresponding chunk in bigobj.  Then update all three bufwads
4803 	 * with the new values we want to write out.
4804 	 */
4805 	for (i = 0; i < s; i++) {
4806 		/* LINTED */
4807 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4808 		/* LINTED */
4809 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4810 		/* LINTED */
4811 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4812 
4813 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
4814 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
4815 
4816 		if (pack->bw_txg > txg)
4817 			fatal(B_FALSE,
4818 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
4819 			    pack->bw_txg, txg);
4820 
4821 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4822 			fatal(B_FALSE, "wrong index: "
4823 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
4824 			    pack->bw_index, n, i);
4825 
4826 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4827 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
4828 			    pack, bigH);
4829 
4830 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4831 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
4832 			    pack, bigT);
4833 
4834 		if (freeit) {
4835 			memset(pack, 0, sizeof (bufwad_t));
4836 		} else {
4837 			pack->bw_index = n + i;
4838 			pack->bw_txg = txg;
4839 			pack->bw_data = 1 + ztest_random(-2ULL);
4840 		}
4841 		*bigH = *pack;
4842 		*bigT = *pack;
4843 	}
4844 
4845 	/*
4846 	 * We've verified all the old bufwads, and made new ones.
4847 	 * Now write them out.
4848 	 */
4849 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4850 
4851 	if (freeit) {
4852 		if (ztest_opts.zo_verbose >= 7) {
4853 			(void) printf("freeing offset %"PRIx64" size %"PRIx64""
4854 			    " txg %"PRIx64"\n",
4855 			    bigoff, bigsize, txg);
4856 		}
4857 		VERIFY0(dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4858 	} else {
4859 		if (ztest_opts.zo_verbose >= 7) {
4860 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
4861 			    " txg %"PRIx64"\n",
4862 			    bigoff, bigsize, txg);
4863 		}
4864 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4865 	}
4866 
4867 	dmu_tx_commit(tx);
4868 
4869 	/*
4870 	 * Sanity check the stuff we just wrote.
4871 	 */
4872 	{
4873 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4874 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4875 
4876 		VERIFY0(dmu_read(os, packobj, packoff,
4877 		    packsize, packcheck, DMU_READ_PREFETCH));
4878 		VERIFY0(dmu_read(os, bigobj, bigoff,
4879 		    bigsize, bigcheck, DMU_READ_PREFETCH));
4880 
4881 		ASSERT0(memcmp(packbuf, packcheck, packsize));
4882 		ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
4883 
4884 		umem_free(packcheck, packsize);
4885 		umem_free(bigcheck, bigsize);
4886 	}
4887 
4888 	umem_free(packbuf, packsize);
4889 	umem_free(bigbuf, bigsize);
4890 	umem_free(od, size);
4891 }
4892 
4893 static void
4894 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4895     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4896 {
4897 	uint64_t i;
4898 	bufwad_t *pack;
4899 	bufwad_t *bigH;
4900 	bufwad_t *bigT;
4901 
4902 	/*
4903 	 * For each index from n to n + s, verify that the existing bufwad
4904 	 * in packobj matches the bufwads at the head and tail of the
4905 	 * corresponding chunk in bigobj.  Then update all three bufwads
4906 	 * with the new values we want to write out.
4907 	 */
4908 	for (i = 0; i < s; i++) {
4909 		/* LINTED */
4910 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4911 		/* LINTED */
4912 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4913 		/* LINTED */
4914 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4915 
4916 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
4917 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
4918 
4919 		if (pack->bw_txg > txg)
4920 			fatal(B_FALSE,
4921 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
4922 			    pack->bw_txg, txg);
4923 
4924 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4925 			fatal(B_FALSE, "wrong index: "
4926 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
4927 			    pack->bw_index, n, i);
4928 
4929 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4930 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
4931 			    pack, bigH);
4932 
4933 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4934 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
4935 			    pack, bigT);
4936 
4937 		pack->bw_index = n + i;
4938 		pack->bw_txg = txg;
4939 		pack->bw_data = 1 + ztest_random(-2ULL);
4940 
4941 		*bigH = *pack;
4942 		*bigT = *pack;
4943 	}
4944 }
4945 
4946 #undef OD_ARRAY_SIZE
4947 #define	OD_ARRAY_SIZE	2
4948 
4949 void
4950 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4951 {
4952 	objset_t *os = zd->zd_os;
4953 	ztest_od_t *od;
4954 	dmu_tx_t *tx;
4955 	uint64_t i;
4956 	int error;
4957 	int size;
4958 	uint64_t n, s, txg;
4959 	bufwad_t *packbuf, *bigbuf;
4960 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4961 	uint64_t blocksize = ztest_random_blocksize();
4962 	uint64_t chunksize = blocksize;
4963 	uint64_t regions = 997;
4964 	uint64_t stride = 123456789ULL;
4965 	uint64_t width = 9;
4966 	dmu_buf_t *bonus_db;
4967 	arc_buf_t **bigbuf_arcbufs;
4968 	dmu_object_info_t doi;
4969 
4970 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4971 	od = umem_alloc(size, UMEM_NOFAIL);
4972 
4973 	/*
4974 	 * This test uses two objects, packobj and bigobj, that are always
4975 	 * updated together (i.e. in the same tx) so that their contents are
4976 	 * in sync and can be compared.  Their contents relate to each other
4977 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4978 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4979 	 * for any index n, there are three bufwads that should be identical:
4980 	 *
4981 	 *	packobj, at offset n * sizeof (bufwad_t)
4982 	 *	bigobj, at the head of the nth chunk
4983 	 *	bigobj, at the tail of the nth chunk
4984 	 *
4985 	 * The chunk size is set equal to bigobj block size so that
4986 	 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
4987 	 */
4988 
4989 	/*
4990 	 * Read the directory info.  If it's the first time, set things up.
4991 	 */
4992 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
4993 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4994 	    chunksize);
4995 
4996 
4997 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4998 		umem_free(od, size);
4999 		return;
5000 	}
5001 
5002 	bigobj = od[0].od_object;
5003 	packobj = od[1].od_object;
5004 	blocksize = od[0].od_blocksize;
5005 	chunksize = blocksize;
5006 	ASSERT3U(chunksize, ==, od[1].od_gen);
5007 
5008 	VERIFY0(dmu_object_info(os, bigobj, &doi));
5009 	VERIFY(ISP2(doi.doi_data_block_size));
5010 	VERIFY3U(chunksize, ==, doi.doi_data_block_size);
5011 	VERIFY3U(chunksize, >=, 2 * sizeof (bufwad_t));
5012 
5013 	/*
5014 	 * Pick a random index and compute the offsets into packobj and bigobj.
5015 	 */
5016 	n = ztest_random(regions) * stride + ztest_random(width);
5017 	s = 1 + ztest_random(width - 1);
5018 
5019 	packoff = n * sizeof (bufwad_t);
5020 	packsize = s * sizeof (bufwad_t);
5021 
5022 	bigoff = n * chunksize;
5023 	bigsize = s * chunksize;
5024 
5025 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
5026 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
5027 
5028 	VERIFY0(dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
5029 
5030 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
5031 
5032 	/*
5033 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
5034 	 * Iteration 1 test zcopy to already referenced dbufs.
5035 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
5036 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
5037 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
5038 	 * Iteration 5 test zcopy when it can't be done.
5039 	 * Iteration 6 one more zcopy write.
5040 	 */
5041 	for (i = 0; i < 7; i++) {
5042 		uint64_t j;
5043 		uint64_t off;
5044 
5045 		/*
5046 		 * In iteration 5 (i == 5) use arcbufs
5047 		 * that don't match bigobj blksz to test
5048 		 * dmu_assign_arcbuf_by_dbuf() when it can't directly
5049 		 * assign an arcbuf to a dbuf.
5050 		 */
5051 		for (j = 0; j < s; j++) {
5052 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5053 				bigbuf_arcbufs[j] =
5054 				    dmu_request_arcbuf(bonus_db, chunksize);
5055 			} else {
5056 				bigbuf_arcbufs[2 * j] =
5057 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5058 				bigbuf_arcbufs[2 * j + 1] =
5059 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5060 			}
5061 		}
5062 
5063 		/*
5064 		 * Get a tx for the mods to both packobj and bigobj.
5065 		 */
5066 		tx = dmu_tx_create(os);
5067 
5068 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
5069 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
5070 
5071 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5072 		if (txg == 0) {
5073 			umem_free(packbuf, packsize);
5074 			umem_free(bigbuf, bigsize);
5075 			for (j = 0; j < s; j++) {
5076 				if (i != 5 ||
5077 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
5078 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
5079 				} else {
5080 					dmu_return_arcbuf(
5081 					    bigbuf_arcbufs[2 * j]);
5082 					dmu_return_arcbuf(
5083 					    bigbuf_arcbufs[2 * j + 1]);
5084 				}
5085 			}
5086 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5087 			umem_free(od, size);
5088 			dmu_buf_rele(bonus_db, FTAG);
5089 			return;
5090 		}
5091 
5092 		/*
5093 		 * 50% of the time don't read objects in the 1st iteration to
5094 		 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
5095 		 * no existing dbufs for the specified offsets.
5096 		 */
5097 		if (i != 0 || ztest_random(2) != 0) {
5098 			error = dmu_read(os, packobj, packoff,
5099 			    packsize, packbuf, DMU_READ_PREFETCH);
5100 			ASSERT0(error);
5101 			error = dmu_read(os, bigobj, bigoff, bigsize,
5102 			    bigbuf, DMU_READ_PREFETCH);
5103 			ASSERT0(error);
5104 		}
5105 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
5106 		    n, chunksize, txg);
5107 
5108 		/*
5109 		 * We've verified all the old bufwads, and made new ones.
5110 		 * Now write them out.
5111 		 */
5112 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
5113 		if (ztest_opts.zo_verbose >= 7) {
5114 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
5115 			    " txg %"PRIx64"\n",
5116 			    bigoff, bigsize, txg);
5117 		}
5118 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
5119 			dmu_buf_t *dbt;
5120 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5121 				memcpy(bigbuf_arcbufs[j]->b_data,
5122 				    (caddr_t)bigbuf + (off - bigoff),
5123 				    chunksize);
5124 			} else {
5125 				memcpy(bigbuf_arcbufs[2 * j]->b_data,
5126 				    (caddr_t)bigbuf + (off - bigoff),
5127 				    chunksize / 2);
5128 				memcpy(bigbuf_arcbufs[2 * j + 1]->b_data,
5129 				    (caddr_t)bigbuf + (off - bigoff) +
5130 				    chunksize / 2,
5131 				    chunksize / 2);
5132 			}
5133 
5134 			if (i == 1) {
5135 				VERIFY(dmu_buf_hold(os, bigobj, off,
5136 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
5137 			}
5138 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5139 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5140 				    off, bigbuf_arcbufs[j], tx));
5141 			} else {
5142 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5143 				    off, bigbuf_arcbufs[2 * j], tx));
5144 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5145 				    off + chunksize / 2,
5146 				    bigbuf_arcbufs[2 * j + 1], tx));
5147 			}
5148 			if (i == 1) {
5149 				dmu_buf_rele(dbt, FTAG);
5150 			}
5151 		}
5152 		dmu_tx_commit(tx);
5153 
5154 		/*
5155 		 * Sanity check the stuff we just wrote.
5156 		 */
5157 		{
5158 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
5159 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
5160 
5161 			VERIFY0(dmu_read(os, packobj, packoff,
5162 			    packsize, packcheck, DMU_READ_PREFETCH));
5163 			VERIFY0(dmu_read(os, bigobj, bigoff,
5164 			    bigsize, bigcheck, DMU_READ_PREFETCH));
5165 
5166 			ASSERT0(memcmp(packbuf, packcheck, packsize));
5167 			ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
5168 
5169 			umem_free(packcheck, packsize);
5170 			umem_free(bigcheck, bigsize);
5171 		}
5172 		if (i == 2) {
5173 			txg_wait_open(dmu_objset_pool(os), 0, B_TRUE);
5174 		} else if (i == 3) {
5175 			txg_wait_synced(dmu_objset_pool(os), 0);
5176 		}
5177 	}
5178 
5179 	dmu_buf_rele(bonus_db, FTAG);
5180 	umem_free(packbuf, packsize);
5181 	umem_free(bigbuf, bigsize);
5182 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5183 	umem_free(od, size);
5184 }
5185 
5186 void
5187 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
5188 {
5189 	(void) id;
5190 	ztest_od_t *od;
5191 
5192 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5193 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
5194 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5195 
5196 	/*
5197 	 * Have multiple threads write to large offsets in an object
5198 	 * to verify that parallel writes to an object -- even to the
5199 	 * same blocks within the object -- doesn't cause any trouble.
5200 	 */
5201 	ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5202 
5203 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
5204 		return;
5205 
5206 	while (ztest_random(10) != 0)
5207 		ztest_io(zd, od->od_object, offset);
5208 
5209 	umem_free(od, sizeof (ztest_od_t));
5210 }
5211 
5212 void
5213 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
5214 {
5215 	ztest_od_t *od;
5216 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
5217 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5218 	uint64_t count = ztest_random(20) + 1;
5219 	uint64_t blocksize = ztest_random_blocksize();
5220 	void *data;
5221 
5222 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5223 
5224 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5225 
5226 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5227 	    !ztest_random(2)) != 0) {
5228 		umem_free(od, sizeof (ztest_od_t));
5229 		return;
5230 	}
5231 
5232 	if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
5233 		umem_free(od, sizeof (ztest_od_t));
5234 		return;
5235 	}
5236 
5237 	ztest_prealloc(zd, od->od_object, offset, count * blocksize);
5238 
5239 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
5240 
5241 	while (ztest_random(count) != 0) {
5242 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
5243 		if (ztest_write(zd, od->od_object, randoff, blocksize,
5244 		    data) != 0)
5245 			break;
5246 		while (ztest_random(4) != 0)
5247 			ztest_io(zd, od->od_object, randoff);
5248 	}
5249 
5250 	umem_free(data, blocksize);
5251 	umem_free(od, sizeof (ztest_od_t));
5252 }
5253 
5254 /*
5255  * Verify that zap_{create,destroy,add,remove,update} work as expected.
5256  */
5257 #define	ZTEST_ZAP_MIN_INTS	1
5258 #define	ZTEST_ZAP_MAX_INTS	4
5259 #define	ZTEST_ZAP_MAX_PROPS	1000
5260 
5261 void
5262 ztest_zap(ztest_ds_t *zd, uint64_t id)
5263 {
5264 	objset_t *os = zd->zd_os;
5265 	ztest_od_t *od;
5266 	uint64_t object;
5267 	uint64_t txg, last_txg;
5268 	uint64_t value[ZTEST_ZAP_MAX_INTS];
5269 	uint64_t zl_ints, zl_intsize, prop;
5270 	int i, ints;
5271 	dmu_tx_t *tx;
5272 	char propname[100], txgname[100];
5273 	int error;
5274 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
5275 
5276 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5277 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5278 
5279 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5280 	    !ztest_random(2)) != 0)
5281 		goto out;
5282 
5283 	object = od->od_object;
5284 
5285 	/*
5286 	 * Generate a known hash collision, and verify that
5287 	 * we can lookup and remove both entries.
5288 	 */
5289 	tx = dmu_tx_create(os);
5290 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5291 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5292 	if (txg == 0)
5293 		goto out;
5294 	for (i = 0; i < 2; i++) {
5295 		value[i] = i;
5296 		VERIFY0(zap_add(os, object, hc[i], sizeof (uint64_t),
5297 		    1, &value[i], tx));
5298 	}
5299 	for (i = 0; i < 2; i++) {
5300 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
5301 		    sizeof (uint64_t), 1, &value[i], tx));
5302 		VERIFY0(
5303 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
5304 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5305 		ASSERT3U(zl_ints, ==, 1);
5306 	}
5307 	for (i = 0; i < 2; i++) {
5308 		VERIFY0(zap_remove(os, object, hc[i], tx));
5309 	}
5310 	dmu_tx_commit(tx);
5311 
5312 	/*
5313 	 * Generate a bunch of random entries.
5314 	 */
5315 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
5316 
5317 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5318 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5319 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5320 	memset(value, 0, sizeof (value));
5321 	last_txg = 0;
5322 
5323 	/*
5324 	 * If these zap entries already exist, validate their contents.
5325 	 */
5326 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5327 	if (error == 0) {
5328 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5329 		ASSERT3U(zl_ints, ==, 1);
5330 
5331 		VERIFY0(zap_lookup(os, object, txgname, zl_intsize,
5332 		    zl_ints, &last_txg));
5333 
5334 		VERIFY0(zap_length(os, object, propname, &zl_intsize,
5335 		    &zl_ints));
5336 
5337 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5338 		ASSERT3U(zl_ints, ==, ints);
5339 
5340 		VERIFY0(zap_lookup(os, object, propname, zl_intsize,
5341 		    zl_ints, value));
5342 
5343 		for (i = 0; i < ints; i++) {
5344 			ASSERT3U(value[i], ==, last_txg + object + i);
5345 		}
5346 	} else {
5347 		ASSERT3U(error, ==, ENOENT);
5348 	}
5349 
5350 	/*
5351 	 * Atomically update two entries in our zap object.
5352 	 * The first is named txg_%llu, and contains the txg
5353 	 * in which the property was last updated.  The second
5354 	 * is named prop_%llu, and the nth element of its value
5355 	 * should be txg + object + n.
5356 	 */
5357 	tx = dmu_tx_create(os);
5358 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5359 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5360 	if (txg == 0)
5361 		goto out;
5362 
5363 	if (last_txg > txg)
5364 		fatal(B_FALSE, "zap future leak: old %"PRIu64" new %"PRIu64"",
5365 		    last_txg, txg);
5366 
5367 	for (i = 0; i < ints; i++)
5368 		value[i] = txg + object + i;
5369 
5370 	VERIFY0(zap_update(os, object, txgname, sizeof (uint64_t),
5371 	    1, &txg, tx));
5372 	VERIFY0(zap_update(os, object, propname, sizeof (uint64_t),
5373 	    ints, value, tx));
5374 
5375 	dmu_tx_commit(tx);
5376 
5377 	/*
5378 	 * Remove a random pair of entries.
5379 	 */
5380 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5381 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5382 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5383 
5384 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5385 
5386 	if (error == ENOENT)
5387 		goto out;
5388 
5389 	ASSERT0(error);
5390 
5391 	tx = dmu_tx_create(os);
5392 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5393 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5394 	if (txg == 0)
5395 		goto out;
5396 	VERIFY0(zap_remove(os, object, txgname, tx));
5397 	VERIFY0(zap_remove(os, object, propname, tx));
5398 	dmu_tx_commit(tx);
5399 out:
5400 	umem_free(od, sizeof (ztest_od_t));
5401 }
5402 
5403 /*
5404  * Test case to test the upgrading of a microzap to fatzap.
5405  */
5406 void
5407 ztest_fzap(ztest_ds_t *zd, uint64_t id)
5408 {
5409 	objset_t *os = zd->zd_os;
5410 	ztest_od_t *od;
5411 	uint64_t object, txg, value;
5412 
5413 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5414 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5415 
5416 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5417 	    !ztest_random(2)) != 0)
5418 		goto out;
5419 	object = od->od_object;
5420 
5421 	/*
5422 	 * Add entries to this ZAP and make sure it spills over
5423 	 * and gets upgraded to a fatzap. Also, since we are adding
5424 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
5425 	 */
5426 	for (value = 0; value < 2050; value++) {
5427 		char name[ZFS_MAX_DATASET_NAME_LEN];
5428 		dmu_tx_t *tx;
5429 		int error;
5430 
5431 		(void) snprintf(name, sizeof (name), "fzap-%"PRIu64"-%"PRIu64"",
5432 		    id, value);
5433 
5434 		tx = dmu_tx_create(os);
5435 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
5436 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5437 		if (txg == 0)
5438 			goto out;
5439 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
5440 		    &value, tx);
5441 		ASSERT(error == 0 || error == EEXIST);
5442 		dmu_tx_commit(tx);
5443 	}
5444 out:
5445 	umem_free(od, sizeof (ztest_od_t));
5446 }
5447 
5448 void
5449 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
5450 {
5451 	(void) id;
5452 	objset_t *os = zd->zd_os;
5453 	ztest_od_t *od;
5454 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
5455 	dmu_tx_t *tx;
5456 	int i, namelen, error;
5457 	int micro = ztest_random(2);
5458 	char name[20], string_value[20];
5459 	void *data;
5460 
5461 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5462 	ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
5463 
5464 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5465 		umem_free(od, sizeof (ztest_od_t));
5466 		return;
5467 	}
5468 
5469 	object = od->od_object;
5470 
5471 	/*
5472 	 * Generate a random name of the form 'xxx.....' where each
5473 	 * x is a random printable character and the dots are dots.
5474 	 * There are 94 such characters, and the name length goes from
5475 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
5476 	 */
5477 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
5478 
5479 	for (i = 0; i < 3; i++)
5480 		name[i] = '!' + ztest_random('~' - '!' + 1);
5481 	for (; i < namelen - 1; i++)
5482 		name[i] = '.';
5483 	name[i] = '\0';
5484 
5485 	if ((namelen & 1) || micro) {
5486 		wsize = sizeof (txg);
5487 		wc = 1;
5488 		data = &txg;
5489 	} else {
5490 		wsize = 1;
5491 		wc = namelen;
5492 		data = string_value;
5493 	}
5494 
5495 	count = -1ULL;
5496 	VERIFY0(zap_count(os, object, &count));
5497 	ASSERT3S(count, !=, -1ULL);
5498 
5499 	/*
5500 	 * Select an operation: length, lookup, add, update, remove.
5501 	 */
5502 	i = ztest_random(5);
5503 
5504 	if (i >= 2) {
5505 		tx = dmu_tx_create(os);
5506 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5507 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5508 		if (txg == 0) {
5509 			umem_free(od, sizeof (ztest_od_t));
5510 			return;
5511 		}
5512 		memcpy(string_value, name, namelen);
5513 	} else {
5514 		tx = NULL;
5515 		txg = 0;
5516 		memset(string_value, 0, namelen);
5517 	}
5518 
5519 	switch (i) {
5520 
5521 	case 0:
5522 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
5523 		if (error == 0) {
5524 			ASSERT3U(wsize, ==, zl_wsize);
5525 			ASSERT3U(wc, ==, zl_wc);
5526 		} else {
5527 			ASSERT3U(error, ==, ENOENT);
5528 		}
5529 		break;
5530 
5531 	case 1:
5532 		error = zap_lookup(os, object, name, wsize, wc, data);
5533 		if (error == 0) {
5534 			if (data == string_value &&
5535 			    memcmp(name, data, namelen) != 0)
5536 				fatal(B_FALSE, "name '%s' != val '%s' len %d",
5537 				    name, (char *)data, namelen);
5538 		} else {
5539 			ASSERT3U(error, ==, ENOENT);
5540 		}
5541 		break;
5542 
5543 	case 2:
5544 		error = zap_add(os, object, name, wsize, wc, data, tx);
5545 		ASSERT(error == 0 || error == EEXIST);
5546 		break;
5547 
5548 	case 3:
5549 		VERIFY0(zap_update(os, object, name, wsize, wc, data, tx));
5550 		break;
5551 
5552 	case 4:
5553 		error = zap_remove(os, object, name, tx);
5554 		ASSERT(error == 0 || error == ENOENT);
5555 		break;
5556 	}
5557 
5558 	if (tx != NULL)
5559 		dmu_tx_commit(tx);
5560 
5561 	umem_free(od, sizeof (ztest_od_t));
5562 }
5563 
5564 /*
5565  * Commit callback data.
5566  */
5567 typedef struct ztest_cb_data {
5568 	list_node_t		zcd_node;
5569 	uint64_t		zcd_txg;
5570 	int			zcd_expected_err;
5571 	boolean_t		zcd_added;
5572 	boolean_t		zcd_called;
5573 	spa_t			*zcd_spa;
5574 } ztest_cb_data_t;
5575 
5576 /* This is the actual commit callback function */
5577 static void
5578 ztest_commit_callback(void *arg, int error)
5579 {
5580 	ztest_cb_data_t *data = arg;
5581 	uint64_t synced_txg;
5582 
5583 	VERIFY3P(data, !=, NULL);
5584 	VERIFY3S(data->zcd_expected_err, ==, error);
5585 	VERIFY(!data->zcd_called);
5586 
5587 	synced_txg = spa_last_synced_txg(data->zcd_spa);
5588 	if (data->zcd_txg > synced_txg)
5589 		fatal(B_FALSE,
5590 		    "commit callback of txg %"PRIu64" called prematurely, "
5591 		    "last synced txg = %"PRIu64"\n",
5592 		    data->zcd_txg, synced_txg);
5593 
5594 	data->zcd_called = B_TRUE;
5595 
5596 	if (error == ECANCELED) {
5597 		ASSERT0(data->zcd_txg);
5598 		ASSERT(!data->zcd_added);
5599 
5600 		/*
5601 		 * The private callback data should be destroyed here, but
5602 		 * since we are going to check the zcd_called field after
5603 		 * dmu_tx_abort(), we will destroy it there.
5604 		 */
5605 		return;
5606 	}
5607 
5608 	ASSERT(data->zcd_added);
5609 	ASSERT3U(data->zcd_txg, !=, 0);
5610 
5611 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5612 
5613 	/* See if this cb was called more quickly */
5614 	if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
5615 		zc_min_txg_delay = synced_txg - data->zcd_txg;
5616 
5617 	/* Remove our callback from the list */
5618 	list_remove(&zcl.zcl_callbacks, data);
5619 
5620 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5621 
5622 	umem_free(data, sizeof (ztest_cb_data_t));
5623 }
5624 
5625 /* Allocate and initialize callback data structure */
5626 static ztest_cb_data_t *
5627 ztest_create_cb_data(objset_t *os, uint64_t txg)
5628 {
5629 	ztest_cb_data_t *cb_data;
5630 
5631 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
5632 
5633 	cb_data->zcd_txg = txg;
5634 	cb_data->zcd_spa = dmu_objset_spa(os);
5635 	list_link_init(&cb_data->zcd_node);
5636 
5637 	return (cb_data);
5638 }
5639 
5640 /*
5641  * Commit callback test.
5642  */
5643 void
5644 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5645 {
5646 	objset_t *os = zd->zd_os;
5647 	ztest_od_t *od;
5648 	dmu_tx_t *tx;
5649 	ztest_cb_data_t *cb_data[3], *tmp_cb;
5650 	uint64_t old_txg, txg;
5651 	int i, error = 0;
5652 
5653 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5654 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5655 
5656 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5657 		umem_free(od, sizeof (ztest_od_t));
5658 		return;
5659 	}
5660 
5661 	tx = dmu_tx_create(os);
5662 
5663 	cb_data[0] = ztest_create_cb_data(os, 0);
5664 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5665 
5666 	dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
5667 
5668 	/* Every once in a while, abort the transaction on purpose */
5669 	if (ztest_random(100) == 0)
5670 		error = -1;
5671 
5672 	if (!error)
5673 		error = dmu_tx_assign(tx, TXG_NOWAIT);
5674 
5675 	txg = error ? 0 : dmu_tx_get_txg(tx);
5676 
5677 	cb_data[0]->zcd_txg = txg;
5678 	cb_data[1] = ztest_create_cb_data(os, txg);
5679 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5680 
5681 	if (error) {
5682 		/*
5683 		 * It's not a strict requirement to call the registered
5684 		 * callbacks from inside dmu_tx_abort(), but that's what
5685 		 * it's supposed to happen in the current implementation
5686 		 * so we will check for that.
5687 		 */
5688 		for (i = 0; i < 2; i++) {
5689 			cb_data[i]->zcd_expected_err = ECANCELED;
5690 			VERIFY(!cb_data[i]->zcd_called);
5691 		}
5692 
5693 		dmu_tx_abort(tx);
5694 
5695 		for (i = 0; i < 2; i++) {
5696 			VERIFY(cb_data[i]->zcd_called);
5697 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5698 		}
5699 
5700 		umem_free(od, sizeof (ztest_od_t));
5701 		return;
5702 	}
5703 
5704 	cb_data[2] = ztest_create_cb_data(os, txg);
5705 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5706 
5707 	/*
5708 	 * Read existing data to make sure there isn't a future leak.
5709 	 */
5710 	VERIFY0(dmu_read(os, od->od_object, 0, sizeof (uint64_t),
5711 	    &old_txg, DMU_READ_PREFETCH));
5712 
5713 	if (old_txg > txg)
5714 		fatal(B_FALSE,
5715 		    "future leak: got %"PRIu64", open txg is %"PRIu64"",
5716 		    old_txg, txg);
5717 
5718 	dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
5719 
5720 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5721 
5722 	/*
5723 	 * Since commit callbacks don't have any ordering requirement and since
5724 	 * it is theoretically possible for a commit callback to be called
5725 	 * after an arbitrary amount of time has elapsed since its txg has been
5726 	 * synced, it is difficult to reliably determine whether a commit
5727 	 * callback hasn't been called due to high load or due to a flawed
5728 	 * implementation.
5729 	 *
5730 	 * In practice, we will assume that if after a certain number of txgs a
5731 	 * commit callback hasn't been called, then most likely there's an
5732 	 * implementation bug..
5733 	 */
5734 	tmp_cb = list_head(&zcl.zcl_callbacks);
5735 	if (tmp_cb != NULL &&
5736 	    tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
5737 		fatal(B_FALSE,
5738 		    "Commit callback threshold exceeded, "
5739 		    "oldest txg: %"PRIu64", open txg: %"PRIu64"\n",
5740 		    tmp_cb->zcd_txg, txg);
5741 	}
5742 
5743 	/*
5744 	 * Let's find the place to insert our callbacks.
5745 	 *
5746 	 * Even though the list is ordered by txg, it is possible for the
5747 	 * insertion point to not be the end because our txg may already be
5748 	 * quiescing at this point and other callbacks in the open txg
5749 	 * (from other objsets) may have sneaked in.
5750 	 */
5751 	tmp_cb = list_tail(&zcl.zcl_callbacks);
5752 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5753 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5754 
5755 	/* Add the 3 callbacks to the list */
5756 	for (i = 0; i < 3; i++) {
5757 		if (tmp_cb == NULL)
5758 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5759 		else
5760 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5761 			    cb_data[i]);
5762 
5763 		cb_data[i]->zcd_added = B_TRUE;
5764 		VERIFY(!cb_data[i]->zcd_called);
5765 
5766 		tmp_cb = cb_data[i];
5767 	}
5768 
5769 	zc_cb_counter += 3;
5770 
5771 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5772 
5773 	dmu_tx_commit(tx);
5774 
5775 	umem_free(od, sizeof (ztest_od_t));
5776 }
5777 
5778 /*
5779  * Visit each object in the dataset. Verify that its properties
5780  * are consistent what was stored in the block tag when it was created,
5781  * and that its unused bonus buffer space has not been overwritten.
5782  */
5783 void
5784 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5785 {
5786 	(void) id;
5787 	objset_t *os = zd->zd_os;
5788 	uint64_t obj;
5789 	int err = 0;
5790 
5791 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5792 		ztest_block_tag_t *bt = NULL;
5793 		dmu_object_info_t doi;
5794 		dmu_buf_t *db;
5795 
5796 		ztest_object_lock(zd, obj, RL_READER);
5797 		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) {
5798 			ztest_object_unlock(zd, obj);
5799 			continue;
5800 		}
5801 
5802 		dmu_object_info_from_db(db, &doi);
5803 		if (doi.doi_bonus_size >= sizeof (*bt))
5804 			bt = ztest_bt_bonus(db);
5805 
5806 		if (bt && bt->bt_magic == BT_MAGIC) {
5807 			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5808 			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
5809 			    bt->bt_crtxg);
5810 			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5811 		}
5812 
5813 		dmu_buf_rele(db, FTAG);
5814 		ztest_object_unlock(zd, obj);
5815 	}
5816 }
5817 
5818 void
5819 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5820 {
5821 	(void) id;
5822 	zfs_prop_t proplist[] = {
5823 		ZFS_PROP_CHECKSUM,
5824 		ZFS_PROP_COMPRESSION,
5825 		ZFS_PROP_COPIES,
5826 		ZFS_PROP_DEDUP
5827 	};
5828 
5829 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5830 
5831 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5832 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5833 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5834 
5835 	VERIFY0(ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5836 	    ztest_random_blocksize(), (int)ztest_random(2)));
5837 
5838 	(void) pthread_rwlock_unlock(&ztest_name_lock);
5839 }
5840 
5841 void
5842 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5843 {
5844 	(void) zd, (void) id;
5845 	nvlist_t *props = NULL;
5846 
5847 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5848 
5849 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_AUTOTRIM, ztest_random(2));
5850 
5851 	VERIFY0(spa_prop_get(ztest_spa, &props));
5852 
5853 	if (ztest_opts.zo_verbose >= 6)
5854 		dump_nvlist(props, 4);
5855 
5856 	fnvlist_free(props);
5857 
5858 	(void) pthread_rwlock_unlock(&ztest_name_lock);
5859 }
5860 
5861 static int
5862 user_release_one(const char *snapname, const char *holdname)
5863 {
5864 	nvlist_t *snaps, *holds;
5865 	int error;
5866 
5867 	snaps = fnvlist_alloc();
5868 	holds = fnvlist_alloc();
5869 	fnvlist_add_boolean(holds, holdname);
5870 	fnvlist_add_nvlist(snaps, snapname, holds);
5871 	fnvlist_free(holds);
5872 	error = dsl_dataset_user_release(snaps, NULL);
5873 	fnvlist_free(snaps);
5874 	return (error);
5875 }
5876 
5877 /*
5878  * Test snapshot hold/release and deferred destroy.
5879  */
5880 void
5881 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5882 {
5883 	int error;
5884 	objset_t *os = zd->zd_os;
5885 	objset_t *origin;
5886 	char snapname[100];
5887 	char fullname[100];
5888 	char clonename[100];
5889 	char tag[100];
5890 	char osname[ZFS_MAX_DATASET_NAME_LEN];
5891 	nvlist_t *holds;
5892 
5893 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5894 
5895 	dmu_objset_name(os, osname);
5896 
5897 	(void) snprintf(snapname, sizeof (snapname), "sh1_%"PRIu64"", id);
5898 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5899 	(void) snprintf(clonename, sizeof (clonename), "%s/ch1_%"PRIu64"",
5900 	    osname, id);
5901 	(void) snprintf(tag, sizeof (tag), "tag_%"PRIu64"", id);
5902 
5903 	/*
5904 	 * Clean up from any previous run.
5905 	 */
5906 	error = dsl_destroy_head(clonename);
5907 	if (error != ENOENT)
5908 		ASSERT0(error);
5909 	error = user_release_one(fullname, tag);
5910 	if (error != ESRCH && error != ENOENT)
5911 		ASSERT0(error);
5912 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5913 	if (error != ENOENT)
5914 		ASSERT0(error);
5915 
5916 	/*
5917 	 * Create snapshot, clone it, mark snap for deferred destroy,
5918 	 * destroy clone, verify snap was also destroyed.
5919 	 */
5920 	error = dmu_objset_snapshot_one(osname, snapname);
5921 	if (error) {
5922 		if (error == ENOSPC) {
5923 			ztest_record_enospc("dmu_objset_snapshot");
5924 			goto out;
5925 		}
5926 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
5927 	}
5928 
5929 	error = dmu_objset_clone(clonename, fullname);
5930 	if (error) {
5931 		if (error == ENOSPC) {
5932 			ztest_record_enospc("dmu_objset_clone");
5933 			goto out;
5934 		}
5935 		fatal(B_FALSE, "dmu_objset_clone(%s) = %d", clonename, error);
5936 	}
5937 
5938 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5939 	if (error) {
5940 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5941 		    fullname, error);
5942 	}
5943 
5944 	error = dsl_destroy_head(clonename);
5945 	if (error)
5946 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clonename, error);
5947 
5948 	error = dmu_objset_hold(fullname, FTAG, &origin);
5949 	if (error != ENOENT)
5950 		fatal(B_FALSE, "dmu_objset_hold(%s) = %d", fullname, error);
5951 
5952 	/*
5953 	 * Create snapshot, add temporary hold, verify that we can't
5954 	 * destroy a held snapshot, mark for deferred destroy,
5955 	 * release hold, verify snapshot was destroyed.
5956 	 */
5957 	error = dmu_objset_snapshot_one(osname, snapname);
5958 	if (error) {
5959 		if (error == ENOSPC) {
5960 			ztest_record_enospc("dmu_objset_snapshot");
5961 			goto out;
5962 		}
5963 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
5964 	}
5965 
5966 	holds = fnvlist_alloc();
5967 	fnvlist_add_string(holds, fullname, tag);
5968 	error = dsl_dataset_user_hold(holds, 0, NULL);
5969 	fnvlist_free(holds);
5970 
5971 	if (error == ENOSPC) {
5972 		ztest_record_enospc("dsl_dataset_user_hold");
5973 		goto out;
5974 	} else if (error) {
5975 		fatal(B_FALSE, "dsl_dataset_user_hold(%s, %s) = %u",
5976 		    fullname, tag, error);
5977 	}
5978 
5979 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5980 	if (error != EBUSY) {
5981 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5982 		    fullname, error);
5983 	}
5984 
5985 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5986 	if (error) {
5987 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5988 		    fullname, error);
5989 	}
5990 
5991 	error = user_release_one(fullname, tag);
5992 	if (error)
5993 		fatal(B_FALSE, "user_release_one(%s, %s) = %d",
5994 		    fullname, tag, error);
5995 
5996 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5997 
5998 out:
5999 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6000 }
6001 
6002 /*
6003  * Inject random faults into the on-disk data.
6004  */
6005 void
6006 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
6007 {
6008 	(void) zd, (void) id;
6009 	ztest_shared_t *zs = ztest_shared;
6010 	spa_t *spa = ztest_spa;
6011 	int fd;
6012 	uint64_t offset;
6013 	uint64_t leaves;
6014 	uint64_t bad = 0x1990c0ffeedecadeull;
6015 	uint64_t top, leaf;
6016 	char *path0;
6017 	char *pathrand;
6018 	size_t fsize;
6019 	int bshift = SPA_MAXBLOCKSHIFT + 2;
6020 	int iters = 1000;
6021 	int maxfaults;
6022 	int mirror_save;
6023 	vdev_t *vd0 = NULL;
6024 	uint64_t guid0 = 0;
6025 	boolean_t islog = B_FALSE;
6026 
6027 	path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6028 	pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6029 
6030 	mutex_enter(&ztest_vdev_lock);
6031 
6032 	/*
6033 	 * Device removal is in progress, fault injection must be disabled
6034 	 * until it completes and the pool is scrubbed.  The fault injection
6035 	 * strategy for damaging blocks does not take in to account evacuated
6036 	 * blocks which may have already been damaged.
6037 	 */
6038 	if (ztest_device_removal_active) {
6039 		mutex_exit(&ztest_vdev_lock);
6040 		goto out;
6041 	}
6042 
6043 	maxfaults = MAXFAULTS(zs);
6044 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raid_children;
6045 	mirror_save = zs->zs_mirrors;
6046 	mutex_exit(&ztest_vdev_lock);
6047 
6048 	ASSERT3U(leaves, >=, 1);
6049 
6050 	/*
6051 	 * While ztest is running the number of leaves will not change.  This
6052 	 * is critical for the fault injection logic as it determines where
6053 	 * errors can be safely injected such that they are always repairable.
6054 	 *
6055 	 * When restarting ztest a different number of leaves may be requested
6056 	 * which will shift the regions to be damaged.  This is fine as long
6057 	 * as the pool has been scrubbed prior to using the new mapping.
6058 	 * Failure to do can result in non-repairable damage being injected.
6059 	 */
6060 	if (ztest_pool_scrubbed == B_FALSE)
6061 		goto out;
6062 
6063 	/*
6064 	 * Grab the name lock as reader. There are some operations
6065 	 * which don't like to have their vdevs changed while
6066 	 * they are in progress (i.e. spa_change_guid). Those
6067 	 * operations will have grabbed the name lock as writer.
6068 	 */
6069 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6070 
6071 	/*
6072 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
6073 	 */
6074 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6075 
6076 	if (ztest_random(2) == 0) {
6077 		/*
6078 		 * Inject errors on a normal data device or slog device.
6079 		 */
6080 		top = ztest_random_vdev_top(spa, B_TRUE);
6081 		leaf = ztest_random(leaves) + zs->zs_splits;
6082 
6083 		/*
6084 		 * Generate paths to the first leaf in this top-level vdev,
6085 		 * and to the random leaf we selected.  We'll induce transient
6086 		 * write failures and random online/offline activity on leaf 0,
6087 		 * and we'll write random garbage to the randomly chosen leaf.
6088 		 */
6089 		(void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
6090 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6091 		    top * leaves + zs->zs_splits);
6092 		(void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
6093 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6094 		    top * leaves + leaf);
6095 
6096 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
6097 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
6098 			islog = B_TRUE;
6099 
6100 		/*
6101 		 * If the top-level vdev needs to be resilvered
6102 		 * then we only allow faults on the device that is
6103 		 * resilvering.
6104 		 */
6105 		if (vd0 != NULL && maxfaults != 1 &&
6106 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
6107 		    vd0->vdev_resilver_txg != 0)) {
6108 			/*
6109 			 * Make vd0 explicitly claim to be unreadable,
6110 			 * or unwritable, or reach behind its back
6111 			 * and close the underlying fd.  We can do this if
6112 			 * maxfaults == 0 because we'll fail and reexecute,
6113 			 * and we can do it if maxfaults >= 2 because we'll
6114 			 * have enough redundancy.  If maxfaults == 1, the
6115 			 * combination of this with injection of random data
6116 			 * corruption below exceeds the pool's fault tolerance.
6117 			 */
6118 			vdev_file_t *vf = vd0->vdev_tsd;
6119 
6120 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
6121 			    (long long)vd0->vdev_id, (int)maxfaults);
6122 
6123 			if (vf != NULL && ztest_random(3) == 0) {
6124 				(void) close(vf->vf_file->f_fd);
6125 				vf->vf_file->f_fd = -1;
6126 			} else if (ztest_random(2) == 0) {
6127 				vd0->vdev_cant_read = B_TRUE;
6128 			} else {
6129 				vd0->vdev_cant_write = B_TRUE;
6130 			}
6131 			guid0 = vd0->vdev_guid;
6132 		}
6133 	} else {
6134 		/*
6135 		 * Inject errors on an l2cache device.
6136 		 */
6137 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
6138 
6139 		if (sav->sav_count == 0) {
6140 			spa_config_exit(spa, SCL_STATE, FTAG);
6141 			(void) pthread_rwlock_unlock(&ztest_name_lock);
6142 			goto out;
6143 		}
6144 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
6145 		guid0 = vd0->vdev_guid;
6146 		(void) strcpy(path0, vd0->vdev_path);
6147 		(void) strcpy(pathrand, vd0->vdev_path);
6148 
6149 		leaf = 0;
6150 		leaves = 1;
6151 		maxfaults = INT_MAX;	/* no limit on cache devices */
6152 	}
6153 
6154 	spa_config_exit(spa, SCL_STATE, FTAG);
6155 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6156 
6157 	/*
6158 	 * If we can tolerate two or more faults, or we're dealing
6159 	 * with a slog, randomly online/offline vd0.
6160 	 */
6161 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
6162 		if (ztest_random(10) < 6) {
6163 			int flags = (ztest_random(2) == 0 ?
6164 			    ZFS_OFFLINE_TEMPORARY : 0);
6165 
6166 			/*
6167 			 * We have to grab the zs_name_lock as writer to
6168 			 * prevent a race between offlining a slog and
6169 			 * destroying a dataset. Offlining the slog will
6170 			 * grab a reference on the dataset which may cause
6171 			 * dsl_destroy_head() to fail with EBUSY thus
6172 			 * leaving the dataset in an inconsistent state.
6173 			 */
6174 			if (islog)
6175 				(void) pthread_rwlock_wrlock(&ztest_name_lock);
6176 
6177 			VERIFY3U(vdev_offline(spa, guid0, flags), !=, EBUSY);
6178 
6179 			if (islog)
6180 				(void) pthread_rwlock_unlock(&ztest_name_lock);
6181 		} else {
6182 			/*
6183 			 * Ideally we would like to be able to randomly
6184 			 * call vdev_[on|off]line without holding locks
6185 			 * to force unpredictable failures but the side
6186 			 * effects of vdev_[on|off]line prevent us from
6187 			 * doing so. We grab the ztest_vdev_lock here to
6188 			 * prevent a race between injection testing and
6189 			 * aux_vdev removal.
6190 			 */
6191 			mutex_enter(&ztest_vdev_lock);
6192 			(void) vdev_online(spa, guid0, 0, NULL);
6193 			mutex_exit(&ztest_vdev_lock);
6194 		}
6195 	}
6196 
6197 	if (maxfaults == 0)
6198 		goto out;
6199 
6200 	/*
6201 	 * We have at least single-fault tolerance, so inject data corruption.
6202 	 */
6203 	fd = open(pathrand, O_RDWR);
6204 
6205 	if (fd == -1) /* we hit a gap in the device namespace */
6206 		goto out;
6207 
6208 	fsize = lseek(fd, 0, SEEK_END);
6209 
6210 	while (--iters != 0) {
6211 		/*
6212 		 * The offset must be chosen carefully to ensure that
6213 		 * we do not inject a given logical block with errors
6214 		 * on two different leaf devices, because ZFS can not
6215 		 * tolerate that (if maxfaults==1).
6216 		 *
6217 		 * To achieve this we divide each leaf device into
6218 		 * chunks of size (# leaves * SPA_MAXBLOCKSIZE * 4).
6219 		 * Each chunk is further divided into error-injection
6220 		 * ranges (can accept errors) and clear ranges (we do
6221 		 * not inject errors in those). Each error-injection
6222 		 * range can accept errors only for a single leaf vdev.
6223 		 * Error-injection ranges are separated by clear ranges.
6224 		 *
6225 		 * For example, with 3 leaves, each chunk looks like:
6226 		 *    0 to  32M: injection range for leaf 0
6227 		 *  32M to  64M: clear range - no injection allowed
6228 		 *  64M to  96M: injection range for leaf 1
6229 		 *  96M to 128M: clear range - no injection allowed
6230 		 * 128M to 160M: injection range for leaf 2
6231 		 * 160M to 192M: clear range - no injection allowed
6232 		 *
6233 		 * Each clear range must be large enough such that a
6234 		 * single block cannot straddle it. This way a block
6235 		 * can't be a target in two different injection ranges
6236 		 * (on different leaf vdevs).
6237 		 */
6238 		offset = ztest_random(fsize / (leaves << bshift)) *
6239 		    (leaves << bshift) + (leaf << bshift) +
6240 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
6241 
6242 		/*
6243 		 * Only allow damage to the labels at one end of the vdev.
6244 		 *
6245 		 * If all labels are damaged, the device will be totally
6246 		 * inaccessible, which will result in loss of data,
6247 		 * because we also damage (parts of) the other side of
6248 		 * the mirror/raidz.
6249 		 *
6250 		 * Additionally, we will always have both an even and an
6251 		 * odd label, so that we can handle crashes in the
6252 		 * middle of vdev_config_sync().
6253 		 */
6254 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
6255 			continue;
6256 
6257 		/*
6258 		 * The two end labels are stored at the "end" of the disk, but
6259 		 * the end of the disk (vdev_psize) is aligned to
6260 		 * sizeof (vdev_label_t).
6261 		 */
6262 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
6263 		if ((leaf & 1) == 1 &&
6264 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
6265 			continue;
6266 
6267 		mutex_enter(&ztest_vdev_lock);
6268 		if (mirror_save != zs->zs_mirrors) {
6269 			mutex_exit(&ztest_vdev_lock);
6270 			(void) close(fd);
6271 			goto out;
6272 		}
6273 
6274 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
6275 			fatal(B_TRUE,
6276 			    "can't inject bad word at 0x%"PRIx64" in %s",
6277 			    offset, pathrand);
6278 
6279 		mutex_exit(&ztest_vdev_lock);
6280 
6281 		if (ztest_opts.zo_verbose >= 7)
6282 			(void) printf("injected bad word into %s,"
6283 			    " offset 0x%"PRIx64"\n", pathrand, offset);
6284 	}
6285 
6286 	(void) close(fd);
6287 out:
6288 	umem_free(path0, MAXPATHLEN);
6289 	umem_free(pathrand, MAXPATHLEN);
6290 }
6291 
6292 /*
6293  * By design ztest will never inject uncorrectable damage in to the pool.
6294  * Issue a scrub, wait for it to complete, and verify there is never any
6295  * persistent damage.
6296  *
6297  * Only after a full scrub has been completed is it safe to start injecting
6298  * data corruption.  See the comment in zfs_fault_inject().
6299  */
6300 static int
6301 ztest_scrub_impl(spa_t *spa)
6302 {
6303 	int error = spa_scan(spa, POOL_SCAN_SCRUB);
6304 	if (error)
6305 		return (error);
6306 
6307 	while (dsl_scan_scrubbing(spa_get_dsl(spa)))
6308 		txg_wait_synced(spa_get_dsl(spa), 0);
6309 
6310 	if (spa_get_errlog_size(spa) > 0)
6311 		return (ECKSUM);
6312 
6313 	ztest_pool_scrubbed = B_TRUE;
6314 
6315 	return (0);
6316 }
6317 
6318 /*
6319  * Scrub the pool.
6320  */
6321 void
6322 ztest_scrub(ztest_ds_t *zd, uint64_t id)
6323 {
6324 	(void) zd, (void) id;
6325 	spa_t *spa = ztest_spa;
6326 	int error;
6327 
6328 	/*
6329 	 * Scrub in progress by device removal.
6330 	 */
6331 	if (ztest_device_removal_active)
6332 		return;
6333 
6334 	/*
6335 	 * Start a scrub, wait a moment, then force a restart.
6336 	 */
6337 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
6338 	(void) poll(NULL, 0, 100);
6339 
6340 	error = ztest_scrub_impl(spa);
6341 	if (error == EBUSY)
6342 		error = 0;
6343 	ASSERT0(error);
6344 }
6345 
6346 /*
6347  * Change the guid for the pool.
6348  */
6349 void
6350 ztest_reguid(ztest_ds_t *zd, uint64_t id)
6351 {
6352 	(void) zd, (void) id;
6353 	spa_t *spa = ztest_spa;
6354 	uint64_t orig, load;
6355 	int error;
6356 
6357 	if (ztest_opts.zo_mmp_test)
6358 		return;
6359 
6360 	orig = spa_guid(spa);
6361 	load = spa_load_guid(spa);
6362 
6363 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
6364 	error = spa_change_guid(spa);
6365 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6366 
6367 	if (error != 0)
6368 		return;
6369 
6370 	if (ztest_opts.zo_verbose >= 4) {
6371 		(void) printf("Changed guid old %"PRIu64" -> %"PRIu64"\n",
6372 		    orig, spa_guid(spa));
6373 	}
6374 
6375 	VERIFY3U(orig, !=, spa_guid(spa));
6376 	VERIFY3U(load, ==, spa_load_guid(spa));
6377 }
6378 
6379 void
6380 ztest_blake3(ztest_ds_t *zd, uint64_t id)
6381 {
6382 	(void) zd, (void) id;
6383 	hrtime_t end = gethrtime() + NANOSEC;
6384 	zio_cksum_salt_t salt;
6385 	void *salt_ptr = &salt.zcs_bytes;
6386 	struct abd *abd_data, *abd_meta;
6387 	void *buf, *templ;
6388 	int i, *ptr;
6389 	uint32_t size;
6390 	BLAKE3_CTX ctx;
6391 
6392 	size = ztest_random_blocksize();
6393 	buf = umem_alloc(size, UMEM_NOFAIL);
6394 	abd_data = abd_alloc(size, B_FALSE);
6395 	abd_meta = abd_alloc(size, B_TRUE);
6396 
6397 	for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6398 		*ptr = ztest_random(UINT_MAX);
6399 	memset(salt_ptr, 'A', 32);
6400 
6401 	abd_copy_from_buf_off(abd_data, buf, 0, size);
6402 	abd_copy_from_buf_off(abd_meta, buf, 0, size);
6403 
6404 	while (gethrtime() <= end) {
6405 		int run_count = 100;
6406 		zio_cksum_t zc_ref1, zc_ref2;
6407 		zio_cksum_t zc_res1, zc_res2;
6408 
6409 		void *ref1 = &zc_ref1;
6410 		void *ref2 = &zc_ref2;
6411 		void *res1 = &zc_res1;
6412 		void *res2 = &zc_res2;
6413 
6414 		/* BLAKE3_KEY_LEN = 32 */
6415 		VERIFY0(blake3_set_impl_name("generic"));
6416 		templ = abd_checksum_blake3_tmpl_init(&salt);
6417 		Blake3_InitKeyed(&ctx, salt_ptr);
6418 		Blake3_Update(&ctx, buf, size);
6419 		Blake3_Final(&ctx, ref1);
6420 		zc_ref2 = zc_ref1;
6421 		ZIO_CHECKSUM_BSWAP(&zc_ref2);
6422 		abd_checksum_blake3_tmpl_free(templ);
6423 
6424 		VERIFY0(blake3_set_impl_name("cycle"));
6425 		while (run_count-- > 0) {
6426 
6427 			/* Test current implementation */
6428 			Blake3_InitKeyed(&ctx, salt_ptr);
6429 			Blake3_Update(&ctx, buf, size);
6430 			Blake3_Final(&ctx, res1);
6431 			zc_res2 = zc_res1;
6432 			ZIO_CHECKSUM_BSWAP(&zc_res2);
6433 
6434 			VERIFY0(memcmp(ref1, res1, 32));
6435 			VERIFY0(memcmp(ref2, res2, 32));
6436 
6437 			/* Test ABD - data */
6438 			templ = abd_checksum_blake3_tmpl_init(&salt);
6439 			abd_checksum_blake3_native(abd_data, size,
6440 			    templ, &zc_res1);
6441 			abd_checksum_blake3_byteswap(abd_data, size,
6442 			    templ, &zc_res2);
6443 
6444 			VERIFY0(memcmp(ref1, res1, 32));
6445 			VERIFY0(memcmp(ref2, res2, 32));
6446 
6447 			/* Test ABD - metadata */
6448 			abd_checksum_blake3_native(abd_meta, size,
6449 			    templ, &zc_res1);
6450 			abd_checksum_blake3_byteswap(abd_meta, size,
6451 			    templ, &zc_res2);
6452 			abd_checksum_blake3_tmpl_free(templ);
6453 
6454 			VERIFY0(memcmp(ref1, res1, 32));
6455 			VERIFY0(memcmp(ref2, res2, 32));
6456 
6457 		}
6458 	}
6459 
6460 	abd_free(abd_data);
6461 	abd_free(abd_meta);
6462 	umem_free(buf, size);
6463 }
6464 
6465 void
6466 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
6467 {
6468 	(void) zd, (void) id;
6469 	hrtime_t end = gethrtime() + NANOSEC;
6470 
6471 	while (gethrtime() <= end) {
6472 		int run_count = 100;
6473 		void *buf;
6474 		struct abd *abd_data, *abd_meta;
6475 		uint32_t size;
6476 		int *ptr;
6477 		int i;
6478 		zio_cksum_t zc_ref;
6479 		zio_cksum_t zc_ref_byteswap;
6480 
6481 		size = ztest_random_blocksize();
6482 
6483 		buf = umem_alloc(size, UMEM_NOFAIL);
6484 		abd_data = abd_alloc(size, B_FALSE);
6485 		abd_meta = abd_alloc(size, B_TRUE);
6486 
6487 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6488 			*ptr = ztest_random(UINT_MAX);
6489 
6490 		abd_copy_from_buf_off(abd_data, buf, 0, size);
6491 		abd_copy_from_buf_off(abd_meta, buf, 0, size);
6492 
6493 		VERIFY0(fletcher_4_impl_set("scalar"));
6494 		fletcher_4_native(buf, size, NULL, &zc_ref);
6495 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
6496 
6497 		VERIFY0(fletcher_4_impl_set("cycle"));
6498 		while (run_count-- > 0) {
6499 			zio_cksum_t zc;
6500 			zio_cksum_t zc_byteswap;
6501 
6502 			fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
6503 			fletcher_4_native(buf, size, NULL, &zc);
6504 
6505 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6506 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6507 			    sizeof (zc_byteswap)));
6508 
6509 			/* Test ABD - data */
6510 			abd_fletcher_4_byteswap(abd_data, size, NULL,
6511 			    &zc_byteswap);
6512 			abd_fletcher_4_native(abd_data, size, NULL, &zc);
6513 
6514 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6515 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6516 			    sizeof (zc_byteswap)));
6517 
6518 			/* Test ABD - metadata */
6519 			abd_fletcher_4_byteswap(abd_meta, size, NULL,
6520 			    &zc_byteswap);
6521 			abd_fletcher_4_native(abd_meta, size, NULL, &zc);
6522 
6523 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6524 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6525 			    sizeof (zc_byteswap)));
6526 
6527 		}
6528 
6529 		umem_free(buf, size);
6530 		abd_free(abd_data);
6531 		abd_free(abd_meta);
6532 	}
6533 }
6534 
6535 void
6536 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
6537 {
6538 	(void) zd, (void) id;
6539 	void *buf;
6540 	size_t size;
6541 	int *ptr;
6542 	int i;
6543 	zio_cksum_t zc_ref;
6544 	zio_cksum_t zc_ref_bswap;
6545 
6546 	hrtime_t end = gethrtime() + NANOSEC;
6547 
6548 	while (gethrtime() <= end) {
6549 		int run_count = 100;
6550 
6551 		size = ztest_random_blocksize();
6552 		buf = umem_alloc(size, UMEM_NOFAIL);
6553 
6554 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6555 			*ptr = ztest_random(UINT_MAX);
6556 
6557 		VERIFY0(fletcher_4_impl_set("scalar"));
6558 		fletcher_4_native(buf, size, NULL, &zc_ref);
6559 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
6560 
6561 		VERIFY0(fletcher_4_impl_set("cycle"));
6562 
6563 		while (run_count-- > 0) {
6564 			zio_cksum_t zc;
6565 			zio_cksum_t zc_bswap;
6566 			size_t pos = 0;
6567 
6568 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6569 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6570 
6571 			while (pos < size) {
6572 				size_t inc = 64 * ztest_random(size / 67);
6573 				/* sometimes add few bytes to test non-simd */
6574 				if (ztest_random(100) < 10)
6575 					inc += P2ALIGN(ztest_random(64),
6576 					    sizeof (uint32_t));
6577 
6578 				if (inc > (size - pos))
6579 					inc = size - pos;
6580 
6581 				fletcher_4_incremental_native(buf + pos, inc,
6582 				    &zc);
6583 				fletcher_4_incremental_byteswap(buf + pos, inc,
6584 				    &zc_bswap);
6585 
6586 				pos += inc;
6587 			}
6588 
6589 			VERIFY3U(pos, ==, size);
6590 
6591 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6592 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6593 
6594 			/*
6595 			 * verify if incremental on the whole buffer is
6596 			 * equivalent to non-incremental version
6597 			 */
6598 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6599 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6600 
6601 			fletcher_4_incremental_native(buf, size, &zc);
6602 			fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
6603 
6604 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6605 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6606 		}
6607 
6608 		umem_free(buf, size);
6609 	}
6610 }
6611 
6612 static int
6613 ztest_set_global_vars(void)
6614 {
6615 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
6616 		char *kv = ztest_opts.zo_gvars[i];
6617 		VERIFY3U(strlen(kv), <=, ZO_GVARS_MAX_ARGLEN);
6618 		VERIFY3U(strlen(kv), >, 0);
6619 		int err = set_global_var(kv);
6620 		if (ztest_opts.zo_verbose > 0) {
6621 			(void) printf("setting global var %s ... %s\n", kv,
6622 			    err ? "failed" : "ok");
6623 		}
6624 		if (err != 0) {
6625 			(void) fprintf(stderr,
6626 			    "failed to set global var '%s'\n", kv);
6627 			return (err);
6628 		}
6629 	}
6630 	return (0);
6631 }
6632 
6633 static char **
6634 ztest_global_vars_to_zdb_args(void)
6635 {
6636 	char **args = calloc(2*ztest_opts.zo_gvars_count + 1, sizeof (char *));
6637 	char **cur = args;
6638 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
6639 		char *kv = ztest_opts.zo_gvars[i];
6640 		*cur = "-o";
6641 		cur++;
6642 		*cur = strdup(kv);
6643 		cur++;
6644 	}
6645 	ASSERT3P(cur, ==, &args[2*ztest_opts.zo_gvars_count]);
6646 	*cur = NULL;
6647 	return (args);
6648 }
6649 
6650 /* The end of strings is indicated by a NULL element */
6651 static char *
6652 join_strings(char **strings, const char *sep)
6653 {
6654 	size_t totallen = 0;
6655 	for (char **sp = strings; *sp != NULL; sp++) {
6656 		totallen += strlen(*sp);
6657 		totallen += strlen(sep);
6658 	}
6659 	if (totallen > 0) {
6660 		ASSERT(totallen >= strlen(sep));
6661 		totallen -= strlen(sep);
6662 	}
6663 
6664 	size_t buflen = totallen + 1;
6665 	char *o = malloc(buflen); /* trailing 0 byte */
6666 	o[0] = '\0';
6667 	for (char **sp = strings; *sp != NULL; sp++) {
6668 		size_t would;
6669 		would = strlcat(o, *sp, buflen);
6670 		VERIFY3U(would, <, buflen);
6671 		if (*(sp+1) == NULL) {
6672 			break;
6673 		}
6674 		would = strlcat(o, sep, buflen);
6675 		VERIFY3U(would, <, buflen);
6676 	}
6677 	ASSERT3S(strlen(o), ==, totallen);
6678 	return (o);
6679 }
6680 
6681 static int
6682 ztest_check_path(char *path)
6683 {
6684 	struct stat s;
6685 	/* return true on success */
6686 	return (!stat(path, &s));
6687 }
6688 
6689 static void
6690 ztest_get_zdb_bin(char *bin, int len)
6691 {
6692 	char *zdb_path;
6693 	/*
6694 	 * Try to use $ZDB and in-tree zdb path. If not successful, just
6695 	 * let popen to search through PATH.
6696 	 */
6697 	if ((zdb_path = getenv("ZDB"))) {
6698 		strlcpy(bin, zdb_path, len); /* In env */
6699 		if (!ztest_check_path(bin)) {
6700 			ztest_dump_core = 0;
6701 			fatal(B_TRUE, "invalid ZDB '%s'", bin);
6702 		}
6703 		return;
6704 	}
6705 
6706 	VERIFY3P(realpath(getexecname(), bin), !=, NULL);
6707 	if (strstr(bin, ".libs/ztest")) {
6708 		strstr(bin, ".libs/ztest")[0] = '\0'; /* In-tree */
6709 		strcat(bin, "zdb");
6710 		if (ztest_check_path(bin))
6711 			return;
6712 	}
6713 	strcpy(bin, "zdb");
6714 }
6715 
6716 static vdev_t *
6717 ztest_random_concrete_vdev_leaf(vdev_t *vd)
6718 {
6719 	if (vd == NULL)
6720 		return (NULL);
6721 
6722 	if (vd->vdev_children == 0)
6723 		return (vd);
6724 
6725 	vdev_t *eligible[vd->vdev_children];
6726 	int eligible_idx = 0, i;
6727 	for (i = 0; i < vd->vdev_children; i++) {
6728 		vdev_t *cvd = vd->vdev_child[i];
6729 		if (cvd->vdev_top->vdev_removing)
6730 			continue;
6731 		if (cvd->vdev_children > 0 ||
6732 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
6733 			eligible[eligible_idx++] = cvd;
6734 		}
6735 	}
6736 	VERIFY3S(eligible_idx, >, 0);
6737 
6738 	uint64_t child_no = ztest_random(eligible_idx);
6739 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
6740 }
6741 
6742 void
6743 ztest_initialize(ztest_ds_t *zd, uint64_t id)
6744 {
6745 	(void) zd, (void) id;
6746 	spa_t *spa = ztest_spa;
6747 	int error = 0;
6748 
6749 	mutex_enter(&ztest_vdev_lock);
6750 
6751 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6752 
6753 	/* Random leaf vdev */
6754 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6755 	if (rand_vd == NULL) {
6756 		spa_config_exit(spa, SCL_VDEV, FTAG);
6757 		mutex_exit(&ztest_vdev_lock);
6758 		return;
6759 	}
6760 
6761 	/*
6762 	 * The random vdev we've selected may change as soon as we
6763 	 * drop the spa_config_lock. We create local copies of things
6764 	 * we're interested in.
6765 	 */
6766 	uint64_t guid = rand_vd->vdev_guid;
6767 	char *path = strdup(rand_vd->vdev_path);
6768 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
6769 
6770 	zfs_dbgmsg("vd %px, guid %llu", rand_vd, (u_longlong_t)guid);
6771 	spa_config_exit(spa, SCL_VDEV, FTAG);
6772 
6773 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
6774 
6775 	nvlist_t *vdev_guids = fnvlist_alloc();
6776 	nvlist_t *vdev_errlist = fnvlist_alloc();
6777 	fnvlist_add_uint64(vdev_guids, path, guid);
6778 	error = spa_vdev_initialize(spa, vdev_guids, cmd, vdev_errlist);
6779 	fnvlist_free(vdev_guids);
6780 	fnvlist_free(vdev_errlist);
6781 
6782 	switch (cmd) {
6783 	case POOL_INITIALIZE_CANCEL:
6784 		if (ztest_opts.zo_verbose >= 4) {
6785 			(void) printf("Cancel initialize %s", path);
6786 			if (!active)
6787 				(void) printf(" failed (no initialize active)");
6788 			(void) printf("\n");
6789 		}
6790 		break;
6791 	case POOL_INITIALIZE_START:
6792 		if (ztest_opts.zo_verbose >= 4) {
6793 			(void) printf("Start initialize %s", path);
6794 			if (active && error == 0)
6795 				(void) printf(" failed (already active)");
6796 			else if (error != 0)
6797 				(void) printf(" failed (error %d)", error);
6798 			(void) printf("\n");
6799 		}
6800 		break;
6801 	case POOL_INITIALIZE_SUSPEND:
6802 		if (ztest_opts.zo_verbose >= 4) {
6803 			(void) printf("Suspend initialize %s", path);
6804 			if (!active)
6805 				(void) printf(" failed (no initialize active)");
6806 			(void) printf("\n");
6807 		}
6808 		break;
6809 	}
6810 	free(path);
6811 	mutex_exit(&ztest_vdev_lock);
6812 }
6813 
6814 void
6815 ztest_trim(ztest_ds_t *zd, uint64_t id)
6816 {
6817 	(void) zd, (void) id;
6818 	spa_t *spa = ztest_spa;
6819 	int error = 0;
6820 
6821 	mutex_enter(&ztest_vdev_lock);
6822 
6823 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6824 
6825 	/* Random leaf vdev */
6826 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6827 	if (rand_vd == NULL) {
6828 		spa_config_exit(spa, SCL_VDEV, FTAG);
6829 		mutex_exit(&ztest_vdev_lock);
6830 		return;
6831 	}
6832 
6833 	/*
6834 	 * The random vdev we've selected may change as soon as we
6835 	 * drop the spa_config_lock. We create local copies of things
6836 	 * we're interested in.
6837 	 */
6838 	uint64_t guid = rand_vd->vdev_guid;
6839 	char *path = strdup(rand_vd->vdev_path);
6840 	boolean_t active = rand_vd->vdev_trim_thread != NULL;
6841 
6842 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, (u_longlong_t)guid);
6843 	spa_config_exit(spa, SCL_VDEV, FTAG);
6844 
6845 	uint64_t cmd = ztest_random(POOL_TRIM_FUNCS);
6846 	uint64_t rate = 1 << ztest_random(30);
6847 	boolean_t partial = (ztest_random(5) > 0);
6848 	boolean_t secure = (ztest_random(5) > 0);
6849 
6850 	nvlist_t *vdev_guids = fnvlist_alloc();
6851 	nvlist_t *vdev_errlist = fnvlist_alloc();
6852 	fnvlist_add_uint64(vdev_guids, path, guid);
6853 	error = spa_vdev_trim(spa, vdev_guids, cmd, rate, partial,
6854 	    secure, vdev_errlist);
6855 	fnvlist_free(vdev_guids);
6856 	fnvlist_free(vdev_errlist);
6857 
6858 	switch (cmd) {
6859 	case POOL_TRIM_CANCEL:
6860 		if (ztest_opts.zo_verbose >= 4) {
6861 			(void) printf("Cancel TRIM %s", path);
6862 			if (!active)
6863 				(void) printf(" failed (no TRIM active)");
6864 			(void) printf("\n");
6865 		}
6866 		break;
6867 	case POOL_TRIM_START:
6868 		if (ztest_opts.zo_verbose >= 4) {
6869 			(void) printf("Start TRIM %s", path);
6870 			if (active && error == 0)
6871 				(void) printf(" failed (already active)");
6872 			else if (error != 0)
6873 				(void) printf(" failed (error %d)", error);
6874 			(void) printf("\n");
6875 		}
6876 		break;
6877 	case POOL_TRIM_SUSPEND:
6878 		if (ztest_opts.zo_verbose >= 4) {
6879 			(void) printf("Suspend TRIM %s", path);
6880 			if (!active)
6881 				(void) printf(" failed (no TRIM active)");
6882 			(void) printf("\n");
6883 		}
6884 		break;
6885 	}
6886 	free(path);
6887 	mutex_exit(&ztest_vdev_lock);
6888 }
6889 
6890 /*
6891  * Verify pool integrity by running zdb.
6892  */
6893 static void
6894 ztest_run_zdb(char *pool)
6895 {
6896 	int status;
6897 	char *bin;
6898 	char *zdb;
6899 	char *zbuf;
6900 	const int len = MAXPATHLEN + MAXNAMELEN + 20;
6901 	FILE *fp;
6902 
6903 	bin = umem_alloc(len, UMEM_NOFAIL);
6904 	zdb = umem_alloc(len, UMEM_NOFAIL);
6905 	zbuf = umem_alloc(1024, UMEM_NOFAIL);
6906 
6907 	ztest_get_zdb_bin(bin, len);
6908 
6909 	char **set_gvars_args = ztest_global_vars_to_zdb_args();
6910 	char *set_gvars_args_joined = join_strings(set_gvars_args, " ");
6911 	free(set_gvars_args);
6912 
6913 	size_t would = snprintf(zdb, len,
6914 	    "%s -bcc%s%s -G -d -Y -e -y %s -p %s %s",
6915 	    bin,
6916 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
6917 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
6918 	    set_gvars_args_joined,
6919 	    ztest_opts.zo_dir,
6920 	    pool);
6921 	ASSERT3U(would, <, len);
6922 
6923 	free(set_gvars_args_joined);
6924 
6925 	if (ztest_opts.zo_verbose >= 5)
6926 		(void) printf("Executing %s\n", zdb);
6927 
6928 	fp = popen(zdb, "r");
6929 
6930 	while (fgets(zbuf, 1024, fp) != NULL)
6931 		if (ztest_opts.zo_verbose >= 3)
6932 			(void) printf("%s", zbuf);
6933 
6934 	status = pclose(fp);
6935 
6936 	if (status == 0)
6937 		goto out;
6938 
6939 	ztest_dump_core = 0;
6940 	if (WIFEXITED(status))
6941 		fatal(B_FALSE, "'%s' exit code %d", zdb, WEXITSTATUS(status));
6942 	else
6943 		fatal(B_FALSE, "'%s' died with signal %d",
6944 		    zdb, WTERMSIG(status));
6945 out:
6946 	umem_free(bin, len);
6947 	umem_free(zdb, len);
6948 	umem_free(zbuf, 1024);
6949 }
6950 
6951 static void
6952 ztest_walk_pool_directory(char *header)
6953 {
6954 	spa_t *spa = NULL;
6955 
6956 	if (ztest_opts.zo_verbose >= 6)
6957 		(void) printf("%s\n", header);
6958 
6959 	mutex_enter(&spa_namespace_lock);
6960 	while ((spa = spa_next(spa)) != NULL)
6961 		if (ztest_opts.zo_verbose >= 6)
6962 			(void) printf("\t%s\n", spa_name(spa));
6963 	mutex_exit(&spa_namespace_lock);
6964 }
6965 
6966 static void
6967 ztest_spa_import_export(char *oldname, char *newname)
6968 {
6969 	nvlist_t *config, *newconfig;
6970 	uint64_t pool_guid;
6971 	spa_t *spa;
6972 	int error;
6973 
6974 	if (ztest_opts.zo_verbose >= 4) {
6975 		(void) printf("import/export: old = %s, new = %s\n",
6976 		    oldname, newname);
6977 	}
6978 
6979 	/*
6980 	 * Clean up from previous runs.
6981 	 */
6982 	(void) spa_destroy(newname);
6983 
6984 	/*
6985 	 * Get the pool's configuration and guid.
6986 	 */
6987 	VERIFY0(spa_open(oldname, &spa, FTAG));
6988 
6989 	/*
6990 	 * Kick off a scrub to tickle scrub/export races.
6991 	 */
6992 	if (ztest_random(2) == 0)
6993 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
6994 
6995 	pool_guid = spa_guid(spa);
6996 	spa_close(spa, FTAG);
6997 
6998 	ztest_walk_pool_directory("pools before export");
6999 
7000 	/*
7001 	 * Export it.
7002 	 */
7003 	VERIFY0(spa_export(oldname, &config, B_FALSE, B_FALSE));
7004 
7005 	ztest_walk_pool_directory("pools after export");
7006 
7007 	/*
7008 	 * Try to import it.
7009 	 */
7010 	newconfig = spa_tryimport(config);
7011 	ASSERT3P(newconfig, !=, NULL);
7012 	fnvlist_free(newconfig);
7013 
7014 	/*
7015 	 * Import it under the new name.
7016 	 */
7017 	error = spa_import(newname, config, NULL, 0);
7018 	if (error != 0) {
7019 		dump_nvlist(config, 0);
7020 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
7021 		    oldname, newname, error);
7022 	}
7023 
7024 	ztest_walk_pool_directory("pools after import");
7025 
7026 	/*
7027 	 * Try to import it again -- should fail with EEXIST.
7028 	 */
7029 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
7030 
7031 	/*
7032 	 * Try to import it under a different name -- should fail with EEXIST.
7033 	 */
7034 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
7035 
7036 	/*
7037 	 * Verify that the pool is no longer visible under the old name.
7038 	 */
7039 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
7040 
7041 	/*
7042 	 * Verify that we can open and close the pool using the new name.
7043 	 */
7044 	VERIFY0(spa_open(newname, &spa, FTAG));
7045 	ASSERT3U(pool_guid, ==, spa_guid(spa));
7046 	spa_close(spa, FTAG);
7047 
7048 	fnvlist_free(config);
7049 }
7050 
7051 static void
7052 ztest_resume(spa_t *spa)
7053 {
7054 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
7055 		(void) printf("resuming from suspended state\n");
7056 	spa_vdev_state_enter(spa, SCL_NONE);
7057 	vdev_clear(spa, NULL);
7058 	(void) spa_vdev_state_exit(spa, NULL, 0);
7059 	(void) zio_resume(spa);
7060 }
7061 
7062 static __attribute__((noreturn)) void
7063 ztest_resume_thread(void *arg)
7064 {
7065 	spa_t *spa = arg;
7066 
7067 	while (!ztest_exiting) {
7068 		if (spa_suspended(spa))
7069 			ztest_resume(spa);
7070 		(void) poll(NULL, 0, 100);
7071 
7072 		/*
7073 		 * Periodically change the zfs_compressed_arc_enabled setting.
7074 		 */
7075 		if (ztest_random(10) == 0)
7076 			zfs_compressed_arc_enabled = ztest_random(2);
7077 
7078 		/*
7079 		 * Periodically change the zfs_abd_scatter_enabled setting.
7080 		 */
7081 		if (ztest_random(10) == 0)
7082 			zfs_abd_scatter_enabled = ztest_random(2);
7083 	}
7084 
7085 	thread_exit();
7086 }
7087 
7088 static __attribute__((noreturn)) void
7089 ztest_deadman_thread(void *arg)
7090 {
7091 	ztest_shared_t *zs = arg;
7092 	spa_t *spa = ztest_spa;
7093 	hrtime_t delay, overdue, last_run = gethrtime();
7094 
7095 	delay = (zs->zs_thread_stop - zs->zs_thread_start) +
7096 	    MSEC2NSEC(zfs_deadman_synctime_ms);
7097 
7098 	while (!ztest_exiting) {
7099 		/*
7100 		 * Wait for the delay timer while checking occasionally
7101 		 * if we should stop.
7102 		 */
7103 		if (gethrtime() < last_run + delay) {
7104 			(void) poll(NULL, 0, 1000);
7105 			continue;
7106 		}
7107 
7108 		/*
7109 		 * If the pool is suspended then fail immediately. Otherwise,
7110 		 * check to see if the pool is making any progress. If
7111 		 * vdev_deadman() discovers that there hasn't been any recent
7112 		 * I/Os then it will end up aborting the tests.
7113 		 */
7114 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
7115 			fatal(B_FALSE,
7116 			    "aborting test after %lu seconds because "
7117 			    "pool has transitioned to a suspended state.",
7118 			    zfs_deadman_synctime_ms / 1000);
7119 		}
7120 		vdev_deadman(spa->spa_root_vdev, FTAG);
7121 
7122 		/*
7123 		 * If the process doesn't complete within a grace period of
7124 		 * zfs_deadman_synctime_ms over the expected finish time,
7125 		 * then it may be hung and is terminated.
7126 		 */
7127 		overdue = zs->zs_proc_stop + MSEC2NSEC(zfs_deadman_synctime_ms);
7128 		if (gethrtime() > overdue) {
7129 			fatal(B_FALSE,
7130 			    "aborting test after %llu seconds because "
7131 			    "the process is overdue for termination.",
7132 			    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7133 		}
7134 
7135 		(void) printf("ztest has been running for %lld seconds\n",
7136 		    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7137 
7138 		last_run = gethrtime();
7139 		delay = MSEC2NSEC(zfs_deadman_checktime_ms);
7140 	}
7141 
7142 	thread_exit();
7143 }
7144 
7145 static void
7146 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
7147 {
7148 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
7149 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
7150 	hrtime_t functime = gethrtime();
7151 	int i;
7152 
7153 	for (i = 0; i < zi->zi_iters; i++)
7154 		zi->zi_func(zd, id);
7155 
7156 	functime = gethrtime() - functime;
7157 
7158 	atomic_add_64(&zc->zc_count, 1);
7159 	atomic_add_64(&zc->zc_time, functime);
7160 
7161 	if (ztest_opts.zo_verbose >= 4)
7162 		(void) printf("%6.2f sec in %s\n",
7163 		    (double)functime / NANOSEC, zi->zi_funcname);
7164 }
7165 
7166 static __attribute__((noreturn)) void
7167 ztest_thread(void *arg)
7168 {
7169 	int rand;
7170 	uint64_t id = (uintptr_t)arg;
7171 	ztest_shared_t *zs = ztest_shared;
7172 	uint64_t call_next;
7173 	hrtime_t now;
7174 	ztest_info_t *zi;
7175 	ztest_shared_callstate_t *zc;
7176 
7177 	while ((now = gethrtime()) < zs->zs_thread_stop) {
7178 		/*
7179 		 * See if it's time to force a crash.
7180 		 */
7181 		if (now > zs->zs_thread_kill)
7182 			ztest_kill(zs);
7183 
7184 		/*
7185 		 * If we're getting ENOSPC with some regularity, stop.
7186 		 */
7187 		if (zs->zs_enospc_count > 10)
7188 			break;
7189 
7190 		/*
7191 		 * Pick a random function to execute.
7192 		 */
7193 		rand = ztest_random(ZTEST_FUNCS);
7194 		zi = &ztest_info[rand];
7195 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
7196 		call_next = zc->zc_next;
7197 
7198 		if (now >= call_next &&
7199 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
7200 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
7201 			ztest_execute(rand, zi, id);
7202 		}
7203 	}
7204 
7205 	thread_exit();
7206 }
7207 
7208 static void
7209 ztest_dataset_name(char *dsname, char *pool, int d)
7210 {
7211 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
7212 }
7213 
7214 static void
7215 ztest_dataset_destroy(int d)
7216 {
7217 	char name[ZFS_MAX_DATASET_NAME_LEN];
7218 	int t;
7219 
7220 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7221 
7222 	if (ztest_opts.zo_verbose >= 3)
7223 		(void) printf("Destroying %s to free up space\n", name);
7224 
7225 	/*
7226 	 * Cleanup any non-standard clones and snapshots.  In general,
7227 	 * ztest thread t operates on dataset (t % zopt_datasets),
7228 	 * so there may be more than one thing to clean up.
7229 	 */
7230 	for (t = d; t < ztest_opts.zo_threads;
7231 	    t += ztest_opts.zo_datasets)
7232 		ztest_dsl_dataset_cleanup(name, t);
7233 
7234 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
7235 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
7236 }
7237 
7238 static void
7239 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
7240 {
7241 	uint64_t usedobjs, dirobjs, scratch;
7242 
7243 	/*
7244 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
7245 	 * Therefore, the number of objects in use should equal the
7246 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
7247 	 * If not, we have an object leak.
7248 	 *
7249 	 * Note that we can only check this in ztest_dataset_open(),
7250 	 * when the open-context and syncing-context values agree.
7251 	 * That's because zap_count() returns the open-context value,
7252 	 * while dmu_objset_space() returns the rootbp fill count.
7253 	 */
7254 	VERIFY0(zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
7255 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
7256 	ASSERT3U(dirobjs + 1, ==, usedobjs);
7257 }
7258 
7259 static int
7260 ztest_dataset_open(int d)
7261 {
7262 	ztest_ds_t *zd = &ztest_ds[d];
7263 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
7264 	objset_t *os;
7265 	zilog_t *zilog;
7266 	char name[ZFS_MAX_DATASET_NAME_LEN];
7267 	int error;
7268 
7269 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7270 
7271 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
7272 
7273 	error = ztest_dataset_create(name);
7274 	if (error == ENOSPC) {
7275 		(void) pthread_rwlock_unlock(&ztest_name_lock);
7276 		ztest_record_enospc(FTAG);
7277 		return (error);
7278 	}
7279 	ASSERT(error == 0 || error == EEXIST);
7280 
7281 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
7282 	    B_TRUE, zd, &os));
7283 	(void) pthread_rwlock_unlock(&ztest_name_lock);
7284 
7285 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
7286 
7287 	zilog = zd->zd_zilog;
7288 
7289 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
7290 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
7291 		fatal(B_FALSE, "missing log records: "
7292 		    "claimed %"PRIu64" < committed %"PRIu64"",
7293 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
7294 
7295 	ztest_dataset_dirobj_verify(zd);
7296 
7297 	zil_replay(os, zd, ztest_replay_vector);
7298 
7299 	ztest_dataset_dirobj_verify(zd);
7300 
7301 	if (ztest_opts.zo_verbose >= 6)
7302 		(void) printf("%s replay %"PRIu64" blocks, "
7303 		    "%"PRIu64" records, seq %"PRIu64"\n",
7304 		    zd->zd_name,
7305 		    zilog->zl_parse_blk_count,
7306 		    zilog->zl_parse_lr_count,
7307 		    zilog->zl_replaying_seq);
7308 
7309 	zilog = zil_open(os, ztest_get_data);
7310 
7311 	if (zilog->zl_replaying_seq != 0 &&
7312 	    zilog->zl_replaying_seq < committed_seq)
7313 		fatal(B_FALSE, "missing log records: "
7314 		    "replayed %"PRIu64" < committed %"PRIu64"",
7315 		    zilog->zl_replaying_seq, committed_seq);
7316 
7317 	return (0);
7318 }
7319 
7320 static void
7321 ztest_dataset_close(int d)
7322 {
7323 	ztest_ds_t *zd = &ztest_ds[d];
7324 
7325 	zil_close(zd->zd_zilog);
7326 	dmu_objset_disown(zd->zd_os, B_TRUE, zd);
7327 
7328 	ztest_zd_fini(zd);
7329 }
7330 
7331 static int
7332 ztest_replay_zil_cb(const char *name, void *arg)
7333 {
7334 	(void) arg;
7335 	objset_t *os;
7336 	ztest_ds_t *zdtmp;
7337 
7338 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_ANY, B_TRUE,
7339 	    B_TRUE, FTAG, &os));
7340 
7341 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
7342 
7343 	ztest_zd_init(zdtmp, NULL, os);
7344 	zil_replay(os, zdtmp, ztest_replay_vector);
7345 	ztest_zd_fini(zdtmp);
7346 
7347 	if (dmu_objset_zil(os)->zl_parse_lr_count != 0 &&
7348 	    ztest_opts.zo_verbose >= 6) {
7349 		zilog_t *zilog = dmu_objset_zil(os);
7350 
7351 		(void) printf("%s replay %"PRIu64" blocks, "
7352 		    "%"PRIu64" records, seq %"PRIu64"\n",
7353 		    name,
7354 		    zilog->zl_parse_blk_count,
7355 		    zilog->zl_parse_lr_count,
7356 		    zilog->zl_replaying_seq);
7357 	}
7358 
7359 	umem_free(zdtmp, sizeof (ztest_ds_t));
7360 
7361 	dmu_objset_disown(os, B_TRUE, FTAG);
7362 	return (0);
7363 }
7364 
7365 static void
7366 ztest_freeze(void)
7367 {
7368 	ztest_ds_t *zd = &ztest_ds[0];
7369 	spa_t *spa;
7370 	int numloops = 0;
7371 
7372 	if (ztest_opts.zo_verbose >= 3)
7373 		(void) printf("testing spa_freeze()...\n");
7374 
7375 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7376 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7377 	VERIFY0(ztest_dataset_open(0));
7378 	ztest_spa = spa;
7379 
7380 	/*
7381 	 * Force the first log block to be transactionally allocated.
7382 	 * We have to do this before we freeze the pool -- otherwise
7383 	 * the log chain won't be anchored.
7384 	 */
7385 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
7386 		ztest_dmu_object_alloc_free(zd, 0);
7387 		zil_commit(zd->zd_zilog, 0);
7388 	}
7389 
7390 	txg_wait_synced(spa_get_dsl(spa), 0);
7391 
7392 	/*
7393 	 * Freeze the pool.  This stops spa_sync() from doing anything,
7394 	 * so that the only way to record changes from now on is the ZIL.
7395 	 */
7396 	spa_freeze(spa);
7397 
7398 	/*
7399 	 * Because it is hard to predict how much space a write will actually
7400 	 * require beforehand, we leave ourselves some fudge space to write over
7401 	 * capacity.
7402 	 */
7403 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
7404 
7405 	/*
7406 	 * Run tests that generate log records but don't alter the pool config
7407 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
7408 	 * We do a txg_wait_synced() after each iteration to force the txg
7409 	 * to increase well beyond the last synced value in the uberblock.
7410 	 * The ZIL should be OK with that.
7411 	 *
7412 	 * Run a random number of times less than zo_maxloops and ensure we do
7413 	 * not run out of space on the pool.
7414 	 */
7415 	while (ztest_random(10) != 0 &&
7416 	    numloops++ < ztest_opts.zo_maxloops &&
7417 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
7418 		ztest_od_t od;
7419 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
7420 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
7421 		ztest_io(zd, od.od_object,
7422 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
7423 		txg_wait_synced(spa_get_dsl(spa), 0);
7424 	}
7425 
7426 	/*
7427 	 * Commit all of the changes we just generated.
7428 	 */
7429 	zil_commit(zd->zd_zilog, 0);
7430 	txg_wait_synced(spa_get_dsl(spa), 0);
7431 
7432 	/*
7433 	 * Close our dataset and close the pool.
7434 	 */
7435 	ztest_dataset_close(0);
7436 	spa_close(spa, FTAG);
7437 	kernel_fini();
7438 
7439 	/*
7440 	 * Open and close the pool and dataset to induce log replay.
7441 	 */
7442 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7443 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7444 	ASSERT3U(spa_freeze_txg(spa), ==, UINT64_MAX);
7445 	VERIFY0(ztest_dataset_open(0));
7446 	ztest_spa = spa;
7447 	txg_wait_synced(spa_get_dsl(spa), 0);
7448 	ztest_dataset_close(0);
7449 	ztest_reguid(NULL, 0);
7450 
7451 	spa_close(spa, FTAG);
7452 	kernel_fini();
7453 }
7454 
7455 static void
7456 ztest_import_impl(void)
7457 {
7458 	importargs_t args = { 0 };
7459 	nvlist_t *cfg = NULL;
7460 	int nsearch = 1;
7461 	char *searchdirs[nsearch];
7462 	int flags = ZFS_IMPORT_MISSING_LOG;
7463 
7464 	searchdirs[0] = ztest_opts.zo_dir;
7465 	args.paths = nsearch;
7466 	args.path = searchdirs;
7467 	args.can_be_active = B_FALSE;
7468 
7469 	VERIFY0(zpool_find_config(NULL, ztest_opts.zo_pool, &cfg, &args,
7470 	    &libzpool_config_ops));
7471 	VERIFY0(spa_import(ztest_opts.zo_pool, cfg, NULL, flags));
7472 	fnvlist_free(cfg);
7473 }
7474 
7475 /*
7476  * Import a storage pool with the given name.
7477  */
7478 static void
7479 ztest_import(ztest_shared_t *zs)
7480 {
7481 	spa_t *spa;
7482 
7483 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7484 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7485 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7486 
7487 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7488 
7489 	ztest_import_impl();
7490 
7491 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7492 	zs->zs_metaslab_sz =
7493 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7494 	spa_close(spa, FTAG);
7495 
7496 	kernel_fini();
7497 
7498 	if (!ztest_opts.zo_mmp_test) {
7499 		ztest_run_zdb(ztest_opts.zo_pool);
7500 		ztest_freeze();
7501 		ztest_run_zdb(ztest_opts.zo_pool);
7502 	}
7503 
7504 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7505 	mutex_destroy(&ztest_vdev_lock);
7506 	mutex_destroy(&ztest_checkpoint_lock);
7507 }
7508 
7509 /*
7510  * Kick off threads to run tests on all datasets in parallel.
7511  */
7512 static void
7513 ztest_run(ztest_shared_t *zs)
7514 {
7515 	spa_t *spa;
7516 	objset_t *os;
7517 	kthread_t *resume_thread, *deadman_thread;
7518 	kthread_t **run_threads;
7519 	uint64_t object;
7520 	int error;
7521 	int t, d;
7522 
7523 	ztest_exiting = B_FALSE;
7524 
7525 	/*
7526 	 * Initialize parent/child shared state.
7527 	 */
7528 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7529 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7530 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7531 
7532 	zs->zs_thread_start = gethrtime();
7533 	zs->zs_thread_stop =
7534 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
7535 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
7536 	zs->zs_thread_kill = zs->zs_thread_stop;
7537 	if (ztest_random(100) < ztest_opts.zo_killrate) {
7538 		zs->zs_thread_kill -=
7539 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
7540 	}
7541 
7542 	mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
7543 
7544 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
7545 	    offsetof(ztest_cb_data_t, zcd_node));
7546 
7547 	/*
7548 	 * Open our pool.  It may need to be imported first depending on
7549 	 * what tests were running when the previous pass was terminated.
7550 	 */
7551 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7552 	error = spa_open(ztest_opts.zo_pool, &spa, FTAG);
7553 	if (error) {
7554 		VERIFY3S(error, ==, ENOENT);
7555 		ztest_import_impl();
7556 		VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7557 		zs->zs_metaslab_sz =
7558 		    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7559 	}
7560 
7561 	metaslab_preload_limit = ztest_random(20) + 1;
7562 	ztest_spa = spa;
7563 
7564 	VERIFY0(vdev_raidz_impl_set("cycle"));
7565 
7566 	dmu_objset_stats_t dds;
7567 	VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
7568 	    DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
7569 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
7570 	dmu_objset_fast_stat(os, &dds);
7571 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
7572 	zs->zs_guid = dds.dds_guid;
7573 	dmu_objset_disown(os, B_TRUE, FTAG);
7574 
7575 	/*
7576 	 * Create a thread to periodically resume suspended I/O.
7577 	 */
7578 	resume_thread = thread_create(NULL, 0, ztest_resume_thread,
7579 	    spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
7580 
7581 	/*
7582 	 * Create a deadman thread and set to panic if we hang.
7583 	 */
7584 	deadman_thread = thread_create(NULL, 0, ztest_deadman_thread,
7585 	    zs, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
7586 
7587 	spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
7588 
7589 	/*
7590 	 * Verify that we can safely inquire about any object,
7591 	 * whether it's allocated or not.  To make it interesting,
7592 	 * we probe a 5-wide window around each power of two.
7593 	 * This hits all edge cases, including zero and the max.
7594 	 */
7595 	for (t = 0; t < 64; t++) {
7596 		for (d = -5; d <= 5; d++) {
7597 			error = dmu_object_info(spa->spa_meta_objset,
7598 			    (1ULL << t) + d, NULL);
7599 			ASSERT(error == 0 || error == ENOENT ||
7600 			    error == EINVAL);
7601 		}
7602 	}
7603 
7604 	/*
7605 	 * If we got any ENOSPC errors on the previous run, destroy something.
7606 	 */
7607 	if (zs->zs_enospc_count != 0) {
7608 		int d = ztest_random(ztest_opts.zo_datasets);
7609 		ztest_dataset_destroy(d);
7610 	}
7611 	zs->zs_enospc_count = 0;
7612 
7613 	/*
7614 	 * If we were in the middle of ztest_device_removal() and were killed
7615 	 * we need to ensure the removal and scrub complete before running
7616 	 * any tests that check ztest_device_removal_active. The removal will
7617 	 * be restarted automatically when the spa is opened, but we need to
7618 	 * initiate the scrub manually if it is not already in progress. Note
7619 	 * that we always run the scrub whenever an indirect vdev exists
7620 	 * because we have no way of knowing for sure if ztest_device_removal()
7621 	 * fully completed its scrub before the pool was reimported.
7622 	 */
7623 	if (spa->spa_removing_phys.sr_state == DSS_SCANNING ||
7624 	    spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
7625 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
7626 			txg_wait_synced(spa_get_dsl(spa), 0);
7627 
7628 		error = ztest_scrub_impl(spa);
7629 		if (error == EBUSY)
7630 			error = 0;
7631 		ASSERT0(error);
7632 	}
7633 
7634 	run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
7635 	    UMEM_NOFAIL);
7636 
7637 	if (ztest_opts.zo_verbose >= 4)
7638 		(void) printf("starting main threads...\n");
7639 
7640 	/*
7641 	 * Replay all logs of all datasets in the pool. This is primarily for
7642 	 * temporary datasets which wouldn't otherwise get replayed, which
7643 	 * can trigger failures when attempting to offline a SLOG in
7644 	 * ztest_fault_inject().
7645 	 */
7646 	(void) dmu_objset_find(ztest_opts.zo_pool, ztest_replay_zil_cb,
7647 	    NULL, DS_FIND_CHILDREN);
7648 
7649 	/*
7650 	 * Kick off all the tests that run in parallel.
7651 	 */
7652 	for (t = 0; t < ztest_opts.zo_threads; t++) {
7653 		if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
7654 			umem_free(run_threads, ztest_opts.zo_threads *
7655 			    sizeof (kthread_t *));
7656 			return;
7657 		}
7658 
7659 		run_threads[t] = thread_create(NULL, 0, ztest_thread,
7660 		    (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
7661 		    defclsyspri);
7662 	}
7663 
7664 	/*
7665 	 * Wait for all of the tests to complete.
7666 	 */
7667 	for (t = 0; t < ztest_opts.zo_threads; t++)
7668 		VERIFY0(thread_join(run_threads[t]));
7669 
7670 	/*
7671 	 * Close all datasets. This must be done after all the threads
7672 	 * are joined so we can be sure none of the datasets are in-use
7673 	 * by any of the threads.
7674 	 */
7675 	for (t = 0; t < ztest_opts.zo_threads; t++) {
7676 		if (t < ztest_opts.zo_datasets)
7677 			ztest_dataset_close(t);
7678 	}
7679 
7680 	txg_wait_synced(spa_get_dsl(spa), 0);
7681 
7682 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7683 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
7684 
7685 	umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
7686 
7687 	/* Kill the resume and deadman threads */
7688 	ztest_exiting = B_TRUE;
7689 	VERIFY0(thread_join(resume_thread));
7690 	VERIFY0(thread_join(deadman_thread));
7691 	ztest_resume(spa);
7692 
7693 	/*
7694 	 * Right before closing the pool, kick off a bunch of async I/O;
7695 	 * spa_close() should wait for it to complete.
7696 	 */
7697 	for (object = 1; object < 50; object++) {
7698 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
7699 		    ZIO_PRIORITY_SYNC_READ);
7700 	}
7701 
7702 	/* Verify that at least one commit cb was called in a timely fashion */
7703 	if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
7704 		VERIFY0(zc_min_txg_delay);
7705 
7706 	spa_close(spa, FTAG);
7707 
7708 	/*
7709 	 * Verify that we can loop over all pools.
7710 	 */
7711 	mutex_enter(&spa_namespace_lock);
7712 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
7713 		if (ztest_opts.zo_verbose > 3)
7714 			(void) printf("spa_next: found %s\n", spa_name(spa));
7715 	mutex_exit(&spa_namespace_lock);
7716 
7717 	/*
7718 	 * Verify that we can export the pool and reimport it under a
7719 	 * different name.
7720 	 */
7721 	if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
7722 		char name[ZFS_MAX_DATASET_NAME_LEN];
7723 		(void) snprintf(name, sizeof (name), "%s_import",
7724 		    ztest_opts.zo_pool);
7725 		ztest_spa_import_export(ztest_opts.zo_pool, name);
7726 		ztest_spa_import_export(name, ztest_opts.zo_pool);
7727 	}
7728 
7729 	kernel_fini();
7730 
7731 	list_destroy(&zcl.zcl_callbacks);
7732 	mutex_destroy(&zcl.zcl_callbacks_lock);
7733 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7734 	mutex_destroy(&ztest_vdev_lock);
7735 	mutex_destroy(&ztest_checkpoint_lock);
7736 }
7737 
7738 static void
7739 print_time(hrtime_t t, char *timebuf)
7740 {
7741 	hrtime_t s = t / NANOSEC;
7742 	hrtime_t m = s / 60;
7743 	hrtime_t h = m / 60;
7744 	hrtime_t d = h / 24;
7745 
7746 	s -= m * 60;
7747 	m -= h * 60;
7748 	h -= d * 24;
7749 
7750 	timebuf[0] = '\0';
7751 
7752 	if (d)
7753 		(void) sprintf(timebuf,
7754 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
7755 	else if (h)
7756 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
7757 	else if (m)
7758 		(void) sprintf(timebuf, "%llum%02llus", m, s);
7759 	else
7760 		(void) sprintf(timebuf, "%llus", s);
7761 }
7762 
7763 static nvlist_t *
7764 make_random_props(void)
7765 {
7766 	nvlist_t *props;
7767 
7768 	props = fnvlist_alloc();
7769 
7770 	if (ztest_random(2) == 0)
7771 		return (props);
7772 
7773 	fnvlist_add_uint64(props,
7774 	    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 1);
7775 
7776 	return (props);
7777 }
7778 
7779 /*
7780  * Create a storage pool with the given name and initial vdev size.
7781  * Then test spa_freeze() functionality.
7782  */
7783 static void
7784 ztest_init(ztest_shared_t *zs)
7785 {
7786 	spa_t *spa;
7787 	nvlist_t *nvroot, *props;
7788 	int i;
7789 
7790 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7791 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7792 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7793 
7794 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7795 
7796 	/*
7797 	 * Create the storage pool.
7798 	 */
7799 	(void) spa_destroy(ztest_opts.zo_pool);
7800 	ztest_shared->zs_vdev_next_leaf = 0;
7801 	zs->zs_splits = 0;
7802 	zs->zs_mirrors = ztest_opts.zo_mirrors;
7803 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
7804 	    NULL, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
7805 	props = make_random_props();
7806 
7807 	/*
7808 	 * We don't expect the pool to suspend unless maxfaults == 0,
7809 	 * in which case ztest_fault_inject() temporarily takes away
7810 	 * the only valid replica.
7811 	 */
7812 	fnvlist_add_uint64(props,
7813 	    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
7814 	    MAXFAULTS(zs) ? ZIO_FAILURE_MODE_PANIC : ZIO_FAILURE_MODE_WAIT);
7815 
7816 	for (i = 0; i < SPA_FEATURES; i++) {
7817 		char *buf;
7818 
7819 		if (!spa_feature_table[i].fi_zfs_mod_supported)
7820 			continue;
7821 
7822 		/*
7823 		 * 75% chance of using the log space map feature. We want ztest
7824 		 * to exercise both the code paths that use the log space map
7825 		 * feature and the ones that don't.
7826 		 */
7827 		if (i == SPA_FEATURE_LOG_SPACEMAP && ztest_random(4) == 0)
7828 			continue;
7829 
7830 		VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
7831 		    spa_feature_table[i].fi_uname));
7832 		fnvlist_add_uint64(props, buf, 0);
7833 		free(buf);
7834 	}
7835 
7836 	VERIFY0(spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
7837 	fnvlist_free(nvroot);
7838 	fnvlist_free(props);
7839 
7840 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7841 	zs->zs_metaslab_sz =
7842 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7843 	spa_close(spa, FTAG);
7844 
7845 	kernel_fini();
7846 
7847 	if (!ztest_opts.zo_mmp_test) {
7848 		ztest_run_zdb(ztest_opts.zo_pool);
7849 		ztest_freeze();
7850 		ztest_run_zdb(ztest_opts.zo_pool);
7851 	}
7852 
7853 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7854 	mutex_destroy(&ztest_vdev_lock);
7855 	mutex_destroy(&ztest_checkpoint_lock);
7856 }
7857 
7858 static void
7859 setup_data_fd(void)
7860 {
7861 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
7862 
7863 	ztest_fd_data = mkstemp(ztest_name_data);
7864 	ASSERT3S(ztest_fd_data, >=, 0);
7865 	(void) unlink(ztest_name_data);
7866 }
7867 
7868 static int
7869 shared_data_size(ztest_shared_hdr_t *hdr)
7870 {
7871 	int size;
7872 
7873 	size = hdr->zh_hdr_size;
7874 	size += hdr->zh_opts_size;
7875 	size += hdr->zh_size;
7876 	size += hdr->zh_stats_size * hdr->zh_stats_count;
7877 	size += hdr->zh_ds_size * hdr->zh_ds_count;
7878 
7879 	return (size);
7880 }
7881 
7882 static void
7883 setup_hdr(void)
7884 {
7885 	int size;
7886 	ztest_shared_hdr_t *hdr;
7887 
7888 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7889 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7890 	ASSERT3P(hdr, !=, MAP_FAILED);
7891 
7892 	VERIFY0(ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
7893 
7894 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
7895 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
7896 	hdr->zh_size = sizeof (ztest_shared_t);
7897 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
7898 	hdr->zh_stats_count = ZTEST_FUNCS;
7899 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
7900 	hdr->zh_ds_count = ztest_opts.zo_datasets;
7901 
7902 	size = shared_data_size(hdr);
7903 	VERIFY0(ftruncate(ztest_fd_data, size));
7904 
7905 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7906 }
7907 
7908 static void
7909 setup_data(void)
7910 {
7911 	int size, offset;
7912 	ztest_shared_hdr_t *hdr;
7913 	uint8_t *buf;
7914 
7915 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7916 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
7917 	ASSERT3P(hdr, !=, MAP_FAILED);
7918 
7919 	size = shared_data_size(hdr);
7920 
7921 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7922 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
7923 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7924 	ASSERT3P(hdr, !=, MAP_FAILED);
7925 	buf = (uint8_t *)hdr;
7926 
7927 	offset = hdr->zh_hdr_size;
7928 	ztest_shared_opts = (void *)&buf[offset];
7929 	offset += hdr->zh_opts_size;
7930 	ztest_shared = (void *)&buf[offset];
7931 	offset += hdr->zh_size;
7932 	ztest_shared_callstate = (void *)&buf[offset];
7933 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
7934 	ztest_shared_ds = (void *)&buf[offset];
7935 }
7936 
7937 static boolean_t
7938 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
7939 {
7940 	pid_t pid;
7941 	int status;
7942 	char *cmdbuf = NULL;
7943 
7944 	pid = fork();
7945 
7946 	if (cmd == NULL) {
7947 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
7948 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
7949 		cmd = cmdbuf;
7950 	}
7951 
7952 	if (pid == -1)
7953 		fatal(B_TRUE, "fork failed");
7954 
7955 	if (pid == 0) {	/* child */
7956 		char fd_data_str[12];
7957 
7958 		VERIFY3S(11, >=,
7959 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
7960 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
7961 
7962 		if (libpath != NULL) {
7963 			const char *curlp = getenv("LD_LIBRARY_PATH");
7964 			if (curlp == NULL)
7965 				VERIFY0(setenv("LD_LIBRARY_PATH", libpath, 1));
7966 			else {
7967 				char *newlp = NULL;
7968 				VERIFY3S(-1, !=,
7969 				    asprintf(&newlp, "%s:%s", libpath, curlp));
7970 				VERIFY0(setenv("LD_LIBRARY_PATH", newlp, 1));
7971 			}
7972 		}
7973 		(void) execl(cmd, cmd, (char *)NULL);
7974 		ztest_dump_core = B_FALSE;
7975 		fatal(B_TRUE, "exec failed: %s", cmd);
7976 	}
7977 
7978 	if (cmdbuf != NULL) {
7979 		umem_free(cmdbuf, MAXPATHLEN);
7980 		cmd = NULL;
7981 	}
7982 
7983 	while (waitpid(pid, &status, 0) != pid)
7984 		continue;
7985 	if (statusp != NULL)
7986 		*statusp = status;
7987 
7988 	if (WIFEXITED(status)) {
7989 		if (WEXITSTATUS(status) != 0) {
7990 			(void) fprintf(stderr, "child exited with code %d\n",
7991 			    WEXITSTATUS(status));
7992 			exit(2);
7993 		}
7994 		return (B_FALSE);
7995 	} else if (WIFSIGNALED(status)) {
7996 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
7997 			(void) fprintf(stderr, "child died with signal %d\n",
7998 			    WTERMSIG(status));
7999 			exit(3);
8000 		}
8001 		return (B_TRUE);
8002 	} else {
8003 		(void) fprintf(stderr, "something strange happened to child\n");
8004 		exit(4);
8005 	}
8006 }
8007 
8008 static void
8009 ztest_run_init(void)
8010 {
8011 	int i;
8012 
8013 	ztest_shared_t *zs = ztest_shared;
8014 
8015 	/*
8016 	 * Blow away any existing copy of zpool.cache
8017 	 */
8018 	(void) remove(spa_config_path);
8019 
8020 	if (ztest_opts.zo_init == 0) {
8021 		if (ztest_opts.zo_verbose >= 1)
8022 			(void) printf("Importing pool %s\n",
8023 			    ztest_opts.zo_pool);
8024 		ztest_import(zs);
8025 		return;
8026 	}
8027 
8028 	/*
8029 	 * Create and initialize our storage pool.
8030 	 */
8031 	for (i = 1; i <= ztest_opts.zo_init; i++) {
8032 		memset(zs, 0, sizeof (*zs));
8033 		if (ztest_opts.zo_verbose >= 3 &&
8034 		    ztest_opts.zo_init != 1) {
8035 			(void) printf("ztest_init(), pass %d\n", i);
8036 		}
8037 		ztest_init(zs);
8038 	}
8039 }
8040 
8041 int
8042 main(int argc, char **argv)
8043 {
8044 	int kills = 0;
8045 	int iters = 0;
8046 	int older = 0;
8047 	int newer = 0;
8048 	ztest_shared_t *zs;
8049 	ztest_info_t *zi;
8050 	ztest_shared_callstate_t *zc;
8051 	char timebuf[100];
8052 	char numbuf[NN_NUMBUF_SZ];
8053 	char *cmd;
8054 	boolean_t hasalt;
8055 	int f, err;
8056 	char *fd_data_str = getenv("ZTEST_FD_DATA");
8057 	struct sigaction action;
8058 
8059 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
8060 
8061 	dprintf_setup(&argc, argv);
8062 	zfs_deadman_synctime_ms = 300000;
8063 	zfs_deadman_checktime_ms = 30000;
8064 	/*
8065 	 * As two-word space map entries may not come up often (especially
8066 	 * if pool and vdev sizes are small) we want to force at least some
8067 	 * of them so the feature get tested.
8068 	 */
8069 	zfs_force_some_double_word_sm_entries = B_TRUE;
8070 
8071 	/*
8072 	 * Verify that even extensively damaged split blocks with many
8073 	 * segments can be reconstructed in a reasonable amount of time
8074 	 * when reconstruction is known to be possible.
8075 	 *
8076 	 * Note: the lower this value is, the more damage we inflict, and
8077 	 * the more time ztest spends in recovering that damage. We chose
8078 	 * to induce damage 1/100th of the time so recovery is tested but
8079 	 * not so frequently that ztest doesn't get to test other code paths.
8080 	 */
8081 	zfs_reconstruct_indirect_damage_fraction = 100;
8082 
8083 	action.sa_handler = sig_handler;
8084 	sigemptyset(&action.sa_mask);
8085 	action.sa_flags = 0;
8086 
8087 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
8088 		(void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
8089 		    strerror(errno));
8090 		exit(EXIT_FAILURE);
8091 	}
8092 
8093 	if (sigaction(SIGABRT, &action, NULL) < 0) {
8094 		(void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
8095 		    strerror(errno));
8096 		exit(EXIT_FAILURE);
8097 	}
8098 
8099 	/*
8100 	 * Force random_get_bytes() to use /dev/urandom in order to prevent
8101 	 * ztest from needlessly depleting the system entropy pool.
8102 	 */
8103 	random_path = "/dev/urandom";
8104 	ztest_fd_rand = open(random_path, O_RDONLY | O_CLOEXEC);
8105 	ASSERT3S(ztest_fd_rand, >=, 0);
8106 
8107 	if (!fd_data_str) {
8108 		process_options(argc, argv);
8109 
8110 		setup_data_fd();
8111 		setup_hdr();
8112 		setup_data();
8113 		memcpy(ztest_shared_opts, &ztest_opts,
8114 		    sizeof (*ztest_shared_opts));
8115 	} else {
8116 		ztest_fd_data = atoi(fd_data_str);
8117 		setup_data();
8118 		memcpy(&ztest_opts, ztest_shared_opts, sizeof (ztest_opts));
8119 	}
8120 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
8121 
8122 	err = ztest_set_global_vars();
8123 	if (err != 0 && !fd_data_str) {
8124 		/* error message done by ztest_set_global_vars */
8125 		exit(EXIT_FAILURE);
8126 	} else {
8127 		/* children should not be spawned if setting gvars fails */
8128 		VERIFY3S(err, ==, 0);
8129 	}
8130 
8131 	/* Override location of zpool.cache */
8132 	VERIFY3S(asprintf((char **)&spa_config_path, "%s/zpool.cache",
8133 	    ztest_opts.zo_dir), !=, -1);
8134 
8135 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
8136 	    UMEM_NOFAIL);
8137 	zs = ztest_shared;
8138 
8139 	if (fd_data_str) {
8140 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
8141 		metaslab_df_alloc_threshold =
8142 		    zs->zs_metaslab_df_alloc_threshold;
8143 
8144 		if (zs->zs_do_init)
8145 			ztest_run_init();
8146 		else
8147 			ztest_run(zs);
8148 		exit(0);
8149 	}
8150 
8151 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
8152 
8153 	if (ztest_opts.zo_verbose >= 1) {
8154 		(void) printf("%"PRIu64" vdevs, %d datasets, %d threads,"
8155 		    "%d %s disks, %"PRIu64" seconds...\n\n",
8156 		    ztest_opts.zo_vdevs,
8157 		    ztest_opts.zo_datasets,
8158 		    ztest_opts.zo_threads,
8159 		    ztest_opts.zo_raid_children,
8160 		    ztest_opts.zo_raid_type,
8161 		    ztest_opts.zo_time);
8162 	}
8163 
8164 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
8165 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
8166 
8167 	zs->zs_do_init = B_TRUE;
8168 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
8169 		if (ztest_opts.zo_verbose >= 1) {
8170 			(void) printf("Executing older ztest for "
8171 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
8172 		}
8173 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
8174 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
8175 	} else {
8176 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
8177 	}
8178 	zs->zs_do_init = B_FALSE;
8179 
8180 	zs->zs_proc_start = gethrtime();
8181 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
8182 
8183 	for (f = 0; f < ZTEST_FUNCS; f++) {
8184 		zi = &ztest_info[f];
8185 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
8186 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
8187 			zc->zc_next = UINT64_MAX;
8188 		else
8189 			zc->zc_next = zs->zs_proc_start +
8190 			    ztest_random(2 * zi->zi_interval[0] + 1);
8191 	}
8192 
8193 	/*
8194 	 * Run the tests in a loop.  These tests include fault injection
8195 	 * to verify that self-healing data works, and forced crashes
8196 	 * to verify that we never lose on-disk consistency.
8197 	 */
8198 	while (gethrtime() < zs->zs_proc_stop) {
8199 		int status;
8200 		boolean_t killed;
8201 
8202 		/*
8203 		 * Initialize the workload counters for each function.
8204 		 */
8205 		for (f = 0; f < ZTEST_FUNCS; f++) {
8206 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
8207 			zc->zc_count = 0;
8208 			zc->zc_time = 0;
8209 		}
8210 
8211 		/* Set the allocation switch size */
8212 		zs->zs_metaslab_df_alloc_threshold =
8213 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
8214 
8215 		if (!hasalt || ztest_random(2) == 0) {
8216 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8217 				(void) printf("Executing newer ztest: %s\n",
8218 				    cmd);
8219 			}
8220 			newer++;
8221 			killed = exec_child(cmd, NULL, B_TRUE, &status);
8222 		} else {
8223 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8224 				(void) printf("Executing older ztest: %s\n",
8225 				    ztest_opts.zo_alt_ztest);
8226 			}
8227 			older++;
8228 			killed = exec_child(ztest_opts.zo_alt_ztest,
8229 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
8230 		}
8231 
8232 		if (killed)
8233 			kills++;
8234 		iters++;
8235 
8236 		if (ztest_opts.zo_verbose >= 1) {
8237 			hrtime_t now = gethrtime();
8238 
8239 			now = MIN(now, zs->zs_proc_stop);
8240 			print_time(zs->zs_proc_stop - now, timebuf);
8241 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
8242 
8243 			(void) printf("Pass %3d, %8s, %3"PRIu64" ENOSPC, "
8244 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
8245 			    iters,
8246 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
8247 			    zs->zs_enospc_count,
8248 			    100.0 * zs->zs_alloc / zs->zs_space,
8249 			    numbuf,
8250 			    100.0 * (now - zs->zs_proc_start) /
8251 			    (ztest_opts.zo_time * NANOSEC), timebuf);
8252 		}
8253 
8254 		if (ztest_opts.zo_verbose >= 2) {
8255 			(void) printf("\nWorkload summary:\n\n");
8256 			(void) printf("%7s %9s   %s\n",
8257 			    "Calls", "Time", "Function");
8258 			(void) printf("%7s %9s   %s\n",
8259 			    "-----", "----", "--------");
8260 			for (f = 0; f < ZTEST_FUNCS; f++) {
8261 				zi = &ztest_info[f];
8262 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
8263 				print_time(zc->zc_time, timebuf);
8264 				(void) printf("%7"PRIu64" %9s   %s\n",
8265 				    zc->zc_count, timebuf,
8266 				    zi->zi_funcname);
8267 			}
8268 			(void) printf("\n");
8269 		}
8270 
8271 		if (!ztest_opts.zo_mmp_test)
8272 			ztest_run_zdb(ztest_opts.zo_pool);
8273 	}
8274 
8275 	if (ztest_opts.zo_verbose >= 1) {
8276 		if (hasalt) {
8277 			(void) printf("%d runs of older ztest: %s\n", older,
8278 			    ztest_opts.zo_alt_ztest);
8279 			(void) printf("%d runs of newer ztest: %s\n", newer,
8280 			    cmd);
8281 		}
8282 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
8283 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
8284 	}
8285 
8286 	umem_free(cmd, MAXNAMELEN);
8287 
8288 	return (0);
8289 }
8290