xref: /freebsd/sys/contrib/openzfs/cmd/ztest.c (revision 2b8331622f0b212cf3bb4fc4914a501e5321d506)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or https://opensource.org/licenses/CDDL-1.0.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Joyent, Inc.
28  * Copyright (c) 2017, Intel Corporation.
29  */
30 
31 /*
32  * The objective of this program is to provide a DMU/ZAP/SPA stress test
33  * that runs entirely in userland, is easy to use, and easy to extend.
34  *
35  * The overall design of the ztest program is as follows:
36  *
37  * (1) For each major functional area (e.g. adding vdevs to a pool,
38  *     creating and destroying datasets, reading and writing objects, etc)
39  *     we have a simple routine to test that functionality.  These
40  *     individual routines do not have to do anything "stressful".
41  *
42  * (2) We turn these simple functionality tests into a stress test by
43  *     running them all in parallel, with as many threads as desired,
44  *     and spread across as many datasets, objects, and vdevs as desired.
45  *
46  * (3) While all this is happening, we inject faults into the pool to
47  *     verify that self-healing data really works.
48  *
49  * (4) Every time we open a dataset, we change its checksum and compression
50  *     functions.  Thus even individual objects vary from block to block
51  *     in which checksum they use and whether they're compressed.
52  *
53  * (5) To verify that we never lose on-disk consistency after a crash,
54  *     we run the entire test in a child of the main process.
55  *     At random times, the child self-immolates with a SIGKILL.
56  *     This is the software equivalent of pulling the power cord.
57  *     The parent then runs the test again, using the existing
58  *     storage pool, as many times as desired. If backwards compatibility
59  *     testing is enabled ztest will sometimes run the "older" version
60  *     of ztest after a SIGKILL.
61  *
62  * (6) To verify that we don't have future leaks or temporal incursions,
63  *     many of the functional tests record the transaction group number
64  *     as part of their data.  When reading old data, they verify that
65  *     the transaction group number is less than the current, open txg.
66  *     If you add a new test, please do this if applicable.
67  *
68  * (7) Threads are created with a reduced stack size, for sanity checking.
69  *     Therefore, it's important not to allocate huge buffers on the stack.
70  *
71  * When run with no arguments, ztest runs for about five minutes and
72  * produces no output if successful.  To get a little bit of information,
73  * specify -V.  To get more information, specify -VV, and so on.
74  *
75  * To turn this into an overnight stress test, use -T to specify run time.
76  *
77  * You can ask more vdevs [-v], datasets [-d], or threads [-t]
78  * to increase the pool capacity, fanout, and overall stress level.
79  *
80  * Use the -k option to set the desired frequency of kills.
81  *
82  * When ztest invokes itself it passes all relevant information through a
83  * temporary file which is mmap-ed in the child process. This allows shared
84  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
85  * stored at offset 0 of this file and contains information on the size and
86  * number of shared structures in the file. The information stored in this file
87  * must remain backwards compatible with older versions of ztest so that
88  * ztest can invoke them during backwards compatibility testing (-B).
89  */
90 
91 #include <sys/zfs_context.h>
92 #include <sys/spa.h>
93 #include <sys/dmu.h>
94 #include <sys/txg.h>
95 #include <sys/dbuf.h>
96 #include <sys/zap.h>
97 #include <sys/dmu_objset.h>
98 #include <sys/poll.h>
99 #include <sys/stat.h>
100 #include <sys/time.h>
101 #include <sys/wait.h>
102 #include <sys/mman.h>
103 #include <sys/resource.h>
104 #include <sys/zio.h>
105 #include <sys/zil.h>
106 #include <sys/zil_impl.h>
107 #include <sys/vdev_draid.h>
108 #include <sys/vdev_impl.h>
109 #include <sys/vdev_file.h>
110 #include <sys/vdev_initialize.h>
111 #include <sys/vdev_raidz.h>
112 #include <sys/vdev_trim.h>
113 #include <sys/spa_impl.h>
114 #include <sys/metaslab_impl.h>
115 #include <sys/dsl_prop.h>
116 #include <sys/dsl_dataset.h>
117 #include <sys/dsl_destroy.h>
118 #include <sys/dsl_scan.h>
119 #include <sys/zio_checksum.h>
120 #include <sys/zfs_refcount.h>
121 #include <sys/zfeature.h>
122 #include <sys/dsl_userhold.h>
123 #include <sys/abd.h>
124 #include <sys/blake3.h>
125 #include <stdio.h>
126 #include <stdlib.h>
127 #include <unistd.h>
128 #include <getopt.h>
129 #include <signal.h>
130 #include <umem.h>
131 #include <ctype.h>
132 #include <math.h>
133 #include <sys/fs/zfs.h>
134 #include <zfs_fletcher.h>
135 #include <libnvpair.h>
136 #include <libzutil.h>
137 #include <sys/crypto/icp.h>
138 #include <sys/zfs_impl.h>
139 #if (__GLIBC__ && !__UCLIBC__)
140 #include <execinfo.h> /* for backtrace() */
141 #endif
142 
143 static int ztest_fd_data = -1;
144 static int ztest_fd_rand = -1;
145 
146 typedef struct ztest_shared_hdr {
147 	uint64_t	zh_hdr_size;
148 	uint64_t	zh_opts_size;
149 	uint64_t	zh_size;
150 	uint64_t	zh_stats_size;
151 	uint64_t	zh_stats_count;
152 	uint64_t	zh_ds_size;
153 	uint64_t	zh_ds_count;
154 } ztest_shared_hdr_t;
155 
156 static ztest_shared_hdr_t *ztest_shared_hdr;
157 
158 enum ztest_class_state {
159 	ZTEST_VDEV_CLASS_OFF,
160 	ZTEST_VDEV_CLASS_ON,
161 	ZTEST_VDEV_CLASS_RND
162 };
163 
164 #define	ZO_GVARS_MAX_ARGLEN	((size_t)64)
165 #define	ZO_GVARS_MAX_COUNT	((size_t)10)
166 
167 typedef struct ztest_shared_opts {
168 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
169 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
170 	char zo_alt_ztest[MAXNAMELEN];
171 	char zo_alt_libpath[MAXNAMELEN];
172 	uint64_t zo_vdevs;
173 	uint64_t zo_vdevtime;
174 	size_t zo_vdev_size;
175 	int zo_ashift;
176 	int zo_mirrors;
177 	int zo_raid_children;
178 	int zo_raid_parity;
179 	char zo_raid_type[8];
180 	int zo_draid_data;
181 	int zo_draid_spares;
182 	int zo_datasets;
183 	int zo_threads;
184 	uint64_t zo_passtime;
185 	uint64_t zo_killrate;
186 	int zo_verbose;
187 	int zo_init;
188 	uint64_t zo_time;
189 	uint64_t zo_maxloops;
190 	uint64_t zo_metaslab_force_ganging;
191 	int zo_mmp_test;
192 	int zo_special_vdevs;
193 	int zo_dump_dbgmsg;
194 	int zo_gvars_count;
195 	char zo_gvars[ZO_GVARS_MAX_COUNT][ZO_GVARS_MAX_ARGLEN];
196 } ztest_shared_opts_t;
197 
198 /* Default values for command line options. */
199 #define	DEFAULT_POOL "ztest"
200 #define	DEFAULT_VDEV_DIR "/tmp"
201 #define	DEFAULT_VDEV_COUNT 5
202 #define	DEFAULT_VDEV_SIZE (SPA_MINDEVSIZE * 4)	/* 256m default size */
203 #define	DEFAULT_VDEV_SIZE_STR "256M"
204 #define	DEFAULT_ASHIFT SPA_MINBLOCKSHIFT
205 #define	DEFAULT_MIRRORS 2
206 #define	DEFAULT_RAID_CHILDREN 4
207 #define	DEFAULT_RAID_PARITY 1
208 #define	DEFAULT_DRAID_DATA 4
209 #define	DEFAULT_DRAID_SPARES 1
210 #define	DEFAULT_DATASETS_COUNT 7
211 #define	DEFAULT_THREADS 23
212 #define	DEFAULT_RUN_TIME 300 /* 300 seconds */
213 #define	DEFAULT_RUN_TIME_STR "300 sec"
214 #define	DEFAULT_PASS_TIME 60 /* 60 seconds */
215 #define	DEFAULT_PASS_TIME_STR "60 sec"
216 #define	DEFAULT_KILL_RATE 70 /* 70% kill rate */
217 #define	DEFAULT_KILLRATE_STR "70%"
218 #define	DEFAULT_INITS 1
219 #define	DEFAULT_MAX_LOOPS 50 /* 5 minutes */
220 #define	DEFAULT_FORCE_GANGING (64 << 10)
221 #define	DEFAULT_FORCE_GANGING_STR "64K"
222 
223 /* Simplifying assumption: -1 is not a valid default. */
224 #define	NO_DEFAULT -1
225 
226 static const ztest_shared_opts_t ztest_opts_defaults = {
227 	.zo_pool = DEFAULT_POOL,
228 	.zo_dir = DEFAULT_VDEV_DIR,
229 	.zo_alt_ztest = { '\0' },
230 	.zo_alt_libpath = { '\0' },
231 	.zo_vdevs = DEFAULT_VDEV_COUNT,
232 	.zo_ashift = DEFAULT_ASHIFT,
233 	.zo_mirrors = DEFAULT_MIRRORS,
234 	.zo_raid_children = DEFAULT_RAID_CHILDREN,
235 	.zo_raid_parity = DEFAULT_RAID_PARITY,
236 	.zo_raid_type = VDEV_TYPE_RAIDZ,
237 	.zo_vdev_size = DEFAULT_VDEV_SIZE,
238 	.zo_draid_data = DEFAULT_DRAID_DATA,	/* data drives */
239 	.zo_draid_spares = DEFAULT_DRAID_SPARES, /* distributed spares */
240 	.zo_datasets = DEFAULT_DATASETS_COUNT,
241 	.zo_threads = DEFAULT_THREADS,
242 	.zo_passtime = DEFAULT_PASS_TIME,
243 	.zo_killrate = DEFAULT_KILL_RATE,
244 	.zo_verbose = 0,
245 	.zo_mmp_test = 0,
246 	.zo_init = DEFAULT_INITS,
247 	.zo_time = DEFAULT_RUN_TIME,
248 	.zo_maxloops = DEFAULT_MAX_LOOPS, /* max loops during spa_freeze() */
249 	.zo_metaslab_force_ganging = DEFAULT_FORCE_GANGING,
250 	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
251 	.zo_gvars_count = 0,
252 };
253 
254 extern uint64_t metaslab_force_ganging;
255 extern uint64_t metaslab_df_alloc_threshold;
256 extern uint64_t zfs_deadman_synctime_ms;
257 extern uint_t metaslab_preload_limit;
258 extern int zfs_compressed_arc_enabled;
259 extern int zfs_abd_scatter_enabled;
260 extern uint_t dmu_object_alloc_chunk_shift;
261 extern boolean_t zfs_force_some_double_word_sm_entries;
262 extern unsigned long zio_decompress_fail_fraction;
263 extern unsigned long zfs_reconstruct_indirect_damage_fraction;
264 
265 
266 static ztest_shared_opts_t *ztest_shared_opts;
267 static ztest_shared_opts_t ztest_opts;
268 static const char *const ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
269 
270 typedef struct ztest_shared_ds {
271 	uint64_t	zd_seq;
272 } ztest_shared_ds_t;
273 
274 static ztest_shared_ds_t *ztest_shared_ds;
275 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
276 
277 #define	BT_MAGIC	0x123456789abcdefULL
278 #define	MAXFAULTS(zs) \
279 	(MAX((zs)->zs_mirrors, 1) * (ztest_opts.zo_raid_parity + 1) - 1)
280 
281 enum ztest_io_type {
282 	ZTEST_IO_WRITE_TAG,
283 	ZTEST_IO_WRITE_PATTERN,
284 	ZTEST_IO_WRITE_ZEROES,
285 	ZTEST_IO_TRUNCATE,
286 	ZTEST_IO_SETATTR,
287 	ZTEST_IO_REWRITE,
288 	ZTEST_IO_TYPES
289 };
290 
291 typedef struct ztest_block_tag {
292 	uint64_t	bt_magic;
293 	uint64_t	bt_objset;
294 	uint64_t	bt_object;
295 	uint64_t	bt_dnodesize;
296 	uint64_t	bt_offset;
297 	uint64_t	bt_gen;
298 	uint64_t	bt_txg;
299 	uint64_t	bt_crtxg;
300 } ztest_block_tag_t;
301 
302 typedef struct bufwad {
303 	uint64_t	bw_index;
304 	uint64_t	bw_txg;
305 	uint64_t	bw_data;
306 } bufwad_t;
307 
308 /*
309  * It would be better to use a rangelock_t per object.  Unfortunately
310  * the rangelock_t is not a drop-in replacement for rl_t, because we
311  * still need to map from object ID to rangelock_t.
312  */
313 typedef enum {
314 	RL_READER,
315 	RL_WRITER,
316 	RL_APPEND
317 } rl_type_t;
318 
319 typedef struct rll {
320 	void		*rll_writer;
321 	int		rll_readers;
322 	kmutex_t	rll_lock;
323 	kcondvar_t	rll_cv;
324 } rll_t;
325 
326 typedef struct rl {
327 	uint64_t	rl_object;
328 	uint64_t	rl_offset;
329 	uint64_t	rl_size;
330 	rll_t		*rl_lock;
331 } rl_t;
332 
333 #define	ZTEST_RANGE_LOCKS	64
334 #define	ZTEST_OBJECT_LOCKS	64
335 
336 /*
337  * Object descriptor.  Used as a template for object lookup/create/remove.
338  */
339 typedef struct ztest_od {
340 	uint64_t	od_dir;
341 	uint64_t	od_object;
342 	dmu_object_type_t od_type;
343 	dmu_object_type_t od_crtype;
344 	uint64_t	od_blocksize;
345 	uint64_t	od_crblocksize;
346 	uint64_t	od_crdnodesize;
347 	uint64_t	od_gen;
348 	uint64_t	od_crgen;
349 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
350 } ztest_od_t;
351 
352 /*
353  * Per-dataset state.
354  */
355 typedef struct ztest_ds {
356 	ztest_shared_ds_t *zd_shared;
357 	objset_t	*zd_os;
358 	pthread_rwlock_t zd_zilog_lock;
359 	zilog_t		*zd_zilog;
360 	ztest_od_t	*zd_od;		/* debugging aid */
361 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
362 	kmutex_t	zd_dirobj_lock;
363 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
364 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
365 } ztest_ds_t;
366 
367 /*
368  * Per-iteration state.
369  */
370 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
371 
372 typedef struct ztest_info {
373 	ztest_func_t	*zi_func;	/* test function */
374 	uint64_t	zi_iters;	/* iterations per execution */
375 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
376 	const char	*zi_funcname;	/* name of test function */
377 } ztest_info_t;
378 
379 typedef struct ztest_shared_callstate {
380 	uint64_t	zc_count;	/* per-pass count */
381 	uint64_t	zc_time;	/* per-pass time */
382 	uint64_t	zc_next;	/* next time to call this function */
383 } ztest_shared_callstate_t;
384 
385 static ztest_shared_callstate_t *ztest_shared_callstate;
386 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
387 
388 ztest_func_t ztest_dmu_read_write;
389 ztest_func_t ztest_dmu_write_parallel;
390 ztest_func_t ztest_dmu_object_alloc_free;
391 ztest_func_t ztest_dmu_object_next_chunk;
392 ztest_func_t ztest_dmu_commit_callbacks;
393 ztest_func_t ztest_zap;
394 ztest_func_t ztest_zap_parallel;
395 ztest_func_t ztest_zil_commit;
396 ztest_func_t ztest_zil_remount;
397 ztest_func_t ztest_dmu_read_write_zcopy;
398 ztest_func_t ztest_dmu_objset_create_destroy;
399 ztest_func_t ztest_dmu_prealloc;
400 ztest_func_t ztest_fzap;
401 ztest_func_t ztest_dmu_snapshot_create_destroy;
402 ztest_func_t ztest_dsl_prop_get_set;
403 ztest_func_t ztest_spa_prop_get_set;
404 ztest_func_t ztest_spa_create_destroy;
405 ztest_func_t ztest_fault_inject;
406 ztest_func_t ztest_dmu_snapshot_hold;
407 ztest_func_t ztest_mmp_enable_disable;
408 ztest_func_t ztest_scrub;
409 ztest_func_t ztest_dsl_dataset_promote_busy;
410 ztest_func_t ztest_vdev_attach_detach;
411 ztest_func_t ztest_vdev_LUN_growth;
412 ztest_func_t ztest_vdev_add_remove;
413 ztest_func_t ztest_vdev_class_add;
414 ztest_func_t ztest_vdev_aux_add_remove;
415 ztest_func_t ztest_split_pool;
416 ztest_func_t ztest_reguid;
417 ztest_func_t ztest_spa_upgrade;
418 ztest_func_t ztest_device_removal;
419 ztest_func_t ztest_spa_checkpoint_create_discard;
420 ztest_func_t ztest_initialize;
421 ztest_func_t ztest_trim;
422 ztest_func_t ztest_blake3;
423 ztest_func_t ztest_fletcher;
424 ztest_func_t ztest_fletcher_incr;
425 ztest_func_t ztest_verify_dnode_bt;
426 
427 static uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
428 static uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
429 static uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
430 static uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
431 static uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
432 
433 #define	ZTI_INIT(func, iters, interval) \
434 	{   .zi_func = (func), \
435 	    .zi_iters = (iters), \
436 	    .zi_interval = (interval), \
437 	    .zi_funcname = # func }
438 
439 static ztest_info_t ztest_info[] = {
440 	ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
441 	ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
442 	ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
443 	ZTI_INIT(ztest_dmu_object_next_chunk, 1, &zopt_sometimes),
444 	ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
445 	ZTI_INIT(ztest_zap, 30, &zopt_always),
446 	ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
447 	ZTI_INIT(ztest_split_pool, 1, &zopt_sometimes),
448 	ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
449 	ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
450 	ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
451 	ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
452 	ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
453 	ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
454 #if 0
455 	ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
456 #endif
457 	ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
458 	ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
459 	ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
460 	ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
461 	ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
462 	ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
463 	ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
464 	ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
465 	ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
466 	ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
467 	ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
468 	ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
469 	ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
470 	ZTI_INIT(ztest_vdev_class_add, 1, &ztest_opts.zo_vdevtime),
471 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
472 	ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
473 	ZTI_INIT(ztest_spa_checkpoint_create_discard, 1, &zopt_rarely),
474 	ZTI_INIT(ztest_initialize, 1, &zopt_sometimes),
475 	ZTI_INIT(ztest_trim, 1, &zopt_sometimes),
476 	ZTI_INIT(ztest_blake3, 1, &zopt_rarely),
477 	ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
478 	ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
479 	ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
480 };
481 
482 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
483 
484 /*
485  * The following struct is used to hold a list of uncalled commit callbacks.
486  * The callbacks are ordered by txg number.
487  */
488 typedef struct ztest_cb_list {
489 	kmutex_t	zcl_callbacks_lock;
490 	list_t		zcl_callbacks;
491 } ztest_cb_list_t;
492 
493 /*
494  * Stuff we need to share writably between parent and child.
495  */
496 typedef struct ztest_shared {
497 	boolean_t	zs_do_init;
498 	hrtime_t	zs_proc_start;
499 	hrtime_t	zs_proc_stop;
500 	hrtime_t	zs_thread_start;
501 	hrtime_t	zs_thread_stop;
502 	hrtime_t	zs_thread_kill;
503 	uint64_t	zs_enospc_count;
504 	uint64_t	zs_vdev_next_leaf;
505 	uint64_t	zs_vdev_aux;
506 	uint64_t	zs_alloc;
507 	uint64_t	zs_space;
508 	uint64_t	zs_splits;
509 	uint64_t	zs_mirrors;
510 	uint64_t	zs_metaslab_sz;
511 	uint64_t	zs_metaslab_df_alloc_threshold;
512 	uint64_t	zs_guid;
513 } ztest_shared_t;
514 
515 #define	ID_PARALLEL	-1ULL
516 
517 static char ztest_dev_template[] = "%s/%s.%llua";
518 static char ztest_aux_template[] = "%s/%s.%s.%llu";
519 static ztest_shared_t *ztest_shared;
520 
521 static spa_t *ztest_spa = NULL;
522 static ztest_ds_t *ztest_ds;
523 
524 static kmutex_t ztest_vdev_lock;
525 static boolean_t ztest_device_removal_active = B_FALSE;
526 static boolean_t ztest_pool_scrubbed = B_FALSE;
527 static kmutex_t ztest_checkpoint_lock;
528 
529 /*
530  * The ztest_name_lock protects the pool and dataset namespace used by
531  * the individual tests. To modify the namespace, consumers must grab
532  * this lock as writer. Grabbing the lock as reader will ensure that the
533  * namespace does not change while the lock is held.
534  */
535 static pthread_rwlock_t ztest_name_lock;
536 
537 static boolean_t ztest_dump_core = B_TRUE;
538 static boolean_t ztest_exiting;
539 
540 /* Global commit callback list */
541 static ztest_cb_list_t zcl;
542 /* Commit cb delay */
543 static uint64_t zc_min_txg_delay = UINT64_MAX;
544 static int zc_cb_counter = 0;
545 
546 /*
547  * Minimum number of commit callbacks that need to be registered for us to check
548  * whether the minimum txg delay is acceptable.
549  */
550 #define	ZTEST_COMMIT_CB_MIN_REG	100
551 
552 /*
553  * If a number of txgs equal to this threshold have been created after a commit
554  * callback has been registered but not called, then we assume there is an
555  * implementation bug.
556  */
557 #define	ZTEST_COMMIT_CB_THRESH	(TXG_CONCURRENT_STATES + 1000)
558 
559 enum ztest_object {
560 	ZTEST_META_DNODE = 0,
561 	ZTEST_DIROBJ,
562 	ZTEST_OBJECTS
563 };
564 
565 static __attribute__((noreturn)) void usage(boolean_t requested);
566 static int ztest_scrub_impl(spa_t *spa);
567 
568 /*
569  * These libumem hooks provide a reasonable set of defaults for the allocator's
570  * debugging facilities.
571  */
572 const char *
573 _umem_debug_init(void)
574 {
575 	return ("default,verbose"); /* $UMEM_DEBUG setting */
576 }
577 
578 const char *
579 _umem_logging_init(void)
580 {
581 	return ("fail,contents"); /* $UMEM_LOGGING setting */
582 }
583 
584 static void
585 dump_debug_buffer(void)
586 {
587 	ssize_t ret __attribute__((unused));
588 
589 	if (!ztest_opts.zo_dump_dbgmsg)
590 		return;
591 
592 	/*
593 	 * We use write() instead of printf() so that this function
594 	 * is safe to call from a signal handler.
595 	 */
596 	ret = write(STDOUT_FILENO, "\n", 1);
597 	zfs_dbgmsg_print("ztest");
598 }
599 
600 #define	BACKTRACE_SZ	100
601 
602 static void sig_handler(int signo)
603 {
604 	struct sigaction action;
605 #if (__GLIBC__ && !__UCLIBC__) /* backtrace() is a GNU extension */
606 	int nptrs;
607 	void *buffer[BACKTRACE_SZ];
608 
609 	nptrs = backtrace(buffer, BACKTRACE_SZ);
610 	backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
611 #endif
612 	dump_debug_buffer();
613 
614 	/*
615 	 * Restore default action and re-raise signal so SIGSEGV and
616 	 * SIGABRT can trigger a core dump.
617 	 */
618 	action.sa_handler = SIG_DFL;
619 	sigemptyset(&action.sa_mask);
620 	action.sa_flags = 0;
621 	(void) sigaction(signo, &action, NULL);
622 	raise(signo);
623 }
624 
625 #define	FATAL_MSG_SZ	1024
626 
627 static const char *fatal_msg;
628 
629 static __attribute__((format(printf, 2, 3))) __attribute__((noreturn)) void
630 fatal(int do_perror, const char *message, ...)
631 {
632 	va_list args;
633 	int save_errno = errno;
634 	char *buf;
635 
636 	(void) fflush(stdout);
637 	buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
638 	if (buf == NULL)
639 		goto out;
640 
641 	va_start(args, message);
642 	(void) sprintf(buf, "ztest: ");
643 	/* LINTED */
644 	(void) vsprintf(buf + strlen(buf), message, args);
645 	va_end(args);
646 	if (do_perror) {
647 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
648 		    ": %s", strerror(save_errno));
649 	}
650 	(void) fprintf(stderr, "%s\n", buf);
651 	fatal_msg = buf;			/* to ease debugging */
652 
653 out:
654 	if (ztest_dump_core)
655 		abort();
656 	else
657 		dump_debug_buffer();
658 
659 	exit(3);
660 }
661 
662 static int
663 str2shift(const char *buf)
664 {
665 	const char *ends = "BKMGTPEZ";
666 	int i;
667 
668 	if (buf[0] == '\0')
669 		return (0);
670 	for (i = 0; i < strlen(ends); i++) {
671 		if (toupper(buf[0]) == ends[i])
672 			break;
673 	}
674 	if (i == strlen(ends)) {
675 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
676 		    buf);
677 		usage(B_FALSE);
678 	}
679 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
680 		return (10*i);
681 	}
682 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
683 	usage(B_FALSE);
684 }
685 
686 static uint64_t
687 nicenumtoull(const char *buf)
688 {
689 	char *end;
690 	uint64_t val;
691 
692 	val = strtoull(buf, &end, 0);
693 	if (end == buf) {
694 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
695 		usage(B_FALSE);
696 	} else if (end[0] == '.') {
697 		double fval = strtod(buf, &end);
698 		fval *= pow(2, str2shift(end));
699 		/*
700 		 * UINT64_MAX is not exactly representable as a double.
701 		 * The closest representation is UINT64_MAX + 1, so we
702 		 * use a >= comparison instead of > for the bounds check.
703 		 */
704 		if (fval >= (double)UINT64_MAX) {
705 			(void) fprintf(stderr, "ztest: value too large: %s\n",
706 			    buf);
707 			usage(B_FALSE);
708 		}
709 		val = (uint64_t)fval;
710 	} else {
711 		int shift = str2shift(end);
712 		if (shift >= 64 || (val << shift) >> shift != val) {
713 			(void) fprintf(stderr, "ztest: value too large: %s\n",
714 			    buf);
715 			usage(B_FALSE);
716 		}
717 		val <<= shift;
718 	}
719 	return (val);
720 }
721 
722 typedef struct ztest_option {
723 	const char	short_opt;
724 	const char	*long_opt;
725 	const char	*long_opt_param;
726 	const char	*comment;
727 	unsigned int	default_int;
728 	const char	*default_str;
729 } ztest_option_t;
730 
731 /*
732  * The following option_table is used for generating the usage info as well as
733  * the long and short option information for calling getopt_long().
734  */
735 static ztest_option_t option_table[] = {
736 	{ 'v',	"vdevs", "INTEGER", "Number of vdevs", DEFAULT_VDEV_COUNT,
737 	    NULL},
738 	{ 's',	"vdev-size", "INTEGER", "Size of each vdev",
739 	    NO_DEFAULT, DEFAULT_VDEV_SIZE_STR},
740 	{ 'a',	"alignment-shift", "INTEGER",
741 	    "Alignment shift; use 0 for random", DEFAULT_ASHIFT, NULL},
742 	{ 'm',	"mirror-copies", "INTEGER", "Number of mirror copies",
743 	    DEFAULT_MIRRORS, NULL},
744 	{ 'r',	"raid-disks", "INTEGER", "Number of raidz/draid disks",
745 	    DEFAULT_RAID_CHILDREN, NULL},
746 	{ 'R',	"raid-parity", "INTEGER", "Raid parity",
747 	    DEFAULT_RAID_PARITY, NULL},
748 	{ 'K',	"raid-kind", "raidz|draid|random", "Raid kind",
749 	    NO_DEFAULT, "random"},
750 	{ 'D',	"draid-data", "INTEGER", "Number of draid data drives",
751 	    DEFAULT_DRAID_DATA, NULL},
752 	{ 'S',	"draid-spares", "INTEGER", "Number of draid spares",
753 	    DEFAULT_DRAID_SPARES, NULL},
754 	{ 'd',	"datasets", "INTEGER", "Number of datasets",
755 	    DEFAULT_DATASETS_COUNT, NULL},
756 	{ 't',	"threads", "INTEGER", "Number of ztest threads",
757 	    DEFAULT_THREADS, NULL},
758 	{ 'g',	"gang-block-threshold", "INTEGER",
759 	    "Metaslab gang block threshold",
760 	    NO_DEFAULT, DEFAULT_FORCE_GANGING_STR},
761 	{ 'i',	"init-count", "INTEGER", "Number of times to initialize pool",
762 	    DEFAULT_INITS, NULL},
763 	{ 'k',	"kill-percentage", "INTEGER", "Kill percentage",
764 	    NO_DEFAULT, DEFAULT_KILLRATE_STR},
765 	{ 'p',	"pool-name", "STRING", "Pool name",
766 	    NO_DEFAULT, DEFAULT_POOL},
767 	{ 'f',	"vdev-file-directory", "PATH", "File directory for vdev files",
768 	    NO_DEFAULT, DEFAULT_VDEV_DIR},
769 	{ 'M',	"multi-host", NULL,
770 	    "Multi-host; simulate pool imported on remote host",
771 	    NO_DEFAULT, NULL},
772 	{ 'E',	"use-existing-pool", NULL,
773 	    "Use existing pool instead of creating new one", NO_DEFAULT, NULL},
774 	{ 'T',	"run-time", "INTEGER", "Total run time",
775 	    NO_DEFAULT, DEFAULT_RUN_TIME_STR},
776 	{ 'P',	"pass-time", "INTEGER", "Time per pass",
777 	    NO_DEFAULT, DEFAULT_PASS_TIME_STR},
778 	{ 'F',	"freeze-loops", "INTEGER", "Max loops in spa_freeze()",
779 	    DEFAULT_MAX_LOOPS, NULL},
780 	{ 'B',	"alt-ztest", "PATH", "Alternate ztest path",
781 	    NO_DEFAULT, NULL},
782 	{ 'C',	"vdev-class-state", "on|off|random", "vdev class state",
783 	    NO_DEFAULT, "random"},
784 	{ 'o',	"option", "\"OPTION=INTEGER\"",
785 	    "Set global variable to an unsigned 32-bit integer value",
786 	    NO_DEFAULT, NULL},
787 	{ 'G',	"dump-debug-msg", NULL,
788 	    "Dump zfs_dbgmsg buffer before exiting due to an error",
789 	    NO_DEFAULT, NULL},
790 	{ 'V',	"verbose", NULL,
791 	    "Verbose (use multiple times for ever more verbosity)",
792 	    NO_DEFAULT, NULL},
793 	{ 'h',	"help",	NULL, "Show this help",
794 	    NO_DEFAULT, NULL},
795 	{0, 0, 0, 0, 0, 0}
796 };
797 
798 static struct option *long_opts = NULL;
799 static char *short_opts = NULL;
800 
801 static void
802 init_options(void)
803 {
804 	ASSERT3P(long_opts, ==, NULL);
805 	ASSERT3P(short_opts, ==, NULL);
806 
807 	int count = sizeof (option_table) / sizeof (option_table[0]);
808 	long_opts = umem_alloc(sizeof (struct option) * count, UMEM_NOFAIL);
809 
810 	short_opts = umem_alloc(sizeof (char) * 2 * count, UMEM_NOFAIL);
811 	int short_opt_index = 0;
812 
813 	for (int i = 0; i < count; i++) {
814 		long_opts[i].val = option_table[i].short_opt;
815 		long_opts[i].name = option_table[i].long_opt;
816 		long_opts[i].has_arg = option_table[i].long_opt_param != NULL
817 		    ? required_argument : no_argument;
818 		long_opts[i].flag = NULL;
819 		short_opts[short_opt_index++] = option_table[i].short_opt;
820 		if (option_table[i].long_opt_param != NULL) {
821 			short_opts[short_opt_index++] = ':';
822 		}
823 	}
824 }
825 
826 static void
827 fini_options(void)
828 {
829 	int count = sizeof (option_table) / sizeof (option_table[0]);
830 
831 	umem_free(long_opts, sizeof (struct option) * count);
832 	umem_free(short_opts, sizeof (char) * 2 * count);
833 
834 	long_opts = NULL;
835 	short_opts = NULL;
836 }
837 
838 static __attribute__((noreturn)) void
839 usage(boolean_t requested)
840 {
841 	char option[80];
842 	FILE *fp = requested ? stdout : stderr;
843 
844 	(void) fprintf(fp, "Usage: %s [OPTIONS...]\n", DEFAULT_POOL);
845 	for (int i = 0; option_table[i].short_opt != 0; i++) {
846 		if (option_table[i].long_opt_param != NULL) {
847 			(void) sprintf(option, "  -%c --%s=%s",
848 			    option_table[i].short_opt,
849 			    option_table[i].long_opt,
850 			    option_table[i].long_opt_param);
851 		} else {
852 			(void) sprintf(option, "  -%c --%s",
853 			    option_table[i].short_opt,
854 			    option_table[i].long_opt);
855 		}
856 		(void) fprintf(fp, "  %-40s%s", option,
857 		    option_table[i].comment);
858 
859 		if (option_table[i].long_opt_param != NULL) {
860 			if (option_table[i].default_str != NULL) {
861 				(void) fprintf(fp, " (default: %s)",
862 				    option_table[i].default_str);
863 			} else if (option_table[i].default_int != NO_DEFAULT) {
864 				(void) fprintf(fp, " (default: %u)",
865 				    option_table[i].default_int);
866 			}
867 		}
868 		(void) fprintf(fp, "\n");
869 	}
870 	exit(requested ? 0 : 1);
871 }
872 
873 static uint64_t
874 ztest_random(uint64_t range)
875 {
876 	uint64_t r;
877 
878 	ASSERT3S(ztest_fd_rand, >=, 0);
879 
880 	if (range == 0)
881 		return (0);
882 
883 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
884 		fatal(B_TRUE, "short read from /dev/urandom");
885 
886 	return (r % range);
887 }
888 
889 static void
890 ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
891 {
892 	char name[32];
893 	char *value;
894 	int state = ZTEST_VDEV_CLASS_RND;
895 
896 	(void) strlcpy(name, input, sizeof (name));
897 
898 	value = strchr(name, '=');
899 	if (value == NULL) {
900 		(void) fprintf(stderr, "missing value in property=value "
901 		    "'-C' argument (%s)\n", input);
902 		usage(B_FALSE);
903 	}
904 	*(value) = '\0';
905 	value++;
906 
907 	if (strcmp(value, "on") == 0) {
908 		state = ZTEST_VDEV_CLASS_ON;
909 	} else if (strcmp(value, "off") == 0) {
910 		state = ZTEST_VDEV_CLASS_OFF;
911 	} else if (strcmp(value, "random") == 0) {
912 		state = ZTEST_VDEV_CLASS_RND;
913 	} else {
914 		(void) fprintf(stderr, "invalid property value '%s'\n", value);
915 		usage(B_FALSE);
916 	}
917 
918 	if (strcmp(name, "special") == 0) {
919 		zo->zo_special_vdevs = state;
920 	} else {
921 		(void) fprintf(stderr, "invalid property name '%s'\n", name);
922 		usage(B_FALSE);
923 	}
924 	if (zo->zo_verbose >= 3)
925 		(void) printf("%s vdev state is '%s'\n", name, value);
926 }
927 
928 static void
929 process_options(int argc, char **argv)
930 {
931 	char *path;
932 	ztest_shared_opts_t *zo = &ztest_opts;
933 
934 	int opt;
935 	uint64_t value;
936 	const char *raid_kind = "random";
937 
938 	memcpy(zo, &ztest_opts_defaults, sizeof (*zo));
939 
940 	init_options();
941 
942 	while ((opt = getopt_long(argc, argv, short_opts, long_opts,
943 	    NULL)) != EOF) {
944 		value = 0;
945 		switch (opt) {
946 		case 'v':
947 		case 's':
948 		case 'a':
949 		case 'm':
950 		case 'r':
951 		case 'R':
952 		case 'D':
953 		case 'S':
954 		case 'd':
955 		case 't':
956 		case 'g':
957 		case 'i':
958 		case 'k':
959 		case 'T':
960 		case 'P':
961 		case 'F':
962 			value = nicenumtoull(optarg);
963 		}
964 		switch (opt) {
965 		case 'v':
966 			zo->zo_vdevs = value;
967 			break;
968 		case 's':
969 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
970 			break;
971 		case 'a':
972 			zo->zo_ashift = value;
973 			break;
974 		case 'm':
975 			zo->zo_mirrors = value;
976 			break;
977 		case 'r':
978 			zo->zo_raid_children = MAX(1, value);
979 			break;
980 		case 'R':
981 			zo->zo_raid_parity = MIN(MAX(value, 1), 3);
982 			break;
983 		case 'K':
984 			raid_kind = optarg;
985 			break;
986 		case 'D':
987 			zo->zo_draid_data = MAX(1, value);
988 			break;
989 		case 'S':
990 			zo->zo_draid_spares = MAX(1, value);
991 			break;
992 		case 'd':
993 			zo->zo_datasets = MAX(1, value);
994 			break;
995 		case 't':
996 			zo->zo_threads = MAX(1, value);
997 			break;
998 		case 'g':
999 			zo->zo_metaslab_force_ganging =
1000 			    MAX(SPA_MINBLOCKSIZE << 1, value);
1001 			break;
1002 		case 'i':
1003 			zo->zo_init = value;
1004 			break;
1005 		case 'k':
1006 			zo->zo_killrate = value;
1007 			break;
1008 		case 'p':
1009 			(void) strlcpy(zo->zo_pool, optarg,
1010 			    sizeof (zo->zo_pool));
1011 			break;
1012 		case 'f':
1013 			path = realpath(optarg, NULL);
1014 			if (path == NULL) {
1015 				(void) fprintf(stderr, "error: %s: %s\n",
1016 				    optarg, strerror(errno));
1017 				usage(B_FALSE);
1018 			} else {
1019 				(void) strlcpy(zo->zo_dir, path,
1020 				    sizeof (zo->zo_dir));
1021 				free(path);
1022 			}
1023 			break;
1024 		case 'M':
1025 			zo->zo_mmp_test = 1;
1026 			break;
1027 		case 'V':
1028 			zo->zo_verbose++;
1029 			break;
1030 		case 'E':
1031 			zo->zo_init = 0;
1032 			break;
1033 		case 'T':
1034 			zo->zo_time = value;
1035 			break;
1036 		case 'P':
1037 			zo->zo_passtime = MAX(1, value);
1038 			break;
1039 		case 'F':
1040 			zo->zo_maxloops = MAX(1, value);
1041 			break;
1042 		case 'B':
1043 			(void) strlcpy(zo->zo_alt_ztest, optarg,
1044 			    sizeof (zo->zo_alt_ztest));
1045 			break;
1046 		case 'C':
1047 			ztest_parse_name_value(optarg, zo);
1048 			break;
1049 		case 'o':
1050 			if (zo->zo_gvars_count >= ZO_GVARS_MAX_COUNT) {
1051 				(void) fprintf(stderr,
1052 				    "max global var count (%zu) exceeded\n",
1053 				    ZO_GVARS_MAX_COUNT);
1054 				usage(B_FALSE);
1055 			}
1056 			char *v = zo->zo_gvars[zo->zo_gvars_count];
1057 			if (strlcpy(v, optarg, ZO_GVARS_MAX_ARGLEN) >=
1058 			    ZO_GVARS_MAX_ARGLEN) {
1059 				(void) fprintf(stderr,
1060 				    "global var option '%s' is too long\n",
1061 				    optarg);
1062 				usage(B_FALSE);
1063 			}
1064 			zo->zo_gvars_count++;
1065 			break;
1066 		case 'G':
1067 			zo->zo_dump_dbgmsg = 1;
1068 			break;
1069 		case 'h':
1070 			usage(B_TRUE);
1071 			break;
1072 		case '?':
1073 		default:
1074 			usage(B_FALSE);
1075 			break;
1076 		}
1077 	}
1078 
1079 	fini_options();
1080 
1081 	/* When raid choice is 'random' add a draid pool 50% of the time */
1082 	if (strcmp(raid_kind, "random") == 0) {
1083 		raid_kind = (ztest_random(2) == 0) ? "draid" : "raidz";
1084 
1085 		if (ztest_opts.zo_verbose >= 3)
1086 			(void) printf("choosing RAID type '%s'\n", raid_kind);
1087 	}
1088 
1089 	if (strcmp(raid_kind, "draid") == 0) {
1090 		uint64_t min_devsize;
1091 
1092 		/* With fewer disk use 256M, otherwise 128M is OK */
1093 		min_devsize = (ztest_opts.zo_raid_children < 16) ?
1094 		    (256ULL << 20) : (128ULL << 20);
1095 
1096 		/* No top-level mirrors with dRAID for now */
1097 		zo->zo_mirrors = 0;
1098 
1099 		/* Use more appropriate defaults for dRAID */
1100 		if (zo->zo_vdevs == ztest_opts_defaults.zo_vdevs)
1101 			zo->zo_vdevs = 1;
1102 		if (zo->zo_raid_children ==
1103 		    ztest_opts_defaults.zo_raid_children)
1104 			zo->zo_raid_children = 16;
1105 		if (zo->zo_ashift < 12)
1106 			zo->zo_ashift = 12;
1107 		if (zo->zo_vdev_size < min_devsize)
1108 			zo->zo_vdev_size = min_devsize;
1109 
1110 		if (zo->zo_draid_data + zo->zo_raid_parity >
1111 		    zo->zo_raid_children - zo->zo_draid_spares) {
1112 			(void) fprintf(stderr, "error: too few draid "
1113 			    "children (%d) for stripe width (%d)\n",
1114 			    zo->zo_raid_children,
1115 			    zo->zo_draid_data + zo->zo_raid_parity);
1116 			usage(B_FALSE);
1117 		}
1118 
1119 		(void) strlcpy(zo->zo_raid_type, VDEV_TYPE_DRAID,
1120 		    sizeof (zo->zo_raid_type));
1121 
1122 	} else /* using raidz */ {
1123 		ASSERT0(strcmp(raid_kind, "raidz"));
1124 
1125 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
1126 		    zo->zo_raid_children - 1);
1127 	}
1128 
1129 	zo->zo_vdevtime =
1130 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
1131 	    UINT64_MAX >> 2);
1132 
1133 	if (*zo->zo_alt_ztest) {
1134 		const char *invalid_what = "ztest";
1135 		char *val = zo->zo_alt_ztest;
1136 		if (0 != access(val, X_OK) ||
1137 		    (strrchr(val, '/') == NULL && (errno == EINVAL)))
1138 			goto invalid;
1139 
1140 		int dirlen = strrchr(val, '/') - val;
1141 		strlcpy(zo->zo_alt_libpath, val,
1142 		    MIN(sizeof (zo->zo_alt_libpath), dirlen + 1));
1143 		invalid_what = "library path", val = zo->zo_alt_libpath;
1144 		if (strrchr(val, '/') == NULL && (errno == EINVAL))
1145 			goto invalid;
1146 		*strrchr(val, '/') = '\0';
1147 		strlcat(val, "/lib", sizeof (zo->zo_alt_libpath));
1148 
1149 		if (0 != access(zo->zo_alt_libpath, X_OK))
1150 			goto invalid;
1151 		return;
1152 
1153 invalid:
1154 		ztest_dump_core = B_FALSE;
1155 		fatal(B_TRUE, "invalid alternate %s %s", invalid_what, val);
1156 	}
1157 }
1158 
1159 static void
1160 ztest_kill(ztest_shared_t *zs)
1161 {
1162 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
1163 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
1164 
1165 	/*
1166 	 * Before we kill ourselves, make sure that the config is updated.
1167 	 * See comment above spa_write_cachefile().
1168 	 */
1169 	mutex_enter(&spa_namespace_lock);
1170 	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE, B_FALSE);
1171 	mutex_exit(&spa_namespace_lock);
1172 
1173 	(void) raise(SIGKILL);
1174 }
1175 
1176 static void
1177 ztest_record_enospc(const char *s)
1178 {
1179 	(void) s;
1180 	ztest_shared->zs_enospc_count++;
1181 }
1182 
1183 static uint64_t
1184 ztest_get_ashift(void)
1185 {
1186 	if (ztest_opts.zo_ashift == 0)
1187 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
1188 	return (ztest_opts.zo_ashift);
1189 }
1190 
1191 static boolean_t
1192 ztest_is_draid_spare(const char *name)
1193 {
1194 	uint64_t spare_id = 0, parity = 0, vdev_id = 0;
1195 
1196 	if (sscanf(name, VDEV_TYPE_DRAID "%"PRIu64"-%"PRIu64"-%"PRIu64"",
1197 	    &parity, &vdev_id, &spare_id) == 3) {
1198 		return (B_TRUE);
1199 	}
1200 
1201 	return (B_FALSE);
1202 }
1203 
1204 static nvlist_t *
1205 make_vdev_file(const char *path, const char *aux, const char *pool,
1206     size_t size, uint64_t ashift)
1207 {
1208 	char *pathbuf = NULL;
1209 	uint64_t vdev;
1210 	nvlist_t *file;
1211 	boolean_t draid_spare = B_FALSE;
1212 
1213 
1214 	if (ashift == 0)
1215 		ashift = ztest_get_ashift();
1216 
1217 	if (path == NULL) {
1218 		pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1219 		path = pathbuf;
1220 
1221 		if (aux != NULL) {
1222 			vdev = ztest_shared->zs_vdev_aux;
1223 			(void) snprintf(pathbuf, MAXPATHLEN,
1224 			    ztest_aux_template, ztest_opts.zo_dir,
1225 			    pool == NULL ? ztest_opts.zo_pool : pool,
1226 			    aux, vdev);
1227 		} else {
1228 			vdev = ztest_shared->zs_vdev_next_leaf++;
1229 			(void) snprintf(pathbuf, MAXPATHLEN,
1230 			    ztest_dev_template, ztest_opts.zo_dir,
1231 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
1232 		}
1233 	} else {
1234 		draid_spare = ztest_is_draid_spare(path);
1235 	}
1236 
1237 	if (size != 0 && !draid_spare) {
1238 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
1239 		if (fd == -1)
1240 			fatal(B_TRUE, "can't open %s", path);
1241 		if (ftruncate(fd, size) != 0)
1242 			fatal(B_TRUE, "can't ftruncate %s", path);
1243 		(void) close(fd);
1244 	}
1245 
1246 	file = fnvlist_alloc();
1247 	fnvlist_add_string(file, ZPOOL_CONFIG_TYPE,
1248 	    draid_spare ? VDEV_TYPE_DRAID_SPARE : VDEV_TYPE_FILE);
1249 	fnvlist_add_string(file, ZPOOL_CONFIG_PATH, path);
1250 	fnvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift);
1251 	umem_free(pathbuf, MAXPATHLEN);
1252 
1253 	return (file);
1254 }
1255 
1256 static nvlist_t *
1257 make_vdev_raid(const char *path, const char *aux, const char *pool, size_t size,
1258     uint64_t ashift, int r)
1259 {
1260 	nvlist_t *raid, **child;
1261 	int c;
1262 
1263 	if (r < 2)
1264 		return (make_vdev_file(path, aux, pool, size, ashift));
1265 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
1266 
1267 	for (c = 0; c < r; c++)
1268 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
1269 
1270 	raid = fnvlist_alloc();
1271 	fnvlist_add_string(raid, ZPOOL_CONFIG_TYPE,
1272 	    ztest_opts.zo_raid_type);
1273 	fnvlist_add_uint64(raid, ZPOOL_CONFIG_NPARITY,
1274 	    ztest_opts.zo_raid_parity);
1275 	fnvlist_add_nvlist_array(raid, ZPOOL_CONFIG_CHILDREN,
1276 	    (const nvlist_t **)child, r);
1277 
1278 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0) {
1279 		uint64_t ndata = ztest_opts.zo_draid_data;
1280 		uint64_t nparity = ztest_opts.zo_raid_parity;
1281 		uint64_t nspares = ztest_opts.zo_draid_spares;
1282 		uint64_t children = ztest_opts.zo_raid_children;
1283 		uint64_t ngroups = 1;
1284 
1285 		/*
1286 		 * Calculate the minimum number of groups required to fill a
1287 		 * slice. This is the LCM of the stripe width (data + parity)
1288 		 * and the number of data drives (children - spares).
1289 		 */
1290 		while (ngroups * (ndata + nparity) % (children - nspares) != 0)
1291 			ngroups++;
1292 
1293 		/* Store the basic dRAID configuration. */
1294 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NDATA, ndata);
1295 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NSPARES, nspares);
1296 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups);
1297 	}
1298 
1299 	for (c = 0; c < r; c++)
1300 		fnvlist_free(child[c]);
1301 
1302 	umem_free(child, r * sizeof (nvlist_t *));
1303 
1304 	return (raid);
1305 }
1306 
1307 static nvlist_t *
1308 make_vdev_mirror(const char *path, const char *aux, const char *pool,
1309     size_t size, uint64_t ashift, int r, int m)
1310 {
1311 	nvlist_t *mirror, **child;
1312 	int c;
1313 
1314 	if (m < 1)
1315 		return (make_vdev_raid(path, aux, pool, size, ashift, r));
1316 
1317 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1318 
1319 	for (c = 0; c < m; c++)
1320 		child[c] = make_vdev_raid(path, aux, pool, size, ashift, r);
1321 
1322 	mirror = fnvlist_alloc();
1323 	fnvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, VDEV_TYPE_MIRROR);
1324 	fnvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1325 	    (const nvlist_t **)child, m);
1326 
1327 	for (c = 0; c < m; c++)
1328 		fnvlist_free(child[c]);
1329 
1330 	umem_free(child, m * sizeof (nvlist_t *));
1331 
1332 	return (mirror);
1333 }
1334 
1335 static nvlist_t *
1336 make_vdev_root(const char *path, const char *aux, const char *pool, size_t size,
1337     uint64_t ashift, const char *class, int r, int m, int t)
1338 {
1339 	nvlist_t *root, **child;
1340 	int c;
1341 	boolean_t log;
1342 
1343 	ASSERT3S(t, >, 0);
1344 
1345 	log = (class != NULL && strcmp(class, "log") == 0);
1346 
1347 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1348 
1349 	for (c = 0; c < t; c++) {
1350 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1351 		    r, m);
1352 		fnvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, log);
1353 
1354 		if (class != NULL && class[0] != '\0') {
1355 			ASSERT(m > 1 || log);   /* expecting a mirror */
1356 			fnvlist_add_string(child[c],
1357 			    ZPOOL_CONFIG_ALLOCATION_BIAS, class);
1358 		}
1359 	}
1360 
1361 	root = fnvlist_alloc();
1362 	fnvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
1363 	fnvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1364 	    (const nvlist_t **)child, t);
1365 
1366 	for (c = 0; c < t; c++)
1367 		fnvlist_free(child[c]);
1368 
1369 	umem_free(child, t * sizeof (nvlist_t *));
1370 
1371 	return (root);
1372 }
1373 
1374 /*
1375  * Find a random spa version. Returns back a random spa version in the
1376  * range [initial_version, SPA_VERSION_FEATURES].
1377  */
1378 static uint64_t
1379 ztest_random_spa_version(uint64_t initial_version)
1380 {
1381 	uint64_t version = initial_version;
1382 
1383 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1384 		version = version +
1385 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1386 	}
1387 
1388 	if (version > SPA_VERSION_BEFORE_FEATURES)
1389 		version = SPA_VERSION_FEATURES;
1390 
1391 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1392 	return (version);
1393 }
1394 
1395 static int
1396 ztest_random_blocksize(void)
1397 {
1398 	ASSERT3U(ztest_spa->spa_max_ashift, !=, 0);
1399 
1400 	/*
1401 	 * Choose a block size >= the ashift.
1402 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1403 	 */
1404 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1405 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1406 		maxbs = 20;
1407 	uint64_t block_shift =
1408 	    ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1409 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1410 }
1411 
1412 static int
1413 ztest_random_dnodesize(void)
1414 {
1415 	int slots;
1416 	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1417 
1418 	if (max_slots == DNODE_MIN_SLOTS)
1419 		return (DNODE_MIN_SIZE);
1420 
1421 	/*
1422 	 * Weight the random distribution more heavily toward smaller
1423 	 * dnode sizes since that is more likely to reflect real-world
1424 	 * usage.
1425 	 */
1426 	ASSERT3U(max_slots, >, 4);
1427 	switch (ztest_random(10)) {
1428 	case 0:
1429 		slots = 5 + ztest_random(max_slots - 4);
1430 		break;
1431 	case 1 ... 4:
1432 		slots = 2 + ztest_random(3);
1433 		break;
1434 	default:
1435 		slots = 1;
1436 		break;
1437 	}
1438 
1439 	return (slots << DNODE_SHIFT);
1440 }
1441 
1442 static int
1443 ztest_random_ibshift(void)
1444 {
1445 	return (DN_MIN_INDBLKSHIFT +
1446 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1447 }
1448 
1449 static uint64_t
1450 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1451 {
1452 	uint64_t top;
1453 	vdev_t *rvd = spa->spa_root_vdev;
1454 	vdev_t *tvd;
1455 
1456 	ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
1457 
1458 	do {
1459 		top = ztest_random(rvd->vdev_children);
1460 		tvd = rvd->vdev_child[top];
1461 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1462 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1463 
1464 	return (top);
1465 }
1466 
1467 static uint64_t
1468 ztest_random_dsl_prop(zfs_prop_t prop)
1469 {
1470 	uint64_t value;
1471 
1472 	do {
1473 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1474 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1475 
1476 	return (value);
1477 }
1478 
1479 static int
1480 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1481     boolean_t inherit)
1482 {
1483 	const char *propname = zfs_prop_to_name(prop);
1484 	const char *valname;
1485 	char *setpoint;
1486 	uint64_t curval;
1487 	int error;
1488 
1489 	error = dsl_prop_set_int(osname, propname,
1490 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1491 
1492 	if (error == ENOSPC) {
1493 		ztest_record_enospc(FTAG);
1494 		return (error);
1495 	}
1496 	ASSERT0(error);
1497 
1498 	setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1499 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1500 
1501 	if (ztest_opts.zo_verbose >= 6) {
1502 		int err;
1503 
1504 		err = zfs_prop_index_to_string(prop, curval, &valname);
1505 		if (err)
1506 			(void) printf("%s %s = %llu at '%s'\n", osname,
1507 			    propname, (unsigned long long)curval, setpoint);
1508 		else
1509 			(void) printf("%s %s = %s at '%s'\n",
1510 			    osname, propname, valname, setpoint);
1511 	}
1512 	umem_free(setpoint, MAXPATHLEN);
1513 
1514 	return (error);
1515 }
1516 
1517 static int
1518 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1519 {
1520 	spa_t *spa = ztest_spa;
1521 	nvlist_t *props = NULL;
1522 	int error;
1523 
1524 	props = fnvlist_alloc();
1525 	fnvlist_add_uint64(props, zpool_prop_to_name(prop), value);
1526 
1527 	error = spa_prop_set(spa, props);
1528 
1529 	fnvlist_free(props);
1530 
1531 	if (error == ENOSPC) {
1532 		ztest_record_enospc(FTAG);
1533 		return (error);
1534 	}
1535 	ASSERT0(error);
1536 
1537 	return (error);
1538 }
1539 
1540 static int
1541 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1542     boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp)
1543 {
1544 	int err;
1545 	char *cp = NULL;
1546 	char ddname[ZFS_MAX_DATASET_NAME_LEN];
1547 
1548 	strlcpy(ddname, name, sizeof (ddname));
1549 	cp = strchr(ddname, '@');
1550 	if (cp != NULL)
1551 		*cp = '\0';
1552 
1553 	err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1554 	while (decrypt && err == EACCES) {
1555 		dsl_crypto_params_t *dcp;
1556 		nvlist_t *crypto_args = fnvlist_alloc();
1557 
1558 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
1559 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1560 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1561 		    crypto_args, &dcp));
1562 		err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1563 		/*
1564 		 * Note: if there was an error loading, the wkey was not
1565 		 * consumed, and needs to be freed.
1566 		 */
1567 		dsl_crypto_params_free(dcp, (err != 0));
1568 		fnvlist_free(crypto_args);
1569 
1570 		if (err == EINVAL) {
1571 			/*
1572 			 * We couldn't load a key for this dataset so try
1573 			 * the parent. This loop will eventually hit the
1574 			 * encryption root since ztest only makes clones
1575 			 * as children of their origin datasets.
1576 			 */
1577 			cp = strrchr(ddname, '/');
1578 			if (cp == NULL)
1579 				return (err);
1580 
1581 			*cp = '\0';
1582 			err = EACCES;
1583 			continue;
1584 		} else if (err != 0) {
1585 			break;
1586 		}
1587 
1588 		err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1589 		break;
1590 	}
1591 
1592 	return (err);
1593 }
1594 
1595 static void
1596 ztest_rll_init(rll_t *rll)
1597 {
1598 	rll->rll_writer = NULL;
1599 	rll->rll_readers = 0;
1600 	mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1601 	cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1602 }
1603 
1604 static void
1605 ztest_rll_destroy(rll_t *rll)
1606 {
1607 	ASSERT3P(rll->rll_writer, ==, NULL);
1608 	ASSERT0(rll->rll_readers);
1609 	mutex_destroy(&rll->rll_lock);
1610 	cv_destroy(&rll->rll_cv);
1611 }
1612 
1613 static void
1614 ztest_rll_lock(rll_t *rll, rl_type_t type)
1615 {
1616 	mutex_enter(&rll->rll_lock);
1617 
1618 	if (type == RL_READER) {
1619 		while (rll->rll_writer != NULL)
1620 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1621 		rll->rll_readers++;
1622 	} else {
1623 		while (rll->rll_writer != NULL || rll->rll_readers)
1624 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1625 		rll->rll_writer = curthread;
1626 	}
1627 
1628 	mutex_exit(&rll->rll_lock);
1629 }
1630 
1631 static void
1632 ztest_rll_unlock(rll_t *rll)
1633 {
1634 	mutex_enter(&rll->rll_lock);
1635 
1636 	if (rll->rll_writer) {
1637 		ASSERT0(rll->rll_readers);
1638 		rll->rll_writer = NULL;
1639 	} else {
1640 		ASSERT3S(rll->rll_readers, >, 0);
1641 		ASSERT3P(rll->rll_writer, ==, NULL);
1642 		rll->rll_readers--;
1643 	}
1644 
1645 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1646 		cv_broadcast(&rll->rll_cv);
1647 
1648 	mutex_exit(&rll->rll_lock);
1649 }
1650 
1651 static void
1652 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1653 {
1654 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1655 
1656 	ztest_rll_lock(rll, type);
1657 }
1658 
1659 static void
1660 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1661 {
1662 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1663 
1664 	ztest_rll_unlock(rll);
1665 }
1666 
1667 static rl_t *
1668 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1669     uint64_t size, rl_type_t type)
1670 {
1671 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1672 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1673 	rl_t *rl;
1674 
1675 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1676 	rl->rl_object = object;
1677 	rl->rl_offset = offset;
1678 	rl->rl_size = size;
1679 	rl->rl_lock = rll;
1680 
1681 	ztest_rll_lock(rll, type);
1682 
1683 	return (rl);
1684 }
1685 
1686 static void
1687 ztest_range_unlock(rl_t *rl)
1688 {
1689 	rll_t *rll = rl->rl_lock;
1690 
1691 	ztest_rll_unlock(rll);
1692 
1693 	umem_free(rl, sizeof (*rl));
1694 }
1695 
1696 static void
1697 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1698 {
1699 	zd->zd_os = os;
1700 	zd->zd_zilog = dmu_objset_zil(os);
1701 	zd->zd_shared = szd;
1702 	dmu_objset_name(os, zd->zd_name);
1703 	int l;
1704 
1705 	if (zd->zd_shared != NULL)
1706 		zd->zd_shared->zd_seq = 0;
1707 
1708 	VERIFY0(pthread_rwlock_init(&zd->zd_zilog_lock, NULL));
1709 	mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1710 
1711 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1712 		ztest_rll_init(&zd->zd_object_lock[l]);
1713 
1714 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1715 		ztest_rll_init(&zd->zd_range_lock[l]);
1716 }
1717 
1718 static void
1719 ztest_zd_fini(ztest_ds_t *zd)
1720 {
1721 	int l;
1722 
1723 	mutex_destroy(&zd->zd_dirobj_lock);
1724 	(void) pthread_rwlock_destroy(&zd->zd_zilog_lock);
1725 
1726 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1727 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1728 
1729 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1730 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1731 }
1732 
1733 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1734 
1735 static uint64_t
1736 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1737 {
1738 	uint64_t txg;
1739 	int error;
1740 
1741 	/*
1742 	 * Attempt to assign tx to some transaction group.
1743 	 */
1744 	error = dmu_tx_assign(tx, txg_how);
1745 	if (error) {
1746 		if (error == ERESTART) {
1747 			ASSERT3U(txg_how, ==, TXG_NOWAIT);
1748 			dmu_tx_wait(tx);
1749 		} else {
1750 			ASSERT3U(error, ==, ENOSPC);
1751 			ztest_record_enospc(tag);
1752 		}
1753 		dmu_tx_abort(tx);
1754 		return (0);
1755 	}
1756 	txg = dmu_tx_get_txg(tx);
1757 	ASSERT3U(txg, !=, 0);
1758 	return (txg);
1759 }
1760 
1761 static void
1762 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1763     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1764     uint64_t crtxg)
1765 {
1766 	bt->bt_magic = BT_MAGIC;
1767 	bt->bt_objset = dmu_objset_id(os);
1768 	bt->bt_object = object;
1769 	bt->bt_dnodesize = dnodesize;
1770 	bt->bt_offset = offset;
1771 	bt->bt_gen = gen;
1772 	bt->bt_txg = txg;
1773 	bt->bt_crtxg = crtxg;
1774 }
1775 
1776 static void
1777 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1778     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1779     uint64_t crtxg)
1780 {
1781 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1782 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1783 	ASSERT3U(bt->bt_object, ==, object);
1784 	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1785 	ASSERT3U(bt->bt_offset, ==, offset);
1786 	ASSERT3U(bt->bt_gen, <=, gen);
1787 	ASSERT3U(bt->bt_txg, <=, txg);
1788 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1789 }
1790 
1791 static ztest_block_tag_t *
1792 ztest_bt_bonus(dmu_buf_t *db)
1793 {
1794 	dmu_object_info_t doi;
1795 	ztest_block_tag_t *bt;
1796 
1797 	dmu_object_info_from_db(db, &doi);
1798 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1799 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1800 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1801 
1802 	return (bt);
1803 }
1804 
1805 /*
1806  * Generate a token to fill up unused bonus buffer space.  Try to make
1807  * it unique to the object, generation, and offset to verify that data
1808  * is not getting overwritten by data from other dnodes.
1809  */
1810 #define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1811 	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1812 
1813 /*
1814  * Fill up the unused bonus buffer region before the block tag with a
1815  * verifiable pattern. Filling the whole bonus area with non-zero data
1816  * helps ensure that all dnode traversal code properly skips the
1817  * interior regions of large dnodes.
1818  */
1819 static void
1820 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1821     objset_t *os, uint64_t gen)
1822 {
1823 	uint64_t *bonusp;
1824 
1825 	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1826 
1827 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1828 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1829 		    gen, bonusp - (uint64_t *)db->db_data);
1830 		*bonusp = token;
1831 	}
1832 }
1833 
1834 /*
1835  * Verify that the unused area of a bonus buffer is filled with the
1836  * expected tokens.
1837  */
1838 static void
1839 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1840     objset_t *os, uint64_t gen)
1841 {
1842 	uint64_t *bonusp;
1843 
1844 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1845 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1846 		    gen, bonusp - (uint64_t *)db->db_data);
1847 		VERIFY3U(*bonusp, ==, token);
1848 	}
1849 }
1850 
1851 /*
1852  * ZIL logging ops
1853  */
1854 
1855 #define	lrz_type	lr_mode
1856 #define	lrz_blocksize	lr_uid
1857 #define	lrz_ibshift	lr_gid
1858 #define	lrz_bonustype	lr_rdev
1859 #define	lrz_dnodesize	lr_crtime[1]
1860 
1861 static void
1862 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1863 {
1864 	char *name = (void *)(lr + 1);		/* name follows lr */
1865 	size_t namesize = strlen(name) + 1;
1866 	itx_t *itx;
1867 
1868 	if (zil_replaying(zd->zd_zilog, tx))
1869 		return;
1870 
1871 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1872 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1873 	    sizeof (*lr) + namesize - sizeof (lr_t));
1874 
1875 	zil_itx_assign(zd->zd_zilog, itx, tx);
1876 }
1877 
1878 static void
1879 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1880 {
1881 	char *name = (void *)(lr + 1);		/* name follows lr */
1882 	size_t namesize = strlen(name) + 1;
1883 	itx_t *itx;
1884 
1885 	if (zil_replaying(zd->zd_zilog, tx))
1886 		return;
1887 
1888 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1889 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1890 	    sizeof (*lr) + namesize - sizeof (lr_t));
1891 
1892 	itx->itx_oid = object;
1893 	zil_itx_assign(zd->zd_zilog, itx, tx);
1894 }
1895 
1896 static void
1897 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1898 {
1899 	itx_t *itx;
1900 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1901 
1902 	if (zil_replaying(zd->zd_zilog, tx))
1903 		return;
1904 
1905 	if (lr->lr_length > zil_max_log_data(zd->zd_zilog, sizeof (lr_write_t)))
1906 		write_state = WR_INDIRECT;
1907 
1908 	itx = zil_itx_create(TX_WRITE,
1909 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1910 
1911 	if (write_state == WR_COPIED &&
1912 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1913 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1914 		zil_itx_destroy(itx);
1915 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1916 		write_state = WR_NEED_COPY;
1917 	}
1918 	itx->itx_private = zd;
1919 	itx->itx_wr_state = write_state;
1920 	itx->itx_sync = (ztest_random(8) == 0);
1921 
1922 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1923 	    sizeof (*lr) - sizeof (lr_t));
1924 
1925 	zil_itx_assign(zd->zd_zilog, itx, tx);
1926 }
1927 
1928 static void
1929 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1930 {
1931 	itx_t *itx;
1932 
1933 	if (zil_replaying(zd->zd_zilog, tx))
1934 		return;
1935 
1936 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1937 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1938 	    sizeof (*lr) - sizeof (lr_t));
1939 
1940 	itx->itx_sync = B_FALSE;
1941 	zil_itx_assign(zd->zd_zilog, itx, tx);
1942 }
1943 
1944 static void
1945 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1946 {
1947 	itx_t *itx;
1948 
1949 	if (zil_replaying(zd->zd_zilog, tx))
1950 		return;
1951 
1952 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1953 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1954 	    sizeof (*lr) - sizeof (lr_t));
1955 
1956 	itx->itx_sync = B_FALSE;
1957 	zil_itx_assign(zd->zd_zilog, itx, tx);
1958 }
1959 
1960 /*
1961  * ZIL replay ops
1962  */
1963 static int
1964 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1965 {
1966 	ztest_ds_t *zd = arg1;
1967 	lr_create_t *lr = arg2;
1968 	char *name = (void *)(lr + 1);		/* name follows lr */
1969 	objset_t *os = zd->zd_os;
1970 	ztest_block_tag_t *bbt;
1971 	dmu_buf_t *db;
1972 	dmu_tx_t *tx;
1973 	uint64_t txg;
1974 	int error = 0;
1975 	int bonuslen;
1976 
1977 	if (byteswap)
1978 		byteswap_uint64_array(lr, sizeof (*lr));
1979 
1980 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
1981 	ASSERT3S(name[0], !=, '\0');
1982 
1983 	tx = dmu_tx_create(os);
1984 
1985 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1986 
1987 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1988 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1989 	} else {
1990 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1991 	}
1992 
1993 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1994 	if (txg == 0)
1995 		return (ENOSPC);
1996 
1997 	ASSERT3U(dmu_objset_zil(os)->zl_replay, ==, !!lr->lr_foid);
1998 	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1999 
2000 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
2001 		if (lr->lr_foid == 0) {
2002 			lr->lr_foid = zap_create_dnsize(os,
2003 			    lr->lrz_type, lr->lrz_bonustype,
2004 			    bonuslen, lr->lrz_dnodesize, tx);
2005 		} else {
2006 			error = zap_create_claim_dnsize(os, lr->lr_foid,
2007 			    lr->lrz_type, lr->lrz_bonustype,
2008 			    bonuslen, lr->lrz_dnodesize, tx);
2009 		}
2010 	} else {
2011 		if (lr->lr_foid == 0) {
2012 			lr->lr_foid = dmu_object_alloc_dnsize(os,
2013 			    lr->lrz_type, 0, lr->lrz_bonustype,
2014 			    bonuslen, lr->lrz_dnodesize, tx);
2015 		} else {
2016 			error = dmu_object_claim_dnsize(os, lr->lr_foid,
2017 			    lr->lrz_type, 0, lr->lrz_bonustype,
2018 			    bonuslen, lr->lrz_dnodesize, tx);
2019 		}
2020 	}
2021 
2022 	if (error) {
2023 		ASSERT3U(error, ==, EEXIST);
2024 		ASSERT(zd->zd_zilog->zl_replay);
2025 		dmu_tx_commit(tx);
2026 		return (error);
2027 	}
2028 
2029 	ASSERT3U(lr->lr_foid, !=, 0);
2030 
2031 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
2032 		VERIFY0(dmu_object_set_blocksize(os, lr->lr_foid,
2033 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
2034 
2035 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2036 	bbt = ztest_bt_bonus(db);
2037 	dmu_buf_will_dirty(db, tx);
2038 	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
2039 	    lr->lr_gen, txg, txg);
2040 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
2041 	dmu_buf_rele(db, FTAG);
2042 
2043 	VERIFY0(zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
2044 	    &lr->lr_foid, tx));
2045 
2046 	(void) ztest_log_create(zd, tx, lr);
2047 
2048 	dmu_tx_commit(tx);
2049 
2050 	return (0);
2051 }
2052 
2053 static int
2054 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
2055 {
2056 	ztest_ds_t *zd = arg1;
2057 	lr_remove_t *lr = arg2;
2058 	char *name = (void *)(lr + 1);		/* name follows lr */
2059 	objset_t *os = zd->zd_os;
2060 	dmu_object_info_t doi;
2061 	dmu_tx_t *tx;
2062 	uint64_t object, txg;
2063 
2064 	if (byteswap)
2065 		byteswap_uint64_array(lr, sizeof (*lr));
2066 
2067 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
2068 	ASSERT3S(name[0], !=, '\0');
2069 
2070 	VERIFY0(
2071 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
2072 	ASSERT3U(object, !=, 0);
2073 
2074 	ztest_object_lock(zd, object, RL_WRITER);
2075 
2076 	VERIFY0(dmu_object_info(os, object, &doi));
2077 
2078 	tx = dmu_tx_create(os);
2079 
2080 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
2081 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2082 
2083 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2084 	if (txg == 0) {
2085 		ztest_object_unlock(zd, object);
2086 		return (ENOSPC);
2087 	}
2088 
2089 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
2090 		VERIFY0(zap_destroy(os, object, tx));
2091 	} else {
2092 		VERIFY0(dmu_object_free(os, object, tx));
2093 	}
2094 
2095 	VERIFY0(zap_remove(os, lr->lr_doid, name, tx));
2096 
2097 	(void) ztest_log_remove(zd, tx, lr, object);
2098 
2099 	dmu_tx_commit(tx);
2100 
2101 	ztest_object_unlock(zd, object);
2102 
2103 	return (0);
2104 }
2105 
2106 static int
2107 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
2108 {
2109 	ztest_ds_t *zd = arg1;
2110 	lr_write_t *lr = arg2;
2111 	objset_t *os = zd->zd_os;
2112 	void *data = lr + 1;			/* data follows lr */
2113 	uint64_t offset, length;
2114 	ztest_block_tag_t *bt = data;
2115 	ztest_block_tag_t *bbt;
2116 	uint64_t gen, txg, lrtxg, crtxg;
2117 	dmu_object_info_t doi;
2118 	dmu_tx_t *tx;
2119 	dmu_buf_t *db;
2120 	arc_buf_t *abuf = NULL;
2121 	rl_t *rl;
2122 
2123 	if (byteswap)
2124 		byteswap_uint64_array(lr, sizeof (*lr));
2125 
2126 	offset = lr->lr_offset;
2127 	length = lr->lr_length;
2128 
2129 	/* If it's a dmu_sync() block, write the whole block */
2130 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
2131 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
2132 		if (length < blocksize) {
2133 			offset -= offset % blocksize;
2134 			length = blocksize;
2135 		}
2136 	}
2137 
2138 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
2139 		byteswap_uint64_array(bt, sizeof (*bt));
2140 
2141 	if (bt->bt_magic != BT_MAGIC)
2142 		bt = NULL;
2143 
2144 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
2145 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
2146 
2147 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2148 
2149 	dmu_object_info_from_db(db, &doi);
2150 
2151 	bbt = ztest_bt_bonus(db);
2152 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2153 	gen = bbt->bt_gen;
2154 	crtxg = bbt->bt_crtxg;
2155 	lrtxg = lr->lr_common.lrc_txg;
2156 
2157 	tx = dmu_tx_create(os);
2158 
2159 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
2160 
2161 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
2162 	    P2PHASE(offset, length) == 0)
2163 		abuf = dmu_request_arcbuf(db, length);
2164 
2165 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2166 	if (txg == 0) {
2167 		if (abuf != NULL)
2168 			dmu_return_arcbuf(abuf);
2169 		dmu_buf_rele(db, FTAG);
2170 		ztest_range_unlock(rl);
2171 		ztest_object_unlock(zd, lr->lr_foid);
2172 		return (ENOSPC);
2173 	}
2174 
2175 	if (bt != NULL) {
2176 		/*
2177 		 * Usually, verify the old data before writing new data --
2178 		 * but not always, because we also want to verify correct
2179 		 * behavior when the data was not recently read into cache.
2180 		 */
2181 		ASSERT(doi.doi_data_block_size);
2182 		ASSERT0(offset % doi.doi_data_block_size);
2183 		if (ztest_random(4) != 0) {
2184 			int prefetch = ztest_random(2) ?
2185 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
2186 			ztest_block_tag_t rbt;
2187 
2188 			VERIFY(dmu_read(os, lr->lr_foid, offset,
2189 			    sizeof (rbt), &rbt, prefetch) == 0);
2190 			if (rbt.bt_magic == BT_MAGIC) {
2191 				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
2192 				    offset, gen, txg, crtxg);
2193 			}
2194 		}
2195 
2196 		/*
2197 		 * Writes can appear to be newer than the bonus buffer because
2198 		 * the ztest_get_data() callback does a dmu_read() of the
2199 		 * open-context data, which may be different than the data
2200 		 * as it was when the write was generated.
2201 		 */
2202 		if (zd->zd_zilog->zl_replay) {
2203 			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
2204 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
2205 			    bt->bt_crtxg);
2206 		}
2207 
2208 		/*
2209 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
2210 		 * so that all of the usual ASSERTs will work.
2211 		 */
2212 		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
2213 		    crtxg);
2214 	}
2215 
2216 	if (abuf == NULL) {
2217 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
2218 	} else {
2219 		memcpy(abuf->b_data, data, length);
2220 		VERIFY0(dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx));
2221 	}
2222 
2223 	(void) ztest_log_write(zd, tx, lr);
2224 
2225 	dmu_buf_rele(db, FTAG);
2226 
2227 	dmu_tx_commit(tx);
2228 
2229 	ztest_range_unlock(rl);
2230 	ztest_object_unlock(zd, lr->lr_foid);
2231 
2232 	return (0);
2233 }
2234 
2235 static int
2236 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
2237 {
2238 	ztest_ds_t *zd = arg1;
2239 	lr_truncate_t *lr = arg2;
2240 	objset_t *os = zd->zd_os;
2241 	dmu_tx_t *tx;
2242 	uint64_t txg;
2243 	rl_t *rl;
2244 
2245 	if (byteswap)
2246 		byteswap_uint64_array(lr, sizeof (*lr));
2247 
2248 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
2249 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
2250 	    RL_WRITER);
2251 
2252 	tx = dmu_tx_create(os);
2253 
2254 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
2255 
2256 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2257 	if (txg == 0) {
2258 		ztest_range_unlock(rl);
2259 		ztest_object_unlock(zd, lr->lr_foid);
2260 		return (ENOSPC);
2261 	}
2262 
2263 	VERIFY0(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2264 	    lr->lr_length, tx));
2265 
2266 	(void) ztest_log_truncate(zd, tx, lr);
2267 
2268 	dmu_tx_commit(tx);
2269 
2270 	ztest_range_unlock(rl);
2271 	ztest_object_unlock(zd, lr->lr_foid);
2272 
2273 	return (0);
2274 }
2275 
2276 static int
2277 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
2278 {
2279 	ztest_ds_t *zd = arg1;
2280 	lr_setattr_t *lr = arg2;
2281 	objset_t *os = zd->zd_os;
2282 	dmu_tx_t *tx;
2283 	dmu_buf_t *db;
2284 	ztest_block_tag_t *bbt;
2285 	uint64_t txg, lrtxg, crtxg, dnodesize;
2286 
2287 	if (byteswap)
2288 		byteswap_uint64_array(lr, sizeof (*lr));
2289 
2290 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
2291 
2292 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2293 
2294 	tx = dmu_tx_create(os);
2295 	dmu_tx_hold_bonus(tx, lr->lr_foid);
2296 
2297 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2298 	if (txg == 0) {
2299 		dmu_buf_rele(db, FTAG);
2300 		ztest_object_unlock(zd, lr->lr_foid);
2301 		return (ENOSPC);
2302 	}
2303 
2304 	bbt = ztest_bt_bonus(db);
2305 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2306 	crtxg = bbt->bt_crtxg;
2307 	lrtxg = lr->lr_common.lrc_txg;
2308 	dnodesize = bbt->bt_dnodesize;
2309 
2310 	if (zd->zd_zilog->zl_replay) {
2311 		ASSERT3U(lr->lr_size, !=, 0);
2312 		ASSERT3U(lr->lr_mode, !=, 0);
2313 		ASSERT3U(lrtxg, !=, 0);
2314 	} else {
2315 		/*
2316 		 * Randomly change the size and increment the generation.
2317 		 */
2318 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2319 		    sizeof (*bbt);
2320 		lr->lr_mode = bbt->bt_gen + 1;
2321 		ASSERT0(lrtxg);
2322 	}
2323 
2324 	/*
2325 	 * Verify that the current bonus buffer is not newer than our txg.
2326 	 */
2327 	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2328 	    MAX(txg, lrtxg), crtxg);
2329 
2330 	dmu_buf_will_dirty(db, tx);
2331 
2332 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2333 	ASSERT3U(lr->lr_size, <=, db->db_size);
2334 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2335 	bbt = ztest_bt_bonus(db);
2336 
2337 	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2338 	    txg, crtxg);
2339 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2340 	dmu_buf_rele(db, FTAG);
2341 
2342 	(void) ztest_log_setattr(zd, tx, lr);
2343 
2344 	dmu_tx_commit(tx);
2345 
2346 	ztest_object_unlock(zd, lr->lr_foid);
2347 
2348 	return (0);
2349 }
2350 
2351 static zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2352 	NULL,			/* 0 no such transaction type */
2353 	ztest_replay_create,	/* TX_CREATE */
2354 	NULL,			/* TX_MKDIR */
2355 	NULL,			/* TX_MKXATTR */
2356 	NULL,			/* TX_SYMLINK */
2357 	ztest_replay_remove,	/* TX_REMOVE */
2358 	NULL,			/* TX_RMDIR */
2359 	NULL,			/* TX_LINK */
2360 	NULL,			/* TX_RENAME */
2361 	ztest_replay_write,	/* TX_WRITE */
2362 	ztest_replay_truncate,	/* TX_TRUNCATE */
2363 	ztest_replay_setattr,	/* TX_SETATTR */
2364 	NULL,			/* TX_ACL */
2365 	NULL,			/* TX_CREATE_ACL */
2366 	NULL,			/* TX_CREATE_ATTR */
2367 	NULL,			/* TX_CREATE_ACL_ATTR */
2368 	NULL,			/* TX_MKDIR_ACL */
2369 	NULL,			/* TX_MKDIR_ATTR */
2370 	NULL,			/* TX_MKDIR_ACL_ATTR */
2371 	NULL,			/* TX_WRITE2 */
2372 	NULL,			/* TX_SETSAXATTR */
2373 	NULL,			/* TX_RENAME_EXCHANGE */
2374 	NULL,			/* TX_RENAME_WHITEOUT */
2375 };
2376 
2377 /*
2378  * ZIL get_data callbacks
2379  */
2380 
2381 static void
2382 ztest_get_done(zgd_t *zgd, int error)
2383 {
2384 	(void) error;
2385 	ztest_ds_t *zd = zgd->zgd_private;
2386 	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
2387 
2388 	if (zgd->zgd_db)
2389 		dmu_buf_rele(zgd->zgd_db, zgd);
2390 
2391 	ztest_range_unlock((rl_t *)zgd->zgd_lr);
2392 	ztest_object_unlock(zd, object);
2393 
2394 	umem_free(zgd, sizeof (*zgd));
2395 }
2396 
2397 static int
2398 ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
2399     struct lwb *lwb, zio_t *zio)
2400 {
2401 	(void) arg2;
2402 	ztest_ds_t *zd = arg;
2403 	objset_t *os = zd->zd_os;
2404 	uint64_t object = lr->lr_foid;
2405 	uint64_t offset = lr->lr_offset;
2406 	uint64_t size = lr->lr_length;
2407 	uint64_t txg = lr->lr_common.lrc_txg;
2408 	uint64_t crtxg;
2409 	dmu_object_info_t doi;
2410 	dmu_buf_t *db;
2411 	zgd_t *zgd;
2412 	int error;
2413 
2414 	ASSERT3P(lwb, !=, NULL);
2415 	ASSERT3P(zio, !=, NULL);
2416 	ASSERT3U(size, !=, 0);
2417 
2418 	ztest_object_lock(zd, object, RL_READER);
2419 	error = dmu_bonus_hold(os, object, FTAG, &db);
2420 	if (error) {
2421 		ztest_object_unlock(zd, object);
2422 		return (error);
2423 	}
2424 
2425 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
2426 
2427 	if (crtxg == 0 || crtxg > txg) {
2428 		dmu_buf_rele(db, FTAG);
2429 		ztest_object_unlock(zd, object);
2430 		return (ENOENT);
2431 	}
2432 
2433 	dmu_object_info_from_db(db, &doi);
2434 	dmu_buf_rele(db, FTAG);
2435 	db = NULL;
2436 
2437 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2438 	zgd->zgd_lwb = lwb;
2439 	zgd->zgd_private = zd;
2440 
2441 	if (buf != NULL) {	/* immediate write */
2442 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2443 		    object, offset, size, RL_READER);
2444 
2445 		error = dmu_read(os, object, offset, size, buf,
2446 		    DMU_READ_NO_PREFETCH);
2447 		ASSERT0(error);
2448 	} else {
2449 		size = doi.doi_data_block_size;
2450 		if (ISP2(size)) {
2451 			offset = P2ALIGN(offset, size);
2452 		} else {
2453 			ASSERT3U(offset, <, size);
2454 			offset = 0;
2455 		}
2456 
2457 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2458 		    object, offset, size, RL_READER);
2459 
2460 		error = dmu_buf_hold(os, object, offset, zgd, &db,
2461 		    DMU_READ_NO_PREFETCH);
2462 
2463 		if (error == 0) {
2464 			blkptr_t *bp = &lr->lr_blkptr;
2465 
2466 			zgd->zgd_db = db;
2467 			zgd->zgd_bp = bp;
2468 
2469 			ASSERT3U(db->db_offset, ==, offset);
2470 			ASSERT3U(db->db_size, ==, size);
2471 
2472 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2473 			    ztest_get_done, zgd);
2474 
2475 			if (error == 0)
2476 				return (0);
2477 		}
2478 	}
2479 
2480 	ztest_get_done(zgd, error);
2481 
2482 	return (error);
2483 }
2484 
2485 static void *
2486 ztest_lr_alloc(size_t lrsize, char *name)
2487 {
2488 	char *lr;
2489 	size_t namesize = name ? strlen(name) + 1 : 0;
2490 
2491 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2492 
2493 	if (name)
2494 		memcpy(lr + lrsize, name, namesize);
2495 
2496 	return (lr);
2497 }
2498 
2499 static void
2500 ztest_lr_free(void *lr, size_t lrsize, char *name)
2501 {
2502 	size_t namesize = name ? strlen(name) + 1 : 0;
2503 
2504 	umem_free(lr, lrsize + namesize);
2505 }
2506 
2507 /*
2508  * Lookup a bunch of objects.  Returns the number of objects not found.
2509  */
2510 static int
2511 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2512 {
2513 	int missing = 0;
2514 	int error;
2515 	int i;
2516 
2517 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2518 
2519 	for (i = 0; i < count; i++, od++) {
2520 		od->od_object = 0;
2521 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2522 		    sizeof (uint64_t), 1, &od->od_object);
2523 		if (error) {
2524 			ASSERT3S(error, ==, ENOENT);
2525 			ASSERT0(od->od_object);
2526 			missing++;
2527 		} else {
2528 			dmu_buf_t *db;
2529 			ztest_block_tag_t *bbt;
2530 			dmu_object_info_t doi;
2531 
2532 			ASSERT3U(od->od_object, !=, 0);
2533 			ASSERT0(missing);	/* there should be no gaps */
2534 
2535 			ztest_object_lock(zd, od->od_object, RL_READER);
2536 			VERIFY0(dmu_bonus_hold(zd->zd_os, od->od_object,
2537 			    FTAG, &db));
2538 			dmu_object_info_from_db(db, &doi);
2539 			bbt = ztest_bt_bonus(db);
2540 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2541 			od->od_type = doi.doi_type;
2542 			od->od_blocksize = doi.doi_data_block_size;
2543 			od->od_gen = bbt->bt_gen;
2544 			dmu_buf_rele(db, FTAG);
2545 			ztest_object_unlock(zd, od->od_object);
2546 		}
2547 	}
2548 
2549 	return (missing);
2550 }
2551 
2552 static int
2553 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2554 {
2555 	int missing = 0;
2556 	int i;
2557 
2558 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2559 
2560 	for (i = 0; i < count; i++, od++) {
2561 		if (missing) {
2562 			od->od_object = 0;
2563 			missing++;
2564 			continue;
2565 		}
2566 
2567 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2568 
2569 		lr->lr_doid = od->od_dir;
2570 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2571 		lr->lrz_type = od->od_crtype;
2572 		lr->lrz_blocksize = od->od_crblocksize;
2573 		lr->lrz_ibshift = ztest_random_ibshift();
2574 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2575 		lr->lrz_dnodesize = od->od_crdnodesize;
2576 		lr->lr_gen = od->od_crgen;
2577 		lr->lr_crtime[0] = time(NULL);
2578 
2579 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2580 			ASSERT0(missing);
2581 			od->od_object = 0;
2582 			missing++;
2583 		} else {
2584 			od->od_object = lr->lr_foid;
2585 			od->od_type = od->od_crtype;
2586 			od->od_blocksize = od->od_crblocksize;
2587 			od->od_gen = od->od_crgen;
2588 			ASSERT3U(od->od_object, !=, 0);
2589 		}
2590 
2591 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2592 	}
2593 
2594 	return (missing);
2595 }
2596 
2597 static int
2598 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2599 {
2600 	int missing = 0;
2601 	int error;
2602 	int i;
2603 
2604 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2605 
2606 	od += count - 1;
2607 
2608 	for (i = count - 1; i >= 0; i--, od--) {
2609 		if (missing) {
2610 			missing++;
2611 			continue;
2612 		}
2613 
2614 		/*
2615 		 * No object was found.
2616 		 */
2617 		if (od->od_object == 0)
2618 			continue;
2619 
2620 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2621 
2622 		lr->lr_doid = od->od_dir;
2623 
2624 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2625 			ASSERT3U(error, ==, ENOSPC);
2626 			missing++;
2627 		} else {
2628 			od->od_object = 0;
2629 		}
2630 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2631 	}
2632 
2633 	return (missing);
2634 }
2635 
2636 static int
2637 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2638     void *data)
2639 {
2640 	lr_write_t *lr;
2641 	int error;
2642 
2643 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2644 
2645 	lr->lr_foid = object;
2646 	lr->lr_offset = offset;
2647 	lr->lr_length = size;
2648 	lr->lr_blkoff = 0;
2649 	BP_ZERO(&lr->lr_blkptr);
2650 
2651 	memcpy(lr + 1, data, size);
2652 
2653 	error = ztest_replay_write(zd, lr, B_FALSE);
2654 
2655 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2656 
2657 	return (error);
2658 }
2659 
2660 static int
2661 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2662 {
2663 	lr_truncate_t *lr;
2664 	int error;
2665 
2666 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2667 
2668 	lr->lr_foid = object;
2669 	lr->lr_offset = offset;
2670 	lr->lr_length = size;
2671 
2672 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2673 
2674 	ztest_lr_free(lr, sizeof (*lr), NULL);
2675 
2676 	return (error);
2677 }
2678 
2679 static int
2680 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2681 {
2682 	lr_setattr_t *lr;
2683 	int error;
2684 
2685 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2686 
2687 	lr->lr_foid = object;
2688 	lr->lr_size = 0;
2689 	lr->lr_mode = 0;
2690 
2691 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2692 
2693 	ztest_lr_free(lr, sizeof (*lr), NULL);
2694 
2695 	return (error);
2696 }
2697 
2698 static void
2699 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2700 {
2701 	objset_t *os = zd->zd_os;
2702 	dmu_tx_t *tx;
2703 	uint64_t txg;
2704 	rl_t *rl;
2705 
2706 	txg_wait_synced(dmu_objset_pool(os), 0);
2707 
2708 	ztest_object_lock(zd, object, RL_READER);
2709 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2710 
2711 	tx = dmu_tx_create(os);
2712 
2713 	dmu_tx_hold_write(tx, object, offset, size);
2714 
2715 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2716 
2717 	if (txg != 0) {
2718 		dmu_prealloc(os, object, offset, size, tx);
2719 		dmu_tx_commit(tx);
2720 		txg_wait_synced(dmu_objset_pool(os), txg);
2721 	} else {
2722 		(void) dmu_free_long_range(os, object, offset, size);
2723 	}
2724 
2725 	ztest_range_unlock(rl);
2726 	ztest_object_unlock(zd, object);
2727 }
2728 
2729 static void
2730 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2731 {
2732 	int err;
2733 	ztest_block_tag_t wbt;
2734 	dmu_object_info_t doi;
2735 	enum ztest_io_type io_type;
2736 	uint64_t blocksize;
2737 	void *data;
2738 
2739 	VERIFY0(dmu_object_info(zd->zd_os, object, &doi));
2740 	blocksize = doi.doi_data_block_size;
2741 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2742 
2743 	/*
2744 	 * Pick an i/o type at random, biased toward writing block tags.
2745 	 */
2746 	io_type = ztest_random(ZTEST_IO_TYPES);
2747 	if (ztest_random(2) == 0)
2748 		io_type = ZTEST_IO_WRITE_TAG;
2749 
2750 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2751 
2752 	switch (io_type) {
2753 
2754 	case ZTEST_IO_WRITE_TAG:
2755 		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2756 		    offset, 0, 0, 0);
2757 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2758 		break;
2759 
2760 	case ZTEST_IO_WRITE_PATTERN:
2761 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2762 		if (ztest_random(2) == 0) {
2763 			/*
2764 			 * Induce fletcher2 collisions to ensure that
2765 			 * zio_ddt_collision() detects and resolves them
2766 			 * when using fletcher2-verify for deduplication.
2767 			 */
2768 			((uint64_t *)data)[0] ^= 1ULL << 63;
2769 			((uint64_t *)data)[4] ^= 1ULL << 63;
2770 		}
2771 		(void) ztest_write(zd, object, offset, blocksize, data);
2772 		break;
2773 
2774 	case ZTEST_IO_WRITE_ZEROES:
2775 		memset(data, 0, blocksize);
2776 		(void) ztest_write(zd, object, offset, blocksize, data);
2777 		break;
2778 
2779 	case ZTEST_IO_TRUNCATE:
2780 		(void) ztest_truncate(zd, object, offset, blocksize);
2781 		break;
2782 
2783 	case ZTEST_IO_SETATTR:
2784 		(void) ztest_setattr(zd, object);
2785 		break;
2786 	default:
2787 		break;
2788 
2789 	case ZTEST_IO_REWRITE:
2790 		(void) pthread_rwlock_rdlock(&ztest_name_lock);
2791 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2792 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2793 		    B_FALSE);
2794 		ASSERT(err == 0 || err == ENOSPC);
2795 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2796 		    ZFS_PROP_COMPRESSION,
2797 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2798 		    B_FALSE);
2799 		ASSERT(err == 0 || err == ENOSPC);
2800 		(void) pthread_rwlock_unlock(&ztest_name_lock);
2801 
2802 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2803 		    DMU_READ_NO_PREFETCH));
2804 
2805 		(void) ztest_write(zd, object, offset, blocksize, data);
2806 		break;
2807 	}
2808 
2809 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2810 
2811 	umem_free(data, blocksize);
2812 }
2813 
2814 /*
2815  * Initialize an object description template.
2816  */
2817 static void
2818 ztest_od_init(ztest_od_t *od, uint64_t id, const char *tag, uint64_t index,
2819     dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2820     uint64_t gen)
2821 {
2822 	od->od_dir = ZTEST_DIROBJ;
2823 	od->od_object = 0;
2824 
2825 	od->od_crtype = type;
2826 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2827 	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2828 	od->od_crgen = gen;
2829 
2830 	od->od_type = DMU_OT_NONE;
2831 	od->od_blocksize = 0;
2832 	od->od_gen = 0;
2833 
2834 	(void) snprintf(od->od_name, sizeof (od->od_name),
2835 	    "%s(%"PRId64")[%"PRIu64"]",
2836 	    tag, id, index);
2837 }
2838 
2839 /*
2840  * Lookup or create the objects for a test using the od template.
2841  * If the objects do not all exist, or if 'remove' is specified,
2842  * remove any existing objects and create new ones.  Otherwise,
2843  * use the existing objects.
2844  */
2845 static int
2846 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2847 {
2848 	int count = size / sizeof (*od);
2849 	int rv = 0;
2850 
2851 	mutex_enter(&zd->zd_dirobj_lock);
2852 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2853 	    (ztest_remove(zd, od, count) != 0 ||
2854 	    ztest_create(zd, od, count) != 0))
2855 		rv = -1;
2856 	zd->zd_od = od;
2857 	mutex_exit(&zd->zd_dirobj_lock);
2858 
2859 	return (rv);
2860 }
2861 
2862 void
2863 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2864 {
2865 	(void) id;
2866 	zilog_t *zilog = zd->zd_zilog;
2867 
2868 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2869 
2870 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2871 
2872 	/*
2873 	 * Remember the committed values in zd, which is in parent/child
2874 	 * shared memory.  If we die, the next iteration of ztest_run()
2875 	 * will verify that the log really does contain this record.
2876 	 */
2877 	mutex_enter(&zilog->zl_lock);
2878 	ASSERT3P(zd->zd_shared, !=, NULL);
2879 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2880 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2881 	mutex_exit(&zilog->zl_lock);
2882 
2883 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2884 }
2885 
2886 /*
2887  * This function is designed to simulate the operations that occur during a
2888  * mount/unmount operation.  We hold the dataset across these operations in an
2889  * attempt to expose any implicit assumptions about ZIL management.
2890  */
2891 void
2892 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2893 {
2894 	(void) id;
2895 	objset_t *os = zd->zd_os;
2896 
2897 	/*
2898 	 * We hold the ztest_vdev_lock so we don't cause problems with
2899 	 * other threads that wish to remove a log device, such as
2900 	 * ztest_device_removal().
2901 	 */
2902 	mutex_enter(&ztest_vdev_lock);
2903 
2904 	/*
2905 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2906 	 * updating the zil (i.e. adding in-memory log records) and the
2907 	 * zd_zilog_lock to block any I/O.
2908 	 */
2909 	mutex_enter(&zd->zd_dirobj_lock);
2910 	(void) pthread_rwlock_wrlock(&zd->zd_zilog_lock);
2911 
2912 	/* zfsvfs_teardown() */
2913 	zil_close(zd->zd_zilog);
2914 
2915 	/* zfsvfs_setup() */
2916 	VERIFY3P(zil_open(os, ztest_get_data, NULL), ==, zd->zd_zilog);
2917 	zil_replay(os, zd, ztest_replay_vector);
2918 
2919 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2920 	mutex_exit(&zd->zd_dirobj_lock);
2921 	mutex_exit(&ztest_vdev_lock);
2922 }
2923 
2924 /*
2925  * Verify that we can't destroy an active pool, create an existing pool,
2926  * or create a pool with a bad vdev spec.
2927  */
2928 void
2929 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2930 {
2931 	(void) zd, (void) id;
2932 	ztest_shared_opts_t *zo = &ztest_opts;
2933 	spa_t *spa;
2934 	nvlist_t *nvroot;
2935 
2936 	if (zo->zo_mmp_test)
2937 		return;
2938 
2939 	/*
2940 	 * Attempt to create using a bad file.
2941 	 */
2942 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2943 	VERIFY3U(ENOENT, ==,
2944 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2945 	fnvlist_free(nvroot);
2946 
2947 	/*
2948 	 * Attempt to create using a bad mirror.
2949 	 */
2950 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
2951 	VERIFY3U(ENOENT, ==,
2952 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2953 	fnvlist_free(nvroot);
2954 
2955 	/*
2956 	 * Attempt to create an existing pool.  It shouldn't matter
2957 	 * what's in the nvroot; we should fail with EEXIST.
2958 	 */
2959 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
2960 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2961 	VERIFY3U(EEXIST, ==,
2962 	    spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2963 	fnvlist_free(nvroot);
2964 
2965 	/*
2966 	 * We open a reference to the spa and then we try to export it
2967 	 * expecting one of the following errors:
2968 	 *
2969 	 * EBUSY
2970 	 *	Because of the reference we just opened.
2971 	 *
2972 	 * ZFS_ERR_EXPORT_IN_PROGRESS
2973 	 *	For the case that there is another ztest thread doing
2974 	 *	an export concurrently.
2975 	 */
2976 	VERIFY0(spa_open(zo->zo_pool, &spa, FTAG));
2977 	int error = spa_destroy(zo->zo_pool);
2978 	if (error != EBUSY && error != ZFS_ERR_EXPORT_IN_PROGRESS) {
2979 		fatal(B_FALSE, "spa_destroy(%s) returned unexpected value %d",
2980 		    spa->spa_name, error);
2981 	}
2982 	spa_close(spa, FTAG);
2983 
2984 	(void) pthread_rwlock_unlock(&ztest_name_lock);
2985 }
2986 
2987 /*
2988  * Start and then stop the MMP threads to ensure the startup and shutdown code
2989  * works properly.  Actual protection and property-related code tested via ZTS.
2990  */
2991 void
2992 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2993 {
2994 	(void) zd, (void) id;
2995 	ztest_shared_opts_t *zo = &ztest_opts;
2996 	spa_t *spa = ztest_spa;
2997 
2998 	if (zo->zo_mmp_test)
2999 		return;
3000 
3001 	/*
3002 	 * Since enabling MMP involves setting a property, it could not be done
3003 	 * while the pool is suspended.
3004 	 */
3005 	if (spa_suspended(spa))
3006 		return;
3007 
3008 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3009 	mutex_enter(&spa->spa_props_lock);
3010 
3011 	zfs_multihost_fail_intervals = 0;
3012 
3013 	if (!spa_multihost(spa)) {
3014 		spa->spa_multihost = B_TRUE;
3015 		mmp_thread_start(spa);
3016 	}
3017 
3018 	mutex_exit(&spa->spa_props_lock);
3019 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3020 
3021 	txg_wait_synced(spa_get_dsl(spa), 0);
3022 	mmp_signal_all_threads();
3023 	txg_wait_synced(spa_get_dsl(spa), 0);
3024 
3025 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3026 	mutex_enter(&spa->spa_props_lock);
3027 
3028 	if (spa_multihost(spa)) {
3029 		mmp_thread_stop(spa);
3030 		spa->spa_multihost = B_FALSE;
3031 	}
3032 
3033 	mutex_exit(&spa->spa_props_lock);
3034 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3035 }
3036 
3037 void
3038 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
3039 {
3040 	(void) zd, (void) id;
3041 	spa_t *spa;
3042 	uint64_t initial_version = SPA_VERSION_INITIAL;
3043 	uint64_t version, newversion;
3044 	nvlist_t *nvroot, *props;
3045 	char *name;
3046 
3047 	if (ztest_opts.zo_mmp_test)
3048 		return;
3049 
3050 	/* dRAID added after feature flags, skip upgrade test. */
3051 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0)
3052 		return;
3053 
3054 	mutex_enter(&ztest_vdev_lock);
3055 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
3056 
3057 	/*
3058 	 * Clean up from previous runs.
3059 	 */
3060 	(void) spa_destroy(name);
3061 
3062 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
3063 	    NULL, ztest_opts.zo_raid_children, ztest_opts.zo_mirrors, 1);
3064 
3065 	/*
3066 	 * If we're configuring a RAIDZ device then make sure that the
3067 	 * initial version is capable of supporting that feature.
3068 	 */
3069 	switch (ztest_opts.zo_raid_parity) {
3070 	case 0:
3071 	case 1:
3072 		initial_version = SPA_VERSION_INITIAL;
3073 		break;
3074 	case 2:
3075 		initial_version = SPA_VERSION_RAIDZ2;
3076 		break;
3077 	case 3:
3078 		initial_version = SPA_VERSION_RAIDZ3;
3079 		break;
3080 	}
3081 
3082 	/*
3083 	 * Create a pool with a spa version that can be upgraded. Pick
3084 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
3085 	 */
3086 	do {
3087 		version = ztest_random_spa_version(initial_version);
3088 	} while (version > SPA_VERSION_BEFORE_FEATURES);
3089 
3090 	props = fnvlist_alloc();
3091 	fnvlist_add_uint64(props,
3092 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
3093 	VERIFY0(spa_create(name, nvroot, props, NULL, NULL));
3094 	fnvlist_free(nvroot);
3095 	fnvlist_free(props);
3096 
3097 	VERIFY0(spa_open(name, &spa, FTAG));
3098 	VERIFY3U(spa_version(spa), ==, version);
3099 	newversion = ztest_random_spa_version(version + 1);
3100 
3101 	if (ztest_opts.zo_verbose >= 4) {
3102 		(void) printf("upgrading spa version from "
3103 		    "%"PRIu64" to %"PRIu64"\n",
3104 		    version, newversion);
3105 	}
3106 
3107 	spa_upgrade(spa, newversion);
3108 	VERIFY3U(spa_version(spa), >, version);
3109 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
3110 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
3111 	spa_close(spa, FTAG);
3112 
3113 	kmem_strfree(name);
3114 	mutex_exit(&ztest_vdev_lock);
3115 }
3116 
3117 static void
3118 ztest_spa_checkpoint(spa_t *spa)
3119 {
3120 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3121 
3122 	int error = spa_checkpoint(spa->spa_name);
3123 
3124 	switch (error) {
3125 	case 0:
3126 	case ZFS_ERR_DEVRM_IN_PROGRESS:
3127 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3128 	case ZFS_ERR_CHECKPOINT_EXISTS:
3129 		break;
3130 	case ENOSPC:
3131 		ztest_record_enospc(FTAG);
3132 		break;
3133 	default:
3134 		fatal(B_FALSE, "spa_checkpoint(%s) = %d", spa->spa_name, error);
3135 	}
3136 }
3137 
3138 static void
3139 ztest_spa_discard_checkpoint(spa_t *spa)
3140 {
3141 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3142 
3143 	int error = spa_checkpoint_discard(spa->spa_name);
3144 
3145 	switch (error) {
3146 	case 0:
3147 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3148 	case ZFS_ERR_NO_CHECKPOINT:
3149 		break;
3150 	default:
3151 		fatal(B_FALSE, "spa_discard_checkpoint(%s) = %d",
3152 		    spa->spa_name, error);
3153 	}
3154 
3155 }
3156 
3157 void
3158 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
3159 {
3160 	(void) zd, (void) id;
3161 	spa_t *spa = ztest_spa;
3162 
3163 	mutex_enter(&ztest_checkpoint_lock);
3164 	if (ztest_random(2) == 0) {
3165 		ztest_spa_checkpoint(spa);
3166 	} else {
3167 		ztest_spa_discard_checkpoint(spa);
3168 	}
3169 	mutex_exit(&ztest_checkpoint_lock);
3170 }
3171 
3172 
3173 static vdev_t *
3174 vdev_lookup_by_path(vdev_t *vd, const char *path)
3175 {
3176 	vdev_t *mvd;
3177 	int c;
3178 
3179 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
3180 		return (vd);
3181 
3182 	for (c = 0; c < vd->vdev_children; c++)
3183 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
3184 		    NULL)
3185 			return (mvd);
3186 
3187 	return (NULL);
3188 }
3189 
3190 static int
3191 spa_num_top_vdevs(spa_t *spa)
3192 {
3193 	vdev_t *rvd = spa->spa_root_vdev;
3194 	ASSERT3U(spa_config_held(spa, SCL_VDEV, RW_READER), ==, SCL_VDEV);
3195 	return (rvd->vdev_children);
3196 }
3197 
3198 /*
3199  * Verify that vdev_add() works as expected.
3200  */
3201 void
3202 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
3203 {
3204 	(void) zd, (void) id;
3205 	ztest_shared_t *zs = ztest_shared;
3206 	spa_t *spa = ztest_spa;
3207 	uint64_t leaves;
3208 	uint64_t guid;
3209 	nvlist_t *nvroot;
3210 	int error;
3211 
3212 	if (ztest_opts.zo_mmp_test)
3213 		return;
3214 
3215 	mutex_enter(&ztest_vdev_lock);
3216 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) *
3217 	    ztest_opts.zo_raid_children;
3218 
3219 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3220 
3221 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3222 
3223 	/*
3224 	 * If we have slogs then remove them 1/4 of the time.
3225 	 */
3226 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
3227 		metaslab_group_t *mg;
3228 
3229 		/*
3230 		 * find the first real slog in log allocation class
3231 		 */
3232 		mg =  spa_log_class(spa)->mc_allocator[0].mca_rotor;
3233 		while (!mg->mg_vd->vdev_islog)
3234 			mg = mg->mg_next;
3235 
3236 		guid = mg->mg_vd->vdev_guid;
3237 
3238 		spa_config_exit(spa, SCL_VDEV, FTAG);
3239 
3240 		/*
3241 		 * We have to grab the zs_name_lock as writer to
3242 		 * prevent a race between removing a slog (dmu_objset_find)
3243 		 * and destroying a dataset. Removing the slog will
3244 		 * grab a reference on the dataset which may cause
3245 		 * dsl_destroy_head() to fail with EBUSY thus
3246 		 * leaving the dataset in an inconsistent state.
3247 		 */
3248 		pthread_rwlock_wrlock(&ztest_name_lock);
3249 		error = spa_vdev_remove(spa, guid, B_FALSE);
3250 		pthread_rwlock_unlock(&ztest_name_lock);
3251 
3252 		switch (error) {
3253 		case 0:
3254 		case EEXIST:	/* Generic zil_reset() error */
3255 		case EBUSY:	/* Replay required */
3256 		case EACCES:	/* Crypto key not loaded */
3257 		case ZFS_ERR_CHECKPOINT_EXISTS:
3258 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3259 			break;
3260 		default:
3261 			fatal(B_FALSE, "spa_vdev_remove() = %d", error);
3262 		}
3263 	} else {
3264 		spa_config_exit(spa, SCL_VDEV, FTAG);
3265 
3266 		/*
3267 		 * Make 1/4 of the devices be log devices
3268 		 */
3269 		nvroot = make_vdev_root(NULL, NULL, NULL,
3270 		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
3271 		    "log" : NULL, ztest_opts.zo_raid_children, zs->zs_mirrors,
3272 		    1);
3273 
3274 		error = spa_vdev_add(spa, nvroot);
3275 		fnvlist_free(nvroot);
3276 
3277 		switch (error) {
3278 		case 0:
3279 			break;
3280 		case ENOSPC:
3281 			ztest_record_enospc("spa_vdev_add");
3282 			break;
3283 		default:
3284 			fatal(B_FALSE, "spa_vdev_add() = %d", error);
3285 		}
3286 	}
3287 
3288 	mutex_exit(&ztest_vdev_lock);
3289 }
3290 
3291 void
3292 ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
3293 {
3294 	(void) zd, (void) id;
3295 	ztest_shared_t *zs = ztest_shared;
3296 	spa_t *spa = ztest_spa;
3297 	uint64_t leaves;
3298 	nvlist_t *nvroot;
3299 	const char *class = (ztest_random(2) == 0) ?
3300 	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
3301 	int error;
3302 
3303 	/*
3304 	 * By default add a special vdev 50% of the time
3305 	 */
3306 	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
3307 	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
3308 	    ztest_random(2) == 0)) {
3309 		return;
3310 	}
3311 
3312 	mutex_enter(&ztest_vdev_lock);
3313 
3314 	/* Only test with mirrors */
3315 	if (zs->zs_mirrors < 2) {
3316 		mutex_exit(&ztest_vdev_lock);
3317 		return;
3318 	}
3319 
3320 	/* requires feature@allocation_classes */
3321 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
3322 		mutex_exit(&ztest_vdev_lock);
3323 		return;
3324 	}
3325 
3326 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) *
3327 	    ztest_opts.zo_raid_children;
3328 
3329 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3330 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3331 	spa_config_exit(spa, SCL_VDEV, FTAG);
3332 
3333 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
3334 	    class, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
3335 
3336 	error = spa_vdev_add(spa, nvroot);
3337 	fnvlist_free(nvroot);
3338 
3339 	if (error == ENOSPC)
3340 		ztest_record_enospc("spa_vdev_add");
3341 	else if (error != 0)
3342 		fatal(B_FALSE, "spa_vdev_add() = %d", error);
3343 
3344 	/*
3345 	 * 50% of the time allow small blocks in the special class
3346 	 */
3347 	if (error == 0 &&
3348 	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
3349 		if (ztest_opts.zo_verbose >= 3)
3350 			(void) printf("Enabling special VDEV small blocks\n");
3351 		error = ztest_dsl_prop_set_uint64(zd->zd_name,
3352 		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
3353 		ASSERT(error == 0 || error == ENOSPC);
3354 	}
3355 
3356 	mutex_exit(&ztest_vdev_lock);
3357 
3358 	if (ztest_opts.zo_verbose >= 3) {
3359 		metaslab_class_t *mc;
3360 
3361 		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3362 			mc = spa_special_class(spa);
3363 		else
3364 			mc = spa_dedup_class(spa);
3365 		(void) printf("Added a %s mirrored vdev (of %d)\n",
3366 		    class, (int)mc->mc_groups);
3367 	}
3368 }
3369 
3370 /*
3371  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3372  */
3373 void
3374 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
3375 {
3376 	(void) zd, (void) id;
3377 	ztest_shared_t *zs = ztest_shared;
3378 	spa_t *spa = ztest_spa;
3379 	vdev_t *rvd = spa->spa_root_vdev;
3380 	spa_aux_vdev_t *sav;
3381 	const char *aux;
3382 	char *path;
3383 	uint64_t guid = 0;
3384 	int error, ignore_err = 0;
3385 
3386 	if (ztest_opts.zo_mmp_test)
3387 		return;
3388 
3389 	path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3390 
3391 	if (ztest_random(2) == 0) {
3392 		sav = &spa->spa_spares;
3393 		aux = ZPOOL_CONFIG_SPARES;
3394 	} else {
3395 		sav = &spa->spa_l2cache;
3396 		aux = ZPOOL_CONFIG_L2CACHE;
3397 	}
3398 
3399 	mutex_enter(&ztest_vdev_lock);
3400 
3401 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3402 
3403 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
3404 		/*
3405 		 * Pick a random device to remove.
3406 		 */
3407 		vdev_t *svd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3408 
3409 		/* dRAID spares cannot be removed; try anyways to see ENOTSUP */
3410 		if (strstr(svd->vdev_path, VDEV_TYPE_DRAID) != NULL)
3411 			ignore_err = ENOTSUP;
3412 
3413 		guid = svd->vdev_guid;
3414 	} else {
3415 		/*
3416 		 * Find an unused device we can add.
3417 		 */
3418 		zs->zs_vdev_aux = 0;
3419 		for (;;) {
3420 			int c;
3421 			(void) snprintf(path, MAXPATHLEN, ztest_aux_template,
3422 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3423 			    zs->zs_vdev_aux);
3424 			for (c = 0; c < sav->sav_count; c++)
3425 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
3426 				    path) == 0)
3427 					break;
3428 			if (c == sav->sav_count &&
3429 			    vdev_lookup_by_path(rvd, path) == NULL)
3430 				break;
3431 			zs->zs_vdev_aux++;
3432 		}
3433 	}
3434 
3435 	spa_config_exit(spa, SCL_VDEV, FTAG);
3436 
3437 	if (guid == 0) {
3438 		/*
3439 		 * Add a new device.
3440 		 */
3441 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3442 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
3443 		error = spa_vdev_add(spa, nvroot);
3444 
3445 		switch (error) {
3446 		case 0:
3447 			break;
3448 		default:
3449 			fatal(B_FALSE, "spa_vdev_add(%p) = %d", nvroot, error);
3450 		}
3451 		fnvlist_free(nvroot);
3452 	} else {
3453 		/*
3454 		 * Remove an existing device.  Sometimes, dirty its
3455 		 * vdev state first to make sure we handle removal
3456 		 * of devices that have pending state changes.
3457 		 */
3458 		if (ztest_random(2) == 0)
3459 			(void) vdev_online(spa, guid, 0, NULL);
3460 
3461 		error = spa_vdev_remove(spa, guid, B_FALSE);
3462 
3463 		switch (error) {
3464 		case 0:
3465 		case EBUSY:
3466 		case ZFS_ERR_CHECKPOINT_EXISTS:
3467 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3468 			break;
3469 		default:
3470 			if (error != ignore_err)
3471 				fatal(B_FALSE,
3472 				    "spa_vdev_remove(%"PRIu64") = %d",
3473 				    guid, error);
3474 		}
3475 	}
3476 
3477 	mutex_exit(&ztest_vdev_lock);
3478 
3479 	umem_free(path, MAXPATHLEN);
3480 }
3481 
3482 /*
3483  * split a pool if it has mirror tlvdevs
3484  */
3485 void
3486 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3487 {
3488 	(void) zd, (void) id;
3489 	ztest_shared_t *zs = ztest_shared;
3490 	spa_t *spa = ztest_spa;
3491 	vdev_t *rvd = spa->spa_root_vdev;
3492 	nvlist_t *tree, **child, *config, *split, **schild;
3493 	uint_t c, children, schildren = 0, lastlogid = 0;
3494 	int error = 0;
3495 
3496 	if (ztest_opts.zo_mmp_test)
3497 		return;
3498 
3499 	mutex_enter(&ztest_vdev_lock);
3500 
3501 	/* ensure we have a usable config; mirrors of raidz aren't supported */
3502 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raid_children > 1) {
3503 		mutex_exit(&ztest_vdev_lock);
3504 		return;
3505 	}
3506 
3507 	/* clean up the old pool, if any */
3508 	(void) spa_destroy("splitp");
3509 
3510 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3511 
3512 	/* generate a config from the existing config */
3513 	mutex_enter(&spa->spa_props_lock);
3514 	tree = fnvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE);
3515 	mutex_exit(&spa->spa_props_lock);
3516 
3517 	VERIFY0(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN,
3518 	    &child, &children));
3519 
3520 	schild = umem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
3521 	    UMEM_NOFAIL);
3522 	for (c = 0; c < children; c++) {
3523 		vdev_t *tvd = rvd->vdev_child[c];
3524 		nvlist_t **mchild;
3525 		uint_t mchildren;
3526 
3527 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3528 			schild[schildren] = fnvlist_alloc();
3529 			fnvlist_add_string(schild[schildren],
3530 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE);
3531 			fnvlist_add_uint64(schild[schildren],
3532 			    ZPOOL_CONFIG_IS_HOLE, 1);
3533 			if (lastlogid == 0)
3534 				lastlogid = schildren;
3535 			++schildren;
3536 			continue;
3537 		}
3538 		lastlogid = 0;
3539 		VERIFY0(nvlist_lookup_nvlist_array(child[c],
3540 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren));
3541 		schild[schildren++] = fnvlist_dup(mchild[0]);
3542 	}
3543 
3544 	/* OK, create a config that can be used to split */
3545 	split = fnvlist_alloc();
3546 	fnvlist_add_string(split, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
3547 	fnvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN,
3548 	    (const nvlist_t **)schild, lastlogid != 0 ? lastlogid : schildren);
3549 
3550 	config = fnvlist_alloc();
3551 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split);
3552 
3553 	for (c = 0; c < schildren; c++)
3554 		fnvlist_free(schild[c]);
3555 	umem_free(schild, rvd->vdev_children * sizeof (nvlist_t *));
3556 	fnvlist_free(split);
3557 
3558 	spa_config_exit(spa, SCL_VDEV, FTAG);
3559 
3560 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
3561 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3562 	(void) pthread_rwlock_unlock(&ztest_name_lock);
3563 
3564 	fnvlist_free(config);
3565 
3566 	if (error == 0) {
3567 		(void) printf("successful split - results:\n");
3568 		mutex_enter(&spa_namespace_lock);
3569 		show_pool_stats(spa);
3570 		show_pool_stats(spa_lookup("splitp"));
3571 		mutex_exit(&spa_namespace_lock);
3572 		++zs->zs_splits;
3573 		--zs->zs_mirrors;
3574 	}
3575 	mutex_exit(&ztest_vdev_lock);
3576 }
3577 
3578 /*
3579  * Verify that we can attach and detach devices.
3580  */
3581 void
3582 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3583 {
3584 	(void) zd, (void) id;
3585 	ztest_shared_t *zs = ztest_shared;
3586 	spa_t *spa = ztest_spa;
3587 	spa_aux_vdev_t *sav = &spa->spa_spares;
3588 	vdev_t *rvd = spa->spa_root_vdev;
3589 	vdev_t *oldvd, *newvd, *pvd;
3590 	nvlist_t *root;
3591 	uint64_t leaves;
3592 	uint64_t leaf, top;
3593 	uint64_t ashift = ztest_get_ashift();
3594 	uint64_t oldguid, pguid;
3595 	uint64_t oldsize, newsize;
3596 	char *oldpath, *newpath;
3597 	int replacing;
3598 	int oldvd_has_siblings = B_FALSE;
3599 	int newvd_is_spare = B_FALSE;
3600 	int newvd_is_dspare = B_FALSE;
3601 	int oldvd_is_log;
3602 	int oldvd_is_special;
3603 	int error, expected_error;
3604 
3605 	if (ztest_opts.zo_mmp_test)
3606 		return;
3607 
3608 	oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3609 	newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3610 
3611 	mutex_enter(&ztest_vdev_lock);
3612 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raid_children;
3613 
3614 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3615 
3616 	/*
3617 	 * If a vdev is in the process of being removed, its removal may
3618 	 * finish while we are in progress, leading to an unexpected error
3619 	 * value.  Don't bother trying to attach while we are in the middle
3620 	 * of removal.
3621 	 */
3622 	if (ztest_device_removal_active) {
3623 		spa_config_exit(spa, SCL_ALL, FTAG);
3624 		goto out;
3625 	}
3626 
3627 	/*
3628 	 * Decide whether to do an attach or a replace.
3629 	 */
3630 	replacing = ztest_random(2);
3631 
3632 	/*
3633 	 * Pick a random top-level vdev.
3634 	 */
3635 	top = ztest_random_vdev_top(spa, B_TRUE);
3636 
3637 	/*
3638 	 * Pick a random leaf within it.
3639 	 */
3640 	leaf = ztest_random(leaves);
3641 
3642 	/*
3643 	 * Locate this vdev.
3644 	 */
3645 	oldvd = rvd->vdev_child[top];
3646 
3647 	/* pick a child from the mirror */
3648 	if (zs->zs_mirrors >= 1) {
3649 		ASSERT3P(oldvd->vdev_ops, ==, &vdev_mirror_ops);
3650 		ASSERT3U(oldvd->vdev_children, >=, zs->zs_mirrors);
3651 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raid_children];
3652 	}
3653 
3654 	/* pick a child out of the raidz group */
3655 	if (ztest_opts.zo_raid_children > 1) {
3656 		if (strcmp(oldvd->vdev_ops->vdev_op_type, "raidz") == 0)
3657 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_raidz_ops);
3658 		else
3659 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_draid_ops);
3660 		ASSERT3U(oldvd->vdev_children, ==, ztest_opts.zo_raid_children);
3661 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raid_children];
3662 	}
3663 
3664 	/*
3665 	 * If we're already doing an attach or replace, oldvd may be a
3666 	 * mirror vdev -- in which case, pick a random child.
3667 	 */
3668 	while (oldvd->vdev_children != 0) {
3669 		oldvd_has_siblings = B_TRUE;
3670 		ASSERT3U(oldvd->vdev_children, >=, 2);
3671 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3672 	}
3673 
3674 	oldguid = oldvd->vdev_guid;
3675 	oldsize = vdev_get_min_asize(oldvd);
3676 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3677 	oldvd_is_special =
3678 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_SPECIAL ||
3679 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_DEDUP;
3680 	(void) strlcpy(oldpath, oldvd->vdev_path, MAXPATHLEN);
3681 	pvd = oldvd->vdev_parent;
3682 	pguid = pvd->vdev_guid;
3683 
3684 	/*
3685 	 * If oldvd has siblings, then half of the time, detach it.  Prior
3686 	 * to the detach the pool is scrubbed in order to prevent creating
3687 	 * unrepairable blocks as a result of the data corruption injection.
3688 	 */
3689 	if (oldvd_has_siblings && ztest_random(2) == 0) {
3690 		spa_config_exit(spa, SCL_ALL, FTAG);
3691 
3692 		error = ztest_scrub_impl(spa);
3693 		if (error)
3694 			goto out;
3695 
3696 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3697 		if (error != 0 && error != ENODEV && error != EBUSY &&
3698 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3699 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3700 			fatal(B_FALSE, "detach (%s) returned %d",
3701 			    oldpath, error);
3702 		goto out;
3703 	}
3704 
3705 	/*
3706 	 * For the new vdev, choose with equal probability between the two
3707 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3708 	 */
3709 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3710 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3711 		newvd_is_spare = B_TRUE;
3712 
3713 		if (newvd->vdev_ops == &vdev_draid_spare_ops)
3714 			newvd_is_dspare = B_TRUE;
3715 
3716 		(void) strlcpy(newpath, newvd->vdev_path, MAXPATHLEN);
3717 	} else {
3718 		(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3719 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3720 		    top * leaves + leaf);
3721 		if (ztest_random(2) == 0)
3722 			newpath[strlen(newpath) - 1] = 'b';
3723 		newvd = vdev_lookup_by_path(rvd, newpath);
3724 	}
3725 
3726 	if (newvd) {
3727 		/*
3728 		 * Reopen to ensure the vdev's asize field isn't stale.
3729 		 */
3730 		vdev_reopen(newvd);
3731 		newsize = vdev_get_min_asize(newvd);
3732 	} else {
3733 		/*
3734 		 * Make newsize a little bigger or smaller than oldsize.
3735 		 * If it's smaller, the attach should fail.
3736 		 * If it's larger, and we're doing a replace,
3737 		 * we should get dynamic LUN growth when we're done.
3738 		 */
3739 		newsize = 10 * oldsize / (9 + ztest_random(3));
3740 	}
3741 
3742 	/*
3743 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3744 	 * unless it's a replace; in that case any non-replacing parent is OK.
3745 	 *
3746 	 * If newvd is already part of the pool, it should fail with EBUSY.
3747 	 *
3748 	 * If newvd is too small, it should fail with EOVERFLOW.
3749 	 *
3750 	 * If newvd is a distributed spare and it's being attached to a
3751 	 * dRAID which is not its parent it should fail with EINVAL.
3752 	 */
3753 	if (pvd->vdev_ops != &vdev_mirror_ops &&
3754 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3755 	    pvd->vdev_ops == &vdev_replacing_ops ||
3756 	    pvd->vdev_ops == &vdev_spare_ops))
3757 		expected_error = ENOTSUP;
3758 	else if (newvd_is_spare &&
3759 	    (!replacing || oldvd_is_log || oldvd_is_special))
3760 		expected_error = ENOTSUP;
3761 	else if (newvd == oldvd)
3762 		expected_error = replacing ? 0 : EBUSY;
3763 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3764 		expected_error = EBUSY;
3765 	else if (!newvd_is_dspare && newsize < oldsize)
3766 		expected_error = EOVERFLOW;
3767 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3768 		expected_error = EDOM;
3769 	else if (newvd_is_dspare && pvd != vdev_draid_spare_get_parent(newvd))
3770 		expected_error = ENOTSUP;
3771 	else
3772 		expected_error = 0;
3773 
3774 	spa_config_exit(spa, SCL_ALL, FTAG);
3775 
3776 	/*
3777 	 * Build the nvlist describing newpath.
3778 	 */
3779 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3780 	    ashift, NULL, 0, 0, 1);
3781 
3782 	/*
3783 	 * When supported select either a healing or sequential resilver.
3784 	 */
3785 	boolean_t rebuilding = B_FALSE;
3786 	if (pvd->vdev_ops == &vdev_mirror_ops ||
3787 	    pvd->vdev_ops ==  &vdev_root_ops) {
3788 		rebuilding = !!ztest_random(2);
3789 	}
3790 
3791 	error = spa_vdev_attach(spa, oldguid, root, replacing, rebuilding);
3792 
3793 	fnvlist_free(root);
3794 
3795 	/*
3796 	 * If our parent was the replacing vdev, but the replace completed,
3797 	 * then instead of failing with ENOTSUP we may either succeed,
3798 	 * fail with ENODEV, or fail with EOVERFLOW.
3799 	 */
3800 	if (expected_error == ENOTSUP &&
3801 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3802 		expected_error = error;
3803 
3804 	/*
3805 	 * If someone grew the LUN, the replacement may be too small.
3806 	 */
3807 	if (error == EOVERFLOW || error == EBUSY)
3808 		expected_error = error;
3809 
3810 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3811 	    error == ZFS_ERR_DISCARDING_CHECKPOINT ||
3812 	    error == ZFS_ERR_RESILVER_IN_PROGRESS ||
3813 	    error == ZFS_ERR_REBUILD_IN_PROGRESS)
3814 		expected_error = error;
3815 
3816 	if (error != expected_error && expected_error != EBUSY) {
3817 		fatal(B_FALSE, "attach (%s %"PRIu64", %s %"PRIu64", %d) "
3818 		    "returned %d, expected %d",
3819 		    oldpath, oldsize, newpath,
3820 		    newsize, replacing, error, expected_error);
3821 	}
3822 out:
3823 	mutex_exit(&ztest_vdev_lock);
3824 
3825 	umem_free(oldpath, MAXPATHLEN);
3826 	umem_free(newpath, MAXPATHLEN);
3827 }
3828 
3829 void
3830 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3831 {
3832 	(void) zd, (void) id;
3833 	spa_t *spa = ztest_spa;
3834 	vdev_t *vd;
3835 	uint64_t guid;
3836 	int error;
3837 
3838 	mutex_enter(&ztest_vdev_lock);
3839 
3840 	if (ztest_device_removal_active) {
3841 		mutex_exit(&ztest_vdev_lock);
3842 		return;
3843 	}
3844 
3845 	/*
3846 	 * Remove a random top-level vdev and wait for removal to finish.
3847 	 */
3848 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3849 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3850 	guid = vd->vdev_guid;
3851 	spa_config_exit(spa, SCL_VDEV, FTAG);
3852 
3853 	error = spa_vdev_remove(spa, guid, B_FALSE);
3854 	if (error == 0) {
3855 		ztest_device_removal_active = B_TRUE;
3856 		mutex_exit(&ztest_vdev_lock);
3857 
3858 		/*
3859 		 * spa->spa_vdev_removal is created in a sync task that
3860 		 * is initiated via dsl_sync_task_nowait(). Since the
3861 		 * task may not run before spa_vdev_remove() returns, we
3862 		 * must wait at least 1 txg to ensure that the removal
3863 		 * struct has been created.
3864 		 */
3865 		txg_wait_synced(spa_get_dsl(spa), 0);
3866 
3867 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
3868 			txg_wait_synced(spa_get_dsl(spa), 0);
3869 	} else {
3870 		mutex_exit(&ztest_vdev_lock);
3871 		return;
3872 	}
3873 
3874 	/*
3875 	 * The pool needs to be scrubbed after completing device removal.
3876 	 * Failure to do so may result in checksum errors due to the
3877 	 * strategy employed by ztest_fault_inject() when selecting which
3878 	 * offset are redundant and can be damaged.
3879 	 */
3880 	error = spa_scan(spa, POOL_SCAN_SCRUB);
3881 	if (error == 0) {
3882 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3883 			txg_wait_synced(spa_get_dsl(spa), 0);
3884 	}
3885 
3886 	mutex_enter(&ztest_vdev_lock);
3887 	ztest_device_removal_active = B_FALSE;
3888 	mutex_exit(&ztest_vdev_lock);
3889 }
3890 
3891 /*
3892  * Callback function which expands the physical size of the vdev.
3893  */
3894 static vdev_t *
3895 grow_vdev(vdev_t *vd, void *arg)
3896 {
3897 	spa_t *spa __maybe_unused = vd->vdev_spa;
3898 	size_t *newsize = arg;
3899 	size_t fsize;
3900 	int fd;
3901 
3902 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
3903 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3904 
3905 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3906 		return (vd);
3907 
3908 	fsize = lseek(fd, 0, SEEK_END);
3909 	VERIFY0(ftruncate(fd, *newsize));
3910 
3911 	if (ztest_opts.zo_verbose >= 6) {
3912 		(void) printf("%s grew from %lu to %lu bytes\n",
3913 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3914 	}
3915 	(void) close(fd);
3916 	return (NULL);
3917 }
3918 
3919 /*
3920  * Callback function which expands a given vdev by calling vdev_online().
3921  */
3922 static vdev_t *
3923 online_vdev(vdev_t *vd, void *arg)
3924 {
3925 	(void) arg;
3926 	spa_t *spa = vd->vdev_spa;
3927 	vdev_t *tvd = vd->vdev_top;
3928 	uint64_t guid = vd->vdev_guid;
3929 	uint64_t generation = spa->spa_config_generation + 1;
3930 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3931 	int error;
3932 
3933 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
3934 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3935 
3936 	/* Calling vdev_online will initialize the new metaslabs */
3937 	spa_config_exit(spa, SCL_STATE, spa);
3938 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3939 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3940 
3941 	/*
3942 	 * If vdev_online returned an error or the underlying vdev_open
3943 	 * failed then we abort the expand. The only way to know that
3944 	 * vdev_open fails is by checking the returned newstate.
3945 	 */
3946 	if (error || newstate != VDEV_STATE_HEALTHY) {
3947 		if (ztest_opts.zo_verbose >= 5) {
3948 			(void) printf("Unable to expand vdev, state %u, "
3949 			    "error %d\n", newstate, error);
3950 		}
3951 		return (vd);
3952 	}
3953 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3954 
3955 	/*
3956 	 * Since we dropped the lock we need to ensure that we're
3957 	 * still talking to the original vdev. It's possible this
3958 	 * vdev may have been detached/replaced while we were
3959 	 * trying to online it.
3960 	 */
3961 	if (generation != spa->spa_config_generation) {
3962 		if (ztest_opts.zo_verbose >= 5) {
3963 			(void) printf("vdev configuration has changed, "
3964 			    "guid %"PRIu64", state %"PRIu64", "
3965 			    "expected gen %"PRIu64", got gen %"PRIu64"\n",
3966 			    guid,
3967 			    tvd->vdev_state,
3968 			    generation,
3969 			    spa->spa_config_generation);
3970 		}
3971 		return (vd);
3972 	}
3973 	return (NULL);
3974 }
3975 
3976 /*
3977  * Traverse the vdev tree calling the supplied function.
3978  * We continue to walk the tree until we either have walked all
3979  * children or we receive a non-NULL return from the callback.
3980  * If a NULL callback is passed, then we just return back the first
3981  * leaf vdev we encounter.
3982  */
3983 static vdev_t *
3984 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3985 {
3986 	uint_t c;
3987 
3988 	if (vd->vdev_ops->vdev_op_leaf) {
3989 		if (func == NULL)
3990 			return (vd);
3991 		else
3992 			return (func(vd, arg));
3993 	}
3994 
3995 	for (c = 0; c < vd->vdev_children; c++) {
3996 		vdev_t *cvd = vd->vdev_child[c];
3997 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3998 			return (cvd);
3999 	}
4000 	return (NULL);
4001 }
4002 
4003 /*
4004  * Verify that dynamic LUN growth works as expected.
4005  */
4006 void
4007 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
4008 {
4009 	(void) zd, (void) id;
4010 	spa_t *spa = ztest_spa;
4011 	vdev_t *vd, *tvd;
4012 	metaslab_class_t *mc;
4013 	metaslab_group_t *mg;
4014 	size_t psize, newsize;
4015 	uint64_t top;
4016 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
4017 
4018 	mutex_enter(&ztest_checkpoint_lock);
4019 	mutex_enter(&ztest_vdev_lock);
4020 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4021 
4022 	/*
4023 	 * If there is a vdev removal in progress, it could complete while
4024 	 * we are running, in which case we would not be able to verify
4025 	 * that the metaslab_class space increased (because it decreases
4026 	 * when the device removal completes).
4027 	 */
4028 	if (ztest_device_removal_active) {
4029 		spa_config_exit(spa, SCL_STATE, spa);
4030 		mutex_exit(&ztest_vdev_lock);
4031 		mutex_exit(&ztest_checkpoint_lock);
4032 		return;
4033 	}
4034 
4035 	top = ztest_random_vdev_top(spa, B_TRUE);
4036 
4037 	tvd = spa->spa_root_vdev->vdev_child[top];
4038 	mg = tvd->vdev_mg;
4039 	mc = mg->mg_class;
4040 	old_ms_count = tvd->vdev_ms_count;
4041 	old_class_space = metaslab_class_get_space(mc);
4042 
4043 	/*
4044 	 * Determine the size of the first leaf vdev associated with
4045 	 * our top-level device.
4046 	 */
4047 	vd = vdev_walk_tree(tvd, NULL, NULL);
4048 	ASSERT3P(vd, !=, NULL);
4049 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4050 
4051 	psize = vd->vdev_psize;
4052 
4053 	/*
4054 	 * We only try to expand the vdev if it's healthy, less than 4x its
4055 	 * original size, and it has a valid psize.
4056 	 */
4057 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
4058 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
4059 		spa_config_exit(spa, SCL_STATE, spa);
4060 		mutex_exit(&ztest_vdev_lock);
4061 		mutex_exit(&ztest_checkpoint_lock);
4062 		return;
4063 	}
4064 	ASSERT3U(psize, >, 0);
4065 	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
4066 	ASSERT3U(newsize, >, psize);
4067 
4068 	if (ztest_opts.zo_verbose >= 6) {
4069 		(void) printf("Expanding LUN %s from %lu to %lu\n",
4070 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
4071 	}
4072 
4073 	/*
4074 	 * Growing the vdev is a two step process:
4075 	 *	1). expand the physical size (i.e. relabel)
4076 	 *	2). online the vdev to create the new metaslabs
4077 	 */
4078 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
4079 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
4080 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
4081 		if (ztest_opts.zo_verbose >= 5) {
4082 			(void) printf("Could not expand LUN because "
4083 			    "the vdev configuration changed.\n");
4084 		}
4085 		spa_config_exit(spa, SCL_STATE, spa);
4086 		mutex_exit(&ztest_vdev_lock);
4087 		mutex_exit(&ztest_checkpoint_lock);
4088 		return;
4089 	}
4090 
4091 	spa_config_exit(spa, SCL_STATE, spa);
4092 
4093 	/*
4094 	 * Expanding the LUN will update the config asynchronously,
4095 	 * thus we must wait for the async thread to complete any
4096 	 * pending tasks before proceeding.
4097 	 */
4098 	for (;;) {
4099 		boolean_t done;
4100 		mutex_enter(&spa->spa_async_lock);
4101 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
4102 		mutex_exit(&spa->spa_async_lock);
4103 		if (done)
4104 			break;
4105 		txg_wait_synced(spa_get_dsl(spa), 0);
4106 		(void) poll(NULL, 0, 100);
4107 	}
4108 
4109 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4110 
4111 	tvd = spa->spa_root_vdev->vdev_child[top];
4112 	new_ms_count = tvd->vdev_ms_count;
4113 	new_class_space = metaslab_class_get_space(mc);
4114 
4115 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
4116 		if (ztest_opts.zo_verbose >= 5) {
4117 			(void) printf("Could not verify LUN expansion due to "
4118 			    "intervening vdev offline or remove.\n");
4119 		}
4120 		spa_config_exit(spa, SCL_STATE, spa);
4121 		mutex_exit(&ztest_vdev_lock);
4122 		mutex_exit(&ztest_checkpoint_lock);
4123 		return;
4124 	}
4125 
4126 	/*
4127 	 * Make sure we were able to grow the vdev.
4128 	 */
4129 	if (new_ms_count <= old_ms_count) {
4130 		fatal(B_FALSE,
4131 		    "LUN expansion failed: ms_count %"PRIu64" < %"PRIu64"\n",
4132 		    old_ms_count, new_ms_count);
4133 	}
4134 
4135 	/*
4136 	 * Make sure we were able to grow the pool.
4137 	 */
4138 	if (new_class_space <= old_class_space) {
4139 		fatal(B_FALSE,
4140 		    "LUN expansion failed: class_space %"PRIu64" < %"PRIu64"\n",
4141 		    old_class_space, new_class_space);
4142 	}
4143 
4144 	if (ztest_opts.zo_verbose >= 5) {
4145 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
4146 
4147 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
4148 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
4149 		(void) printf("%s grew from %s to %s\n",
4150 		    spa->spa_name, oldnumbuf, newnumbuf);
4151 	}
4152 
4153 	spa_config_exit(spa, SCL_STATE, spa);
4154 	mutex_exit(&ztest_vdev_lock);
4155 	mutex_exit(&ztest_checkpoint_lock);
4156 }
4157 
4158 /*
4159  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
4160  */
4161 static void
4162 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
4163 {
4164 	(void) arg, (void) cr;
4165 
4166 	/*
4167 	 * Create the objects common to all ztest datasets.
4168 	 */
4169 	VERIFY0(zap_create_claim(os, ZTEST_DIROBJ,
4170 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx));
4171 }
4172 
4173 static int
4174 ztest_dataset_create(char *dsname)
4175 {
4176 	int err;
4177 	uint64_t rand;
4178 	dsl_crypto_params_t *dcp = NULL;
4179 
4180 	/*
4181 	 * 50% of the time, we create encrypted datasets
4182 	 * using a random cipher suite and a hard-coded
4183 	 * wrapping key.
4184 	 */
4185 	rand = ztest_random(2);
4186 	if (rand != 0) {
4187 		nvlist_t *crypto_args = fnvlist_alloc();
4188 		nvlist_t *props = fnvlist_alloc();
4189 
4190 		/* slight bias towards the default cipher suite */
4191 		rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
4192 		if (rand < ZIO_CRYPT_AES_128_CCM)
4193 			rand = ZIO_CRYPT_ON;
4194 
4195 		fnvlist_add_uint64(props,
4196 		    zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
4197 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
4198 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
4199 
4200 		/*
4201 		 * These parameters aren't really used by the kernel. They
4202 		 * are simply stored so that userspace knows how to load
4203 		 * the wrapping key.
4204 		 */
4205 		fnvlist_add_uint64(props,
4206 		    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
4207 		fnvlist_add_string(props,
4208 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
4209 		fnvlist_add_uint64(props,
4210 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
4211 		fnvlist_add_uint64(props,
4212 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
4213 
4214 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
4215 		    crypto_args, &dcp));
4216 
4217 		/*
4218 		 * Cycle through all available encryption implementations
4219 		 * to verify interoperability.
4220 		 */
4221 		VERIFY0(gcm_impl_set("cycle"));
4222 		VERIFY0(aes_impl_set("cycle"));
4223 
4224 		fnvlist_free(crypto_args);
4225 		fnvlist_free(props);
4226 	}
4227 
4228 	err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
4229 	    ztest_objset_create_cb, NULL);
4230 	dsl_crypto_params_free(dcp, !!err);
4231 
4232 	rand = ztest_random(100);
4233 	if (err || rand < 80)
4234 		return (err);
4235 
4236 	if (ztest_opts.zo_verbose >= 5)
4237 		(void) printf("Setting dataset %s to sync always\n", dsname);
4238 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
4239 	    ZFS_SYNC_ALWAYS, B_FALSE));
4240 }
4241 
4242 static int
4243 ztest_objset_destroy_cb(const char *name, void *arg)
4244 {
4245 	(void) arg;
4246 	objset_t *os;
4247 	dmu_object_info_t doi;
4248 	int error;
4249 
4250 	/*
4251 	 * Verify that the dataset contains a directory object.
4252 	 */
4253 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
4254 	    B_TRUE, FTAG, &os));
4255 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
4256 	if (error != ENOENT) {
4257 		/* We could have crashed in the middle of destroying it */
4258 		ASSERT0(error);
4259 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
4260 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
4261 	}
4262 	dmu_objset_disown(os, B_TRUE, FTAG);
4263 
4264 	/*
4265 	 * Destroy the dataset.
4266 	 */
4267 	if (strchr(name, '@') != NULL) {
4268 		error = dsl_destroy_snapshot(name, B_TRUE);
4269 		if (error != ECHRNG) {
4270 			/*
4271 			 * The program was executed, but encountered a runtime
4272 			 * error, such as insufficient slop, or a hold on the
4273 			 * dataset.
4274 			 */
4275 			ASSERT0(error);
4276 		}
4277 	} else {
4278 		error = dsl_destroy_head(name);
4279 		if (error == ENOSPC) {
4280 			/* There could be checkpoint or insufficient slop */
4281 			ztest_record_enospc(FTAG);
4282 		} else if (error != EBUSY) {
4283 			/* There could be a hold on this dataset */
4284 			ASSERT0(error);
4285 		}
4286 	}
4287 	return (0);
4288 }
4289 
4290 static boolean_t
4291 ztest_snapshot_create(char *osname, uint64_t id)
4292 {
4293 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4294 	int error;
4295 
4296 	(void) snprintf(snapname, sizeof (snapname), "%"PRIu64"", id);
4297 
4298 	error = dmu_objset_snapshot_one(osname, snapname);
4299 	if (error == ENOSPC) {
4300 		ztest_record_enospc(FTAG);
4301 		return (B_FALSE);
4302 	}
4303 	if (error != 0 && error != EEXIST && error != ECHRNG) {
4304 		fatal(B_FALSE, "ztest_snapshot_create(%s@%s) = %d", osname,
4305 		    snapname, error);
4306 	}
4307 	return (B_TRUE);
4308 }
4309 
4310 static boolean_t
4311 ztest_snapshot_destroy(char *osname, uint64_t id)
4312 {
4313 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4314 	int error;
4315 
4316 	(void) snprintf(snapname, sizeof (snapname), "%s@%"PRIu64"",
4317 	    osname, id);
4318 
4319 	error = dsl_destroy_snapshot(snapname, B_FALSE);
4320 	if (error != 0 && error != ENOENT && error != ECHRNG)
4321 		fatal(B_FALSE, "ztest_snapshot_destroy(%s) = %d",
4322 		    snapname, error);
4323 	return (B_TRUE);
4324 }
4325 
4326 void
4327 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
4328 {
4329 	(void) zd;
4330 	ztest_ds_t *zdtmp;
4331 	int iters;
4332 	int error;
4333 	objset_t *os, *os2;
4334 	char name[ZFS_MAX_DATASET_NAME_LEN];
4335 	zilog_t *zilog;
4336 	int i;
4337 
4338 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
4339 
4340 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4341 
4342 	(void) snprintf(name, sizeof (name), "%s/temp_%"PRIu64"",
4343 	    ztest_opts.zo_pool, id);
4344 
4345 	/*
4346 	 * If this dataset exists from a previous run, process its replay log
4347 	 * half of the time.  If we don't replay it, then dsl_destroy_head()
4348 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
4349 	 */
4350 	if (ztest_random(2) == 0 &&
4351 	    ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
4352 	    B_TRUE, FTAG, &os) == 0) {
4353 		ztest_zd_init(zdtmp, NULL, os);
4354 		zil_replay(os, zdtmp, ztest_replay_vector);
4355 		ztest_zd_fini(zdtmp);
4356 		dmu_objset_disown(os, B_TRUE, FTAG);
4357 	}
4358 
4359 	/*
4360 	 * There may be an old instance of the dataset we're about to
4361 	 * create lying around from a previous run.  If so, destroy it
4362 	 * and all of its snapshots.
4363 	 */
4364 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
4365 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
4366 
4367 	/*
4368 	 * Verify that the destroyed dataset is no longer in the namespace.
4369 	 * It may still be present if the destroy above fails with ENOSPC.
4370 	 */
4371 	error = ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, B_TRUE,
4372 	    FTAG, &os);
4373 	if (error == 0) {
4374 		dmu_objset_disown(os, B_TRUE, FTAG);
4375 		ztest_record_enospc(FTAG);
4376 		goto out;
4377 	}
4378 	VERIFY3U(ENOENT, ==, error);
4379 
4380 	/*
4381 	 * Verify that we can create a new dataset.
4382 	 */
4383 	error = ztest_dataset_create(name);
4384 	if (error) {
4385 		if (error == ENOSPC) {
4386 			ztest_record_enospc(FTAG);
4387 			goto out;
4388 		}
4389 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", name, error);
4390 	}
4391 
4392 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
4393 	    FTAG, &os));
4394 
4395 	ztest_zd_init(zdtmp, NULL, os);
4396 
4397 	/*
4398 	 * Open the intent log for it.
4399 	 */
4400 	zilog = zil_open(os, ztest_get_data, NULL);
4401 
4402 	/*
4403 	 * Put some objects in there, do a little I/O to them,
4404 	 * and randomly take a couple of snapshots along the way.
4405 	 */
4406 	iters = ztest_random(5);
4407 	for (i = 0; i < iters; i++) {
4408 		ztest_dmu_object_alloc_free(zdtmp, id);
4409 		if (ztest_random(iters) == 0)
4410 			(void) ztest_snapshot_create(name, i);
4411 	}
4412 
4413 	/*
4414 	 * Verify that we cannot create an existing dataset.
4415 	 */
4416 	VERIFY3U(EEXIST, ==,
4417 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
4418 
4419 	/*
4420 	 * Verify that we can hold an objset that is also owned.
4421 	 */
4422 	VERIFY0(dmu_objset_hold(name, FTAG, &os2));
4423 	dmu_objset_rele(os2, FTAG);
4424 
4425 	/*
4426 	 * Verify that we cannot own an objset that is already owned.
4427 	 */
4428 	VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
4429 	    B_FALSE, B_TRUE, FTAG, &os2));
4430 
4431 	zil_close(zilog);
4432 	dmu_objset_disown(os, B_TRUE, FTAG);
4433 	ztest_zd_fini(zdtmp);
4434 out:
4435 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4436 
4437 	umem_free(zdtmp, sizeof (ztest_ds_t));
4438 }
4439 
4440 /*
4441  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
4442  */
4443 void
4444 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
4445 {
4446 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4447 	(void) ztest_snapshot_destroy(zd->zd_name, id);
4448 	(void) ztest_snapshot_create(zd->zd_name, id);
4449 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4450 }
4451 
4452 /*
4453  * Cleanup non-standard snapshots and clones.
4454  */
4455 static void
4456 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
4457 {
4458 	char *snap1name;
4459 	char *clone1name;
4460 	char *snap2name;
4461 	char *clone2name;
4462 	char *snap3name;
4463 	int error;
4464 
4465 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4466 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4467 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4468 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4469 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4470 
4471 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4472 	    osname, id);
4473 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4474 	    osname, id);
4475 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4476 	    clone1name, id);
4477 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4478 	    osname, id);
4479 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4480 	    clone1name, id);
4481 
4482 	error = dsl_destroy_head(clone2name);
4483 	if (error && error != ENOENT)
4484 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone2name, error);
4485 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
4486 	if (error && error != ENOENT)
4487 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4488 		    snap3name, error);
4489 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
4490 	if (error && error != ENOENT)
4491 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4492 		    snap2name, error);
4493 	error = dsl_destroy_head(clone1name);
4494 	if (error && error != ENOENT)
4495 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone1name, error);
4496 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
4497 	if (error && error != ENOENT)
4498 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4499 		    snap1name, error);
4500 
4501 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4502 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4503 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4504 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4505 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4506 }
4507 
4508 /*
4509  * Verify dsl_dataset_promote handles EBUSY
4510  */
4511 void
4512 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
4513 {
4514 	objset_t *os;
4515 	char *snap1name;
4516 	char *clone1name;
4517 	char *snap2name;
4518 	char *clone2name;
4519 	char *snap3name;
4520 	char *osname = zd->zd_name;
4521 	int error;
4522 
4523 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4524 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4525 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4526 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4527 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4528 
4529 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4530 
4531 	ztest_dsl_dataset_cleanup(osname, id);
4532 
4533 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4534 	    osname, id);
4535 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4536 	    osname, id);
4537 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4538 	    clone1name, id);
4539 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4540 	    osname, id);
4541 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4542 	    clone1name, id);
4543 
4544 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
4545 	if (error && error != EEXIST) {
4546 		if (error == ENOSPC) {
4547 			ztest_record_enospc(FTAG);
4548 			goto out;
4549 		}
4550 		fatal(B_FALSE, "dmu_take_snapshot(%s) = %d", snap1name, error);
4551 	}
4552 
4553 	error = dmu_objset_clone(clone1name, snap1name);
4554 	if (error) {
4555 		if (error == ENOSPC) {
4556 			ztest_record_enospc(FTAG);
4557 			goto out;
4558 		}
4559 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone1name, error);
4560 	}
4561 
4562 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4563 	if (error && error != EEXIST) {
4564 		if (error == ENOSPC) {
4565 			ztest_record_enospc(FTAG);
4566 			goto out;
4567 		}
4568 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap2name, error);
4569 	}
4570 
4571 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4572 	if (error && error != EEXIST) {
4573 		if (error == ENOSPC) {
4574 			ztest_record_enospc(FTAG);
4575 			goto out;
4576 		}
4577 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap3name, error);
4578 	}
4579 
4580 	error = dmu_objset_clone(clone2name, snap3name);
4581 	if (error) {
4582 		if (error == ENOSPC) {
4583 			ztest_record_enospc(FTAG);
4584 			goto out;
4585 		}
4586 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone2name, error);
4587 	}
4588 
4589 	error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
4590 	    FTAG, &os);
4591 	if (error)
4592 		fatal(B_FALSE, "dmu_objset_own(%s) = %d", snap2name, error);
4593 	error = dsl_dataset_promote(clone2name, NULL);
4594 	if (error == ENOSPC) {
4595 		dmu_objset_disown(os, B_TRUE, FTAG);
4596 		ztest_record_enospc(FTAG);
4597 		goto out;
4598 	}
4599 	if (error != EBUSY)
4600 		fatal(B_FALSE, "dsl_dataset_promote(%s), %d, not EBUSY",
4601 		    clone2name, error);
4602 	dmu_objset_disown(os, B_TRUE, FTAG);
4603 
4604 out:
4605 	ztest_dsl_dataset_cleanup(osname, id);
4606 
4607 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4608 
4609 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4610 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4611 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4612 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4613 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4614 }
4615 
4616 #undef OD_ARRAY_SIZE
4617 #define	OD_ARRAY_SIZE	4
4618 
4619 /*
4620  * Verify that dmu_object_{alloc,free} work as expected.
4621  */
4622 void
4623 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4624 {
4625 	ztest_od_t *od;
4626 	int batchsize;
4627 	int size;
4628 	int b;
4629 
4630 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4631 	od = umem_alloc(size, UMEM_NOFAIL);
4632 	batchsize = OD_ARRAY_SIZE;
4633 
4634 	for (b = 0; b < batchsize; b++)
4635 		ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4636 		    0, 0, 0);
4637 
4638 	/*
4639 	 * Destroy the previous batch of objects, create a new batch,
4640 	 * and do some I/O on the new objects.
4641 	 */
4642 	if (ztest_object_init(zd, od, size, B_TRUE) != 0) {
4643 		zd->zd_od = NULL;
4644 		umem_free(od, size);
4645 		return;
4646 	}
4647 
4648 	while (ztest_random(4 * batchsize) != 0)
4649 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
4650 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4651 
4652 	umem_free(od, size);
4653 }
4654 
4655 /*
4656  * Rewind the global allocator to verify object allocation backfilling.
4657  */
4658 void
4659 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
4660 {
4661 	(void) id;
4662 	objset_t *os = zd->zd_os;
4663 	uint_t dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
4664 	uint64_t object;
4665 
4666 	/*
4667 	 * Rewind the global allocator randomly back to a lower object number
4668 	 * to force backfilling and reclamation of recently freed dnodes.
4669 	 */
4670 	mutex_enter(&os->os_obj_lock);
4671 	object = ztest_random(os->os_obj_next_chunk);
4672 	os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
4673 	mutex_exit(&os->os_obj_lock);
4674 }
4675 
4676 #undef OD_ARRAY_SIZE
4677 #define	OD_ARRAY_SIZE	2
4678 
4679 /*
4680  * Verify that dmu_{read,write} work as expected.
4681  */
4682 void
4683 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4684 {
4685 	int size;
4686 	ztest_od_t *od;
4687 
4688 	objset_t *os = zd->zd_os;
4689 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4690 	od = umem_alloc(size, UMEM_NOFAIL);
4691 	dmu_tx_t *tx;
4692 	int freeit, error;
4693 	uint64_t i, n, s, txg;
4694 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4695 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4696 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4697 	uint64_t regions = 997;
4698 	uint64_t stride = 123456789ULL;
4699 	uint64_t width = 40;
4700 	int free_percent = 5;
4701 
4702 	/*
4703 	 * This test uses two objects, packobj and bigobj, that are always
4704 	 * updated together (i.e. in the same tx) so that their contents are
4705 	 * in sync and can be compared.  Their contents relate to each other
4706 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4707 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4708 	 * for any index n, there are three bufwads that should be identical:
4709 	 *
4710 	 *	packobj, at offset n * sizeof (bufwad_t)
4711 	 *	bigobj, at the head of the nth chunk
4712 	 *	bigobj, at the tail of the nth chunk
4713 	 *
4714 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
4715 	 * and it doesn't have any relation to the object blocksize.
4716 	 * The only requirement is that it can hold at least two bufwads.
4717 	 *
4718 	 * Normally, we write the bufwad to each of these locations.
4719 	 * However, free_percent of the time we instead write zeroes to
4720 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
4721 	 * bigobj to packobj, we can verify that the DMU is correctly
4722 	 * tracking which parts of an object are allocated and free,
4723 	 * and that the contents of the allocated blocks are correct.
4724 	 */
4725 
4726 	/*
4727 	 * Read the directory info.  If it's the first time, set things up.
4728 	 */
4729 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
4730 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4731 	    chunksize);
4732 
4733 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
4734 		umem_free(od, size);
4735 		return;
4736 	}
4737 
4738 	bigobj = od[0].od_object;
4739 	packobj = od[1].od_object;
4740 	chunksize = od[0].od_gen;
4741 	ASSERT3U(chunksize, ==, od[1].od_gen);
4742 
4743 	/*
4744 	 * Prefetch a random chunk of the big object.
4745 	 * Our aim here is to get some async reads in flight
4746 	 * for blocks that we may free below; the DMU should
4747 	 * handle this race correctly.
4748 	 */
4749 	n = ztest_random(regions) * stride + ztest_random(width);
4750 	s = 1 + ztest_random(2 * width - 1);
4751 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4752 	    ZIO_PRIORITY_SYNC_READ);
4753 
4754 	/*
4755 	 * Pick a random index and compute the offsets into packobj and bigobj.
4756 	 */
4757 	n = ztest_random(regions) * stride + ztest_random(width);
4758 	s = 1 + ztest_random(width - 1);
4759 
4760 	packoff = n * sizeof (bufwad_t);
4761 	packsize = s * sizeof (bufwad_t);
4762 
4763 	bigoff = n * chunksize;
4764 	bigsize = s * chunksize;
4765 
4766 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4767 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4768 
4769 	/*
4770 	 * free_percent of the time, free a range of bigobj rather than
4771 	 * overwriting it.
4772 	 */
4773 	freeit = (ztest_random(100) < free_percent);
4774 
4775 	/*
4776 	 * Read the current contents of our objects.
4777 	 */
4778 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
4779 	    DMU_READ_PREFETCH);
4780 	ASSERT0(error);
4781 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4782 	    DMU_READ_PREFETCH);
4783 	ASSERT0(error);
4784 
4785 	/*
4786 	 * Get a tx for the mods to both packobj and bigobj.
4787 	 */
4788 	tx = dmu_tx_create(os);
4789 
4790 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
4791 
4792 	if (freeit)
4793 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4794 	else
4795 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4796 
4797 	/* This accounts for setting the checksum/compression. */
4798 	dmu_tx_hold_bonus(tx, bigobj);
4799 
4800 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4801 	if (txg == 0) {
4802 		umem_free(packbuf, packsize);
4803 		umem_free(bigbuf, bigsize);
4804 		umem_free(od, size);
4805 		return;
4806 	}
4807 
4808 	enum zio_checksum cksum;
4809 	do {
4810 		cksum = (enum zio_checksum)
4811 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4812 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4813 	dmu_object_set_checksum(os, bigobj, cksum, tx);
4814 
4815 	enum zio_compress comp;
4816 	do {
4817 		comp = (enum zio_compress)
4818 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4819 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4820 	dmu_object_set_compress(os, bigobj, comp, tx);
4821 
4822 	/*
4823 	 * For each index from n to n + s, verify that the existing bufwad
4824 	 * in packobj matches the bufwads at the head and tail of the
4825 	 * corresponding chunk in bigobj.  Then update all three bufwads
4826 	 * with the new values we want to write out.
4827 	 */
4828 	for (i = 0; i < s; i++) {
4829 		/* LINTED */
4830 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4831 		/* LINTED */
4832 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4833 		/* LINTED */
4834 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4835 
4836 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
4837 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
4838 
4839 		if (pack->bw_txg > txg)
4840 			fatal(B_FALSE,
4841 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
4842 			    pack->bw_txg, txg);
4843 
4844 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4845 			fatal(B_FALSE, "wrong index: "
4846 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
4847 			    pack->bw_index, n, i);
4848 
4849 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4850 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
4851 			    pack, bigH);
4852 
4853 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4854 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
4855 			    pack, bigT);
4856 
4857 		if (freeit) {
4858 			memset(pack, 0, sizeof (bufwad_t));
4859 		} else {
4860 			pack->bw_index = n + i;
4861 			pack->bw_txg = txg;
4862 			pack->bw_data = 1 + ztest_random(-2ULL);
4863 		}
4864 		*bigH = *pack;
4865 		*bigT = *pack;
4866 	}
4867 
4868 	/*
4869 	 * We've verified all the old bufwads, and made new ones.
4870 	 * Now write them out.
4871 	 */
4872 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4873 
4874 	if (freeit) {
4875 		if (ztest_opts.zo_verbose >= 7) {
4876 			(void) printf("freeing offset %"PRIx64" size %"PRIx64""
4877 			    " txg %"PRIx64"\n",
4878 			    bigoff, bigsize, txg);
4879 		}
4880 		VERIFY0(dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4881 	} else {
4882 		if (ztest_opts.zo_verbose >= 7) {
4883 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
4884 			    " txg %"PRIx64"\n",
4885 			    bigoff, bigsize, txg);
4886 		}
4887 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4888 	}
4889 
4890 	dmu_tx_commit(tx);
4891 
4892 	/*
4893 	 * Sanity check the stuff we just wrote.
4894 	 */
4895 	{
4896 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4897 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4898 
4899 		VERIFY0(dmu_read(os, packobj, packoff,
4900 		    packsize, packcheck, DMU_READ_PREFETCH));
4901 		VERIFY0(dmu_read(os, bigobj, bigoff,
4902 		    bigsize, bigcheck, DMU_READ_PREFETCH));
4903 
4904 		ASSERT0(memcmp(packbuf, packcheck, packsize));
4905 		ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
4906 
4907 		umem_free(packcheck, packsize);
4908 		umem_free(bigcheck, bigsize);
4909 	}
4910 
4911 	umem_free(packbuf, packsize);
4912 	umem_free(bigbuf, bigsize);
4913 	umem_free(od, size);
4914 }
4915 
4916 static void
4917 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4918     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4919 {
4920 	uint64_t i;
4921 	bufwad_t *pack;
4922 	bufwad_t *bigH;
4923 	bufwad_t *bigT;
4924 
4925 	/*
4926 	 * For each index from n to n + s, verify that the existing bufwad
4927 	 * in packobj matches the bufwads at the head and tail of the
4928 	 * corresponding chunk in bigobj.  Then update all three bufwads
4929 	 * with the new values we want to write out.
4930 	 */
4931 	for (i = 0; i < s; i++) {
4932 		/* LINTED */
4933 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4934 		/* LINTED */
4935 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4936 		/* LINTED */
4937 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4938 
4939 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
4940 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
4941 
4942 		if (pack->bw_txg > txg)
4943 			fatal(B_FALSE,
4944 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
4945 			    pack->bw_txg, txg);
4946 
4947 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4948 			fatal(B_FALSE, "wrong index: "
4949 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
4950 			    pack->bw_index, n, i);
4951 
4952 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4953 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
4954 			    pack, bigH);
4955 
4956 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4957 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
4958 			    pack, bigT);
4959 
4960 		pack->bw_index = n + i;
4961 		pack->bw_txg = txg;
4962 		pack->bw_data = 1 + ztest_random(-2ULL);
4963 
4964 		*bigH = *pack;
4965 		*bigT = *pack;
4966 	}
4967 }
4968 
4969 #undef OD_ARRAY_SIZE
4970 #define	OD_ARRAY_SIZE	2
4971 
4972 void
4973 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4974 {
4975 	objset_t *os = zd->zd_os;
4976 	ztest_od_t *od;
4977 	dmu_tx_t *tx;
4978 	uint64_t i;
4979 	int error;
4980 	int size;
4981 	uint64_t n, s, txg;
4982 	bufwad_t *packbuf, *bigbuf;
4983 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4984 	uint64_t blocksize = ztest_random_blocksize();
4985 	uint64_t chunksize = blocksize;
4986 	uint64_t regions = 997;
4987 	uint64_t stride = 123456789ULL;
4988 	uint64_t width = 9;
4989 	dmu_buf_t *bonus_db;
4990 	arc_buf_t **bigbuf_arcbufs;
4991 	dmu_object_info_t doi;
4992 
4993 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4994 	od = umem_alloc(size, UMEM_NOFAIL);
4995 
4996 	/*
4997 	 * This test uses two objects, packobj and bigobj, that are always
4998 	 * updated together (i.e. in the same tx) so that their contents are
4999 	 * in sync and can be compared.  Their contents relate to each other
5000 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
5001 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
5002 	 * for any index n, there are three bufwads that should be identical:
5003 	 *
5004 	 *	packobj, at offset n * sizeof (bufwad_t)
5005 	 *	bigobj, at the head of the nth chunk
5006 	 *	bigobj, at the tail of the nth chunk
5007 	 *
5008 	 * The chunk size is set equal to bigobj block size so that
5009 	 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
5010 	 */
5011 
5012 	/*
5013 	 * Read the directory info.  If it's the first time, set things up.
5014 	 */
5015 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5016 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
5017 	    chunksize);
5018 
5019 
5020 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
5021 		umem_free(od, size);
5022 		return;
5023 	}
5024 
5025 	bigobj = od[0].od_object;
5026 	packobj = od[1].od_object;
5027 	blocksize = od[0].od_blocksize;
5028 	chunksize = blocksize;
5029 	ASSERT3U(chunksize, ==, od[1].od_gen);
5030 
5031 	VERIFY0(dmu_object_info(os, bigobj, &doi));
5032 	VERIFY(ISP2(doi.doi_data_block_size));
5033 	VERIFY3U(chunksize, ==, doi.doi_data_block_size);
5034 	VERIFY3U(chunksize, >=, 2 * sizeof (bufwad_t));
5035 
5036 	/*
5037 	 * Pick a random index and compute the offsets into packobj and bigobj.
5038 	 */
5039 	n = ztest_random(regions) * stride + ztest_random(width);
5040 	s = 1 + ztest_random(width - 1);
5041 
5042 	packoff = n * sizeof (bufwad_t);
5043 	packsize = s * sizeof (bufwad_t);
5044 
5045 	bigoff = n * chunksize;
5046 	bigsize = s * chunksize;
5047 
5048 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
5049 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
5050 
5051 	VERIFY0(dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
5052 
5053 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
5054 
5055 	/*
5056 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
5057 	 * Iteration 1 test zcopy to already referenced dbufs.
5058 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
5059 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
5060 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
5061 	 * Iteration 5 test zcopy when it can't be done.
5062 	 * Iteration 6 one more zcopy write.
5063 	 */
5064 	for (i = 0; i < 7; i++) {
5065 		uint64_t j;
5066 		uint64_t off;
5067 
5068 		/*
5069 		 * In iteration 5 (i == 5) use arcbufs
5070 		 * that don't match bigobj blksz to test
5071 		 * dmu_assign_arcbuf_by_dbuf() when it can't directly
5072 		 * assign an arcbuf to a dbuf.
5073 		 */
5074 		for (j = 0; j < s; j++) {
5075 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5076 				bigbuf_arcbufs[j] =
5077 				    dmu_request_arcbuf(bonus_db, chunksize);
5078 			} else {
5079 				bigbuf_arcbufs[2 * j] =
5080 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5081 				bigbuf_arcbufs[2 * j + 1] =
5082 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5083 			}
5084 		}
5085 
5086 		/*
5087 		 * Get a tx for the mods to both packobj and bigobj.
5088 		 */
5089 		tx = dmu_tx_create(os);
5090 
5091 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
5092 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
5093 
5094 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5095 		if (txg == 0) {
5096 			umem_free(packbuf, packsize);
5097 			umem_free(bigbuf, bigsize);
5098 			for (j = 0; j < s; j++) {
5099 				if (i != 5 ||
5100 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
5101 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
5102 				} else {
5103 					dmu_return_arcbuf(
5104 					    bigbuf_arcbufs[2 * j]);
5105 					dmu_return_arcbuf(
5106 					    bigbuf_arcbufs[2 * j + 1]);
5107 				}
5108 			}
5109 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5110 			umem_free(od, size);
5111 			dmu_buf_rele(bonus_db, FTAG);
5112 			return;
5113 		}
5114 
5115 		/*
5116 		 * 50% of the time don't read objects in the 1st iteration to
5117 		 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
5118 		 * no existing dbufs for the specified offsets.
5119 		 */
5120 		if (i != 0 || ztest_random(2) != 0) {
5121 			error = dmu_read(os, packobj, packoff,
5122 			    packsize, packbuf, DMU_READ_PREFETCH);
5123 			ASSERT0(error);
5124 			error = dmu_read(os, bigobj, bigoff, bigsize,
5125 			    bigbuf, DMU_READ_PREFETCH);
5126 			ASSERT0(error);
5127 		}
5128 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
5129 		    n, chunksize, txg);
5130 
5131 		/*
5132 		 * We've verified all the old bufwads, and made new ones.
5133 		 * Now write them out.
5134 		 */
5135 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
5136 		if (ztest_opts.zo_verbose >= 7) {
5137 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
5138 			    " txg %"PRIx64"\n",
5139 			    bigoff, bigsize, txg);
5140 		}
5141 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
5142 			dmu_buf_t *dbt;
5143 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5144 				memcpy(bigbuf_arcbufs[j]->b_data,
5145 				    (caddr_t)bigbuf + (off - bigoff),
5146 				    chunksize);
5147 			} else {
5148 				memcpy(bigbuf_arcbufs[2 * j]->b_data,
5149 				    (caddr_t)bigbuf + (off - bigoff),
5150 				    chunksize / 2);
5151 				memcpy(bigbuf_arcbufs[2 * j + 1]->b_data,
5152 				    (caddr_t)bigbuf + (off - bigoff) +
5153 				    chunksize / 2,
5154 				    chunksize / 2);
5155 			}
5156 
5157 			if (i == 1) {
5158 				VERIFY(dmu_buf_hold(os, bigobj, off,
5159 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
5160 			}
5161 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5162 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5163 				    off, bigbuf_arcbufs[j], tx));
5164 			} else {
5165 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5166 				    off, bigbuf_arcbufs[2 * j], tx));
5167 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5168 				    off + chunksize / 2,
5169 				    bigbuf_arcbufs[2 * j + 1], tx));
5170 			}
5171 			if (i == 1) {
5172 				dmu_buf_rele(dbt, FTAG);
5173 			}
5174 		}
5175 		dmu_tx_commit(tx);
5176 
5177 		/*
5178 		 * Sanity check the stuff we just wrote.
5179 		 */
5180 		{
5181 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
5182 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
5183 
5184 			VERIFY0(dmu_read(os, packobj, packoff,
5185 			    packsize, packcheck, DMU_READ_PREFETCH));
5186 			VERIFY0(dmu_read(os, bigobj, bigoff,
5187 			    bigsize, bigcheck, DMU_READ_PREFETCH));
5188 
5189 			ASSERT0(memcmp(packbuf, packcheck, packsize));
5190 			ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
5191 
5192 			umem_free(packcheck, packsize);
5193 			umem_free(bigcheck, bigsize);
5194 		}
5195 		if (i == 2) {
5196 			txg_wait_open(dmu_objset_pool(os), 0, B_TRUE);
5197 		} else if (i == 3) {
5198 			txg_wait_synced(dmu_objset_pool(os), 0);
5199 		}
5200 	}
5201 
5202 	dmu_buf_rele(bonus_db, FTAG);
5203 	umem_free(packbuf, packsize);
5204 	umem_free(bigbuf, bigsize);
5205 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5206 	umem_free(od, size);
5207 }
5208 
5209 void
5210 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
5211 {
5212 	(void) id;
5213 	ztest_od_t *od;
5214 
5215 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5216 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
5217 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5218 
5219 	/*
5220 	 * Have multiple threads write to large offsets in an object
5221 	 * to verify that parallel writes to an object -- even to the
5222 	 * same blocks within the object -- doesn't cause any trouble.
5223 	 */
5224 	ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5225 
5226 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
5227 		return;
5228 
5229 	while (ztest_random(10) != 0)
5230 		ztest_io(zd, od->od_object, offset);
5231 
5232 	umem_free(od, sizeof (ztest_od_t));
5233 }
5234 
5235 void
5236 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
5237 {
5238 	ztest_od_t *od;
5239 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
5240 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5241 	uint64_t count = ztest_random(20) + 1;
5242 	uint64_t blocksize = ztest_random_blocksize();
5243 	void *data;
5244 
5245 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5246 
5247 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5248 
5249 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5250 	    !ztest_random(2)) != 0) {
5251 		umem_free(od, sizeof (ztest_od_t));
5252 		return;
5253 	}
5254 
5255 	if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
5256 		umem_free(od, sizeof (ztest_od_t));
5257 		return;
5258 	}
5259 
5260 	ztest_prealloc(zd, od->od_object, offset, count * blocksize);
5261 
5262 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
5263 
5264 	while (ztest_random(count) != 0) {
5265 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
5266 		if (ztest_write(zd, od->od_object, randoff, blocksize,
5267 		    data) != 0)
5268 			break;
5269 		while (ztest_random(4) != 0)
5270 			ztest_io(zd, od->od_object, randoff);
5271 	}
5272 
5273 	umem_free(data, blocksize);
5274 	umem_free(od, sizeof (ztest_od_t));
5275 }
5276 
5277 /*
5278  * Verify that zap_{create,destroy,add,remove,update} work as expected.
5279  */
5280 #define	ZTEST_ZAP_MIN_INTS	1
5281 #define	ZTEST_ZAP_MAX_INTS	4
5282 #define	ZTEST_ZAP_MAX_PROPS	1000
5283 
5284 void
5285 ztest_zap(ztest_ds_t *zd, uint64_t id)
5286 {
5287 	objset_t *os = zd->zd_os;
5288 	ztest_od_t *od;
5289 	uint64_t object;
5290 	uint64_t txg, last_txg;
5291 	uint64_t value[ZTEST_ZAP_MAX_INTS];
5292 	uint64_t zl_ints, zl_intsize, prop;
5293 	int i, ints;
5294 	dmu_tx_t *tx;
5295 	char propname[100], txgname[100];
5296 	int error;
5297 	const char *const hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
5298 
5299 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5300 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5301 
5302 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5303 	    !ztest_random(2)) != 0)
5304 		goto out;
5305 
5306 	object = od->od_object;
5307 
5308 	/*
5309 	 * Generate a known hash collision, and verify that
5310 	 * we can lookup and remove both entries.
5311 	 */
5312 	tx = dmu_tx_create(os);
5313 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5314 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5315 	if (txg == 0)
5316 		goto out;
5317 	for (i = 0; i < 2; i++) {
5318 		value[i] = i;
5319 		VERIFY0(zap_add(os, object, hc[i], sizeof (uint64_t),
5320 		    1, &value[i], tx));
5321 	}
5322 	for (i = 0; i < 2; i++) {
5323 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
5324 		    sizeof (uint64_t), 1, &value[i], tx));
5325 		VERIFY0(
5326 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
5327 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5328 		ASSERT3U(zl_ints, ==, 1);
5329 	}
5330 	for (i = 0; i < 2; i++) {
5331 		VERIFY0(zap_remove(os, object, hc[i], tx));
5332 	}
5333 	dmu_tx_commit(tx);
5334 
5335 	/*
5336 	 * Generate a bunch of random entries.
5337 	 */
5338 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
5339 
5340 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5341 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5342 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5343 	memset(value, 0, sizeof (value));
5344 	last_txg = 0;
5345 
5346 	/*
5347 	 * If these zap entries already exist, validate their contents.
5348 	 */
5349 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5350 	if (error == 0) {
5351 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5352 		ASSERT3U(zl_ints, ==, 1);
5353 
5354 		VERIFY0(zap_lookup(os, object, txgname, zl_intsize,
5355 		    zl_ints, &last_txg));
5356 
5357 		VERIFY0(zap_length(os, object, propname, &zl_intsize,
5358 		    &zl_ints));
5359 
5360 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5361 		ASSERT3U(zl_ints, ==, ints);
5362 
5363 		VERIFY0(zap_lookup(os, object, propname, zl_intsize,
5364 		    zl_ints, value));
5365 
5366 		for (i = 0; i < ints; i++) {
5367 			ASSERT3U(value[i], ==, last_txg + object + i);
5368 		}
5369 	} else {
5370 		ASSERT3U(error, ==, ENOENT);
5371 	}
5372 
5373 	/*
5374 	 * Atomically update two entries in our zap object.
5375 	 * The first is named txg_%llu, and contains the txg
5376 	 * in which the property was last updated.  The second
5377 	 * is named prop_%llu, and the nth element of its value
5378 	 * should be txg + object + n.
5379 	 */
5380 	tx = dmu_tx_create(os);
5381 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5382 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5383 	if (txg == 0)
5384 		goto out;
5385 
5386 	if (last_txg > txg)
5387 		fatal(B_FALSE, "zap future leak: old %"PRIu64" new %"PRIu64"",
5388 		    last_txg, txg);
5389 
5390 	for (i = 0; i < ints; i++)
5391 		value[i] = txg + object + i;
5392 
5393 	VERIFY0(zap_update(os, object, txgname, sizeof (uint64_t),
5394 	    1, &txg, tx));
5395 	VERIFY0(zap_update(os, object, propname, sizeof (uint64_t),
5396 	    ints, value, tx));
5397 
5398 	dmu_tx_commit(tx);
5399 
5400 	/*
5401 	 * Remove a random pair of entries.
5402 	 */
5403 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5404 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5405 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5406 
5407 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5408 
5409 	if (error == ENOENT)
5410 		goto out;
5411 
5412 	ASSERT0(error);
5413 
5414 	tx = dmu_tx_create(os);
5415 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5416 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5417 	if (txg == 0)
5418 		goto out;
5419 	VERIFY0(zap_remove(os, object, txgname, tx));
5420 	VERIFY0(zap_remove(os, object, propname, tx));
5421 	dmu_tx_commit(tx);
5422 out:
5423 	umem_free(od, sizeof (ztest_od_t));
5424 }
5425 
5426 /*
5427  * Test case to test the upgrading of a microzap to fatzap.
5428  */
5429 void
5430 ztest_fzap(ztest_ds_t *zd, uint64_t id)
5431 {
5432 	objset_t *os = zd->zd_os;
5433 	ztest_od_t *od;
5434 	uint64_t object, txg, value;
5435 
5436 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5437 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5438 
5439 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5440 	    !ztest_random(2)) != 0)
5441 		goto out;
5442 	object = od->od_object;
5443 
5444 	/*
5445 	 * Add entries to this ZAP and make sure it spills over
5446 	 * and gets upgraded to a fatzap. Also, since we are adding
5447 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
5448 	 */
5449 	for (value = 0; value < 2050; value++) {
5450 		char name[ZFS_MAX_DATASET_NAME_LEN];
5451 		dmu_tx_t *tx;
5452 		int error;
5453 
5454 		(void) snprintf(name, sizeof (name), "fzap-%"PRIu64"-%"PRIu64"",
5455 		    id, value);
5456 
5457 		tx = dmu_tx_create(os);
5458 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
5459 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5460 		if (txg == 0)
5461 			goto out;
5462 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
5463 		    &value, tx);
5464 		ASSERT(error == 0 || error == EEXIST);
5465 		dmu_tx_commit(tx);
5466 	}
5467 out:
5468 	umem_free(od, sizeof (ztest_od_t));
5469 }
5470 
5471 void
5472 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
5473 {
5474 	(void) id;
5475 	objset_t *os = zd->zd_os;
5476 	ztest_od_t *od;
5477 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
5478 	dmu_tx_t *tx;
5479 	int i, namelen, error;
5480 	int micro = ztest_random(2);
5481 	char name[20], string_value[20];
5482 	void *data;
5483 
5484 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5485 	ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
5486 
5487 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5488 		umem_free(od, sizeof (ztest_od_t));
5489 		return;
5490 	}
5491 
5492 	object = od->od_object;
5493 
5494 	/*
5495 	 * Generate a random name of the form 'xxx.....' where each
5496 	 * x is a random printable character and the dots are dots.
5497 	 * There are 94 such characters, and the name length goes from
5498 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
5499 	 */
5500 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
5501 
5502 	for (i = 0; i < 3; i++)
5503 		name[i] = '!' + ztest_random('~' - '!' + 1);
5504 	for (; i < namelen - 1; i++)
5505 		name[i] = '.';
5506 	name[i] = '\0';
5507 
5508 	if ((namelen & 1) || micro) {
5509 		wsize = sizeof (txg);
5510 		wc = 1;
5511 		data = &txg;
5512 	} else {
5513 		wsize = 1;
5514 		wc = namelen;
5515 		data = string_value;
5516 	}
5517 
5518 	count = -1ULL;
5519 	VERIFY0(zap_count(os, object, &count));
5520 	ASSERT3S(count, !=, -1ULL);
5521 
5522 	/*
5523 	 * Select an operation: length, lookup, add, update, remove.
5524 	 */
5525 	i = ztest_random(5);
5526 
5527 	if (i >= 2) {
5528 		tx = dmu_tx_create(os);
5529 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5530 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5531 		if (txg == 0) {
5532 			umem_free(od, sizeof (ztest_od_t));
5533 			return;
5534 		}
5535 		memcpy(string_value, name, namelen);
5536 	} else {
5537 		tx = NULL;
5538 		txg = 0;
5539 		memset(string_value, 0, namelen);
5540 	}
5541 
5542 	switch (i) {
5543 
5544 	case 0:
5545 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
5546 		if (error == 0) {
5547 			ASSERT3U(wsize, ==, zl_wsize);
5548 			ASSERT3U(wc, ==, zl_wc);
5549 		} else {
5550 			ASSERT3U(error, ==, ENOENT);
5551 		}
5552 		break;
5553 
5554 	case 1:
5555 		error = zap_lookup(os, object, name, wsize, wc, data);
5556 		if (error == 0) {
5557 			if (data == string_value &&
5558 			    memcmp(name, data, namelen) != 0)
5559 				fatal(B_FALSE, "name '%s' != val '%s' len %d",
5560 				    name, (char *)data, namelen);
5561 		} else {
5562 			ASSERT3U(error, ==, ENOENT);
5563 		}
5564 		break;
5565 
5566 	case 2:
5567 		error = zap_add(os, object, name, wsize, wc, data, tx);
5568 		ASSERT(error == 0 || error == EEXIST);
5569 		break;
5570 
5571 	case 3:
5572 		VERIFY0(zap_update(os, object, name, wsize, wc, data, tx));
5573 		break;
5574 
5575 	case 4:
5576 		error = zap_remove(os, object, name, tx);
5577 		ASSERT(error == 0 || error == ENOENT);
5578 		break;
5579 	}
5580 
5581 	if (tx != NULL)
5582 		dmu_tx_commit(tx);
5583 
5584 	umem_free(od, sizeof (ztest_od_t));
5585 }
5586 
5587 /*
5588  * Commit callback data.
5589  */
5590 typedef struct ztest_cb_data {
5591 	list_node_t		zcd_node;
5592 	uint64_t		zcd_txg;
5593 	int			zcd_expected_err;
5594 	boolean_t		zcd_added;
5595 	boolean_t		zcd_called;
5596 	spa_t			*zcd_spa;
5597 } ztest_cb_data_t;
5598 
5599 /* This is the actual commit callback function */
5600 static void
5601 ztest_commit_callback(void *arg, int error)
5602 {
5603 	ztest_cb_data_t *data = arg;
5604 	uint64_t synced_txg;
5605 
5606 	VERIFY3P(data, !=, NULL);
5607 	VERIFY3S(data->zcd_expected_err, ==, error);
5608 	VERIFY(!data->zcd_called);
5609 
5610 	synced_txg = spa_last_synced_txg(data->zcd_spa);
5611 	if (data->zcd_txg > synced_txg)
5612 		fatal(B_FALSE,
5613 		    "commit callback of txg %"PRIu64" called prematurely, "
5614 		    "last synced txg = %"PRIu64"\n",
5615 		    data->zcd_txg, synced_txg);
5616 
5617 	data->zcd_called = B_TRUE;
5618 
5619 	if (error == ECANCELED) {
5620 		ASSERT0(data->zcd_txg);
5621 		ASSERT(!data->zcd_added);
5622 
5623 		/*
5624 		 * The private callback data should be destroyed here, but
5625 		 * since we are going to check the zcd_called field after
5626 		 * dmu_tx_abort(), we will destroy it there.
5627 		 */
5628 		return;
5629 	}
5630 
5631 	ASSERT(data->zcd_added);
5632 	ASSERT3U(data->zcd_txg, !=, 0);
5633 
5634 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5635 
5636 	/* See if this cb was called more quickly */
5637 	if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
5638 		zc_min_txg_delay = synced_txg - data->zcd_txg;
5639 
5640 	/* Remove our callback from the list */
5641 	list_remove(&zcl.zcl_callbacks, data);
5642 
5643 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5644 
5645 	umem_free(data, sizeof (ztest_cb_data_t));
5646 }
5647 
5648 /* Allocate and initialize callback data structure */
5649 static ztest_cb_data_t *
5650 ztest_create_cb_data(objset_t *os, uint64_t txg)
5651 {
5652 	ztest_cb_data_t *cb_data;
5653 
5654 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
5655 
5656 	cb_data->zcd_txg = txg;
5657 	cb_data->zcd_spa = dmu_objset_spa(os);
5658 	list_link_init(&cb_data->zcd_node);
5659 
5660 	return (cb_data);
5661 }
5662 
5663 /*
5664  * Commit callback test.
5665  */
5666 void
5667 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5668 {
5669 	objset_t *os = zd->zd_os;
5670 	ztest_od_t *od;
5671 	dmu_tx_t *tx;
5672 	ztest_cb_data_t *cb_data[3], *tmp_cb;
5673 	uint64_t old_txg, txg;
5674 	int i, error = 0;
5675 
5676 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5677 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5678 
5679 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5680 		umem_free(od, sizeof (ztest_od_t));
5681 		return;
5682 	}
5683 
5684 	tx = dmu_tx_create(os);
5685 
5686 	cb_data[0] = ztest_create_cb_data(os, 0);
5687 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5688 
5689 	dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
5690 
5691 	/* Every once in a while, abort the transaction on purpose */
5692 	if (ztest_random(100) == 0)
5693 		error = -1;
5694 
5695 	if (!error)
5696 		error = dmu_tx_assign(tx, TXG_NOWAIT);
5697 
5698 	txg = error ? 0 : dmu_tx_get_txg(tx);
5699 
5700 	cb_data[0]->zcd_txg = txg;
5701 	cb_data[1] = ztest_create_cb_data(os, txg);
5702 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5703 
5704 	if (error) {
5705 		/*
5706 		 * It's not a strict requirement to call the registered
5707 		 * callbacks from inside dmu_tx_abort(), but that's what
5708 		 * it's supposed to happen in the current implementation
5709 		 * so we will check for that.
5710 		 */
5711 		for (i = 0; i < 2; i++) {
5712 			cb_data[i]->zcd_expected_err = ECANCELED;
5713 			VERIFY(!cb_data[i]->zcd_called);
5714 		}
5715 
5716 		dmu_tx_abort(tx);
5717 
5718 		for (i = 0; i < 2; i++) {
5719 			VERIFY(cb_data[i]->zcd_called);
5720 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5721 		}
5722 
5723 		umem_free(od, sizeof (ztest_od_t));
5724 		return;
5725 	}
5726 
5727 	cb_data[2] = ztest_create_cb_data(os, txg);
5728 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5729 
5730 	/*
5731 	 * Read existing data to make sure there isn't a future leak.
5732 	 */
5733 	VERIFY0(dmu_read(os, od->od_object, 0, sizeof (uint64_t),
5734 	    &old_txg, DMU_READ_PREFETCH));
5735 
5736 	if (old_txg > txg)
5737 		fatal(B_FALSE,
5738 		    "future leak: got %"PRIu64", open txg is %"PRIu64"",
5739 		    old_txg, txg);
5740 
5741 	dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
5742 
5743 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5744 
5745 	/*
5746 	 * Since commit callbacks don't have any ordering requirement and since
5747 	 * it is theoretically possible for a commit callback to be called
5748 	 * after an arbitrary amount of time has elapsed since its txg has been
5749 	 * synced, it is difficult to reliably determine whether a commit
5750 	 * callback hasn't been called due to high load or due to a flawed
5751 	 * implementation.
5752 	 *
5753 	 * In practice, we will assume that if after a certain number of txgs a
5754 	 * commit callback hasn't been called, then most likely there's an
5755 	 * implementation bug..
5756 	 */
5757 	tmp_cb = list_head(&zcl.zcl_callbacks);
5758 	if (tmp_cb != NULL &&
5759 	    tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
5760 		fatal(B_FALSE,
5761 		    "Commit callback threshold exceeded, "
5762 		    "oldest txg: %"PRIu64", open txg: %"PRIu64"\n",
5763 		    tmp_cb->zcd_txg, txg);
5764 	}
5765 
5766 	/*
5767 	 * Let's find the place to insert our callbacks.
5768 	 *
5769 	 * Even though the list is ordered by txg, it is possible for the
5770 	 * insertion point to not be the end because our txg may already be
5771 	 * quiescing at this point and other callbacks in the open txg
5772 	 * (from other objsets) may have sneaked in.
5773 	 */
5774 	tmp_cb = list_tail(&zcl.zcl_callbacks);
5775 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5776 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5777 
5778 	/* Add the 3 callbacks to the list */
5779 	for (i = 0; i < 3; i++) {
5780 		if (tmp_cb == NULL)
5781 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5782 		else
5783 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5784 			    cb_data[i]);
5785 
5786 		cb_data[i]->zcd_added = B_TRUE;
5787 		VERIFY(!cb_data[i]->zcd_called);
5788 
5789 		tmp_cb = cb_data[i];
5790 	}
5791 
5792 	zc_cb_counter += 3;
5793 
5794 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5795 
5796 	dmu_tx_commit(tx);
5797 
5798 	umem_free(od, sizeof (ztest_od_t));
5799 }
5800 
5801 /*
5802  * Visit each object in the dataset. Verify that its properties
5803  * are consistent what was stored in the block tag when it was created,
5804  * and that its unused bonus buffer space has not been overwritten.
5805  */
5806 void
5807 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5808 {
5809 	(void) id;
5810 	objset_t *os = zd->zd_os;
5811 	uint64_t obj;
5812 	int err = 0;
5813 
5814 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5815 		ztest_block_tag_t *bt = NULL;
5816 		dmu_object_info_t doi;
5817 		dmu_buf_t *db;
5818 
5819 		ztest_object_lock(zd, obj, RL_READER);
5820 		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) {
5821 			ztest_object_unlock(zd, obj);
5822 			continue;
5823 		}
5824 
5825 		dmu_object_info_from_db(db, &doi);
5826 		if (doi.doi_bonus_size >= sizeof (*bt))
5827 			bt = ztest_bt_bonus(db);
5828 
5829 		if (bt && bt->bt_magic == BT_MAGIC) {
5830 			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5831 			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
5832 			    bt->bt_crtxg);
5833 			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5834 		}
5835 
5836 		dmu_buf_rele(db, FTAG);
5837 		ztest_object_unlock(zd, obj);
5838 	}
5839 }
5840 
5841 void
5842 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5843 {
5844 	(void) id;
5845 	zfs_prop_t proplist[] = {
5846 		ZFS_PROP_CHECKSUM,
5847 		ZFS_PROP_COMPRESSION,
5848 		ZFS_PROP_COPIES,
5849 		ZFS_PROP_DEDUP
5850 	};
5851 
5852 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5853 
5854 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) {
5855 		int error = ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5856 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5857 		ASSERT(error == 0 || error == ENOSPC);
5858 	}
5859 
5860 	int error = ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
5861 	    ztest_random_blocksize(), (int)ztest_random(2));
5862 	ASSERT(error == 0 || error == ENOSPC);
5863 
5864 	(void) pthread_rwlock_unlock(&ztest_name_lock);
5865 }
5866 
5867 void
5868 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5869 {
5870 	(void) zd, (void) id;
5871 	nvlist_t *props = NULL;
5872 
5873 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5874 
5875 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_AUTOTRIM, ztest_random(2));
5876 
5877 	VERIFY0(spa_prop_get(ztest_spa, &props));
5878 
5879 	if (ztest_opts.zo_verbose >= 6)
5880 		dump_nvlist(props, 4);
5881 
5882 	fnvlist_free(props);
5883 
5884 	(void) pthread_rwlock_unlock(&ztest_name_lock);
5885 }
5886 
5887 static int
5888 user_release_one(const char *snapname, const char *holdname)
5889 {
5890 	nvlist_t *snaps, *holds;
5891 	int error;
5892 
5893 	snaps = fnvlist_alloc();
5894 	holds = fnvlist_alloc();
5895 	fnvlist_add_boolean(holds, holdname);
5896 	fnvlist_add_nvlist(snaps, snapname, holds);
5897 	fnvlist_free(holds);
5898 	error = dsl_dataset_user_release(snaps, NULL);
5899 	fnvlist_free(snaps);
5900 	return (error);
5901 }
5902 
5903 /*
5904  * Test snapshot hold/release and deferred destroy.
5905  */
5906 void
5907 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5908 {
5909 	int error;
5910 	objset_t *os = zd->zd_os;
5911 	objset_t *origin;
5912 	char snapname[100];
5913 	char fullname[100];
5914 	char clonename[100];
5915 	char tag[100];
5916 	char osname[ZFS_MAX_DATASET_NAME_LEN];
5917 	nvlist_t *holds;
5918 
5919 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
5920 
5921 	dmu_objset_name(os, osname);
5922 
5923 	(void) snprintf(snapname, sizeof (snapname), "sh1_%"PRIu64"", id);
5924 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5925 	(void) snprintf(clonename, sizeof (clonename), "%s/ch1_%"PRIu64"",
5926 	    osname, id);
5927 	(void) snprintf(tag, sizeof (tag), "tag_%"PRIu64"", id);
5928 
5929 	/*
5930 	 * Clean up from any previous run.
5931 	 */
5932 	error = dsl_destroy_head(clonename);
5933 	if (error != ENOENT)
5934 		ASSERT0(error);
5935 	error = user_release_one(fullname, tag);
5936 	if (error != ESRCH && error != ENOENT)
5937 		ASSERT0(error);
5938 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5939 	if (error != ENOENT)
5940 		ASSERT0(error);
5941 
5942 	/*
5943 	 * Create snapshot, clone it, mark snap for deferred destroy,
5944 	 * destroy clone, verify snap was also destroyed.
5945 	 */
5946 	error = dmu_objset_snapshot_one(osname, snapname);
5947 	if (error) {
5948 		if (error == ENOSPC) {
5949 			ztest_record_enospc("dmu_objset_snapshot");
5950 			goto out;
5951 		}
5952 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
5953 	}
5954 
5955 	error = dmu_objset_clone(clonename, fullname);
5956 	if (error) {
5957 		if (error == ENOSPC) {
5958 			ztest_record_enospc("dmu_objset_clone");
5959 			goto out;
5960 		}
5961 		fatal(B_FALSE, "dmu_objset_clone(%s) = %d", clonename, error);
5962 	}
5963 
5964 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5965 	if (error) {
5966 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5967 		    fullname, error);
5968 	}
5969 
5970 	error = dsl_destroy_head(clonename);
5971 	if (error)
5972 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clonename, error);
5973 
5974 	error = dmu_objset_hold(fullname, FTAG, &origin);
5975 	if (error != ENOENT)
5976 		fatal(B_FALSE, "dmu_objset_hold(%s) = %d", fullname, error);
5977 
5978 	/*
5979 	 * Create snapshot, add temporary hold, verify that we can't
5980 	 * destroy a held snapshot, mark for deferred destroy,
5981 	 * release hold, verify snapshot was destroyed.
5982 	 */
5983 	error = dmu_objset_snapshot_one(osname, snapname);
5984 	if (error) {
5985 		if (error == ENOSPC) {
5986 			ztest_record_enospc("dmu_objset_snapshot");
5987 			goto out;
5988 		}
5989 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
5990 	}
5991 
5992 	holds = fnvlist_alloc();
5993 	fnvlist_add_string(holds, fullname, tag);
5994 	error = dsl_dataset_user_hold(holds, 0, NULL);
5995 	fnvlist_free(holds);
5996 
5997 	if (error == ENOSPC) {
5998 		ztest_record_enospc("dsl_dataset_user_hold");
5999 		goto out;
6000 	} else if (error) {
6001 		fatal(B_FALSE, "dsl_dataset_user_hold(%s, %s) = %u",
6002 		    fullname, tag, error);
6003 	}
6004 
6005 	error = dsl_destroy_snapshot(fullname, B_FALSE);
6006 	if (error != EBUSY) {
6007 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
6008 		    fullname, error);
6009 	}
6010 
6011 	error = dsl_destroy_snapshot(fullname, B_TRUE);
6012 	if (error) {
6013 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
6014 		    fullname, error);
6015 	}
6016 
6017 	error = user_release_one(fullname, tag);
6018 	if (error)
6019 		fatal(B_FALSE, "user_release_one(%s, %s) = %d",
6020 		    fullname, tag, error);
6021 
6022 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
6023 
6024 out:
6025 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6026 }
6027 
6028 /*
6029  * Inject random faults into the on-disk data.
6030  */
6031 void
6032 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
6033 {
6034 	(void) zd, (void) id;
6035 	ztest_shared_t *zs = ztest_shared;
6036 	spa_t *spa = ztest_spa;
6037 	int fd;
6038 	uint64_t offset;
6039 	uint64_t leaves;
6040 	uint64_t bad = 0x1990c0ffeedecadeull;
6041 	uint64_t top, leaf;
6042 	char *path0;
6043 	char *pathrand;
6044 	size_t fsize;
6045 	int bshift = SPA_MAXBLOCKSHIFT + 2;
6046 	int iters = 1000;
6047 	int maxfaults;
6048 	int mirror_save;
6049 	vdev_t *vd0 = NULL;
6050 	uint64_t guid0 = 0;
6051 	boolean_t islog = B_FALSE;
6052 
6053 	path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6054 	pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6055 
6056 	mutex_enter(&ztest_vdev_lock);
6057 
6058 	/*
6059 	 * Device removal is in progress, fault injection must be disabled
6060 	 * until it completes and the pool is scrubbed.  The fault injection
6061 	 * strategy for damaging blocks does not take in to account evacuated
6062 	 * blocks which may have already been damaged.
6063 	 */
6064 	if (ztest_device_removal_active) {
6065 		mutex_exit(&ztest_vdev_lock);
6066 		goto out;
6067 	}
6068 
6069 	maxfaults = MAXFAULTS(zs);
6070 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raid_children;
6071 	mirror_save = zs->zs_mirrors;
6072 	mutex_exit(&ztest_vdev_lock);
6073 
6074 	ASSERT3U(leaves, >=, 1);
6075 
6076 	/*
6077 	 * While ztest is running the number of leaves will not change.  This
6078 	 * is critical for the fault injection logic as it determines where
6079 	 * errors can be safely injected such that they are always repairable.
6080 	 *
6081 	 * When restarting ztest a different number of leaves may be requested
6082 	 * which will shift the regions to be damaged.  This is fine as long
6083 	 * as the pool has been scrubbed prior to using the new mapping.
6084 	 * Failure to do can result in non-repairable damage being injected.
6085 	 */
6086 	if (ztest_pool_scrubbed == B_FALSE)
6087 		goto out;
6088 
6089 	/*
6090 	 * Grab the name lock as reader. There are some operations
6091 	 * which don't like to have their vdevs changed while
6092 	 * they are in progress (i.e. spa_change_guid). Those
6093 	 * operations will have grabbed the name lock as writer.
6094 	 */
6095 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6096 
6097 	/*
6098 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
6099 	 */
6100 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6101 
6102 	if (ztest_random(2) == 0) {
6103 		/*
6104 		 * Inject errors on a normal data device or slog device.
6105 		 */
6106 		top = ztest_random_vdev_top(spa, B_TRUE);
6107 		leaf = ztest_random(leaves) + zs->zs_splits;
6108 
6109 		/*
6110 		 * Generate paths to the first leaf in this top-level vdev,
6111 		 * and to the random leaf we selected.  We'll induce transient
6112 		 * write failures and random online/offline activity on leaf 0,
6113 		 * and we'll write random garbage to the randomly chosen leaf.
6114 		 */
6115 		(void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
6116 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6117 		    top * leaves + zs->zs_splits);
6118 		(void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
6119 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6120 		    top * leaves + leaf);
6121 
6122 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
6123 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
6124 			islog = B_TRUE;
6125 
6126 		/*
6127 		 * If the top-level vdev needs to be resilvered
6128 		 * then we only allow faults on the device that is
6129 		 * resilvering.
6130 		 */
6131 		if (vd0 != NULL && maxfaults != 1 &&
6132 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
6133 		    vd0->vdev_resilver_txg != 0)) {
6134 			/*
6135 			 * Make vd0 explicitly claim to be unreadable,
6136 			 * or unwritable, or reach behind its back
6137 			 * and close the underlying fd.  We can do this if
6138 			 * maxfaults == 0 because we'll fail and reexecute,
6139 			 * and we can do it if maxfaults >= 2 because we'll
6140 			 * have enough redundancy.  If maxfaults == 1, the
6141 			 * combination of this with injection of random data
6142 			 * corruption below exceeds the pool's fault tolerance.
6143 			 */
6144 			vdev_file_t *vf = vd0->vdev_tsd;
6145 
6146 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
6147 			    (long long)vd0->vdev_id, (int)maxfaults);
6148 
6149 			if (vf != NULL && ztest_random(3) == 0) {
6150 				(void) close(vf->vf_file->f_fd);
6151 				vf->vf_file->f_fd = -1;
6152 			} else if (ztest_random(2) == 0) {
6153 				vd0->vdev_cant_read = B_TRUE;
6154 			} else {
6155 				vd0->vdev_cant_write = B_TRUE;
6156 			}
6157 			guid0 = vd0->vdev_guid;
6158 		}
6159 	} else {
6160 		/*
6161 		 * Inject errors on an l2cache device.
6162 		 */
6163 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
6164 
6165 		if (sav->sav_count == 0) {
6166 			spa_config_exit(spa, SCL_STATE, FTAG);
6167 			(void) pthread_rwlock_unlock(&ztest_name_lock);
6168 			goto out;
6169 		}
6170 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
6171 		guid0 = vd0->vdev_guid;
6172 		(void) strlcpy(path0, vd0->vdev_path, MAXPATHLEN);
6173 		(void) strlcpy(pathrand, vd0->vdev_path, MAXPATHLEN);
6174 
6175 		leaf = 0;
6176 		leaves = 1;
6177 		maxfaults = INT_MAX;	/* no limit on cache devices */
6178 	}
6179 
6180 	spa_config_exit(spa, SCL_STATE, FTAG);
6181 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6182 
6183 	/*
6184 	 * If we can tolerate two or more faults, or we're dealing
6185 	 * with a slog, randomly online/offline vd0.
6186 	 */
6187 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
6188 		if (ztest_random(10) < 6) {
6189 			int flags = (ztest_random(2) == 0 ?
6190 			    ZFS_OFFLINE_TEMPORARY : 0);
6191 
6192 			/*
6193 			 * We have to grab the zs_name_lock as writer to
6194 			 * prevent a race between offlining a slog and
6195 			 * destroying a dataset. Offlining the slog will
6196 			 * grab a reference on the dataset which may cause
6197 			 * dsl_destroy_head() to fail with EBUSY thus
6198 			 * leaving the dataset in an inconsistent state.
6199 			 */
6200 			if (islog)
6201 				(void) pthread_rwlock_wrlock(&ztest_name_lock);
6202 
6203 			VERIFY3U(vdev_offline(spa, guid0, flags), !=, EBUSY);
6204 
6205 			if (islog)
6206 				(void) pthread_rwlock_unlock(&ztest_name_lock);
6207 		} else {
6208 			/*
6209 			 * Ideally we would like to be able to randomly
6210 			 * call vdev_[on|off]line without holding locks
6211 			 * to force unpredictable failures but the side
6212 			 * effects of vdev_[on|off]line prevent us from
6213 			 * doing so. We grab the ztest_vdev_lock here to
6214 			 * prevent a race between injection testing and
6215 			 * aux_vdev removal.
6216 			 */
6217 			mutex_enter(&ztest_vdev_lock);
6218 			(void) vdev_online(spa, guid0, 0, NULL);
6219 			mutex_exit(&ztest_vdev_lock);
6220 		}
6221 	}
6222 
6223 	if (maxfaults == 0)
6224 		goto out;
6225 
6226 	/*
6227 	 * We have at least single-fault tolerance, so inject data corruption.
6228 	 */
6229 	fd = open(pathrand, O_RDWR);
6230 
6231 	if (fd == -1) /* we hit a gap in the device namespace */
6232 		goto out;
6233 
6234 	fsize = lseek(fd, 0, SEEK_END);
6235 
6236 	while (--iters != 0) {
6237 		/*
6238 		 * The offset must be chosen carefully to ensure that
6239 		 * we do not inject a given logical block with errors
6240 		 * on two different leaf devices, because ZFS can not
6241 		 * tolerate that (if maxfaults==1).
6242 		 *
6243 		 * To achieve this we divide each leaf device into
6244 		 * chunks of size (# leaves * SPA_MAXBLOCKSIZE * 4).
6245 		 * Each chunk is further divided into error-injection
6246 		 * ranges (can accept errors) and clear ranges (we do
6247 		 * not inject errors in those). Each error-injection
6248 		 * range can accept errors only for a single leaf vdev.
6249 		 * Error-injection ranges are separated by clear ranges.
6250 		 *
6251 		 * For example, with 3 leaves, each chunk looks like:
6252 		 *    0 to  32M: injection range for leaf 0
6253 		 *  32M to  64M: clear range - no injection allowed
6254 		 *  64M to  96M: injection range for leaf 1
6255 		 *  96M to 128M: clear range - no injection allowed
6256 		 * 128M to 160M: injection range for leaf 2
6257 		 * 160M to 192M: clear range - no injection allowed
6258 		 *
6259 		 * Each clear range must be large enough such that a
6260 		 * single block cannot straddle it. This way a block
6261 		 * can't be a target in two different injection ranges
6262 		 * (on different leaf vdevs).
6263 		 */
6264 		offset = ztest_random(fsize / (leaves << bshift)) *
6265 		    (leaves << bshift) + (leaf << bshift) +
6266 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
6267 
6268 		/*
6269 		 * Only allow damage to the labels at one end of the vdev.
6270 		 *
6271 		 * If all labels are damaged, the device will be totally
6272 		 * inaccessible, which will result in loss of data,
6273 		 * because we also damage (parts of) the other side of
6274 		 * the mirror/raidz.
6275 		 *
6276 		 * Additionally, we will always have both an even and an
6277 		 * odd label, so that we can handle crashes in the
6278 		 * middle of vdev_config_sync().
6279 		 */
6280 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
6281 			continue;
6282 
6283 		/*
6284 		 * The two end labels are stored at the "end" of the disk, but
6285 		 * the end of the disk (vdev_psize) is aligned to
6286 		 * sizeof (vdev_label_t).
6287 		 */
6288 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
6289 		if ((leaf & 1) == 1 &&
6290 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
6291 			continue;
6292 
6293 		mutex_enter(&ztest_vdev_lock);
6294 		if (mirror_save != zs->zs_mirrors) {
6295 			mutex_exit(&ztest_vdev_lock);
6296 			(void) close(fd);
6297 			goto out;
6298 		}
6299 
6300 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
6301 			fatal(B_TRUE,
6302 			    "can't inject bad word at 0x%"PRIx64" in %s",
6303 			    offset, pathrand);
6304 
6305 		mutex_exit(&ztest_vdev_lock);
6306 
6307 		if (ztest_opts.zo_verbose >= 7)
6308 			(void) printf("injected bad word into %s,"
6309 			    " offset 0x%"PRIx64"\n", pathrand, offset);
6310 	}
6311 
6312 	(void) close(fd);
6313 out:
6314 	umem_free(path0, MAXPATHLEN);
6315 	umem_free(pathrand, MAXPATHLEN);
6316 }
6317 
6318 /*
6319  * By design ztest will never inject uncorrectable damage in to the pool.
6320  * Issue a scrub, wait for it to complete, and verify there is never any
6321  * persistent damage.
6322  *
6323  * Only after a full scrub has been completed is it safe to start injecting
6324  * data corruption.  See the comment in zfs_fault_inject().
6325  */
6326 static int
6327 ztest_scrub_impl(spa_t *spa)
6328 {
6329 	int error = spa_scan(spa, POOL_SCAN_SCRUB);
6330 	if (error)
6331 		return (error);
6332 
6333 	while (dsl_scan_scrubbing(spa_get_dsl(spa)))
6334 		txg_wait_synced(spa_get_dsl(spa), 0);
6335 
6336 	if (spa_approx_errlog_size(spa) > 0)
6337 		return (ECKSUM);
6338 
6339 	ztest_pool_scrubbed = B_TRUE;
6340 
6341 	return (0);
6342 }
6343 
6344 /*
6345  * Scrub the pool.
6346  */
6347 void
6348 ztest_scrub(ztest_ds_t *zd, uint64_t id)
6349 {
6350 	(void) zd, (void) id;
6351 	spa_t *spa = ztest_spa;
6352 	int error;
6353 
6354 	/*
6355 	 * Scrub in progress by device removal.
6356 	 */
6357 	if (ztest_device_removal_active)
6358 		return;
6359 
6360 	/*
6361 	 * Start a scrub, wait a moment, then force a restart.
6362 	 */
6363 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
6364 	(void) poll(NULL, 0, 100);
6365 
6366 	error = ztest_scrub_impl(spa);
6367 	if (error == EBUSY)
6368 		error = 0;
6369 	ASSERT0(error);
6370 }
6371 
6372 /*
6373  * Change the guid for the pool.
6374  */
6375 void
6376 ztest_reguid(ztest_ds_t *zd, uint64_t id)
6377 {
6378 	(void) zd, (void) id;
6379 	spa_t *spa = ztest_spa;
6380 	uint64_t orig, load;
6381 	int error;
6382 
6383 	if (ztest_opts.zo_mmp_test)
6384 		return;
6385 
6386 	orig = spa_guid(spa);
6387 	load = spa_load_guid(spa);
6388 
6389 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
6390 	error = spa_change_guid(spa);
6391 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6392 
6393 	if (error != 0)
6394 		return;
6395 
6396 	if (ztest_opts.zo_verbose >= 4) {
6397 		(void) printf("Changed guid old %"PRIu64" -> %"PRIu64"\n",
6398 		    orig, spa_guid(spa));
6399 	}
6400 
6401 	VERIFY3U(orig, !=, spa_guid(spa));
6402 	VERIFY3U(load, ==, spa_load_guid(spa));
6403 }
6404 
6405 void
6406 ztest_blake3(ztest_ds_t *zd, uint64_t id)
6407 {
6408 	(void) zd, (void) id;
6409 	hrtime_t end = gethrtime() + NANOSEC;
6410 	zio_cksum_salt_t salt;
6411 	void *salt_ptr = &salt.zcs_bytes;
6412 	struct abd *abd_data, *abd_meta;
6413 	void *buf, *templ;
6414 	int i, *ptr;
6415 	uint32_t size;
6416 	BLAKE3_CTX ctx;
6417 	const zfs_impl_t *blake3 = zfs_impl_get_ops("blake3");
6418 
6419 	size = ztest_random_blocksize();
6420 	buf = umem_alloc(size, UMEM_NOFAIL);
6421 	abd_data = abd_alloc(size, B_FALSE);
6422 	abd_meta = abd_alloc(size, B_TRUE);
6423 
6424 	for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6425 		*ptr = ztest_random(UINT_MAX);
6426 	memset(salt_ptr, 'A', 32);
6427 
6428 	abd_copy_from_buf_off(abd_data, buf, 0, size);
6429 	abd_copy_from_buf_off(abd_meta, buf, 0, size);
6430 
6431 	while (gethrtime() <= end) {
6432 		int run_count = 100;
6433 		zio_cksum_t zc_ref1, zc_ref2;
6434 		zio_cksum_t zc_res1, zc_res2;
6435 
6436 		void *ref1 = &zc_ref1;
6437 		void *ref2 = &zc_ref2;
6438 		void *res1 = &zc_res1;
6439 		void *res2 = &zc_res2;
6440 
6441 		/* BLAKE3_KEY_LEN = 32 */
6442 		VERIFY0(blake3->setname("generic"));
6443 		templ = abd_checksum_blake3_tmpl_init(&salt);
6444 		Blake3_InitKeyed(&ctx, salt_ptr);
6445 		Blake3_Update(&ctx, buf, size);
6446 		Blake3_Final(&ctx, ref1);
6447 		zc_ref2 = zc_ref1;
6448 		ZIO_CHECKSUM_BSWAP(&zc_ref2);
6449 		abd_checksum_blake3_tmpl_free(templ);
6450 
6451 		VERIFY0(blake3->setname("cycle"));
6452 		while (run_count-- > 0) {
6453 
6454 			/* Test current implementation */
6455 			Blake3_InitKeyed(&ctx, salt_ptr);
6456 			Blake3_Update(&ctx, buf, size);
6457 			Blake3_Final(&ctx, res1);
6458 			zc_res2 = zc_res1;
6459 			ZIO_CHECKSUM_BSWAP(&zc_res2);
6460 
6461 			VERIFY0(memcmp(ref1, res1, 32));
6462 			VERIFY0(memcmp(ref2, res2, 32));
6463 
6464 			/* Test ABD - data */
6465 			templ = abd_checksum_blake3_tmpl_init(&salt);
6466 			abd_checksum_blake3_native(abd_data, size,
6467 			    templ, &zc_res1);
6468 			abd_checksum_blake3_byteswap(abd_data, size,
6469 			    templ, &zc_res2);
6470 
6471 			VERIFY0(memcmp(ref1, res1, 32));
6472 			VERIFY0(memcmp(ref2, res2, 32));
6473 
6474 			/* Test ABD - metadata */
6475 			abd_checksum_blake3_native(abd_meta, size,
6476 			    templ, &zc_res1);
6477 			abd_checksum_blake3_byteswap(abd_meta, size,
6478 			    templ, &zc_res2);
6479 			abd_checksum_blake3_tmpl_free(templ);
6480 
6481 			VERIFY0(memcmp(ref1, res1, 32));
6482 			VERIFY0(memcmp(ref2, res2, 32));
6483 
6484 		}
6485 	}
6486 
6487 	abd_free(abd_data);
6488 	abd_free(abd_meta);
6489 	umem_free(buf, size);
6490 }
6491 
6492 void
6493 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
6494 {
6495 	(void) zd, (void) id;
6496 	hrtime_t end = gethrtime() + NANOSEC;
6497 
6498 	while (gethrtime() <= end) {
6499 		int run_count = 100;
6500 		void *buf;
6501 		struct abd *abd_data, *abd_meta;
6502 		uint32_t size;
6503 		int *ptr;
6504 		int i;
6505 		zio_cksum_t zc_ref;
6506 		zio_cksum_t zc_ref_byteswap;
6507 
6508 		size = ztest_random_blocksize();
6509 
6510 		buf = umem_alloc(size, UMEM_NOFAIL);
6511 		abd_data = abd_alloc(size, B_FALSE);
6512 		abd_meta = abd_alloc(size, B_TRUE);
6513 
6514 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6515 			*ptr = ztest_random(UINT_MAX);
6516 
6517 		abd_copy_from_buf_off(abd_data, buf, 0, size);
6518 		abd_copy_from_buf_off(abd_meta, buf, 0, size);
6519 
6520 		VERIFY0(fletcher_4_impl_set("scalar"));
6521 		fletcher_4_native(buf, size, NULL, &zc_ref);
6522 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
6523 
6524 		VERIFY0(fletcher_4_impl_set("cycle"));
6525 		while (run_count-- > 0) {
6526 			zio_cksum_t zc;
6527 			zio_cksum_t zc_byteswap;
6528 
6529 			fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
6530 			fletcher_4_native(buf, size, NULL, &zc);
6531 
6532 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6533 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6534 			    sizeof (zc_byteswap)));
6535 
6536 			/* Test ABD - data */
6537 			abd_fletcher_4_byteswap(abd_data, size, NULL,
6538 			    &zc_byteswap);
6539 			abd_fletcher_4_native(abd_data, size, NULL, &zc);
6540 
6541 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6542 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6543 			    sizeof (zc_byteswap)));
6544 
6545 			/* Test ABD - metadata */
6546 			abd_fletcher_4_byteswap(abd_meta, size, NULL,
6547 			    &zc_byteswap);
6548 			abd_fletcher_4_native(abd_meta, size, NULL, &zc);
6549 
6550 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6551 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6552 			    sizeof (zc_byteswap)));
6553 
6554 		}
6555 
6556 		umem_free(buf, size);
6557 		abd_free(abd_data);
6558 		abd_free(abd_meta);
6559 	}
6560 }
6561 
6562 void
6563 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
6564 {
6565 	(void) zd, (void) id;
6566 	void *buf;
6567 	size_t size;
6568 	int *ptr;
6569 	int i;
6570 	zio_cksum_t zc_ref;
6571 	zio_cksum_t zc_ref_bswap;
6572 
6573 	hrtime_t end = gethrtime() + NANOSEC;
6574 
6575 	while (gethrtime() <= end) {
6576 		int run_count = 100;
6577 
6578 		size = ztest_random_blocksize();
6579 		buf = umem_alloc(size, UMEM_NOFAIL);
6580 
6581 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6582 			*ptr = ztest_random(UINT_MAX);
6583 
6584 		VERIFY0(fletcher_4_impl_set("scalar"));
6585 		fletcher_4_native(buf, size, NULL, &zc_ref);
6586 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
6587 
6588 		VERIFY0(fletcher_4_impl_set("cycle"));
6589 
6590 		while (run_count-- > 0) {
6591 			zio_cksum_t zc;
6592 			zio_cksum_t zc_bswap;
6593 			size_t pos = 0;
6594 
6595 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6596 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6597 
6598 			while (pos < size) {
6599 				size_t inc = 64 * ztest_random(size / 67);
6600 				/* sometimes add few bytes to test non-simd */
6601 				if (ztest_random(100) < 10)
6602 					inc += P2ALIGN(ztest_random(64),
6603 					    sizeof (uint32_t));
6604 
6605 				if (inc > (size - pos))
6606 					inc = size - pos;
6607 
6608 				fletcher_4_incremental_native(buf + pos, inc,
6609 				    &zc);
6610 				fletcher_4_incremental_byteswap(buf + pos, inc,
6611 				    &zc_bswap);
6612 
6613 				pos += inc;
6614 			}
6615 
6616 			VERIFY3U(pos, ==, size);
6617 
6618 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6619 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6620 
6621 			/*
6622 			 * verify if incremental on the whole buffer is
6623 			 * equivalent to non-incremental version
6624 			 */
6625 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6626 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6627 
6628 			fletcher_4_incremental_native(buf, size, &zc);
6629 			fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
6630 
6631 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6632 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6633 		}
6634 
6635 		umem_free(buf, size);
6636 	}
6637 }
6638 
6639 static int
6640 ztest_set_global_vars(void)
6641 {
6642 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
6643 		char *kv = ztest_opts.zo_gvars[i];
6644 		VERIFY3U(strlen(kv), <=, ZO_GVARS_MAX_ARGLEN);
6645 		VERIFY3U(strlen(kv), >, 0);
6646 		int err = set_global_var(kv);
6647 		if (ztest_opts.zo_verbose > 0) {
6648 			(void) printf("setting global var %s ... %s\n", kv,
6649 			    err ? "failed" : "ok");
6650 		}
6651 		if (err != 0) {
6652 			(void) fprintf(stderr,
6653 			    "failed to set global var '%s'\n", kv);
6654 			return (err);
6655 		}
6656 	}
6657 	return (0);
6658 }
6659 
6660 static char **
6661 ztest_global_vars_to_zdb_args(void)
6662 {
6663 	char **args = calloc(2*ztest_opts.zo_gvars_count + 1, sizeof (char *));
6664 	char **cur = args;
6665 	if (args == NULL)
6666 		return (NULL);
6667 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
6668 		*cur++ = (char *)"-o";
6669 		*cur++ = ztest_opts.zo_gvars[i];
6670 	}
6671 	ASSERT3P(cur, ==, &args[2*ztest_opts.zo_gvars_count]);
6672 	*cur = NULL;
6673 	return (args);
6674 }
6675 
6676 /* The end of strings is indicated by a NULL element */
6677 static char *
6678 join_strings(char **strings, const char *sep)
6679 {
6680 	size_t totallen = 0;
6681 	for (char **sp = strings; *sp != NULL; sp++) {
6682 		totallen += strlen(*sp);
6683 		totallen += strlen(sep);
6684 	}
6685 	if (totallen > 0) {
6686 		ASSERT(totallen >= strlen(sep));
6687 		totallen -= strlen(sep);
6688 	}
6689 
6690 	size_t buflen = totallen + 1;
6691 	char *o = umem_alloc(buflen, UMEM_NOFAIL); /* trailing 0 byte */
6692 	o[0] = '\0';
6693 	for (char **sp = strings; *sp != NULL; sp++) {
6694 		size_t would;
6695 		would = strlcat(o, *sp, buflen);
6696 		VERIFY3U(would, <, buflen);
6697 		if (*(sp+1) == NULL) {
6698 			break;
6699 		}
6700 		would = strlcat(o, sep, buflen);
6701 		VERIFY3U(would, <, buflen);
6702 	}
6703 	ASSERT3S(strlen(o), ==, totallen);
6704 	return (o);
6705 }
6706 
6707 static int
6708 ztest_check_path(char *path)
6709 {
6710 	struct stat s;
6711 	/* return true on success */
6712 	return (!stat(path, &s));
6713 }
6714 
6715 static void
6716 ztest_get_zdb_bin(char *bin, int len)
6717 {
6718 	char *zdb_path;
6719 	/*
6720 	 * Try to use $ZDB and in-tree zdb path. If not successful, just
6721 	 * let popen to search through PATH.
6722 	 */
6723 	if ((zdb_path = getenv("ZDB"))) {
6724 		strlcpy(bin, zdb_path, len); /* In env */
6725 		if (!ztest_check_path(bin)) {
6726 			ztest_dump_core = 0;
6727 			fatal(B_TRUE, "invalid ZDB '%s'", bin);
6728 		}
6729 		return;
6730 	}
6731 
6732 	VERIFY3P(realpath(getexecname(), bin), !=, NULL);
6733 	if (strstr(bin, ".libs/ztest")) {
6734 		strstr(bin, ".libs/ztest")[0] = '\0'; /* In-tree */
6735 		strcat(bin, "zdb");
6736 		if (ztest_check_path(bin))
6737 			return;
6738 	}
6739 	strcpy(bin, "zdb");
6740 }
6741 
6742 static vdev_t *
6743 ztest_random_concrete_vdev_leaf(vdev_t *vd)
6744 {
6745 	if (vd == NULL)
6746 		return (NULL);
6747 
6748 	if (vd->vdev_children == 0)
6749 		return (vd);
6750 
6751 	vdev_t *eligible[vd->vdev_children];
6752 	int eligible_idx = 0, i;
6753 	for (i = 0; i < vd->vdev_children; i++) {
6754 		vdev_t *cvd = vd->vdev_child[i];
6755 		if (cvd->vdev_top->vdev_removing)
6756 			continue;
6757 		if (cvd->vdev_children > 0 ||
6758 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
6759 			eligible[eligible_idx++] = cvd;
6760 		}
6761 	}
6762 	VERIFY3S(eligible_idx, >, 0);
6763 
6764 	uint64_t child_no = ztest_random(eligible_idx);
6765 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
6766 }
6767 
6768 void
6769 ztest_initialize(ztest_ds_t *zd, uint64_t id)
6770 {
6771 	(void) zd, (void) id;
6772 	spa_t *spa = ztest_spa;
6773 	int error = 0;
6774 
6775 	mutex_enter(&ztest_vdev_lock);
6776 
6777 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6778 
6779 	/* Random leaf vdev */
6780 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6781 	if (rand_vd == NULL) {
6782 		spa_config_exit(spa, SCL_VDEV, FTAG);
6783 		mutex_exit(&ztest_vdev_lock);
6784 		return;
6785 	}
6786 
6787 	/*
6788 	 * The random vdev we've selected may change as soon as we
6789 	 * drop the spa_config_lock. We create local copies of things
6790 	 * we're interested in.
6791 	 */
6792 	uint64_t guid = rand_vd->vdev_guid;
6793 	char *path = strdup(rand_vd->vdev_path);
6794 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
6795 
6796 	zfs_dbgmsg("vd %px, guid %llu", rand_vd, (u_longlong_t)guid);
6797 	spa_config_exit(spa, SCL_VDEV, FTAG);
6798 
6799 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
6800 
6801 	nvlist_t *vdev_guids = fnvlist_alloc();
6802 	nvlist_t *vdev_errlist = fnvlist_alloc();
6803 	fnvlist_add_uint64(vdev_guids, path, guid);
6804 	error = spa_vdev_initialize(spa, vdev_guids, cmd, vdev_errlist);
6805 	fnvlist_free(vdev_guids);
6806 	fnvlist_free(vdev_errlist);
6807 
6808 	switch (cmd) {
6809 	case POOL_INITIALIZE_CANCEL:
6810 		if (ztest_opts.zo_verbose >= 4) {
6811 			(void) printf("Cancel initialize %s", path);
6812 			if (!active)
6813 				(void) printf(" failed (no initialize active)");
6814 			(void) printf("\n");
6815 		}
6816 		break;
6817 	case POOL_INITIALIZE_START:
6818 		if (ztest_opts.zo_verbose >= 4) {
6819 			(void) printf("Start initialize %s", path);
6820 			if (active && error == 0)
6821 				(void) printf(" failed (already active)");
6822 			else if (error != 0)
6823 				(void) printf(" failed (error %d)", error);
6824 			(void) printf("\n");
6825 		}
6826 		break;
6827 	case POOL_INITIALIZE_SUSPEND:
6828 		if (ztest_opts.zo_verbose >= 4) {
6829 			(void) printf("Suspend initialize %s", path);
6830 			if (!active)
6831 				(void) printf(" failed (no initialize active)");
6832 			(void) printf("\n");
6833 		}
6834 		break;
6835 	}
6836 	free(path);
6837 	mutex_exit(&ztest_vdev_lock);
6838 }
6839 
6840 void
6841 ztest_trim(ztest_ds_t *zd, uint64_t id)
6842 {
6843 	(void) zd, (void) id;
6844 	spa_t *spa = ztest_spa;
6845 	int error = 0;
6846 
6847 	mutex_enter(&ztest_vdev_lock);
6848 
6849 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6850 
6851 	/* Random leaf vdev */
6852 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6853 	if (rand_vd == NULL) {
6854 		spa_config_exit(spa, SCL_VDEV, FTAG);
6855 		mutex_exit(&ztest_vdev_lock);
6856 		return;
6857 	}
6858 
6859 	/*
6860 	 * The random vdev we've selected may change as soon as we
6861 	 * drop the spa_config_lock. We create local copies of things
6862 	 * we're interested in.
6863 	 */
6864 	uint64_t guid = rand_vd->vdev_guid;
6865 	char *path = strdup(rand_vd->vdev_path);
6866 	boolean_t active = rand_vd->vdev_trim_thread != NULL;
6867 
6868 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, (u_longlong_t)guid);
6869 	spa_config_exit(spa, SCL_VDEV, FTAG);
6870 
6871 	uint64_t cmd = ztest_random(POOL_TRIM_FUNCS);
6872 	uint64_t rate = 1 << ztest_random(30);
6873 	boolean_t partial = (ztest_random(5) > 0);
6874 	boolean_t secure = (ztest_random(5) > 0);
6875 
6876 	nvlist_t *vdev_guids = fnvlist_alloc();
6877 	nvlist_t *vdev_errlist = fnvlist_alloc();
6878 	fnvlist_add_uint64(vdev_guids, path, guid);
6879 	error = spa_vdev_trim(spa, vdev_guids, cmd, rate, partial,
6880 	    secure, vdev_errlist);
6881 	fnvlist_free(vdev_guids);
6882 	fnvlist_free(vdev_errlist);
6883 
6884 	switch (cmd) {
6885 	case POOL_TRIM_CANCEL:
6886 		if (ztest_opts.zo_verbose >= 4) {
6887 			(void) printf("Cancel TRIM %s", path);
6888 			if (!active)
6889 				(void) printf(" failed (no TRIM active)");
6890 			(void) printf("\n");
6891 		}
6892 		break;
6893 	case POOL_TRIM_START:
6894 		if (ztest_opts.zo_verbose >= 4) {
6895 			(void) printf("Start TRIM %s", path);
6896 			if (active && error == 0)
6897 				(void) printf(" failed (already active)");
6898 			else if (error != 0)
6899 				(void) printf(" failed (error %d)", error);
6900 			(void) printf("\n");
6901 		}
6902 		break;
6903 	case POOL_TRIM_SUSPEND:
6904 		if (ztest_opts.zo_verbose >= 4) {
6905 			(void) printf("Suspend TRIM %s", path);
6906 			if (!active)
6907 				(void) printf(" failed (no TRIM active)");
6908 			(void) printf("\n");
6909 		}
6910 		break;
6911 	}
6912 	free(path);
6913 	mutex_exit(&ztest_vdev_lock);
6914 }
6915 
6916 /*
6917  * Verify pool integrity by running zdb.
6918  */
6919 static void
6920 ztest_run_zdb(const char *pool)
6921 {
6922 	int status;
6923 	char *bin;
6924 	char *zdb;
6925 	char *zbuf;
6926 	const int len = MAXPATHLEN + MAXNAMELEN + 20;
6927 	FILE *fp;
6928 
6929 	bin = umem_alloc(len, UMEM_NOFAIL);
6930 	zdb = umem_alloc(len, UMEM_NOFAIL);
6931 	zbuf = umem_alloc(1024, UMEM_NOFAIL);
6932 
6933 	ztest_get_zdb_bin(bin, len);
6934 
6935 	char **set_gvars_args = ztest_global_vars_to_zdb_args();
6936 	if (set_gvars_args == NULL) {
6937 		fatal(B_FALSE, "Failed to allocate memory in "
6938 		    "ztest_global_vars_to_zdb_args(). Cannot run zdb.\n");
6939 	}
6940 	char *set_gvars_args_joined = join_strings(set_gvars_args, " ");
6941 	free(set_gvars_args);
6942 
6943 	size_t would = snprintf(zdb, len,
6944 	    "%s -bcc%s%s -G -d -Y -e -y %s -p %s %s",
6945 	    bin,
6946 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
6947 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
6948 	    set_gvars_args_joined,
6949 	    ztest_opts.zo_dir,
6950 	    pool);
6951 	ASSERT3U(would, <, len);
6952 
6953 	umem_free(set_gvars_args_joined, strlen(set_gvars_args_joined) + 1);
6954 
6955 	if (ztest_opts.zo_verbose >= 5)
6956 		(void) printf("Executing %s\n", zdb);
6957 
6958 	fp = popen(zdb, "r");
6959 
6960 	while (fgets(zbuf, 1024, fp) != NULL)
6961 		if (ztest_opts.zo_verbose >= 3)
6962 			(void) printf("%s", zbuf);
6963 
6964 	status = pclose(fp);
6965 
6966 	if (status == 0)
6967 		goto out;
6968 
6969 	ztest_dump_core = 0;
6970 	if (WIFEXITED(status))
6971 		fatal(B_FALSE, "'%s' exit code %d", zdb, WEXITSTATUS(status));
6972 	else
6973 		fatal(B_FALSE, "'%s' died with signal %d",
6974 		    zdb, WTERMSIG(status));
6975 out:
6976 	umem_free(bin, len);
6977 	umem_free(zdb, len);
6978 	umem_free(zbuf, 1024);
6979 }
6980 
6981 static void
6982 ztest_walk_pool_directory(const char *header)
6983 {
6984 	spa_t *spa = NULL;
6985 
6986 	if (ztest_opts.zo_verbose >= 6)
6987 		(void) puts(header);
6988 
6989 	mutex_enter(&spa_namespace_lock);
6990 	while ((spa = spa_next(spa)) != NULL)
6991 		if (ztest_opts.zo_verbose >= 6)
6992 			(void) printf("\t%s\n", spa_name(spa));
6993 	mutex_exit(&spa_namespace_lock);
6994 }
6995 
6996 static void
6997 ztest_spa_import_export(char *oldname, char *newname)
6998 {
6999 	nvlist_t *config, *newconfig;
7000 	uint64_t pool_guid;
7001 	spa_t *spa;
7002 	int error;
7003 
7004 	if (ztest_opts.zo_verbose >= 4) {
7005 		(void) printf("import/export: old = %s, new = %s\n",
7006 		    oldname, newname);
7007 	}
7008 
7009 	/*
7010 	 * Clean up from previous runs.
7011 	 */
7012 	(void) spa_destroy(newname);
7013 
7014 	/*
7015 	 * Get the pool's configuration and guid.
7016 	 */
7017 	VERIFY0(spa_open(oldname, &spa, FTAG));
7018 
7019 	/*
7020 	 * Kick off a scrub to tickle scrub/export races.
7021 	 */
7022 	if (ztest_random(2) == 0)
7023 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
7024 
7025 	pool_guid = spa_guid(spa);
7026 	spa_close(spa, FTAG);
7027 
7028 	ztest_walk_pool_directory("pools before export");
7029 
7030 	/*
7031 	 * Export it.
7032 	 */
7033 	VERIFY0(spa_export(oldname, &config, B_FALSE, B_FALSE));
7034 
7035 	ztest_walk_pool_directory("pools after export");
7036 
7037 	/*
7038 	 * Try to import it.
7039 	 */
7040 	newconfig = spa_tryimport(config);
7041 	ASSERT3P(newconfig, !=, NULL);
7042 	fnvlist_free(newconfig);
7043 
7044 	/*
7045 	 * Import it under the new name.
7046 	 */
7047 	error = spa_import(newname, config, NULL, 0);
7048 	if (error != 0) {
7049 		dump_nvlist(config, 0);
7050 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
7051 		    oldname, newname, error);
7052 	}
7053 
7054 	ztest_walk_pool_directory("pools after import");
7055 
7056 	/*
7057 	 * Try to import it again -- should fail with EEXIST.
7058 	 */
7059 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
7060 
7061 	/*
7062 	 * Try to import it under a different name -- should fail with EEXIST.
7063 	 */
7064 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
7065 
7066 	/*
7067 	 * Verify that the pool is no longer visible under the old name.
7068 	 */
7069 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
7070 
7071 	/*
7072 	 * Verify that we can open and close the pool using the new name.
7073 	 */
7074 	VERIFY0(spa_open(newname, &spa, FTAG));
7075 	ASSERT3U(pool_guid, ==, spa_guid(spa));
7076 	spa_close(spa, FTAG);
7077 
7078 	fnvlist_free(config);
7079 }
7080 
7081 static void
7082 ztest_resume(spa_t *spa)
7083 {
7084 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
7085 		(void) printf("resuming from suspended state\n");
7086 	spa_vdev_state_enter(spa, SCL_NONE);
7087 	vdev_clear(spa, NULL);
7088 	(void) spa_vdev_state_exit(spa, NULL, 0);
7089 	(void) zio_resume(spa);
7090 }
7091 
7092 static __attribute__((noreturn)) void
7093 ztest_resume_thread(void *arg)
7094 {
7095 	spa_t *spa = arg;
7096 
7097 	while (!ztest_exiting) {
7098 		if (spa_suspended(spa))
7099 			ztest_resume(spa);
7100 		(void) poll(NULL, 0, 100);
7101 
7102 		/*
7103 		 * Periodically change the zfs_compressed_arc_enabled setting.
7104 		 */
7105 		if (ztest_random(10) == 0)
7106 			zfs_compressed_arc_enabled = ztest_random(2);
7107 
7108 		/*
7109 		 * Periodically change the zfs_abd_scatter_enabled setting.
7110 		 */
7111 		if (ztest_random(10) == 0)
7112 			zfs_abd_scatter_enabled = ztest_random(2);
7113 	}
7114 
7115 	thread_exit();
7116 }
7117 
7118 static __attribute__((noreturn)) void
7119 ztest_deadman_thread(void *arg)
7120 {
7121 	ztest_shared_t *zs = arg;
7122 	spa_t *spa = ztest_spa;
7123 	hrtime_t delay, overdue, last_run = gethrtime();
7124 
7125 	delay = (zs->zs_thread_stop - zs->zs_thread_start) +
7126 	    MSEC2NSEC(zfs_deadman_synctime_ms);
7127 
7128 	while (!ztest_exiting) {
7129 		/*
7130 		 * Wait for the delay timer while checking occasionally
7131 		 * if we should stop.
7132 		 */
7133 		if (gethrtime() < last_run + delay) {
7134 			(void) poll(NULL, 0, 1000);
7135 			continue;
7136 		}
7137 
7138 		/*
7139 		 * If the pool is suspended then fail immediately. Otherwise,
7140 		 * check to see if the pool is making any progress. If
7141 		 * vdev_deadman() discovers that there hasn't been any recent
7142 		 * I/Os then it will end up aborting the tests.
7143 		 */
7144 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
7145 			fatal(B_FALSE,
7146 			    "aborting test after %llu seconds because "
7147 			    "pool has transitioned to a suspended state.",
7148 			    (u_longlong_t)zfs_deadman_synctime_ms / 1000);
7149 		}
7150 		vdev_deadman(spa->spa_root_vdev, FTAG);
7151 
7152 		/*
7153 		 * If the process doesn't complete within a grace period of
7154 		 * zfs_deadman_synctime_ms over the expected finish time,
7155 		 * then it may be hung and is terminated.
7156 		 */
7157 		overdue = zs->zs_proc_stop + MSEC2NSEC(zfs_deadman_synctime_ms);
7158 		if (gethrtime() > overdue) {
7159 			fatal(B_FALSE,
7160 			    "aborting test after %llu seconds because "
7161 			    "the process is overdue for termination.",
7162 			    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7163 		}
7164 
7165 		(void) printf("ztest has been running for %lld seconds\n",
7166 		    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7167 
7168 		last_run = gethrtime();
7169 		delay = MSEC2NSEC(zfs_deadman_checktime_ms);
7170 	}
7171 
7172 	thread_exit();
7173 }
7174 
7175 static void
7176 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
7177 {
7178 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
7179 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
7180 	hrtime_t functime = gethrtime();
7181 	int i;
7182 
7183 	for (i = 0; i < zi->zi_iters; i++)
7184 		zi->zi_func(zd, id);
7185 
7186 	functime = gethrtime() - functime;
7187 
7188 	atomic_add_64(&zc->zc_count, 1);
7189 	atomic_add_64(&zc->zc_time, functime);
7190 
7191 	if (ztest_opts.zo_verbose >= 4)
7192 		(void) printf("%6.2f sec in %s\n",
7193 		    (double)functime / NANOSEC, zi->zi_funcname);
7194 }
7195 
7196 static __attribute__((noreturn)) void
7197 ztest_thread(void *arg)
7198 {
7199 	int rand;
7200 	uint64_t id = (uintptr_t)arg;
7201 	ztest_shared_t *zs = ztest_shared;
7202 	uint64_t call_next;
7203 	hrtime_t now;
7204 	ztest_info_t *zi;
7205 	ztest_shared_callstate_t *zc;
7206 
7207 	while ((now = gethrtime()) < zs->zs_thread_stop) {
7208 		/*
7209 		 * See if it's time to force a crash.
7210 		 */
7211 		if (now > zs->zs_thread_kill)
7212 			ztest_kill(zs);
7213 
7214 		/*
7215 		 * If we're getting ENOSPC with some regularity, stop.
7216 		 */
7217 		if (zs->zs_enospc_count > 10)
7218 			break;
7219 
7220 		/*
7221 		 * Pick a random function to execute.
7222 		 */
7223 		rand = ztest_random(ZTEST_FUNCS);
7224 		zi = &ztest_info[rand];
7225 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
7226 		call_next = zc->zc_next;
7227 
7228 		if (now >= call_next &&
7229 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
7230 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
7231 			ztest_execute(rand, zi, id);
7232 		}
7233 	}
7234 
7235 	thread_exit();
7236 }
7237 
7238 static void
7239 ztest_dataset_name(char *dsname, const char *pool, int d)
7240 {
7241 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
7242 }
7243 
7244 static void
7245 ztest_dataset_destroy(int d)
7246 {
7247 	char name[ZFS_MAX_DATASET_NAME_LEN];
7248 	int t;
7249 
7250 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7251 
7252 	if (ztest_opts.zo_verbose >= 3)
7253 		(void) printf("Destroying %s to free up space\n", name);
7254 
7255 	/*
7256 	 * Cleanup any non-standard clones and snapshots.  In general,
7257 	 * ztest thread t operates on dataset (t % zopt_datasets),
7258 	 * so there may be more than one thing to clean up.
7259 	 */
7260 	for (t = d; t < ztest_opts.zo_threads;
7261 	    t += ztest_opts.zo_datasets)
7262 		ztest_dsl_dataset_cleanup(name, t);
7263 
7264 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
7265 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
7266 }
7267 
7268 static void
7269 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
7270 {
7271 	uint64_t usedobjs, dirobjs, scratch;
7272 
7273 	/*
7274 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
7275 	 * Therefore, the number of objects in use should equal the
7276 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
7277 	 * If not, we have an object leak.
7278 	 *
7279 	 * Note that we can only check this in ztest_dataset_open(),
7280 	 * when the open-context and syncing-context values agree.
7281 	 * That's because zap_count() returns the open-context value,
7282 	 * while dmu_objset_space() returns the rootbp fill count.
7283 	 */
7284 	VERIFY0(zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
7285 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
7286 	ASSERT3U(dirobjs + 1, ==, usedobjs);
7287 }
7288 
7289 static int
7290 ztest_dataset_open(int d)
7291 {
7292 	ztest_ds_t *zd = &ztest_ds[d];
7293 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
7294 	objset_t *os;
7295 	zilog_t *zilog;
7296 	char name[ZFS_MAX_DATASET_NAME_LEN];
7297 	int error;
7298 
7299 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7300 
7301 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
7302 
7303 	error = ztest_dataset_create(name);
7304 	if (error == ENOSPC) {
7305 		(void) pthread_rwlock_unlock(&ztest_name_lock);
7306 		ztest_record_enospc(FTAG);
7307 		return (error);
7308 	}
7309 	ASSERT(error == 0 || error == EEXIST);
7310 
7311 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
7312 	    B_TRUE, zd, &os));
7313 	(void) pthread_rwlock_unlock(&ztest_name_lock);
7314 
7315 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
7316 
7317 	zilog = zd->zd_zilog;
7318 
7319 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
7320 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
7321 		fatal(B_FALSE, "missing log records: "
7322 		    "claimed %"PRIu64" < committed %"PRIu64"",
7323 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
7324 
7325 	ztest_dataset_dirobj_verify(zd);
7326 
7327 	zil_replay(os, zd, ztest_replay_vector);
7328 
7329 	ztest_dataset_dirobj_verify(zd);
7330 
7331 	if (ztest_opts.zo_verbose >= 6)
7332 		(void) printf("%s replay %"PRIu64" blocks, "
7333 		    "%"PRIu64" records, seq %"PRIu64"\n",
7334 		    zd->zd_name,
7335 		    zilog->zl_parse_blk_count,
7336 		    zilog->zl_parse_lr_count,
7337 		    zilog->zl_replaying_seq);
7338 
7339 	zilog = zil_open(os, ztest_get_data, NULL);
7340 
7341 	if (zilog->zl_replaying_seq != 0 &&
7342 	    zilog->zl_replaying_seq < committed_seq)
7343 		fatal(B_FALSE, "missing log records: "
7344 		    "replayed %"PRIu64" < committed %"PRIu64"",
7345 		    zilog->zl_replaying_seq, committed_seq);
7346 
7347 	return (0);
7348 }
7349 
7350 static void
7351 ztest_dataset_close(int d)
7352 {
7353 	ztest_ds_t *zd = &ztest_ds[d];
7354 
7355 	zil_close(zd->zd_zilog);
7356 	dmu_objset_disown(zd->zd_os, B_TRUE, zd);
7357 
7358 	ztest_zd_fini(zd);
7359 }
7360 
7361 static int
7362 ztest_replay_zil_cb(const char *name, void *arg)
7363 {
7364 	(void) arg;
7365 	objset_t *os;
7366 	ztest_ds_t *zdtmp;
7367 
7368 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_ANY, B_TRUE,
7369 	    B_TRUE, FTAG, &os));
7370 
7371 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
7372 
7373 	ztest_zd_init(zdtmp, NULL, os);
7374 	zil_replay(os, zdtmp, ztest_replay_vector);
7375 	ztest_zd_fini(zdtmp);
7376 
7377 	if (dmu_objset_zil(os)->zl_parse_lr_count != 0 &&
7378 	    ztest_opts.zo_verbose >= 6) {
7379 		zilog_t *zilog = dmu_objset_zil(os);
7380 
7381 		(void) printf("%s replay %"PRIu64" blocks, "
7382 		    "%"PRIu64" records, seq %"PRIu64"\n",
7383 		    name,
7384 		    zilog->zl_parse_blk_count,
7385 		    zilog->zl_parse_lr_count,
7386 		    zilog->zl_replaying_seq);
7387 	}
7388 
7389 	umem_free(zdtmp, sizeof (ztest_ds_t));
7390 
7391 	dmu_objset_disown(os, B_TRUE, FTAG);
7392 	return (0);
7393 }
7394 
7395 static void
7396 ztest_freeze(void)
7397 {
7398 	ztest_ds_t *zd = &ztest_ds[0];
7399 	spa_t *spa;
7400 	int numloops = 0;
7401 
7402 	if (ztest_opts.zo_verbose >= 3)
7403 		(void) printf("testing spa_freeze()...\n");
7404 
7405 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7406 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7407 	VERIFY0(ztest_dataset_open(0));
7408 	ztest_spa = spa;
7409 
7410 	/*
7411 	 * Force the first log block to be transactionally allocated.
7412 	 * We have to do this before we freeze the pool -- otherwise
7413 	 * the log chain won't be anchored.
7414 	 */
7415 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
7416 		ztest_dmu_object_alloc_free(zd, 0);
7417 		zil_commit(zd->zd_zilog, 0);
7418 	}
7419 
7420 	txg_wait_synced(spa_get_dsl(spa), 0);
7421 
7422 	/*
7423 	 * Freeze the pool.  This stops spa_sync() from doing anything,
7424 	 * so that the only way to record changes from now on is the ZIL.
7425 	 */
7426 	spa_freeze(spa);
7427 
7428 	/*
7429 	 * Because it is hard to predict how much space a write will actually
7430 	 * require beforehand, we leave ourselves some fudge space to write over
7431 	 * capacity.
7432 	 */
7433 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
7434 
7435 	/*
7436 	 * Run tests that generate log records but don't alter the pool config
7437 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
7438 	 * We do a txg_wait_synced() after each iteration to force the txg
7439 	 * to increase well beyond the last synced value in the uberblock.
7440 	 * The ZIL should be OK with that.
7441 	 *
7442 	 * Run a random number of times less than zo_maxloops and ensure we do
7443 	 * not run out of space on the pool.
7444 	 */
7445 	while (ztest_random(10) != 0 &&
7446 	    numloops++ < ztest_opts.zo_maxloops &&
7447 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
7448 		ztest_od_t od;
7449 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
7450 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
7451 		ztest_io(zd, od.od_object,
7452 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
7453 		txg_wait_synced(spa_get_dsl(spa), 0);
7454 	}
7455 
7456 	/*
7457 	 * Commit all of the changes we just generated.
7458 	 */
7459 	zil_commit(zd->zd_zilog, 0);
7460 	txg_wait_synced(spa_get_dsl(spa), 0);
7461 
7462 	/*
7463 	 * Close our dataset and close the pool.
7464 	 */
7465 	ztest_dataset_close(0);
7466 	spa_close(spa, FTAG);
7467 	kernel_fini();
7468 
7469 	/*
7470 	 * Open and close the pool and dataset to induce log replay.
7471 	 */
7472 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7473 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7474 	ASSERT3U(spa_freeze_txg(spa), ==, UINT64_MAX);
7475 	VERIFY0(ztest_dataset_open(0));
7476 	ztest_spa = spa;
7477 	txg_wait_synced(spa_get_dsl(spa), 0);
7478 	ztest_dataset_close(0);
7479 	ztest_reguid(NULL, 0);
7480 
7481 	spa_close(spa, FTAG);
7482 	kernel_fini();
7483 }
7484 
7485 static void
7486 ztest_import_impl(void)
7487 {
7488 	importargs_t args = { 0 };
7489 	nvlist_t *cfg = NULL;
7490 	int nsearch = 1;
7491 	char *searchdirs[nsearch];
7492 	int flags = ZFS_IMPORT_MISSING_LOG;
7493 
7494 	searchdirs[0] = ztest_opts.zo_dir;
7495 	args.paths = nsearch;
7496 	args.path = searchdirs;
7497 	args.can_be_active = B_FALSE;
7498 
7499 	libpc_handle_t lpch = {
7500 		.lpc_lib_handle = NULL,
7501 		.lpc_ops = &libzpool_config_ops,
7502 		.lpc_printerr = B_TRUE
7503 	};
7504 	VERIFY0(zpool_find_config(&lpch, ztest_opts.zo_pool, &cfg, &args));
7505 	VERIFY0(spa_import(ztest_opts.zo_pool, cfg, NULL, flags));
7506 	fnvlist_free(cfg);
7507 }
7508 
7509 /*
7510  * Import a storage pool with the given name.
7511  */
7512 static void
7513 ztest_import(ztest_shared_t *zs)
7514 {
7515 	spa_t *spa;
7516 
7517 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7518 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7519 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7520 
7521 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7522 
7523 	ztest_import_impl();
7524 
7525 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7526 	zs->zs_metaslab_sz =
7527 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7528 	spa_close(spa, FTAG);
7529 
7530 	kernel_fini();
7531 
7532 	if (!ztest_opts.zo_mmp_test) {
7533 		ztest_run_zdb(ztest_opts.zo_pool);
7534 		ztest_freeze();
7535 		ztest_run_zdb(ztest_opts.zo_pool);
7536 	}
7537 
7538 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7539 	mutex_destroy(&ztest_vdev_lock);
7540 	mutex_destroy(&ztest_checkpoint_lock);
7541 }
7542 
7543 /*
7544  * Kick off threads to run tests on all datasets in parallel.
7545  */
7546 static void
7547 ztest_run(ztest_shared_t *zs)
7548 {
7549 	spa_t *spa;
7550 	objset_t *os;
7551 	kthread_t *resume_thread, *deadman_thread;
7552 	kthread_t **run_threads;
7553 	uint64_t object;
7554 	int error;
7555 	int t, d;
7556 
7557 	ztest_exiting = B_FALSE;
7558 
7559 	/*
7560 	 * Initialize parent/child shared state.
7561 	 */
7562 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7563 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7564 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7565 
7566 	zs->zs_thread_start = gethrtime();
7567 	zs->zs_thread_stop =
7568 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
7569 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
7570 	zs->zs_thread_kill = zs->zs_thread_stop;
7571 	if (ztest_random(100) < ztest_opts.zo_killrate) {
7572 		zs->zs_thread_kill -=
7573 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
7574 	}
7575 
7576 	mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
7577 
7578 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
7579 	    offsetof(ztest_cb_data_t, zcd_node));
7580 
7581 	/*
7582 	 * Open our pool.  It may need to be imported first depending on
7583 	 * what tests were running when the previous pass was terminated.
7584 	 */
7585 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7586 	error = spa_open(ztest_opts.zo_pool, &spa, FTAG);
7587 	if (error) {
7588 		VERIFY3S(error, ==, ENOENT);
7589 		ztest_import_impl();
7590 		VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7591 		zs->zs_metaslab_sz =
7592 		    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7593 	}
7594 
7595 	metaslab_preload_limit = ztest_random(20) + 1;
7596 	ztest_spa = spa;
7597 
7598 	VERIFY0(vdev_raidz_impl_set("cycle"));
7599 
7600 	dmu_objset_stats_t dds;
7601 	VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
7602 	    DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
7603 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
7604 	dmu_objset_fast_stat(os, &dds);
7605 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
7606 	zs->zs_guid = dds.dds_guid;
7607 	dmu_objset_disown(os, B_TRUE, FTAG);
7608 
7609 	/*
7610 	 * Create a thread to periodically resume suspended I/O.
7611 	 */
7612 	resume_thread = thread_create(NULL, 0, ztest_resume_thread,
7613 	    spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
7614 
7615 	/*
7616 	 * Create a deadman thread and set to panic if we hang.
7617 	 */
7618 	deadman_thread = thread_create(NULL, 0, ztest_deadman_thread,
7619 	    zs, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
7620 
7621 	spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
7622 
7623 	/*
7624 	 * Verify that we can safely inquire about any object,
7625 	 * whether it's allocated or not.  To make it interesting,
7626 	 * we probe a 5-wide window around each power of two.
7627 	 * This hits all edge cases, including zero and the max.
7628 	 */
7629 	for (t = 0; t < 64; t++) {
7630 		for (d = -5; d <= 5; d++) {
7631 			error = dmu_object_info(spa->spa_meta_objset,
7632 			    (1ULL << t) + d, NULL);
7633 			ASSERT(error == 0 || error == ENOENT ||
7634 			    error == EINVAL);
7635 		}
7636 	}
7637 
7638 	/*
7639 	 * If we got any ENOSPC errors on the previous run, destroy something.
7640 	 */
7641 	if (zs->zs_enospc_count != 0) {
7642 		int d = ztest_random(ztest_opts.zo_datasets);
7643 		ztest_dataset_destroy(d);
7644 	}
7645 	zs->zs_enospc_count = 0;
7646 
7647 	/*
7648 	 * If we were in the middle of ztest_device_removal() and were killed
7649 	 * we need to ensure the removal and scrub complete before running
7650 	 * any tests that check ztest_device_removal_active. The removal will
7651 	 * be restarted automatically when the spa is opened, but we need to
7652 	 * initiate the scrub manually if it is not already in progress. Note
7653 	 * that we always run the scrub whenever an indirect vdev exists
7654 	 * because we have no way of knowing for sure if ztest_device_removal()
7655 	 * fully completed its scrub before the pool was reimported.
7656 	 */
7657 	if (spa->spa_removing_phys.sr_state == DSS_SCANNING ||
7658 	    spa->spa_removing_phys.sr_prev_indirect_vdev != -1) {
7659 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
7660 			txg_wait_synced(spa_get_dsl(spa), 0);
7661 
7662 		error = ztest_scrub_impl(spa);
7663 		if (error == EBUSY)
7664 			error = 0;
7665 		ASSERT0(error);
7666 	}
7667 
7668 	run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
7669 	    UMEM_NOFAIL);
7670 
7671 	if (ztest_opts.zo_verbose >= 4)
7672 		(void) printf("starting main threads...\n");
7673 
7674 	/*
7675 	 * Replay all logs of all datasets in the pool. This is primarily for
7676 	 * temporary datasets which wouldn't otherwise get replayed, which
7677 	 * can trigger failures when attempting to offline a SLOG in
7678 	 * ztest_fault_inject().
7679 	 */
7680 	(void) dmu_objset_find(ztest_opts.zo_pool, ztest_replay_zil_cb,
7681 	    NULL, DS_FIND_CHILDREN);
7682 
7683 	/*
7684 	 * Kick off all the tests that run in parallel.
7685 	 */
7686 	for (t = 0; t < ztest_opts.zo_threads; t++) {
7687 		if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
7688 			umem_free(run_threads, ztest_opts.zo_threads *
7689 			    sizeof (kthread_t *));
7690 			return;
7691 		}
7692 
7693 		run_threads[t] = thread_create(NULL, 0, ztest_thread,
7694 		    (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
7695 		    defclsyspri);
7696 	}
7697 
7698 	/*
7699 	 * Wait for all of the tests to complete.
7700 	 */
7701 	for (t = 0; t < ztest_opts.zo_threads; t++)
7702 		VERIFY0(thread_join(run_threads[t]));
7703 
7704 	/*
7705 	 * Close all datasets. This must be done after all the threads
7706 	 * are joined so we can be sure none of the datasets are in-use
7707 	 * by any of the threads.
7708 	 */
7709 	for (t = 0; t < ztest_opts.zo_threads; t++) {
7710 		if (t < ztest_opts.zo_datasets)
7711 			ztest_dataset_close(t);
7712 	}
7713 
7714 	txg_wait_synced(spa_get_dsl(spa), 0);
7715 
7716 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
7717 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
7718 
7719 	umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
7720 
7721 	/* Kill the resume and deadman threads */
7722 	ztest_exiting = B_TRUE;
7723 	VERIFY0(thread_join(resume_thread));
7724 	VERIFY0(thread_join(deadman_thread));
7725 	ztest_resume(spa);
7726 
7727 	/*
7728 	 * Right before closing the pool, kick off a bunch of async I/O;
7729 	 * spa_close() should wait for it to complete.
7730 	 */
7731 	for (object = 1; object < 50; object++) {
7732 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
7733 		    ZIO_PRIORITY_SYNC_READ);
7734 	}
7735 
7736 	/* Verify that at least one commit cb was called in a timely fashion */
7737 	if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
7738 		VERIFY0(zc_min_txg_delay);
7739 
7740 	spa_close(spa, FTAG);
7741 
7742 	/*
7743 	 * Verify that we can loop over all pools.
7744 	 */
7745 	mutex_enter(&spa_namespace_lock);
7746 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
7747 		if (ztest_opts.zo_verbose > 3)
7748 			(void) printf("spa_next: found %s\n", spa_name(spa));
7749 	mutex_exit(&spa_namespace_lock);
7750 
7751 	/*
7752 	 * Verify that we can export the pool and reimport it under a
7753 	 * different name.
7754 	 */
7755 	if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
7756 		char name[ZFS_MAX_DATASET_NAME_LEN];
7757 		(void) snprintf(name, sizeof (name), "%s_import",
7758 		    ztest_opts.zo_pool);
7759 		ztest_spa_import_export(ztest_opts.zo_pool, name);
7760 		ztest_spa_import_export(name, ztest_opts.zo_pool);
7761 	}
7762 
7763 	kernel_fini();
7764 
7765 	list_destroy(&zcl.zcl_callbacks);
7766 	mutex_destroy(&zcl.zcl_callbacks_lock);
7767 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7768 	mutex_destroy(&ztest_vdev_lock);
7769 	mutex_destroy(&ztest_checkpoint_lock);
7770 }
7771 
7772 static void
7773 print_time(hrtime_t t, char *timebuf)
7774 {
7775 	hrtime_t s = t / NANOSEC;
7776 	hrtime_t m = s / 60;
7777 	hrtime_t h = m / 60;
7778 	hrtime_t d = h / 24;
7779 
7780 	s -= m * 60;
7781 	m -= h * 60;
7782 	h -= d * 24;
7783 
7784 	timebuf[0] = '\0';
7785 
7786 	if (d)
7787 		(void) sprintf(timebuf,
7788 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
7789 	else if (h)
7790 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
7791 	else if (m)
7792 		(void) sprintf(timebuf, "%llum%02llus", m, s);
7793 	else
7794 		(void) sprintf(timebuf, "%llus", s);
7795 }
7796 
7797 static nvlist_t *
7798 make_random_props(void)
7799 {
7800 	nvlist_t *props;
7801 
7802 	props = fnvlist_alloc();
7803 
7804 	if (ztest_random(2) == 0)
7805 		return (props);
7806 
7807 	fnvlist_add_uint64(props,
7808 	    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 1);
7809 
7810 	return (props);
7811 }
7812 
7813 /*
7814  * Create a storage pool with the given name and initial vdev size.
7815  * Then test spa_freeze() functionality.
7816  */
7817 static void
7818 ztest_init(ztest_shared_t *zs)
7819 {
7820 	spa_t *spa;
7821 	nvlist_t *nvroot, *props;
7822 	int i;
7823 
7824 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7825 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7826 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7827 
7828 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7829 
7830 	/*
7831 	 * Create the storage pool.
7832 	 */
7833 	(void) spa_destroy(ztest_opts.zo_pool);
7834 	ztest_shared->zs_vdev_next_leaf = 0;
7835 	zs->zs_splits = 0;
7836 	zs->zs_mirrors = ztest_opts.zo_mirrors;
7837 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
7838 	    NULL, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
7839 	props = make_random_props();
7840 
7841 	/*
7842 	 * We don't expect the pool to suspend unless maxfaults == 0,
7843 	 * in which case ztest_fault_inject() temporarily takes away
7844 	 * the only valid replica.
7845 	 */
7846 	fnvlist_add_uint64(props,
7847 	    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
7848 	    MAXFAULTS(zs) ? ZIO_FAILURE_MODE_PANIC : ZIO_FAILURE_MODE_WAIT);
7849 
7850 	for (i = 0; i < SPA_FEATURES; i++) {
7851 		char *buf;
7852 
7853 		if (!spa_feature_table[i].fi_zfs_mod_supported)
7854 			continue;
7855 
7856 		/*
7857 		 * 75% chance of using the log space map feature. We want ztest
7858 		 * to exercise both the code paths that use the log space map
7859 		 * feature and the ones that don't.
7860 		 */
7861 		if (i == SPA_FEATURE_LOG_SPACEMAP && ztest_random(4) == 0)
7862 			continue;
7863 
7864 		VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
7865 		    spa_feature_table[i].fi_uname));
7866 		fnvlist_add_uint64(props, buf, 0);
7867 		free(buf);
7868 	}
7869 
7870 	VERIFY0(spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
7871 	fnvlist_free(nvroot);
7872 	fnvlist_free(props);
7873 
7874 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7875 	zs->zs_metaslab_sz =
7876 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7877 	spa_close(spa, FTAG);
7878 
7879 	kernel_fini();
7880 
7881 	if (!ztest_opts.zo_mmp_test) {
7882 		ztest_run_zdb(ztest_opts.zo_pool);
7883 		ztest_freeze();
7884 		ztest_run_zdb(ztest_opts.zo_pool);
7885 	}
7886 
7887 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7888 	mutex_destroy(&ztest_vdev_lock);
7889 	mutex_destroy(&ztest_checkpoint_lock);
7890 }
7891 
7892 static void
7893 setup_data_fd(void)
7894 {
7895 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
7896 
7897 	ztest_fd_data = mkstemp(ztest_name_data);
7898 	ASSERT3S(ztest_fd_data, >=, 0);
7899 	(void) unlink(ztest_name_data);
7900 }
7901 
7902 static int
7903 shared_data_size(ztest_shared_hdr_t *hdr)
7904 {
7905 	int size;
7906 
7907 	size = hdr->zh_hdr_size;
7908 	size += hdr->zh_opts_size;
7909 	size += hdr->zh_size;
7910 	size += hdr->zh_stats_size * hdr->zh_stats_count;
7911 	size += hdr->zh_ds_size * hdr->zh_ds_count;
7912 
7913 	return (size);
7914 }
7915 
7916 static void
7917 setup_hdr(void)
7918 {
7919 	int size;
7920 	ztest_shared_hdr_t *hdr;
7921 
7922 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7923 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7924 	ASSERT3P(hdr, !=, MAP_FAILED);
7925 
7926 	VERIFY0(ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
7927 
7928 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
7929 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
7930 	hdr->zh_size = sizeof (ztest_shared_t);
7931 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
7932 	hdr->zh_stats_count = ZTEST_FUNCS;
7933 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
7934 	hdr->zh_ds_count = ztest_opts.zo_datasets;
7935 
7936 	size = shared_data_size(hdr);
7937 	VERIFY0(ftruncate(ztest_fd_data, size));
7938 
7939 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7940 }
7941 
7942 static void
7943 setup_data(void)
7944 {
7945 	int size, offset;
7946 	ztest_shared_hdr_t *hdr;
7947 	uint8_t *buf;
7948 
7949 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7950 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
7951 	ASSERT3P(hdr, !=, MAP_FAILED);
7952 
7953 	size = shared_data_size(hdr);
7954 
7955 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7956 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
7957 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7958 	ASSERT3P(hdr, !=, MAP_FAILED);
7959 	buf = (uint8_t *)hdr;
7960 
7961 	offset = hdr->zh_hdr_size;
7962 	ztest_shared_opts = (void *)&buf[offset];
7963 	offset += hdr->zh_opts_size;
7964 	ztest_shared = (void *)&buf[offset];
7965 	offset += hdr->zh_size;
7966 	ztest_shared_callstate = (void *)&buf[offset];
7967 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
7968 	ztest_shared_ds = (void *)&buf[offset];
7969 }
7970 
7971 static boolean_t
7972 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
7973 {
7974 	pid_t pid;
7975 	int status;
7976 	char *cmdbuf = NULL;
7977 
7978 	pid = fork();
7979 
7980 	if (cmd == NULL) {
7981 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
7982 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
7983 		cmd = cmdbuf;
7984 	}
7985 
7986 	if (pid == -1)
7987 		fatal(B_TRUE, "fork failed");
7988 
7989 	if (pid == 0) {	/* child */
7990 		char fd_data_str[12];
7991 
7992 		VERIFY3S(11, >=,
7993 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
7994 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
7995 
7996 		if (libpath != NULL) {
7997 			const char *curlp = getenv("LD_LIBRARY_PATH");
7998 			if (curlp == NULL)
7999 				VERIFY0(setenv("LD_LIBRARY_PATH", libpath, 1));
8000 			else {
8001 				char *newlp = NULL;
8002 				VERIFY3S(-1, !=,
8003 				    asprintf(&newlp, "%s:%s", libpath, curlp));
8004 				VERIFY0(setenv("LD_LIBRARY_PATH", newlp, 1));
8005 				free(newlp);
8006 			}
8007 		}
8008 		(void) execl(cmd, cmd, (char *)NULL);
8009 		ztest_dump_core = B_FALSE;
8010 		fatal(B_TRUE, "exec failed: %s", cmd);
8011 	}
8012 
8013 	if (cmdbuf != NULL) {
8014 		umem_free(cmdbuf, MAXPATHLEN);
8015 		cmd = NULL;
8016 	}
8017 
8018 	while (waitpid(pid, &status, 0) != pid)
8019 		continue;
8020 	if (statusp != NULL)
8021 		*statusp = status;
8022 
8023 	if (WIFEXITED(status)) {
8024 		if (WEXITSTATUS(status) != 0) {
8025 			(void) fprintf(stderr, "child exited with code %d\n",
8026 			    WEXITSTATUS(status));
8027 			exit(2);
8028 		}
8029 		return (B_FALSE);
8030 	} else if (WIFSIGNALED(status)) {
8031 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
8032 			(void) fprintf(stderr, "child died with signal %d\n",
8033 			    WTERMSIG(status));
8034 			exit(3);
8035 		}
8036 		return (B_TRUE);
8037 	} else {
8038 		(void) fprintf(stderr, "something strange happened to child\n");
8039 		exit(4);
8040 	}
8041 }
8042 
8043 static void
8044 ztest_run_init(void)
8045 {
8046 	int i;
8047 
8048 	ztest_shared_t *zs = ztest_shared;
8049 
8050 	/*
8051 	 * Blow away any existing copy of zpool.cache
8052 	 */
8053 	(void) remove(spa_config_path);
8054 
8055 	if (ztest_opts.zo_init == 0) {
8056 		if (ztest_opts.zo_verbose >= 1)
8057 			(void) printf("Importing pool %s\n",
8058 			    ztest_opts.zo_pool);
8059 		ztest_import(zs);
8060 		return;
8061 	}
8062 
8063 	/*
8064 	 * Create and initialize our storage pool.
8065 	 */
8066 	for (i = 1; i <= ztest_opts.zo_init; i++) {
8067 		memset(zs, 0, sizeof (*zs));
8068 		if (ztest_opts.zo_verbose >= 3 &&
8069 		    ztest_opts.zo_init != 1) {
8070 			(void) printf("ztest_init(), pass %d\n", i);
8071 		}
8072 		ztest_init(zs);
8073 	}
8074 }
8075 
8076 int
8077 main(int argc, char **argv)
8078 {
8079 	int kills = 0;
8080 	int iters = 0;
8081 	int older = 0;
8082 	int newer = 0;
8083 	ztest_shared_t *zs;
8084 	ztest_info_t *zi;
8085 	ztest_shared_callstate_t *zc;
8086 	char timebuf[100];
8087 	char numbuf[NN_NUMBUF_SZ];
8088 	char *cmd;
8089 	boolean_t hasalt;
8090 	int f, err;
8091 	char *fd_data_str = getenv("ZTEST_FD_DATA");
8092 	struct sigaction action;
8093 
8094 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
8095 
8096 	dprintf_setup(&argc, argv);
8097 	zfs_deadman_synctime_ms = 300000;
8098 	zfs_deadman_checktime_ms = 30000;
8099 	/*
8100 	 * As two-word space map entries may not come up often (especially
8101 	 * if pool and vdev sizes are small) we want to force at least some
8102 	 * of them so the feature get tested.
8103 	 */
8104 	zfs_force_some_double_word_sm_entries = B_TRUE;
8105 
8106 	/*
8107 	 * Verify that even extensively damaged split blocks with many
8108 	 * segments can be reconstructed in a reasonable amount of time
8109 	 * when reconstruction is known to be possible.
8110 	 *
8111 	 * Note: the lower this value is, the more damage we inflict, and
8112 	 * the more time ztest spends in recovering that damage. We chose
8113 	 * to induce damage 1/100th of the time so recovery is tested but
8114 	 * not so frequently that ztest doesn't get to test other code paths.
8115 	 */
8116 	zfs_reconstruct_indirect_damage_fraction = 100;
8117 
8118 	action.sa_handler = sig_handler;
8119 	sigemptyset(&action.sa_mask);
8120 	action.sa_flags = 0;
8121 
8122 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
8123 		(void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
8124 		    strerror(errno));
8125 		exit(EXIT_FAILURE);
8126 	}
8127 
8128 	if (sigaction(SIGABRT, &action, NULL) < 0) {
8129 		(void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
8130 		    strerror(errno));
8131 		exit(EXIT_FAILURE);
8132 	}
8133 
8134 	/*
8135 	 * Force random_get_bytes() to use /dev/urandom in order to prevent
8136 	 * ztest from needlessly depleting the system entropy pool.
8137 	 */
8138 	random_path = "/dev/urandom";
8139 	ztest_fd_rand = open(random_path, O_RDONLY | O_CLOEXEC);
8140 	ASSERT3S(ztest_fd_rand, >=, 0);
8141 
8142 	if (!fd_data_str) {
8143 		process_options(argc, argv);
8144 
8145 		setup_data_fd();
8146 		setup_hdr();
8147 		setup_data();
8148 		memcpy(ztest_shared_opts, &ztest_opts,
8149 		    sizeof (*ztest_shared_opts));
8150 	} else {
8151 		ztest_fd_data = atoi(fd_data_str);
8152 		setup_data();
8153 		memcpy(&ztest_opts, ztest_shared_opts, sizeof (ztest_opts));
8154 	}
8155 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
8156 
8157 	err = ztest_set_global_vars();
8158 	if (err != 0 && !fd_data_str) {
8159 		/* error message done by ztest_set_global_vars */
8160 		exit(EXIT_FAILURE);
8161 	} else {
8162 		/* children should not be spawned if setting gvars fails */
8163 		VERIFY3S(err, ==, 0);
8164 	}
8165 
8166 	/* Override location of zpool.cache */
8167 	VERIFY3S(asprintf((char **)&spa_config_path, "%s/zpool.cache",
8168 	    ztest_opts.zo_dir), !=, -1);
8169 
8170 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
8171 	    UMEM_NOFAIL);
8172 	zs = ztest_shared;
8173 
8174 	if (fd_data_str) {
8175 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
8176 		metaslab_df_alloc_threshold =
8177 		    zs->zs_metaslab_df_alloc_threshold;
8178 
8179 		if (zs->zs_do_init)
8180 			ztest_run_init();
8181 		else
8182 			ztest_run(zs);
8183 		exit(0);
8184 	}
8185 
8186 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
8187 
8188 	if (ztest_opts.zo_verbose >= 1) {
8189 		(void) printf("%"PRIu64" vdevs, %d datasets, %d threads,"
8190 		    "%d %s disks, %"PRIu64" seconds...\n\n",
8191 		    ztest_opts.zo_vdevs,
8192 		    ztest_opts.zo_datasets,
8193 		    ztest_opts.zo_threads,
8194 		    ztest_opts.zo_raid_children,
8195 		    ztest_opts.zo_raid_type,
8196 		    ztest_opts.zo_time);
8197 	}
8198 
8199 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
8200 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
8201 
8202 	zs->zs_do_init = B_TRUE;
8203 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
8204 		if (ztest_opts.zo_verbose >= 1) {
8205 			(void) printf("Executing older ztest for "
8206 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
8207 		}
8208 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
8209 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
8210 	} else {
8211 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
8212 	}
8213 	zs->zs_do_init = B_FALSE;
8214 
8215 	zs->zs_proc_start = gethrtime();
8216 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
8217 
8218 	for (f = 0; f < ZTEST_FUNCS; f++) {
8219 		zi = &ztest_info[f];
8220 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
8221 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
8222 			zc->zc_next = UINT64_MAX;
8223 		else
8224 			zc->zc_next = zs->zs_proc_start +
8225 			    ztest_random(2 * zi->zi_interval[0] + 1);
8226 	}
8227 
8228 	/*
8229 	 * Run the tests in a loop.  These tests include fault injection
8230 	 * to verify that self-healing data works, and forced crashes
8231 	 * to verify that we never lose on-disk consistency.
8232 	 */
8233 	while (gethrtime() < zs->zs_proc_stop) {
8234 		int status;
8235 		boolean_t killed;
8236 
8237 		/*
8238 		 * Initialize the workload counters for each function.
8239 		 */
8240 		for (f = 0; f < ZTEST_FUNCS; f++) {
8241 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
8242 			zc->zc_count = 0;
8243 			zc->zc_time = 0;
8244 		}
8245 
8246 		/* Set the allocation switch size */
8247 		zs->zs_metaslab_df_alloc_threshold =
8248 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
8249 
8250 		if (!hasalt || ztest_random(2) == 0) {
8251 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8252 				(void) printf("Executing newer ztest: %s\n",
8253 				    cmd);
8254 			}
8255 			newer++;
8256 			killed = exec_child(cmd, NULL, B_TRUE, &status);
8257 		} else {
8258 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8259 				(void) printf("Executing older ztest: %s\n",
8260 				    ztest_opts.zo_alt_ztest);
8261 			}
8262 			older++;
8263 			killed = exec_child(ztest_opts.zo_alt_ztest,
8264 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
8265 		}
8266 
8267 		if (killed)
8268 			kills++;
8269 		iters++;
8270 
8271 		if (ztest_opts.zo_verbose >= 1) {
8272 			hrtime_t now = gethrtime();
8273 
8274 			now = MIN(now, zs->zs_proc_stop);
8275 			print_time(zs->zs_proc_stop - now, timebuf);
8276 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
8277 
8278 			(void) printf("Pass %3d, %8s, %3"PRIu64" ENOSPC, "
8279 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
8280 			    iters,
8281 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
8282 			    zs->zs_enospc_count,
8283 			    100.0 * zs->zs_alloc / zs->zs_space,
8284 			    numbuf,
8285 			    100.0 * (now - zs->zs_proc_start) /
8286 			    (ztest_opts.zo_time * NANOSEC), timebuf);
8287 		}
8288 
8289 		if (ztest_opts.zo_verbose >= 2) {
8290 			(void) printf("\nWorkload summary:\n\n");
8291 			(void) printf("%7s %9s   %s\n",
8292 			    "Calls", "Time", "Function");
8293 			(void) printf("%7s %9s   %s\n",
8294 			    "-----", "----", "--------");
8295 			for (f = 0; f < ZTEST_FUNCS; f++) {
8296 				zi = &ztest_info[f];
8297 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
8298 				print_time(zc->zc_time, timebuf);
8299 				(void) printf("%7"PRIu64" %9s   %s\n",
8300 				    zc->zc_count, timebuf,
8301 				    zi->zi_funcname);
8302 			}
8303 			(void) printf("\n");
8304 		}
8305 
8306 		if (!ztest_opts.zo_mmp_test)
8307 			ztest_run_zdb(ztest_opts.zo_pool);
8308 	}
8309 
8310 	if (ztest_opts.zo_verbose >= 1) {
8311 		if (hasalt) {
8312 			(void) printf("%d runs of older ztest: %s\n", older,
8313 			    ztest_opts.zo_alt_ztest);
8314 			(void) printf("%d runs of newer ztest: %s\n", newer,
8315 			    cmd);
8316 		}
8317 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
8318 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
8319 	}
8320 
8321 	umem_free(cmd, MAXNAMELEN);
8322 
8323 	return (0);
8324 }
8325