xref: /freebsd/sys/contrib/openzfs/cmd/ztest.c (revision 6137b5f7b8c183ee8806d79b3f1d8e5e3ddb3df3)
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 	uint64_t	zh_scratch_state_size;
155 } ztest_shared_hdr_t;
156 
157 static ztest_shared_hdr_t *ztest_shared_hdr;
158 
159 enum ztest_class_state {
160 	ZTEST_VDEV_CLASS_OFF,
161 	ZTEST_VDEV_CLASS_ON,
162 	ZTEST_VDEV_CLASS_RND
163 };
164 
165 /* Dedicated RAIDZ Expansion test states */
166 typedef enum {
167 	RAIDZ_EXPAND_NONE,		/* Default is none, must opt-in	*/
168 	RAIDZ_EXPAND_REQUESTED,		/* The '-X' option was used	*/
169 	RAIDZ_EXPAND_STARTED,		/* Testing has commenced	*/
170 	RAIDZ_EXPAND_KILLED,		/* Reached the proccess kill	*/
171 	RAIDZ_EXPAND_CHECKED,		/* Pool scrub verification done	*/
172 } raidz_expand_test_state_t;
173 
174 
175 #define	ZO_GVARS_MAX_ARGLEN	((size_t)64)
176 #define	ZO_GVARS_MAX_COUNT	((size_t)10)
177 
178 typedef struct ztest_shared_opts {
179 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
180 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
181 	char zo_alt_ztest[MAXNAMELEN];
182 	char zo_alt_libpath[MAXNAMELEN];
183 	uint64_t zo_vdevs;
184 	uint64_t zo_vdevtime;
185 	size_t zo_vdev_size;
186 	int zo_ashift;
187 	int zo_mirrors;
188 	int zo_raid_do_expand;
189 	int zo_raid_children;
190 	int zo_raid_parity;
191 	char zo_raid_type[8];
192 	int zo_draid_data;
193 	int zo_draid_spares;
194 	int zo_datasets;
195 	int zo_threads;
196 	uint64_t zo_passtime;
197 	uint64_t zo_killrate;
198 	int zo_verbose;
199 	int zo_init;
200 	uint64_t zo_time;
201 	uint64_t zo_maxloops;
202 	uint64_t zo_metaslab_force_ganging;
203 	raidz_expand_test_state_t zo_raidz_expand_test;
204 	int zo_mmp_test;
205 	int zo_special_vdevs;
206 	int zo_dump_dbgmsg;
207 	int zo_gvars_count;
208 	char zo_gvars[ZO_GVARS_MAX_COUNT][ZO_GVARS_MAX_ARGLEN];
209 } ztest_shared_opts_t;
210 
211 /* Default values for command line options. */
212 #define	DEFAULT_POOL "ztest"
213 #define	DEFAULT_VDEV_DIR "/tmp"
214 #define	DEFAULT_VDEV_COUNT 5
215 #define	DEFAULT_VDEV_SIZE (SPA_MINDEVSIZE * 4)	/* 256m default size */
216 #define	DEFAULT_VDEV_SIZE_STR "256M"
217 #define	DEFAULT_ASHIFT SPA_MINBLOCKSHIFT
218 #define	DEFAULT_MIRRORS 2
219 #define	DEFAULT_RAID_CHILDREN 4
220 #define	DEFAULT_RAID_PARITY 1
221 #define	DEFAULT_DRAID_DATA 4
222 #define	DEFAULT_DRAID_SPARES 1
223 #define	DEFAULT_DATASETS_COUNT 7
224 #define	DEFAULT_THREADS 23
225 #define	DEFAULT_RUN_TIME 300 /* 300 seconds */
226 #define	DEFAULT_RUN_TIME_STR "300 sec"
227 #define	DEFAULT_PASS_TIME 60 /* 60 seconds */
228 #define	DEFAULT_PASS_TIME_STR "60 sec"
229 #define	DEFAULT_KILL_RATE 70 /* 70% kill rate */
230 #define	DEFAULT_KILLRATE_STR "70%"
231 #define	DEFAULT_INITS 1
232 #define	DEFAULT_MAX_LOOPS 50 /* 5 minutes */
233 #define	DEFAULT_FORCE_GANGING (64 << 10)
234 #define	DEFAULT_FORCE_GANGING_STR "64K"
235 
236 /* Simplifying assumption: -1 is not a valid default. */
237 #define	NO_DEFAULT -1
238 
239 static const ztest_shared_opts_t ztest_opts_defaults = {
240 	.zo_pool = DEFAULT_POOL,
241 	.zo_dir = DEFAULT_VDEV_DIR,
242 	.zo_alt_ztest = { '\0' },
243 	.zo_alt_libpath = { '\0' },
244 	.zo_vdevs = DEFAULT_VDEV_COUNT,
245 	.zo_ashift = DEFAULT_ASHIFT,
246 	.zo_mirrors = DEFAULT_MIRRORS,
247 	.zo_raid_children = DEFAULT_RAID_CHILDREN,
248 	.zo_raid_parity = DEFAULT_RAID_PARITY,
249 	.zo_raid_type = VDEV_TYPE_RAIDZ,
250 	.zo_vdev_size = DEFAULT_VDEV_SIZE,
251 	.zo_draid_data = DEFAULT_DRAID_DATA,	/* data drives */
252 	.zo_draid_spares = DEFAULT_DRAID_SPARES, /* distributed spares */
253 	.zo_datasets = DEFAULT_DATASETS_COUNT,
254 	.zo_threads = DEFAULT_THREADS,
255 	.zo_passtime = DEFAULT_PASS_TIME,
256 	.zo_killrate = DEFAULT_KILL_RATE,
257 	.zo_verbose = 0,
258 	.zo_mmp_test = 0,
259 	.zo_init = DEFAULT_INITS,
260 	.zo_time = DEFAULT_RUN_TIME,
261 	.zo_maxloops = DEFAULT_MAX_LOOPS, /* max loops during spa_freeze() */
262 	.zo_metaslab_force_ganging = DEFAULT_FORCE_GANGING,
263 	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
264 	.zo_gvars_count = 0,
265 	.zo_raidz_expand_test = RAIDZ_EXPAND_NONE,
266 };
267 
268 extern uint64_t metaslab_force_ganging;
269 extern uint64_t metaslab_df_alloc_threshold;
270 extern uint64_t zfs_deadman_synctime_ms;
271 extern uint_t metaslab_preload_limit;
272 extern int zfs_compressed_arc_enabled;
273 extern int zfs_abd_scatter_enabled;
274 extern uint_t dmu_object_alloc_chunk_shift;
275 extern boolean_t zfs_force_some_double_word_sm_entries;
276 extern unsigned long zio_decompress_fail_fraction;
277 extern unsigned long zfs_reconstruct_indirect_damage_fraction;
278 extern uint64_t raidz_expand_max_reflow_bytes;
279 extern uint_t raidz_expand_pause_point;
280 
281 
282 static ztest_shared_opts_t *ztest_shared_opts;
283 static ztest_shared_opts_t ztest_opts;
284 static const char *const ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
285 
286 typedef struct ztest_shared_ds {
287 	uint64_t	zd_seq;
288 } ztest_shared_ds_t;
289 
290 static ztest_shared_ds_t *ztest_shared_ds;
291 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
292 
293 typedef struct ztest_scratch_state {
294 	uint64_t	zs_raidz_scratch_verify_pause;
295 } ztest_shared_scratch_state_t;
296 
297 static ztest_shared_scratch_state_t *ztest_scratch_state;
298 
299 #define	BT_MAGIC	0x123456789abcdefULL
300 #define	MAXFAULTS(zs) \
301 	(MAX((zs)->zs_mirrors, 1) * (ztest_opts.zo_raid_parity + 1) - 1)
302 
303 enum ztest_io_type {
304 	ZTEST_IO_WRITE_TAG,
305 	ZTEST_IO_WRITE_PATTERN,
306 	ZTEST_IO_WRITE_ZEROES,
307 	ZTEST_IO_TRUNCATE,
308 	ZTEST_IO_SETATTR,
309 	ZTEST_IO_REWRITE,
310 	ZTEST_IO_TYPES
311 };
312 
313 typedef struct ztest_block_tag {
314 	uint64_t	bt_magic;
315 	uint64_t	bt_objset;
316 	uint64_t	bt_object;
317 	uint64_t	bt_dnodesize;
318 	uint64_t	bt_offset;
319 	uint64_t	bt_gen;
320 	uint64_t	bt_txg;
321 	uint64_t	bt_crtxg;
322 } ztest_block_tag_t;
323 
324 typedef struct bufwad {
325 	uint64_t	bw_index;
326 	uint64_t	bw_txg;
327 	uint64_t	bw_data;
328 } bufwad_t;
329 
330 /*
331  * It would be better to use a rangelock_t per object.  Unfortunately
332  * the rangelock_t is not a drop-in replacement for rl_t, because we
333  * still need to map from object ID to rangelock_t.
334  */
335 typedef enum {
336 	ZTRL_READER,
337 	ZTRL_WRITER,
338 	ZTRL_APPEND
339 } rl_type_t;
340 
341 typedef struct rll {
342 	void		*rll_writer;
343 	int		rll_readers;
344 	kmutex_t	rll_lock;
345 	kcondvar_t	rll_cv;
346 } rll_t;
347 
348 typedef struct rl {
349 	uint64_t	rl_object;
350 	uint64_t	rl_offset;
351 	uint64_t	rl_size;
352 	rll_t		*rl_lock;
353 } rl_t;
354 
355 #define	ZTEST_RANGE_LOCKS	64
356 #define	ZTEST_OBJECT_LOCKS	64
357 
358 /*
359  * Object descriptor.  Used as a template for object lookup/create/remove.
360  */
361 typedef struct ztest_od {
362 	uint64_t	od_dir;
363 	uint64_t	od_object;
364 	dmu_object_type_t od_type;
365 	dmu_object_type_t od_crtype;
366 	uint64_t	od_blocksize;
367 	uint64_t	od_crblocksize;
368 	uint64_t	od_crdnodesize;
369 	uint64_t	od_gen;
370 	uint64_t	od_crgen;
371 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
372 } ztest_od_t;
373 
374 /*
375  * Per-dataset state.
376  */
377 typedef struct ztest_ds {
378 	ztest_shared_ds_t *zd_shared;
379 	objset_t	*zd_os;
380 	pthread_rwlock_t zd_zilog_lock;
381 	zilog_t		*zd_zilog;
382 	ztest_od_t	*zd_od;		/* debugging aid */
383 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
384 	kmutex_t	zd_dirobj_lock;
385 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
386 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
387 } ztest_ds_t;
388 
389 /*
390  * Per-iteration state.
391  */
392 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
393 
394 typedef struct ztest_info {
395 	ztest_func_t	*zi_func;	/* test function */
396 	uint64_t	zi_iters;	/* iterations per execution */
397 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
398 	const char	*zi_funcname;	/* name of test function */
399 } ztest_info_t;
400 
401 typedef struct ztest_shared_callstate {
402 	uint64_t	zc_count;	/* per-pass count */
403 	uint64_t	zc_time;	/* per-pass time */
404 	uint64_t	zc_next;	/* next time to call this function */
405 } ztest_shared_callstate_t;
406 
407 static ztest_shared_callstate_t *ztest_shared_callstate;
408 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
409 
410 ztest_func_t ztest_dmu_read_write;
411 ztest_func_t ztest_dmu_write_parallel;
412 ztest_func_t ztest_dmu_object_alloc_free;
413 ztest_func_t ztest_dmu_object_next_chunk;
414 ztest_func_t ztest_dmu_commit_callbacks;
415 ztest_func_t ztest_zap;
416 ztest_func_t ztest_zap_parallel;
417 ztest_func_t ztest_zil_commit;
418 ztest_func_t ztest_zil_remount;
419 ztest_func_t ztest_dmu_read_write_zcopy;
420 ztest_func_t ztest_dmu_objset_create_destroy;
421 ztest_func_t ztest_dmu_prealloc;
422 ztest_func_t ztest_fzap;
423 ztest_func_t ztest_dmu_snapshot_create_destroy;
424 ztest_func_t ztest_dsl_prop_get_set;
425 ztest_func_t ztest_spa_prop_get_set;
426 ztest_func_t ztest_spa_create_destroy;
427 ztest_func_t ztest_fault_inject;
428 ztest_func_t ztest_dmu_snapshot_hold;
429 ztest_func_t ztest_mmp_enable_disable;
430 ztest_func_t ztest_scrub;
431 ztest_func_t ztest_dsl_dataset_promote_busy;
432 ztest_func_t ztest_vdev_attach_detach;
433 ztest_func_t ztest_vdev_raidz_attach;
434 ztest_func_t ztest_vdev_LUN_growth;
435 ztest_func_t ztest_vdev_add_remove;
436 ztest_func_t ztest_vdev_class_add;
437 ztest_func_t ztest_vdev_aux_add_remove;
438 ztest_func_t ztest_split_pool;
439 ztest_func_t ztest_reguid;
440 ztest_func_t ztest_spa_upgrade;
441 ztest_func_t ztest_device_removal;
442 ztest_func_t ztest_spa_checkpoint_create_discard;
443 ztest_func_t ztest_initialize;
444 ztest_func_t ztest_trim;
445 ztest_func_t ztest_blake3;
446 ztest_func_t ztest_fletcher;
447 ztest_func_t ztest_fletcher_incr;
448 ztest_func_t ztest_verify_dnode_bt;
449 
450 static uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
451 static uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
452 static uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
453 static uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
454 static uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
455 
456 #define	ZTI_INIT(func, iters, interval) \
457 	{   .zi_func = (func), \
458 	    .zi_iters = (iters), \
459 	    .zi_interval = (interval), \
460 	    .zi_funcname = # func }
461 
462 static ztest_info_t ztest_info[] = {
463 	ZTI_INIT(ztest_dmu_read_write, 1, &zopt_always),
464 	ZTI_INIT(ztest_dmu_write_parallel, 10, &zopt_always),
465 	ZTI_INIT(ztest_dmu_object_alloc_free, 1, &zopt_always),
466 	ZTI_INIT(ztest_dmu_object_next_chunk, 1, &zopt_sometimes),
467 	ZTI_INIT(ztest_dmu_commit_callbacks, 1, &zopt_always),
468 	ZTI_INIT(ztest_zap, 30, &zopt_always),
469 	ZTI_INIT(ztest_zap_parallel, 100, &zopt_always),
470 	ZTI_INIT(ztest_split_pool, 1, &zopt_sometimes),
471 	ZTI_INIT(ztest_zil_commit, 1, &zopt_incessant),
472 	ZTI_INIT(ztest_zil_remount, 1, &zopt_sometimes),
473 	ZTI_INIT(ztest_dmu_read_write_zcopy, 1, &zopt_often),
474 	ZTI_INIT(ztest_dmu_objset_create_destroy, 1, &zopt_often),
475 	ZTI_INIT(ztest_dsl_prop_get_set, 1, &zopt_often),
476 	ZTI_INIT(ztest_spa_prop_get_set, 1, &zopt_sometimes),
477 #if 0
478 	ZTI_INIT(ztest_dmu_prealloc, 1, &zopt_sometimes),
479 #endif
480 	ZTI_INIT(ztest_fzap, 1, &zopt_sometimes),
481 	ZTI_INIT(ztest_dmu_snapshot_create_destroy, 1, &zopt_sometimes),
482 	ZTI_INIT(ztest_spa_create_destroy, 1, &zopt_sometimes),
483 	ZTI_INIT(ztest_fault_inject, 1, &zopt_sometimes),
484 	ZTI_INIT(ztest_dmu_snapshot_hold, 1, &zopt_sometimes),
485 	ZTI_INIT(ztest_mmp_enable_disable, 1, &zopt_sometimes),
486 	ZTI_INIT(ztest_reguid, 1, &zopt_rarely),
487 	ZTI_INIT(ztest_scrub, 1, &zopt_rarely),
488 	ZTI_INIT(ztest_spa_upgrade, 1, &zopt_rarely),
489 	ZTI_INIT(ztest_dsl_dataset_promote_busy, 1, &zopt_rarely),
490 	ZTI_INIT(ztest_vdev_attach_detach, 1, &zopt_sometimes),
491 	ZTI_INIT(ztest_vdev_raidz_attach, 1, &zopt_sometimes),
492 	ZTI_INIT(ztest_vdev_LUN_growth, 1, &zopt_rarely),
493 	ZTI_INIT(ztest_vdev_add_remove, 1, &ztest_opts.zo_vdevtime),
494 	ZTI_INIT(ztest_vdev_class_add, 1, &ztest_opts.zo_vdevtime),
495 	ZTI_INIT(ztest_vdev_aux_add_remove, 1, &ztest_opts.zo_vdevtime),
496 	ZTI_INIT(ztest_device_removal, 1, &zopt_sometimes),
497 	ZTI_INIT(ztest_spa_checkpoint_create_discard, 1, &zopt_rarely),
498 	ZTI_INIT(ztest_initialize, 1, &zopt_sometimes),
499 	ZTI_INIT(ztest_trim, 1, &zopt_sometimes),
500 	ZTI_INIT(ztest_blake3, 1, &zopt_rarely),
501 	ZTI_INIT(ztest_fletcher, 1, &zopt_rarely),
502 	ZTI_INIT(ztest_fletcher_incr, 1, &zopt_rarely),
503 	ZTI_INIT(ztest_verify_dnode_bt, 1, &zopt_sometimes),
504 };
505 
506 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
507 
508 /*
509  * The following struct is used to hold a list of uncalled commit callbacks.
510  * The callbacks are ordered by txg number.
511  */
512 typedef struct ztest_cb_list {
513 	kmutex_t	zcl_callbacks_lock;
514 	list_t		zcl_callbacks;
515 } ztest_cb_list_t;
516 
517 /*
518  * Stuff we need to share writably between parent and child.
519  */
520 typedef struct ztest_shared {
521 	boolean_t	zs_do_init;
522 	hrtime_t	zs_proc_start;
523 	hrtime_t	zs_proc_stop;
524 	hrtime_t	zs_thread_start;
525 	hrtime_t	zs_thread_stop;
526 	hrtime_t	zs_thread_kill;
527 	uint64_t	zs_enospc_count;
528 	uint64_t	zs_vdev_next_leaf;
529 	uint64_t	zs_vdev_aux;
530 	uint64_t	zs_alloc;
531 	uint64_t	zs_space;
532 	uint64_t	zs_splits;
533 	uint64_t	zs_mirrors;
534 	uint64_t	zs_metaslab_sz;
535 	uint64_t	zs_metaslab_df_alloc_threshold;
536 	uint64_t	zs_guid;
537 } ztest_shared_t;
538 
539 #define	ID_PARALLEL	-1ULL
540 
541 static char ztest_dev_template[] = "%s/%s.%llua";
542 static char ztest_aux_template[] = "%s/%s.%s.%llu";
543 static ztest_shared_t *ztest_shared;
544 
545 static spa_t *ztest_spa = NULL;
546 static ztest_ds_t *ztest_ds;
547 
548 static kmutex_t ztest_vdev_lock;
549 static boolean_t ztest_device_removal_active = B_FALSE;
550 static boolean_t ztest_pool_scrubbed = B_FALSE;
551 static kmutex_t ztest_checkpoint_lock;
552 
553 /*
554  * The ztest_name_lock protects the pool and dataset namespace used by
555  * the individual tests. To modify the namespace, consumers must grab
556  * this lock as writer. Grabbing the lock as reader will ensure that the
557  * namespace does not change while the lock is held.
558  */
559 static pthread_rwlock_t ztest_name_lock;
560 
561 static boolean_t ztest_dump_core = B_TRUE;
562 static boolean_t ztest_exiting;
563 
564 /* Global commit callback list */
565 static ztest_cb_list_t zcl;
566 /* Commit cb delay */
567 static uint64_t zc_min_txg_delay = UINT64_MAX;
568 static int zc_cb_counter = 0;
569 
570 /*
571  * Minimum number of commit callbacks that need to be registered for us to check
572  * whether the minimum txg delay is acceptable.
573  */
574 #define	ZTEST_COMMIT_CB_MIN_REG	100
575 
576 /*
577  * If a number of txgs equal to this threshold have been created after a commit
578  * callback has been registered but not called, then we assume there is an
579  * implementation bug.
580  */
581 #define	ZTEST_COMMIT_CB_THRESH	(TXG_CONCURRENT_STATES + 1000)
582 
583 enum ztest_object {
584 	ZTEST_META_DNODE = 0,
585 	ZTEST_DIROBJ,
586 	ZTEST_OBJECTS
587 };
588 
589 static __attribute__((noreturn)) void usage(boolean_t requested);
590 static int ztest_scrub_impl(spa_t *spa);
591 
592 /*
593  * These libumem hooks provide a reasonable set of defaults for the allocator's
594  * debugging facilities.
595  */
596 const char *
597 _umem_debug_init(void)
598 {
599 	return ("default,verbose"); /* $UMEM_DEBUG setting */
600 }
601 
602 const char *
603 _umem_logging_init(void)
604 {
605 	return ("fail,contents"); /* $UMEM_LOGGING setting */
606 }
607 
608 static void
609 dump_debug_buffer(void)
610 {
611 	ssize_t ret __attribute__((unused));
612 
613 	if (!ztest_opts.zo_dump_dbgmsg)
614 		return;
615 
616 	/*
617 	 * We use write() instead of printf() so that this function
618 	 * is safe to call from a signal handler.
619 	 */
620 	ret = write(STDOUT_FILENO, "\n", 1);
621 	zfs_dbgmsg_print("ztest");
622 }
623 
624 #define	BACKTRACE_SZ	100
625 
626 static void sig_handler(int signo)
627 {
628 	struct sigaction action;
629 #if (__GLIBC__ && !__UCLIBC__) /* backtrace() is a GNU extension */
630 	int nptrs;
631 	void *buffer[BACKTRACE_SZ];
632 
633 	nptrs = backtrace(buffer, BACKTRACE_SZ);
634 	backtrace_symbols_fd(buffer, nptrs, STDERR_FILENO);
635 #endif
636 	dump_debug_buffer();
637 
638 	/*
639 	 * Restore default action and re-raise signal so SIGSEGV and
640 	 * SIGABRT can trigger a core dump.
641 	 */
642 	action.sa_handler = SIG_DFL;
643 	sigemptyset(&action.sa_mask);
644 	action.sa_flags = 0;
645 	(void) sigaction(signo, &action, NULL);
646 	raise(signo);
647 }
648 
649 #define	FATAL_MSG_SZ	1024
650 
651 static const char *fatal_msg;
652 
653 static __attribute__((format(printf, 2, 3))) __attribute__((noreturn)) void
654 fatal(int do_perror, const char *message, ...)
655 {
656 	va_list args;
657 	int save_errno = errno;
658 	char *buf;
659 
660 	(void) fflush(stdout);
661 	buf = umem_alloc(FATAL_MSG_SZ, UMEM_NOFAIL);
662 	if (buf == NULL)
663 		goto out;
664 
665 	va_start(args, message);
666 	(void) sprintf(buf, "ztest: ");
667 	/* LINTED */
668 	(void) vsprintf(buf + strlen(buf), message, args);
669 	va_end(args);
670 	if (do_perror) {
671 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
672 		    ": %s", strerror(save_errno));
673 	}
674 	(void) fprintf(stderr, "%s\n", buf);
675 	fatal_msg = buf;			/* to ease debugging */
676 
677 out:
678 	if (ztest_dump_core)
679 		abort();
680 	else
681 		dump_debug_buffer();
682 
683 	exit(3);
684 }
685 
686 static int
687 str2shift(const char *buf)
688 {
689 	const char *ends = "BKMGTPEZ";
690 	int i;
691 
692 	if (buf[0] == '\0')
693 		return (0);
694 	for (i = 0; i < strlen(ends); i++) {
695 		if (toupper(buf[0]) == ends[i])
696 			break;
697 	}
698 	if (i == strlen(ends)) {
699 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
700 		    buf);
701 		usage(B_FALSE);
702 	}
703 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
704 		return (10*i);
705 	}
706 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
707 	usage(B_FALSE);
708 }
709 
710 static uint64_t
711 nicenumtoull(const char *buf)
712 {
713 	char *end;
714 	uint64_t val;
715 
716 	val = strtoull(buf, &end, 0);
717 	if (end == buf) {
718 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
719 		usage(B_FALSE);
720 	} else if (end[0] == '.') {
721 		double fval = strtod(buf, &end);
722 		fval *= pow(2, str2shift(end));
723 		/*
724 		 * UINT64_MAX is not exactly representable as a double.
725 		 * The closest representation is UINT64_MAX + 1, so we
726 		 * use a >= comparison instead of > for the bounds check.
727 		 */
728 		if (fval >= (double)UINT64_MAX) {
729 			(void) fprintf(stderr, "ztest: value too large: %s\n",
730 			    buf);
731 			usage(B_FALSE);
732 		}
733 		val = (uint64_t)fval;
734 	} else {
735 		int shift = str2shift(end);
736 		if (shift >= 64 || (val << shift) >> shift != val) {
737 			(void) fprintf(stderr, "ztest: value too large: %s\n",
738 			    buf);
739 			usage(B_FALSE);
740 		}
741 		val <<= shift;
742 	}
743 	return (val);
744 }
745 
746 typedef struct ztest_option {
747 	const char	short_opt;
748 	const char	*long_opt;
749 	const char	*long_opt_param;
750 	const char	*comment;
751 	unsigned int	default_int;
752 	const char	*default_str;
753 } ztest_option_t;
754 
755 /*
756  * The following option_table is used for generating the usage info as well as
757  * the long and short option information for calling getopt_long().
758  */
759 static ztest_option_t option_table[] = {
760 	{ 'v',	"vdevs", "INTEGER", "Number of vdevs", DEFAULT_VDEV_COUNT,
761 	    NULL},
762 	{ 's',	"vdev-size", "INTEGER", "Size of each vdev",
763 	    NO_DEFAULT, DEFAULT_VDEV_SIZE_STR},
764 	{ 'a',	"alignment-shift", "INTEGER",
765 	    "Alignment shift; use 0 for random", DEFAULT_ASHIFT, NULL},
766 	{ 'm',	"mirror-copies", "INTEGER", "Number of mirror copies",
767 	    DEFAULT_MIRRORS, NULL},
768 	{ 'r',	"raid-disks", "INTEGER", "Number of raidz/draid disks",
769 	    DEFAULT_RAID_CHILDREN, NULL},
770 	{ 'R',	"raid-parity", "INTEGER", "Raid parity",
771 	    DEFAULT_RAID_PARITY, NULL},
772 	{ 'K',  "raid-kind", "raidz|eraidz|draid|random", "Raid kind",
773 	    NO_DEFAULT, "random"},
774 	{ 'D',	"draid-data", "INTEGER", "Number of draid data drives",
775 	    DEFAULT_DRAID_DATA, NULL},
776 	{ 'S',	"draid-spares", "INTEGER", "Number of draid spares",
777 	    DEFAULT_DRAID_SPARES, NULL},
778 	{ 'd',	"datasets", "INTEGER", "Number of datasets",
779 	    DEFAULT_DATASETS_COUNT, NULL},
780 	{ 't',	"threads", "INTEGER", "Number of ztest threads",
781 	    DEFAULT_THREADS, NULL},
782 	{ 'g',	"gang-block-threshold", "INTEGER",
783 	    "Metaslab gang block threshold",
784 	    NO_DEFAULT, DEFAULT_FORCE_GANGING_STR},
785 	{ 'i',	"init-count", "INTEGER", "Number of times to initialize pool",
786 	    DEFAULT_INITS, NULL},
787 	{ 'k',	"kill-percentage", "INTEGER", "Kill percentage",
788 	    NO_DEFAULT, DEFAULT_KILLRATE_STR},
789 	{ 'p',	"pool-name", "STRING", "Pool name",
790 	    NO_DEFAULT, DEFAULT_POOL},
791 	{ 'f',	"vdev-file-directory", "PATH", "File directory for vdev files",
792 	    NO_DEFAULT, DEFAULT_VDEV_DIR},
793 	{ 'M',	"multi-host", NULL,
794 	    "Multi-host; simulate pool imported on remote host",
795 	    NO_DEFAULT, NULL},
796 	{ 'E',	"use-existing-pool", NULL,
797 	    "Use existing pool instead of creating new one", NO_DEFAULT, NULL},
798 	{ 'T',	"run-time", "INTEGER", "Total run time",
799 	    NO_DEFAULT, DEFAULT_RUN_TIME_STR},
800 	{ 'P',	"pass-time", "INTEGER", "Time per pass",
801 	    NO_DEFAULT, DEFAULT_PASS_TIME_STR},
802 	{ 'F',	"freeze-loops", "INTEGER", "Max loops in spa_freeze()",
803 	    DEFAULT_MAX_LOOPS, NULL},
804 	{ 'B',	"alt-ztest", "PATH", "Alternate ztest path",
805 	    NO_DEFAULT, NULL},
806 	{ 'C',	"vdev-class-state", "on|off|random", "vdev class state",
807 	    NO_DEFAULT, "random"},
808 	{ 'X', "raidz-expansion", NULL,
809 	    "Perform a dedicated raidz expansion test",
810 	    NO_DEFAULT, NULL},
811 	{ 'o',	"option", "\"OPTION=INTEGER\"",
812 	    "Set global variable to an unsigned 32-bit integer value",
813 	    NO_DEFAULT, NULL},
814 	{ 'G',	"dump-debug-msg", NULL,
815 	    "Dump zfs_dbgmsg buffer before exiting due to an error",
816 	    NO_DEFAULT, NULL},
817 	{ 'V',	"verbose", NULL,
818 	    "Verbose (use multiple times for ever more verbosity)",
819 	    NO_DEFAULT, NULL},
820 	{ 'h',	"help",	NULL, "Show this help",
821 	    NO_DEFAULT, NULL},
822 	{0, 0, 0, 0, 0, 0}
823 };
824 
825 static struct option *long_opts = NULL;
826 static char *short_opts = NULL;
827 
828 static void
829 init_options(void)
830 {
831 	ASSERT3P(long_opts, ==, NULL);
832 	ASSERT3P(short_opts, ==, NULL);
833 
834 	int count = sizeof (option_table) / sizeof (option_table[0]);
835 	long_opts = umem_alloc(sizeof (struct option) * count, UMEM_NOFAIL);
836 
837 	short_opts = umem_alloc(sizeof (char) * 2 * count, UMEM_NOFAIL);
838 	int short_opt_index = 0;
839 
840 	for (int i = 0; i < count; i++) {
841 		long_opts[i].val = option_table[i].short_opt;
842 		long_opts[i].name = option_table[i].long_opt;
843 		long_opts[i].has_arg = option_table[i].long_opt_param != NULL
844 		    ? required_argument : no_argument;
845 		long_opts[i].flag = NULL;
846 		short_opts[short_opt_index++] = option_table[i].short_opt;
847 		if (option_table[i].long_opt_param != NULL) {
848 			short_opts[short_opt_index++] = ':';
849 		}
850 	}
851 }
852 
853 static void
854 fini_options(void)
855 {
856 	int count = sizeof (option_table) / sizeof (option_table[0]);
857 
858 	umem_free(long_opts, sizeof (struct option) * count);
859 	umem_free(short_opts, sizeof (char) * 2 * count);
860 
861 	long_opts = NULL;
862 	short_opts = NULL;
863 }
864 
865 static __attribute__((noreturn)) void
866 usage(boolean_t requested)
867 {
868 	char option[80];
869 	FILE *fp = requested ? stdout : stderr;
870 
871 	(void) fprintf(fp, "Usage: %s [OPTIONS...]\n", DEFAULT_POOL);
872 	for (int i = 0; option_table[i].short_opt != 0; i++) {
873 		if (option_table[i].long_opt_param != NULL) {
874 			(void) sprintf(option, "  -%c --%s=%s",
875 			    option_table[i].short_opt,
876 			    option_table[i].long_opt,
877 			    option_table[i].long_opt_param);
878 		} else {
879 			(void) sprintf(option, "  -%c --%s",
880 			    option_table[i].short_opt,
881 			    option_table[i].long_opt);
882 		}
883 		(void) fprintf(fp, "  %-43s%s", option,
884 		    option_table[i].comment);
885 
886 		if (option_table[i].long_opt_param != NULL) {
887 			if (option_table[i].default_str != NULL) {
888 				(void) fprintf(fp, " (default: %s)",
889 				    option_table[i].default_str);
890 			} else if (option_table[i].default_int != NO_DEFAULT) {
891 				(void) fprintf(fp, " (default: %u)",
892 				    option_table[i].default_int);
893 			}
894 		}
895 		(void) fprintf(fp, "\n");
896 	}
897 	exit(requested ? 0 : 1);
898 }
899 
900 static uint64_t
901 ztest_random(uint64_t range)
902 {
903 	uint64_t r;
904 
905 	ASSERT3S(ztest_fd_rand, >=, 0);
906 
907 	if (range == 0)
908 		return (0);
909 
910 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
911 		fatal(B_TRUE, "short read from /dev/urandom");
912 
913 	return (r % range);
914 }
915 
916 static void
917 ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
918 {
919 	char name[32];
920 	char *value;
921 	int state = ZTEST_VDEV_CLASS_RND;
922 
923 	(void) strlcpy(name, input, sizeof (name));
924 
925 	value = strchr(name, '=');
926 	if (value == NULL) {
927 		(void) fprintf(stderr, "missing value in property=value "
928 		    "'-C' argument (%s)\n", input);
929 		usage(B_FALSE);
930 	}
931 	*(value) = '\0';
932 	value++;
933 
934 	if (strcmp(value, "on") == 0) {
935 		state = ZTEST_VDEV_CLASS_ON;
936 	} else if (strcmp(value, "off") == 0) {
937 		state = ZTEST_VDEV_CLASS_OFF;
938 	} else if (strcmp(value, "random") == 0) {
939 		state = ZTEST_VDEV_CLASS_RND;
940 	} else {
941 		(void) fprintf(stderr, "invalid property value '%s'\n", value);
942 		usage(B_FALSE);
943 	}
944 
945 	if (strcmp(name, "special") == 0) {
946 		zo->zo_special_vdevs = state;
947 	} else {
948 		(void) fprintf(stderr, "invalid property name '%s'\n", name);
949 		usage(B_FALSE);
950 	}
951 	if (zo->zo_verbose >= 3)
952 		(void) printf("%s vdev state is '%s'\n", name, value);
953 }
954 
955 static void
956 process_options(int argc, char **argv)
957 {
958 	char *path;
959 	ztest_shared_opts_t *zo = &ztest_opts;
960 
961 	int opt;
962 	uint64_t value;
963 	const char *raid_kind = "random";
964 
965 	memcpy(zo, &ztest_opts_defaults, sizeof (*zo));
966 
967 	init_options();
968 
969 	while ((opt = getopt_long(argc, argv, short_opts, long_opts,
970 	    NULL)) != EOF) {
971 		value = 0;
972 		switch (opt) {
973 		case 'v':
974 		case 's':
975 		case 'a':
976 		case 'm':
977 		case 'r':
978 		case 'R':
979 		case 'D':
980 		case 'S':
981 		case 'd':
982 		case 't':
983 		case 'g':
984 		case 'i':
985 		case 'k':
986 		case 'T':
987 		case 'P':
988 		case 'F':
989 			value = nicenumtoull(optarg);
990 		}
991 		switch (opt) {
992 		case 'v':
993 			zo->zo_vdevs = value;
994 			break;
995 		case 's':
996 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
997 			break;
998 		case 'a':
999 			zo->zo_ashift = value;
1000 			break;
1001 		case 'm':
1002 			zo->zo_mirrors = value;
1003 			break;
1004 		case 'r':
1005 			zo->zo_raid_children = MAX(1, value);
1006 			break;
1007 		case 'R':
1008 			zo->zo_raid_parity = MIN(MAX(value, 1), 3);
1009 			break;
1010 		case 'K':
1011 			raid_kind = optarg;
1012 			break;
1013 		case 'D':
1014 			zo->zo_draid_data = MAX(1, value);
1015 			break;
1016 		case 'S':
1017 			zo->zo_draid_spares = MAX(1, value);
1018 			break;
1019 		case 'd':
1020 			zo->zo_datasets = MAX(1, value);
1021 			break;
1022 		case 't':
1023 			zo->zo_threads = MAX(1, value);
1024 			break;
1025 		case 'g':
1026 			zo->zo_metaslab_force_ganging =
1027 			    MAX(SPA_MINBLOCKSIZE << 1, value);
1028 			break;
1029 		case 'i':
1030 			zo->zo_init = value;
1031 			break;
1032 		case 'k':
1033 			zo->zo_killrate = value;
1034 			break;
1035 		case 'p':
1036 			(void) strlcpy(zo->zo_pool, optarg,
1037 			    sizeof (zo->zo_pool));
1038 			break;
1039 		case 'f':
1040 			path = realpath(optarg, NULL);
1041 			if (path == NULL) {
1042 				(void) fprintf(stderr, "error: %s: %s\n",
1043 				    optarg, strerror(errno));
1044 				usage(B_FALSE);
1045 			} else {
1046 				(void) strlcpy(zo->zo_dir, path,
1047 				    sizeof (zo->zo_dir));
1048 				free(path);
1049 			}
1050 			break;
1051 		case 'M':
1052 			zo->zo_mmp_test = 1;
1053 			break;
1054 		case 'V':
1055 			zo->zo_verbose++;
1056 			break;
1057 		case 'X':
1058 			zo->zo_raidz_expand_test = RAIDZ_EXPAND_REQUESTED;
1059 			break;
1060 		case 'E':
1061 			zo->zo_init = 0;
1062 			break;
1063 		case 'T':
1064 			zo->zo_time = value;
1065 			break;
1066 		case 'P':
1067 			zo->zo_passtime = MAX(1, value);
1068 			break;
1069 		case 'F':
1070 			zo->zo_maxloops = MAX(1, value);
1071 			break;
1072 		case 'B':
1073 			(void) strlcpy(zo->zo_alt_ztest, optarg,
1074 			    sizeof (zo->zo_alt_ztest));
1075 			break;
1076 		case 'C':
1077 			ztest_parse_name_value(optarg, zo);
1078 			break;
1079 		case 'o':
1080 			if (zo->zo_gvars_count >= ZO_GVARS_MAX_COUNT) {
1081 				(void) fprintf(stderr,
1082 				    "max global var count (%zu) exceeded\n",
1083 				    ZO_GVARS_MAX_COUNT);
1084 				usage(B_FALSE);
1085 			}
1086 			char *v = zo->zo_gvars[zo->zo_gvars_count];
1087 			if (strlcpy(v, optarg, ZO_GVARS_MAX_ARGLEN) >=
1088 			    ZO_GVARS_MAX_ARGLEN) {
1089 				(void) fprintf(stderr,
1090 				    "global var option '%s' is too long\n",
1091 				    optarg);
1092 				usage(B_FALSE);
1093 			}
1094 			zo->zo_gvars_count++;
1095 			break;
1096 		case 'G':
1097 			zo->zo_dump_dbgmsg = 1;
1098 			break;
1099 		case 'h':
1100 			usage(B_TRUE);
1101 			break;
1102 		case '?':
1103 		default:
1104 			usage(B_FALSE);
1105 			break;
1106 		}
1107 	}
1108 
1109 	fini_options();
1110 
1111 	/* Force compatible options for raidz expansion run */
1112 	if (zo->zo_raidz_expand_test == RAIDZ_EXPAND_REQUESTED) {
1113 		zo->zo_mmp_test = 0;
1114 		zo->zo_mirrors = 0;
1115 		zo->zo_vdevs = 1;
1116 		zo->zo_vdev_size = DEFAULT_VDEV_SIZE * 2;
1117 		zo->zo_raid_do_expand = B_FALSE;
1118 		raid_kind = "raidz";
1119 	}
1120 
1121 	if (strcmp(raid_kind, "random") == 0) {
1122 		switch (ztest_random(3)) {
1123 		case 0:
1124 			raid_kind = "raidz";
1125 			break;
1126 		case 1:
1127 			raid_kind = "eraidz";
1128 			break;
1129 		case 2:
1130 			raid_kind = "draid";
1131 			break;
1132 		}
1133 
1134 		if (ztest_opts.zo_verbose >= 3)
1135 			(void) printf("choosing RAID type '%s'\n", raid_kind);
1136 	}
1137 
1138 	if (strcmp(raid_kind, "draid") == 0) {
1139 		uint64_t min_devsize;
1140 
1141 		/* With fewer disk use 256M, otherwise 128M is OK */
1142 		min_devsize = (ztest_opts.zo_raid_children < 16) ?
1143 		    (256ULL << 20) : (128ULL << 20);
1144 
1145 		/* No top-level mirrors with dRAID for now */
1146 		zo->zo_mirrors = 0;
1147 
1148 		/* Use more appropriate defaults for dRAID */
1149 		if (zo->zo_vdevs == ztest_opts_defaults.zo_vdevs)
1150 			zo->zo_vdevs = 1;
1151 		if (zo->zo_raid_children ==
1152 		    ztest_opts_defaults.zo_raid_children)
1153 			zo->zo_raid_children = 16;
1154 		if (zo->zo_ashift < 12)
1155 			zo->zo_ashift = 12;
1156 		if (zo->zo_vdev_size < min_devsize)
1157 			zo->zo_vdev_size = min_devsize;
1158 
1159 		if (zo->zo_draid_data + zo->zo_raid_parity >
1160 		    zo->zo_raid_children - zo->zo_draid_spares) {
1161 			(void) fprintf(stderr, "error: too few draid "
1162 			    "children (%d) for stripe width (%d)\n",
1163 			    zo->zo_raid_children,
1164 			    zo->zo_draid_data + zo->zo_raid_parity);
1165 			usage(B_FALSE);
1166 		}
1167 
1168 		(void) strlcpy(zo->zo_raid_type, VDEV_TYPE_DRAID,
1169 		    sizeof (zo->zo_raid_type));
1170 
1171 	} else if (strcmp(raid_kind, "eraidz") == 0) {
1172 		/* using eraidz (expandable raidz) */
1173 		zo->zo_raid_do_expand = B_TRUE;
1174 
1175 		/* tests expect top-level to be raidz */
1176 		zo->zo_mirrors = 0;
1177 		zo->zo_vdevs = 1;
1178 
1179 		/* Make sure parity is less than data columns */
1180 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
1181 		    zo->zo_raid_children - 1);
1182 
1183 	} else /* using raidz */ {
1184 		ASSERT0(strcmp(raid_kind, "raidz"));
1185 
1186 		zo->zo_raid_parity = MIN(zo->zo_raid_parity,
1187 		    zo->zo_raid_children - 1);
1188 	}
1189 
1190 	zo->zo_vdevtime =
1191 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
1192 	    UINT64_MAX >> 2);
1193 
1194 	if (*zo->zo_alt_ztest) {
1195 		const char *invalid_what = "ztest";
1196 		char *val = zo->zo_alt_ztest;
1197 		if (0 != access(val, X_OK) ||
1198 		    (strrchr(val, '/') == NULL && (errno == EINVAL)))
1199 			goto invalid;
1200 
1201 		int dirlen = strrchr(val, '/') - val;
1202 		strlcpy(zo->zo_alt_libpath, val,
1203 		    MIN(sizeof (zo->zo_alt_libpath), dirlen + 1));
1204 		invalid_what = "library path", val = zo->zo_alt_libpath;
1205 		if (strrchr(val, '/') == NULL && (errno == EINVAL))
1206 			goto invalid;
1207 		*strrchr(val, '/') = '\0';
1208 		strlcat(val, "/lib", sizeof (zo->zo_alt_libpath));
1209 
1210 		if (0 != access(zo->zo_alt_libpath, X_OK))
1211 			goto invalid;
1212 		return;
1213 
1214 invalid:
1215 		ztest_dump_core = B_FALSE;
1216 		fatal(B_TRUE, "invalid alternate %s %s", invalid_what, val);
1217 	}
1218 }
1219 
1220 static void
1221 ztest_kill(ztest_shared_t *zs)
1222 {
1223 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
1224 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
1225 
1226 	/*
1227 	 * Before we kill ourselves, make sure that the config is updated.
1228 	 * See comment above spa_write_cachefile().
1229 	 */
1230 	if (raidz_expand_pause_point != RAIDZ_EXPAND_PAUSE_NONE) {
1231 		if (mutex_tryenter(&spa_namespace_lock)) {
1232 			spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE,
1233 			    B_FALSE);
1234 			mutex_exit(&spa_namespace_lock);
1235 
1236 			ztest_scratch_state->zs_raidz_scratch_verify_pause =
1237 			    raidz_expand_pause_point;
1238 		} else {
1239 			/*
1240 			 * Do not verify scratch object in case if
1241 			 * spa_namespace_lock cannot be acquired,
1242 			 * it can cause deadlock in spa_config_update().
1243 			 */
1244 			raidz_expand_pause_point = RAIDZ_EXPAND_PAUSE_NONE;
1245 
1246 			return;
1247 		}
1248 	} else {
1249 		mutex_enter(&spa_namespace_lock);
1250 		spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE, B_FALSE);
1251 		mutex_exit(&spa_namespace_lock);
1252 	}
1253 
1254 	(void) raise(SIGKILL);
1255 }
1256 
1257 static void
1258 ztest_record_enospc(const char *s)
1259 {
1260 	(void) s;
1261 	ztest_shared->zs_enospc_count++;
1262 }
1263 
1264 static uint64_t
1265 ztest_get_ashift(void)
1266 {
1267 	if (ztest_opts.zo_ashift == 0)
1268 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
1269 	return (ztest_opts.zo_ashift);
1270 }
1271 
1272 static boolean_t
1273 ztest_is_draid_spare(const char *name)
1274 {
1275 	uint64_t spare_id = 0, parity = 0, vdev_id = 0;
1276 
1277 	if (sscanf(name, VDEV_TYPE_DRAID "%"PRIu64"-%"PRIu64"-%"PRIu64"",
1278 	    &parity, &vdev_id, &spare_id) == 3) {
1279 		return (B_TRUE);
1280 	}
1281 
1282 	return (B_FALSE);
1283 }
1284 
1285 static nvlist_t *
1286 make_vdev_file(const char *path, const char *aux, const char *pool,
1287     size_t size, uint64_t ashift)
1288 {
1289 	char *pathbuf = NULL;
1290 	uint64_t vdev;
1291 	nvlist_t *file;
1292 	boolean_t draid_spare = B_FALSE;
1293 
1294 
1295 	if (ashift == 0)
1296 		ashift = ztest_get_ashift();
1297 
1298 	if (path == NULL) {
1299 		pathbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1300 		path = pathbuf;
1301 
1302 		if (aux != NULL) {
1303 			vdev = ztest_shared->zs_vdev_aux;
1304 			(void) snprintf(pathbuf, MAXPATHLEN,
1305 			    ztest_aux_template, ztest_opts.zo_dir,
1306 			    pool == NULL ? ztest_opts.zo_pool : pool,
1307 			    aux, vdev);
1308 		} else {
1309 			vdev = ztest_shared->zs_vdev_next_leaf++;
1310 			(void) snprintf(pathbuf, MAXPATHLEN,
1311 			    ztest_dev_template, ztest_opts.zo_dir,
1312 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
1313 		}
1314 	} else {
1315 		draid_spare = ztest_is_draid_spare(path);
1316 	}
1317 
1318 	if (size != 0 && !draid_spare) {
1319 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
1320 		if (fd == -1)
1321 			fatal(B_TRUE, "can't open %s", path);
1322 		if (ftruncate(fd, size) != 0)
1323 			fatal(B_TRUE, "can't ftruncate %s", path);
1324 		(void) close(fd);
1325 	}
1326 
1327 	file = fnvlist_alloc();
1328 	fnvlist_add_string(file, ZPOOL_CONFIG_TYPE,
1329 	    draid_spare ? VDEV_TYPE_DRAID_SPARE : VDEV_TYPE_FILE);
1330 	fnvlist_add_string(file, ZPOOL_CONFIG_PATH, path);
1331 	fnvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift);
1332 	umem_free(pathbuf, MAXPATHLEN);
1333 
1334 	return (file);
1335 }
1336 
1337 static nvlist_t *
1338 make_vdev_raid(const char *path, const char *aux, const char *pool, size_t size,
1339     uint64_t ashift, int r)
1340 {
1341 	nvlist_t *raid, **child;
1342 	int c;
1343 
1344 	if (r < 2)
1345 		return (make_vdev_file(path, aux, pool, size, ashift));
1346 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
1347 
1348 	for (c = 0; c < r; c++)
1349 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
1350 
1351 	raid = fnvlist_alloc();
1352 	fnvlist_add_string(raid, ZPOOL_CONFIG_TYPE,
1353 	    ztest_opts.zo_raid_type);
1354 	fnvlist_add_uint64(raid, ZPOOL_CONFIG_NPARITY,
1355 	    ztest_opts.zo_raid_parity);
1356 	fnvlist_add_nvlist_array(raid, ZPOOL_CONFIG_CHILDREN,
1357 	    (const nvlist_t **)child, r);
1358 
1359 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0) {
1360 		uint64_t ndata = ztest_opts.zo_draid_data;
1361 		uint64_t nparity = ztest_opts.zo_raid_parity;
1362 		uint64_t nspares = ztest_opts.zo_draid_spares;
1363 		uint64_t children = ztest_opts.zo_raid_children;
1364 		uint64_t ngroups = 1;
1365 
1366 		/*
1367 		 * Calculate the minimum number of groups required to fill a
1368 		 * slice. This is the LCM of the stripe width (data + parity)
1369 		 * and the number of data drives (children - spares).
1370 		 */
1371 		while (ngroups * (ndata + nparity) % (children - nspares) != 0)
1372 			ngroups++;
1373 
1374 		/* Store the basic dRAID configuration. */
1375 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NDATA, ndata);
1376 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NSPARES, nspares);
1377 		fnvlist_add_uint64(raid, ZPOOL_CONFIG_DRAID_NGROUPS, ngroups);
1378 	}
1379 
1380 	for (c = 0; c < r; c++)
1381 		fnvlist_free(child[c]);
1382 
1383 	umem_free(child, r * sizeof (nvlist_t *));
1384 
1385 	return (raid);
1386 }
1387 
1388 static nvlist_t *
1389 make_vdev_mirror(const char *path, const char *aux, const char *pool,
1390     size_t size, uint64_t ashift, int r, int m)
1391 {
1392 	nvlist_t *mirror, **child;
1393 	int c;
1394 
1395 	if (m < 1)
1396 		return (make_vdev_raid(path, aux, pool, size, ashift, r));
1397 
1398 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1399 
1400 	for (c = 0; c < m; c++)
1401 		child[c] = make_vdev_raid(path, aux, pool, size, ashift, r);
1402 
1403 	mirror = fnvlist_alloc();
1404 	fnvlist_add_string(mirror, ZPOOL_CONFIG_TYPE, VDEV_TYPE_MIRROR);
1405 	fnvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1406 	    (const nvlist_t **)child, m);
1407 
1408 	for (c = 0; c < m; c++)
1409 		fnvlist_free(child[c]);
1410 
1411 	umem_free(child, m * sizeof (nvlist_t *));
1412 
1413 	return (mirror);
1414 }
1415 
1416 static nvlist_t *
1417 make_vdev_root(const char *path, const char *aux, const char *pool, size_t size,
1418     uint64_t ashift, const char *class, int r, int m, int t)
1419 {
1420 	nvlist_t *root, **child;
1421 	int c;
1422 	boolean_t log;
1423 
1424 	ASSERT3S(t, >, 0);
1425 
1426 	log = (class != NULL && strcmp(class, "log") == 0);
1427 
1428 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1429 
1430 	for (c = 0; c < t; c++) {
1431 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1432 		    r, m);
1433 		fnvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG, log);
1434 
1435 		if (class != NULL && class[0] != '\0') {
1436 			ASSERT(m > 1 || log);   /* expecting a mirror */
1437 			fnvlist_add_string(child[c],
1438 			    ZPOOL_CONFIG_ALLOCATION_BIAS, class);
1439 		}
1440 	}
1441 
1442 	root = fnvlist_alloc();
1443 	fnvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
1444 	fnvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1445 	    (const nvlist_t **)child, t);
1446 
1447 	for (c = 0; c < t; c++)
1448 		fnvlist_free(child[c]);
1449 
1450 	umem_free(child, t * sizeof (nvlist_t *));
1451 
1452 	return (root);
1453 }
1454 
1455 /*
1456  * Find a random spa version. Returns back a random spa version in the
1457  * range [initial_version, SPA_VERSION_FEATURES].
1458  */
1459 static uint64_t
1460 ztest_random_spa_version(uint64_t initial_version)
1461 {
1462 	uint64_t version = initial_version;
1463 
1464 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1465 		version = version +
1466 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1467 	}
1468 
1469 	if (version > SPA_VERSION_BEFORE_FEATURES)
1470 		version = SPA_VERSION_FEATURES;
1471 
1472 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1473 	return (version);
1474 }
1475 
1476 static int
1477 ztest_random_blocksize(void)
1478 {
1479 	ASSERT3U(ztest_spa->spa_max_ashift, !=, 0);
1480 
1481 	/*
1482 	 * Choose a block size >= the ashift.
1483 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1484 	 */
1485 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1486 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1487 		maxbs = 20;
1488 	uint64_t block_shift =
1489 	    ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1490 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1491 }
1492 
1493 static int
1494 ztest_random_dnodesize(void)
1495 {
1496 	int slots;
1497 	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1498 
1499 	if (max_slots == DNODE_MIN_SLOTS)
1500 		return (DNODE_MIN_SIZE);
1501 
1502 	/*
1503 	 * Weight the random distribution more heavily toward smaller
1504 	 * dnode sizes since that is more likely to reflect real-world
1505 	 * usage.
1506 	 */
1507 	ASSERT3U(max_slots, >, 4);
1508 	switch (ztest_random(10)) {
1509 	case 0:
1510 		slots = 5 + ztest_random(max_slots - 4);
1511 		break;
1512 	case 1 ... 4:
1513 		slots = 2 + ztest_random(3);
1514 		break;
1515 	default:
1516 		slots = 1;
1517 		break;
1518 	}
1519 
1520 	return (slots << DNODE_SHIFT);
1521 }
1522 
1523 static int
1524 ztest_random_ibshift(void)
1525 {
1526 	return (DN_MIN_INDBLKSHIFT +
1527 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1528 }
1529 
1530 static uint64_t
1531 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1532 {
1533 	uint64_t top;
1534 	vdev_t *rvd = spa->spa_root_vdev;
1535 	vdev_t *tvd;
1536 
1537 	ASSERT3U(spa_config_held(spa, SCL_ALL, RW_READER), !=, 0);
1538 
1539 	do {
1540 		top = ztest_random(rvd->vdev_children);
1541 		tvd = rvd->vdev_child[top];
1542 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1543 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1544 
1545 	return (top);
1546 }
1547 
1548 static uint64_t
1549 ztest_random_dsl_prop(zfs_prop_t prop)
1550 {
1551 	uint64_t value;
1552 
1553 	do {
1554 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1555 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1556 
1557 	return (value);
1558 }
1559 
1560 static int
1561 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1562     boolean_t inherit)
1563 {
1564 	const char *propname = zfs_prop_to_name(prop);
1565 	const char *valname;
1566 	char *setpoint;
1567 	uint64_t curval;
1568 	int error;
1569 
1570 	error = dsl_prop_set_int(osname, propname,
1571 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1572 
1573 	if (error == ENOSPC) {
1574 		ztest_record_enospc(FTAG);
1575 		return (error);
1576 	}
1577 	ASSERT0(error);
1578 
1579 	setpoint = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
1580 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1581 
1582 	if (ztest_opts.zo_verbose >= 6) {
1583 		int err;
1584 
1585 		err = zfs_prop_index_to_string(prop, curval, &valname);
1586 		if (err)
1587 			(void) printf("%s %s = %llu at '%s'\n", osname,
1588 			    propname, (unsigned long long)curval, setpoint);
1589 		else
1590 			(void) printf("%s %s = %s at '%s'\n",
1591 			    osname, propname, valname, setpoint);
1592 	}
1593 	umem_free(setpoint, MAXPATHLEN);
1594 
1595 	return (error);
1596 }
1597 
1598 static int
1599 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1600 {
1601 	spa_t *spa = ztest_spa;
1602 	nvlist_t *props = NULL;
1603 	int error;
1604 
1605 	props = fnvlist_alloc();
1606 	fnvlist_add_uint64(props, zpool_prop_to_name(prop), value);
1607 
1608 	error = spa_prop_set(spa, props);
1609 
1610 	fnvlist_free(props);
1611 
1612 	if (error == ENOSPC) {
1613 		ztest_record_enospc(FTAG);
1614 		return (error);
1615 	}
1616 	ASSERT0(error);
1617 
1618 	return (error);
1619 }
1620 
1621 static int
1622 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1623     boolean_t readonly, boolean_t decrypt, const void *tag, objset_t **osp)
1624 {
1625 	int err;
1626 	char *cp = NULL;
1627 	char ddname[ZFS_MAX_DATASET_NAME_LEN];
1628 
1629 	strlcpy(ddname, name, sizeof (ddname));
1630 	cp = strchr(ddname, '@');
1631 	if (cp != NULL)
1632 		*cp = '\0';
1633 
1634 	err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1635 	while (decrypt && err == EACCES) {
1636 		dsl_crypto_params_t *dcp;
1637 		nvlist_t *crypto_args = fnvlist_alloc();
1638 
1639 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
1640 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1641 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1642 		    crypto_args, &dcp));
1643 		err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1644 		/*
1645 		 * Note: if there was an error loading, the wkey was not
1646 		 * consumed, and needs to be freed.
1647 		 */
1648 		dsl_crypto_params_free(dcp, (err != 0));
1649 		fnvlist_free(crypto_args);
1650 
1651 		if (err == EINVAL) {
1652 			/*
1653 			 * We couldn't load a key for this dataset so try
1654 			 * the parent. This loop will eventually hit the
1655 			 * encryption root since ztest only makes clones
1656 			 * as children of their origin datasets.
1657 			 */
1658 			cp = strrchr(ddname, '/');
1659 			if (cp == NULL)
1660 				return (err);
1661 
1662 			*cp = '\0';
1663 			err = EACCES;
1664 			continue;
1665 		} else if (err != 0) {
1666 			break;
1667 		}
1668 
1669 		err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1670 		break;
1671 	}
1672 
1673 	return (err);
1674 }
1675 
1676 static void
1677 ztest_rll_init(rll_t *rll)
1678 {
1679 	rll->rll_writer = NULL;
1680 	rll->rll_readers = 0;
1681 	mutex_init(&rll->rll_lock, NULL, MUTEX_DEFAULT, NULL);
1682 	cv_init(&rll->rll_cv, NULL, CV_DEFAULT, NULL);
1683 }
1684 
1685 static void
1686 ztest_rll_destroy(rll_t *rll)
1687 {
1688 	ASSERT3P(rll->rll_writer, ==, NULL);
1689 	ASSERT0(rll->rll_readers);
1690 	mutex_destroy(&rll->rll_lock);
1691 	cv_destroy(&rll->rll_cv);
1692 }
1693 
1694 static void
1695 ztest_rll_lock(rll_t *rll, rl_type_t type)
1696 {
1697 	mutex_enter(&rll->rll_lock);
1698 
1699 	if (type == ZTRL_READER) {
1700 		while (rll->rll_writer != NULL)
1701 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1702 		rll->rll_readers++;
1703 	} else {
1704 		while (rll->rll_writer != NULL || rll->rll_readers)
1705 			(void) cv_wait(&rll->rll_cv, &rll->rll_lock);
1706 		rll->rll_writer = curthread;
1707 	}
1708 
1709 	mutex_exit(&rll->rll_lock);
1710 }
1711 
1712 static void
1713 ztest_rll_unlock(rll_t *rll)
1714 {
1715 	mutex_enter(&rll->rll_lock);
1716 
1717 	if (rll->rll_writer) {
1718 		ASSERT0(rll->rll_readers);
1719 		rll->rll_writer = NULL;
1720 	} else {
1721 		ASSERT3S(rll->rll_readers, >, 0);
1722 		ASSERT3P(rll->rll_writer, ==, NULL);
1723 		rll->rll_readers--;
1724 	}
1725 
1726 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1727 		cv_broadcast(&rll->rll_cv);
1728 
1729 	mutex_exit(&rll->rll_lock);
1730 }
1731 
1732 static void
1733 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1734 {
1735 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1736 
1737 	ztest_rll_lock(rll, type);
1738 }
1739 
1740 static void
1741 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1742 {
1743 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1744 
1745 	ztest_rll_unlock(rll);
1746 }
1747 
1748 static rl_t *
1749 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1750     uint64_t size, rl_type_t type)
1751 {
1752 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1753 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1754 	rl_t *rl;
1755 
1756 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1757 	rl->rl_object = object;
1758 	rl->rl_offset = offset;
1759 	rl->rl_size = size;
1760 	rl->rl_lock = rll;
1761 
1762 	ztest_rll_lock(rll, type);
1763 
1764 	return (rl);
1765 }
1766 
1767 static void
1768 ztest_range_unlock(rl_t *rl)
1769 {
1770 	rll_t *rll = rl->rl_lock;
1771 
1772 	ztest_rll_unlock(rll);
1773 
1774 	umem_free(rl, sizeof (*rl));
1775 }
1776 
1777 static void
1778 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1779 {
1780 	zd->zd_os = os;
1781 	zd->zd_zilog = dmu_objset_zil(os);
1782 	zd->zd_shared = szd;
1783 	dmu_objset_name(os, zd->zd_name);
1784 	int l;
1785 
1786 	if (zd->zd_shared != NULL)
1787 		zd->zd_shared->zd_seq = 0;
1788 
1789 	VERIFY0(pthread_rwlock_init(&zd->zd_zilog_lock, NULL));
1790 	mutex_init(&zd->zd_dirobj_lock, NULL, MUTEX_DEFAULT, NULL);
1791 
1792 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1793 		ztest_rll_init(&zd->zd_object_lock[l]);
1794 
1795 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1796 		ztest_rll_init(&zd->zd_range_lock[l]);
1797 }
1798 
1799 static void
1800 ztest_zd_fini(ztest_ds_t *zd)
1801 {
1802 	int l;
1803 
1804 	mutex_destroy(&zd->zd_dirobj_lock);
1805 	(void) pthread_rwlock_destroy(&zd->zd_zilog_lock);
1806 
1807 	for (l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1808 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1809 
1810 	for (l = 0; l < ZTEST_RANGE_LOCKS; l++)
1811 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1812 }
1813 
1814 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1815 
1816 static uint64_t
1817 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1818 {
1819 	uint64_t txg;
1820 	int error;
1821 
1822 	/*
1823 	 * Attempt to assign tx to some transaction group.
1824 	 */
1825 	error = dmu_tx_assign(tx, txg_how);
1826 	if (error) {
1827 		if (error == ERESTART) {
1828 			ASSERT3U(txg_how, ==, TXG_NOWAIT);
1829 			dmu_tx_wait(tx);
1830 		} else {
1831 			ASSERT3U(error, ==, ENOSPC);
1832 			ztest_record_enospc(tag);
1833 		}
1834 		dmu_tx_abort(tx);
1835 		return (0);
1836 	}
1837 	txg = dmu_tx_get_txg(tx);
1838 	ASSERT3U(txg, !=, 0);
1839 	return (txg);
1840 }
1841 
1842 static void
1843 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1844     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1845     uint64_t crtxg)
1846 {
1847 	bt->bt_magic = BT_MAGIC;
1848 	bt->bt_objset = dmu_objset_id(os);
1849 	bt->bt_object = object;
1850 	bt->bt_dnodesize = dnodesize;
1851 	bt->bt_offset = offset;
1852 	bt->bt_gen = gen;
1853 	bt->bt_txg = txg;
1854 	bt->bt_crtxg = crtxg;
1855 }
1856 
1857 static void
1858 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1859     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1860     uint64_t crtxg)
1861 {
1862 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1863 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1864 	ASSERT3U(bt->bt_object, ==, object);
1865 	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1866 	ASSERT3U(bt->bt_offset, ==, offset);
1867 	ASSERT3U(bt->bt_gen, <=, gen);
1868 	ASSERT3U(bt->bt_txg, <=, txg);
1869 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1870 }
1871 
1872 static ztest_block_tag_t *
1873 ztest_bt_bonus(dmu_buf_t *db)
1874 {
1875 	dmu_object_info_t doi;
1876 	ztest_block_tag_t *bt;
1877 
1878 	dmu_object_info_from_db(db, &doi);
1879 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1880 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1881 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1882 
1883 	return (bt);
1884 }
1885 
1886 /*
1887  * Generate a token to fill up unused bonus buffer space.  Try to make
1888  * it unique to the object, generation, and offset to verify that data
1889  * is not getting overwritten by data from other dnodes.
1890  */
1891 #define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset) \
1892 	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1893 
1894 /*
1895  * Fill up the unused bonus buffer region before the block tag with a
1896  * verifiable pattern. Filling the whole bonus area with non-zero data
1897  * helps ensure that all dnode traversal code properly skips the
1898  * interior regions of large dnodes.
1899  */
1900 static void
1901 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1902     objset_t *os, uint64_t gen)
1903 {
1904 	uint64_t *bonusp;
1905 
1906 	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1907 
1908 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1909 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1910 		    gen, bonusp - (uint64_t *)db->db_data);
1911 		*bonusp = token;
1912 	}
1913 }
1914 
1915 /*
1916  * Verify that the unused area of a bonus buffer is filled with the
1917  * expected tokens.
1918  */
1919 static void
1920 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1921     objset_t *os, uint64_t gen)
1922 {
1923 	uint64_t *bonusp;
1924 
1925 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1926 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1927 		    gen, bonusp - (uint64_t *)db->db_data);
1928 		VERIFY3U(*bonusp, ==, token);
1929 	}
1930 }
1931 
1932 /*
1933  * ZIL logging ops
1934  */
1935 
1936 #define	lrz_type	lr_mode
1937 #define	lrz_blocksize	lr_uid
1938 #define	lrz_ibshift	lr_gid
1939 #define	lrz_bonustype	lr_rdev
1940 #define	lrz_dnodesize	lr_crtime[1]
1941 
1942 static void
1943 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1944 {
1945 	char *name = (void *)(lr + 1);		/* name follows lr */
1946 	size_t namesize = strlen(name) + 1;
1947 	itx_t *itx;
1948 
1949 	if (zil_replaying(zd->zd_zilog, tx))
1950 		return;
1951 
1952 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1953 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1954 	    sizeof (*lr) + namesize - sizeof (lr_t));
1955 
1956 	zil_itx_assign(zd->zd_zilog, itx, tx);
1957 }
1958 
1959 static void
1960 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1961 {
1962 	char *name = (void *)(lr + 1);		/* name follows lr */
1963 	size_t namesize = strlen(name) + 1;
1964 	itx_t *itx;
1965 
1966 	if (zil_replaying(zd->zd_zilog, tx))
1967 		return;
1968 
1969 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1970 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
1971 	    sizeof (*lr) + namesize - sizeof (lr_t));
1972 
1973 	itx->itx_oid = object;
1974 	zil_itx_assign(zd->zd_zilog, itx, tx);
1975 }
1976 
1977 static void
1978 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1979 {
1980 	itx_t *itx;
1981 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1982 
1983 	if (zil_replaying(zd->zd_zilog, tx))
1984 		return;
1985 
1986 	if (lr->lr_length > zil_max_log_data(zd->zd_zilog, sizeof (lr_write_t)))
1987 		write_state = WR_INDIRECT;
1988 
1989 	itx = zil_itx_create(TX_WRITE,
1990 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1991 
1992 	if (write_state == WR_COPIED &&
1993 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1994 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1995 		zil_itx_destroy(itx);
1996 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1997 		write_state = WR_NEED_COPY;
1998 	}
1999 	itx->itx_private = zd;
2000 	itx->itx_wr_state = write_state;
2001 	itx->itx_sync = (ztest_random(8) == 0);
2002 
2003 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
2004 	    sizeof (*lr) - sizeof (lr_t));
2005 
2006 	zil_itx_assign(zd->zd_zilog, itx, tx);
2007 }
2008 
2009 static void
2010 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
2011 {
2012 	itx_t *itx;
2013 
2014 	if (zil_replaying(zd->zd_zilog, tx))
2015 		return;
2016 
2017 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
2018 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
2019 	    sizeof (*lr) - sizeof (lr_t));
2020 
2021 	itx->itx_sync = B_FALSE;
2022 	zil_itx_assign(zd->zd_zilog, itx, tx);
2023 }
2024 
2025 static void
2026 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
2027 {
2028 	itx_t *itx;
2029 
2030 	if (zil_replaying(zd->zd_zilog, tx))
2031 		return;
2032 
2033 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
2034 	memcpy(&itx->itx_lr + 1, &lr->lr_common + 1,
2035 	    sizeof (*lr) - sizeof (lr_t));
2036 
2037 	itx->itx_sync = B_FALSE;
2038 	zil_itx_assign(zd->zd_zilog, itx, tx);
2039 }
2040 
2041 /*
2042  * ZIL replay ops
2043  */
2044 static int
2045 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
2046 {
2047 	ztest_ds_t *zd = arg1;
2048 	lr_create_t *lr = arg2;
2049 	char *name = (void *)(lr + 1);		/* name follows lr */
2050 	objset_t *os = zd->zd_os;
2051 	ztest_block_tag_t *bbt;
2052 	dmu_buf_t *db;
2053 	dmu_tx_t *tx;
2054 	uint64_t txg;
2055 	int error = 0;
2056 	int bonuslen;
2057 
2058 	if (byteswap)
2059 		byteswap_uint64_array(lr, sizeof (*lr));
2060 
2061 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
2062 	ASSERT3S(name[0], !=, '\0');
2063 
2064 	tx = dmu_tx_create(os);
2065 
2066 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
2067 
2068 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
2069 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
2070 	} else {
2071 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
2072 	}
2073 
2074 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2075 	if (txg == 0)
2076 		return (ENOSPC);
2077 
2078 	ASSERT3U(dmu_objset_zil(os)->zl_replay, ==, !!lr->lr_foid);
2079 	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
2080 
2081 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
2082 		if (lr->lr_foid == 0) {
2083 			lr->lr_foid = zap_create_dnsize(os,
2084 			    lr->lrz_type, lr->lrz_bonustype,
2085 			    bonuslen, lr->lrz_dnodesize, tx);
2086 		} else {
2087 			error = zap_create_claim_dnsize(os, lr->lr_foid,
2088 			    lr->lrz_type, lr->lrz_bonustype,
2089 			    bonuslen, lr->lrz_dnodesize, tx);
2090 		}
2091 	} else {
2092 		if (lr->lr_foid == 0) {
2093 			lr->lr_foid = dmu_object_alloc_dnsize(os,
2094 			    lr->lrz_type, 0, lr->lrz_bonustype,
2095 			    bonuslen, lr->lrz_dnodesize, tx);
2096 		} else {
2097 			error = dmu_object_claim_dnsize(os, lr->lr_foid,
2098 			    lr->lrz_type, 0, lr->lrz_bonustype,
2099 			    bonuslen, lr->lrz_dnodesize, tx);
2100 		}
2101 	}
2102 
2103 	if (error) {
2104 		ASSERT3U(error, ==, EEXIST);
2105 		ASSERT(zd->zd_zilog->zl_replay);
2106 		dmu_tx_commit(tx);
2107 		return (error);
2108 	}
2109 
2110 	ASSERT3U(lr->lr_foid, !=, 0);
2111 
2112 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
2113 		VERIFY0(dmu_object_set_blocksize(os, lr->lr_foid,
2114 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
2115 
2116 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2117 	bbt = ztest_bt_bonus(db);
2118 	dmu_buf_will_dirty(db, tx);
2119 	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
2120 	    lr->lr_gen, txg, txg);
2121 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
2122 	dmu_buf_rele(db, FTAG);
2123 
2124 	VERIFY0(zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
2125 	    &lr->lr_foid, tx));
2126 
2127 	(void) ztest_log_create(zd, tx, lr);
2128 
2129 	dmu_tx_commit(tx);
2130 
2131 	return (0);
2132 }
2133 
2134 static int
2135 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
2136 {
2137 	ztest_ds_t *zd = arg1;
2138 	lr_remove_t *lr = arg2;
2139 	char *name = (void *)(lr + 1);		/* name follows lr */
2140 	objset_t *os = zd->zd_os;
2141 	dmu_object_info_t doi;
2142 	dmu_tx_t *tx;
2143 	uint64_t object, txg;
2144 
2145 	if (byteswap)
2146 		byteswap_uint64_array(lr, sizeof (*lr));
2147 
2148 	ASSERT3U(lr->lr_doid, ==, ZTEST_DIROBJ);
2149 	ASSERT3S(name[0], !=, '\0');
2150 
2151 	VERIFY0(
2152 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
2153 	ASSERT3U(object, !=, 0);
2154 
2155 	ztest_object_lock(zd, object, ZTRL_WRITER);
2156 
2157 	VERIFY0(dmu_object_info(os, object, &doi));
2158 
2159 	tx = dmu_tx_create(os);
2160 
2161 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
2162 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
2163 
2164 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2165 	if (txg == 0) {
2166 		ztest_object_unlock(zd, object);
2167 		return (ENOSPC);
2168 	}
2169 
2170 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
2171 		VERIFY0(zap_destroy(os, object, tx));
2172 	} else {
2173 		VERIFY0(dmu_object_free(os, object, tx));
2174 	}
2175 
2176 	VERIFY0(zap_remove(os, lr->lr_doid, name, tx));
2177 
2178 	(void) ztest_log_remove(zd, tx, lr, object);
2179 
2180 	dmu_tx_commit(tx);
2181 
2182 	ztest_object_unlock(zd, object);
2183 
2184 	return (0);
2185 }
2186 
2187 static int
2188 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
2189 {
2190 	ztest_ds_t *zd = arg1;
2191 	lr_write_t *lr = arg2;
2192 	objset_t *os = zd->zd_os;
2193 	void *data = lr + 1;			/* data follows lr */
2194 	uint64_t offset, length;
2195 	ztest_block_tag_t *bt = data;
2196 	ztest_block_tag_t *bbt;
2197 	uint64_t gen, txg, lrtxg, crtxg;
2198 	dmu_object_info_t doi;
2199 	dmu_tx_t *tx;
2200 	dmu_buf_t *db;
2201 	arc_buf_t *abuf = NULL;
2202 	rl_t *rl;
2203 
2204 	if (byteswap)
2205 		byteswap_uint64_array(lr, sizeof (*lr));
2206 
2207 	offset = lr->lr_offset;
2208 	length = lr->lr_length;
2209 
2210 	/* If it's a dmu_sync() block, write the whole block */
2211 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
2212 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
2213 		if (length < blocksize) {
2214 			offset -= offset % blocksize;
2215 			length = blocksize;
2216 		}
2217 	}
2218 
2219 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
2220 		byteswap_uint64_array(bt, sizeof (*bt));
2221 
2222 	if (bt->bt_magic != BT_MAGIC)
2223 		bt = NULL;
2224 
2225 	ztest_object_lock(zd, lr->lr_foid, ZTRL_READER);
2226 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, ZTRL_WRITER);
2227 
2228 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2229 
2230 	dmu_object_info_from_db(db, &doi);
2231 
2232 	bbt = ztest_bt_bonus(db);
2233 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2234 	gen = bbt->bt_gen;
2235 	crtxg = bbt->bt_crtxg;
2236 	lrtxg = lr->lr_common.lrc_txg;
2237 
2238 	tx = dmu_tx_create(os);
2239 
2240 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
2241 
2242 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
2243 	    P2PHASE(offset, length) == 0)
2244 		abuf = dmu_request_arcbuf(db, length);
2245 
2246 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2247 	if (txg == 0) {
2248 		if (abuf != NULL)
2249 			dmu_return_arcbuf(abuf);
2250 		dmu_buf_rele(db, FTAG);
2251 		ztest_range_unlock(rl);
2252 		ztest_object_unlock(zd, lr->lr_foid);
2253 		return (ENOSPC);
2254 	}
2255 
2256 	if (bt != NULL) {
2257 		/*
2258 		 * Usually, verify the old data before writing new data --
2259 		 * but not always, because we also want to verify correct
2260 		 * behavior when the data was not recently read into cache.
2261 		 */
2262 		ASSERT(doi.doi_data_block_size);
2263 		ASSERT0(offset % doi.doi_data_block_size);
2264 		if (ztest_random(4) != 0) {
2265 			int prefetch = ztest_random(2) ?
2266 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
2267 			ztest_block_tag_t rbt;
2268 
2269 			VERIFY(dmu_read(os, lr->lr_foid, offset,
2270 			    sizeof (rbt), &rbt, prefetch) == 0);
2271 			if (rbt.bt_magic == BT_MAGIC) {
2272 				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
2273 				    offset, gen, txg, crtxg);
2274 			}
2275 		}
2276 
2277 		/*
2278 		 * Writes can appear to be newer than the bonus buffer because
2279 		 * the ztest_get_data() callback does a dmu_read() of the
2280 		 * open-context data, which may be different than the data
2281 		 * as it was when the write was generated.
2282 		 */
2283 		if (zd->zd_zilog->zl_replay) {
2284 			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
2285 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
2286 			    bt->bt_crtxg);
2287 		}
2288 
2289 		/*
2290 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
2291 		 * so that all of the usual ASSERTs will work.
2292 		 */
2293 		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
2294 		    crtxg);
2295 	}
2296 
2297 	if (abuf == NULL) {
2298 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
2299 	} else {
2300 		memcpy(abuf->b_data, data, length);
2301 		VERIFY0(dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx));
2302 	}
2303 
2304 	(void) ztest_log_write(zd, tx, lr);
2305 
2306 	dmu_buf_rele(db, FTAG);
2307 
2308 	dmu_tx_commit(tx);
2309 
2310 	ztest_range_unlock(rl);
2311 	ztest_object_unlock(zd, lr->lr_foid);
2312 
2313 	return (0);
2314 }
2315 
2316 static int
2317 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
2318 {
2319 	ztest_ds_t *zd = arg1;
2320 	lr_truncate_t *lr = arg2;
2321 	objset_t *os = zd->zd_os;
2322 	dmu_tx_t *tx;
2323 	uint64_t txg;
2324 	rl_t *rl;
2325 
2326 	if (byteswap)
2327 		byteswap_uint64_array(lr, sizeof (*lr));
2328 
2329 	ztest_object_lock(zd, lr->lr_foid, ZTRL_READER);
2330 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
2331 	    ZTRL_WRITER);
2332 
2333 	tx = dmu_tx_create(os);
2334 
2335 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
2336 
2337 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2338 	if (txg == 0) {
2339 		ztest_range_unlock(rl);
2340 		ztest_object_unlock(zd, lr->lr_foid);
2341 		return (ENOSPC);
2342 	}
2343 
2344 	VERIFY0(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
2345 	    lr->lr_length, tx));
2346 
2347 	(void) ztest_log_truncate(zd, tx, lr);
2348 
2349 	dmu_tx_commit(tx);
2350 
2351 	ztest_range_unlock(rl);
2352 	ztest_object_unlock(zd, lr->lr_foid);
2353 
2354 	return (0);
2355 }
2356 
2357 static int
2358 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
2359 {
2360 	ztest_ds_t *zd = arg1;
2361 	lr_setattr_t *lr = arg2;
2362 	objset_t *os = zd->zd_os;
2363 	dmu_tx_t *tx;
2364 	dmu_buf_t *db;
2365 	ztest_block_tag_t *bbt;
2366 	uint64_t txg, lrtxg, crtxg, dnodesize;
2367 
2368 	if (byteswap)
2369 		byteswap_uint64_array(lr, sizeof (*lr));
2370 
2371 	ztest_object_lock(zd, lr->lr_foid, ZTRL_WRITER);
2372 
2373 	VERIFY0(dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
2374 
2375 	tx = dmu_tx_create(os);
2376 	dmu_tx_hold_bonus(tx, lr->lr_foid);
2377 
2378 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2379 	if (txg == 0) {
2380 		dmu_buf_rele(db, FTAG);
2381 		ztest_object_unlock(zd, lr->lr_foid);
2382 		return (ENOSPC);
2383 	}
2384 
2385 	bbt = ztest_bt_bonus(db);
2386 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2387 	crtxg = bbt->bt_crtxg;
2388 	lrtxg = lr->lr_common.lrc_txg;
2389 	dnodesize = bbt->bt_dnodesize;
2390 
2391 	if (zd->zd_zilog->zl_replay) {
2392 		ASSERT3U(lr->lr_size, !=, 0);
2393 		ASSERT3U(lr->lr_mode, !=, 0);
2394 		ASSERT3U(lrtxg, !=, 0);
2395 	} else {
2396 		/*
2397 		 * Randomly change the size and increment the generation.
2398 		 */
2399 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2400 		    sizeof (*bbt);
2401 		lr->lr_mode = bbt->bt_gen + 1;
2402 		ASSERT0(lrtxg);
2403 	}
2404 
2405 	/*
2406 	 * Verify that the current bonus buffer is not newer than our txg.
2407 	 */
2408 	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2409 	    MAX(txg, lrtxg), crtxg);
2410 
2411 	dmu_buf_will_dirty(db, tx);
2412 
2413 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2414 	ASSERT3U(lr->lr_size, <=, db->db_size);
2415 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2416 	bbt = ztest_bt_bonus(db);
2417 
2418 	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2419 	    txg, crtxg);
2420 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2421 	dmu_buf_rele(db, FTAG);
2422 
2423 	(void) ztest_log_setattr(zd, tx, lr);
2424 
2425 	dmu_tx_commit(tx);
2426 
2427 	ztest_object_unlock(zd, lr->lr_foid);
2428 
2429 	return (0);
2430 }
2431 
2432 static zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2433 	NULL,			/* 0 no such transaction type */
2434 	ztest_replay_create,	/* TX_CREATE */
2435 	NULL,			/* TX_MKDIR */
2436 	NULL,			/* TX_MKXATTR */
2437 	NULL,			/* TX_SYMLINK */
2438 	ztest_replay_remove,	/* TX_REMOVE */
2439 	NULL,			/* TX_RMDIR */
2440 	NULL,			/* TX_LINK */
2441 	NULL,			/* TX_RENAME */
2442 	ztest_replay_write,	/* TX_WRITE */
2443 	ztest_replay_truncate,	/* TX_TRUNCATE */
2444 	ztest_replay_setattr,	/* TX_SETATTR */
2445 	NULL,			/* TX_ACL */
2446 	NULL,			/* TX_CREATE_ACL */
2447 	NULL,			/* TX_CREATE_ATTR */
2448 	NULL,			/* TX_CREATE_ACL_ATTR */
2449 	NULL,			/* TX_MKDIR_ACL */
2450 	NULL,			/* TX_MKDIR_ATTR */
2451 	NULL,			/* TX_MKDIR_ACL_ATTR */
2452 	NULL,			/* TX_WRITE2 */
2453 	NULL,			/* TX_SETSAXATTR */
2454 	NULL,			/* TX_RENAME_EXCHANGE */
2455 	NULL,			/* TX_RENAME_WHITEOUT */
2456 };
2457 
2458 /*
2459  * ZIL get_data callbacks
2460  */
2461 
2462 static void
2463 ztest_get_done(zgd_t *zgd, int error)
2464 {
2465 	(void) error;
2466 	ztest_ds_t *zd = zgd->zgd_private;
2467 	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
2468 
2469 	if (zgd->zgd_db)
2470 		dmu_buf_rele(zgd->zgd_db, zgd);
2471 
2472 	ztest_range_unlock((rl_t *)zgd->zgd_lr);
2473 	ztest_object_unlock(zd, object);
2474 
2475 	umem_free(zgd, sizeof (*zgd));
2476 }
2477 
2478 static int
2479 ztest_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
2480     struct lwb *lwb, zio_t *zio)
2481 {
2482 	(void) arg2;
2483 	ztest_ds_t *zd = arg;
2484 	objset_t *os = zd->zd_os;
2485 	uint64_t object = lr->lr_foid;
2486 	uint64_t offset = lr->lr_offset;
2487 	uint64_t size = lr->lr_length;
2488 	uint64_t txg = lr->lr_common.lrc_txg;
2489 	uint64_t crtxg;
2490 	dmu_object_info_t doi;
2491 	dmu_buf_t *db;
2492 	zgd_t *zgd;
2493 	int error;
2494 
2495 	ASSERT3P(lwb, !=, NULL);
2496 	ASSERT3U(size, !=, 0);
2497 
2498 	ztest_object_lock(zd, object, ZTRL_READER);
2499 	error = dmu_bonus_hold(os, object, FTAG, &db);
2500 	if (error) {
2501 		ztest_object_unlock(zd, object);
2502 		return (error);
2503 	}
2504 
2505 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
2506 
2507 	if (crtxg == 0 || crtxg > txg) {
2508 		dmu_buf_rele(db, FTAG);
2509 		ztest_object_unlock(zd, object);
2510 		return (ENOENT);
2511 	}
2512 
2513 	dmu_object_info_from_db(db, &doi);
2514 	dmu_buf_rele(db, FTAG);
2515 	db = NULL;
2516 
2517 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2518 	zgd->zgd_lwb = lwb;
2519 	zgd->zgd_private = zd;
2520 
2521 	if (buf != NULL) {	/* immediate write */
2522 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2523 		    object, offset, size, ZTRL_READER);
2524 
2525 		error = dmu_read(os, object, offset, size, buf,
2526 		    DMU_READ_NO_PREFETCH);
2527 		ASSERT0(error);
2528 	} else {
2529 		ASSERT3P(zio, !=, NULL);
2530 		size = doi.doi_data_block_size;
2531 		if (ISP2(size)) {
2532 			offset = P2ALIGN(offset, size);
2533 		} else {
2534 			ASSERT3U(offset, <, size);
2535 			offset = 0;
2536 		}
2537 
2538 		zgd->zgd_lr = (struct zfs_locked_range *)ztest_range_lock(zd,
2539 		    object, offset, size, ZTRL_READER);
2540 
2541 		error = dmu_buf_hold_noread(os, object, offset, zgd, &db);
2542 
2543 		if (error == 0) {
2544 			blkptr_t *bp = &lr->lr_blkptr;
2545 
2546 			zgd->zgd_db = db;
2547 			zgd->zgd_bp = bp;
2548 
2549 			ASSERT3U(db->db_offset, ==, offset);
2550 			ASSERT3U(db->db_size, ==, size);
2551 
2552 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2553 			    ztest_get_done, zgd);
2554 
2555 			if (error == 0)
2556 				return (0);
2557 		}
2558 	}
2559 
2560 	ztest_get_done(zgd, error);
2561 
2562 	return (error);
2563 }
2564 
2565 static void *
2566 ztest_lr_alloc(size_t lrsize, char *name)
2567 {
2568 	char *lr;
2569 	size_t namesize = name ? strlen(name) + 1 : 0;
2570 
2571 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2572 
2573 	if (name)
2574 		memcpy(lr + lrsize, name, namesize);
2575 
2576 	return (lr);
2577 }
2578 
2579 static void
2580 ztest_lr_free(void *lr, size_t lrsize, char *name)
2581 {
2582 	size_t namesize = name ? strlen(name) + 1 : 0;
2583 
2584 	umem_free(lr, lrsize + namesize);
2585 }
2586 
2587 /*
2588  * Lookup a bunch of objects.  Returns the number of objects not found.
2589  */
2590 static int
2591 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2592 {
2593 	int missing = 0;
2594 	int error;
2595 	int i;
2596 
2597 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2598 
2599 	for (i = 0; i < count; i++, od++) {
2600 		od->od_object = 0;
2601 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2602 		    sizeof (uint64_t), 1, &od->od_object);
2603 		if (error) {
2604 			ASSERT3S(error, ==, ENOENT);
2605 			ASSERT0(od->od_object);
2606 			missing++;
2607 		} else {
2608 			dmu_buf_t *db;
2609 			ztest_block_tag_t *bbt;
2610 			dmu_object_info_t doi;
2611 
2612 			ASSERT3U(od->od_object, !=, 0);
2613 			ASSERT0(missing);	/* there should be no gaps */
2614 
2615 			ztest_object_lock(zd, od->od_object, ZTRL_READER);
2616 			VERIFY0(dmu_bonus_hold(zd->zd_os, od->od_object,
2617 			    FTAG, &db));
2618 			dmu_object_info_from_db(db, &doi);
2619 			bbt = ztest_bt_bonus(db);
2620 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2621 			od->od_type = doi.doi_type;
2622 			od->od_blocksize = doi.doi_data_block_size;
2623 			od->od_gen = bbt->bt_gen;
2624 			dmu_buf_rele(db, FTAG);
2625 			ztest_object_unlock(zd, od->od_object);
2626 		}
2627 	}
2628 
2629 	return (missing);
2630 }
2631 
2632 static int
2633 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2634 {
2635 	int missing = 0;
2636 	int i;
2637 
2638 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2639 
2640 	for (i = 0; i < count; i++, od++) {
2641 		if (missing) {
2642 			od->od_object = 0;
2643 			missing++;
2644 			continue;
2645 		}
2646 
2647 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2648 
2649 		lr->lr_doid = od->od_dir;
2650 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2651 		lr->lrz_type = od->od_crtype;
2652 		lr->lrz_blocksize = od->od_crblocksize;
2653 		lr->lrz_ibshift = ztest_random_ibshift();
2654 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2655 		lr->lrz_dnodesize = od->od_crdnodesize;
2656 		lr->lr_gen = od->od_crgen;
2657 		lr->lr_crtime[0] = time(NULL);
2658 
2659 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2660 			ASSERT0(missing);
2661 			od->od_object = 0;
2662 			missing++;
2663 		} else {
2664 			od->od_object = lr->lr_foid;
2665 			od->od_type = od->od_crtype;
2666 			od->od_blocksize = od->od_crblocksize;
2667 			od->od_gen = od->od_crgen;
2668 			ASSERT3U(od->od_object, !=, 0);
2669 		}
2670 
2671 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2672 	}
2673 
2674 	return (missing);
2675 }
2676 
2677 static int
2678 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2679 {
2680 	int missing = 0;
2681 	int error;
2682 	int i;
2683 
2684 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2685 
2686 	od += count - 1;
2687 
2688 	for (i = count - 1; i >= 0; i--, od--) {
2689 		if (missing) {
2690 			missing++;
2691 			continue;
2692 		}
2693 
2694 		/*
2695 		 * No object was found.
2696 		 */
2697 		if (od->od_object == 0)
2698 			continue;
2699 
2700 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2701 
2702 		lr->lr_doid = od->od_dir;
2703 
2704 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2705 			ASSERT3U(error, ==, ENOSPC);
2706 			missing++;
2707 		} else {
2708 			od->od_object = 0;
2709 		}
2710 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2711 	}
2712 
2713 	return (missing);
2714 }
2715 
2716 static int
2717 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2718     const void *data)
2719 {
2720 	lr_write_t *lr;
2721 	int error;
2722 
2723 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2724 
2725 	lr->lr_foid = object;
2726 	lr->lr_offset = offset;
2727 	lr->lr_length = size;
2728 	lr->lr_blkoff = 0;
2729 	BP_ZERO(&lr->lr_blkptr);
2730 
2731 	memcpy(lr + 1, data, size);
2732 
2733 	error = ztest_replay_write(zd, lr, B_FALSE);
2734 
2735 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2736 
2737 	return (error);
2738 }
2739 
2740 static int
2741 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2742 {
2743 	lr_truncate_t *lr;
2744 	int error;
2745 
2746 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2747 
2748 	lr->lr_foid = object;
2749 	lr->lr_offset = offset;
2750 	lr->lr_length = size;
2751 
2752 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2753 
2754 	ztest_lr_free(lr, sizeof (*lr), NULL);
2755 
2756 	return (error);
2757 }
2758 
2759 static int
2760 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2761 {
2762 	lr_setattr_t *lr;
2763 	int error;
2764 
2765 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2766 
2767 	lr->lr_foid = object;
2768 	lr->lr_size = 0;
2769 	lr->lr_mode = 0;
2770 
2771 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2772 
2773 	ztest_lr_free(lr, sizeof (*lr), NULL);
2774 
2775 	return (error);
2776 }
2777 
2778 static void
2779 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2780 {
2781 	objset_t *os = zd->zd_os;
2782 	dmu_tx_t *tx;
2783 	uint64_t txg;
2784 	rl_t *rl;
2785 
2786 	txg_wait_synced(dmu_objset_pool(os), 0);
2787 
2788 	ztest_object_lock(zd, object, ZTRL_READER);
2789 	rl = ztest_range_lock(zd, object, offset, size, ZTRL_WRITER);
2790 
2791 	tx = dmu_tx_create(os);
2792 
2793 	dmu_tx_hold_write(tx, object, offset, size);
2794 
2795 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2796 
2797 	if (txg != 0) {
2798 		dmu_prealloc(os, object, offset, size, tx);
2799 		dmu_tx_commit(tx);
2800 		txg_wait_synced(dmu_objset_pool(os), txg);
2801 	} else {
2802 		(void) dmu_free_long_range(os, object, offset, size);
2803 	}
2804 
2805 	ztest_range_unlock(rl);
2806 	ztest_object_unlock(zd, object);
2807 }
2808 
2809 static void
2810 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2811 {
2812 	int err;
2813 	ztest_block_tag_t wbt;
2814 	dmu_object_info_t doi;
2815 	enum ztest_io_type io_type;
2816 	uint64_t blocksize;
2817 	void *data;
2818 
2819 	VERIFY0(dmu_object_info(zd->zd_os, object, &doi));
2820 	blocksize = doi.doi_data_block_size;
2821 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2822 
2823 	/*
2824 	 * Pick an i/o type at random, biased toward writing block tags.
2825 	 */
2826 	io_type = ztest_random(ZTEST_IO_TYPES);
2827 	if (ztest_random(2) == 0)
2828 		io_type = ZTEST_IO_WRITE_TAG;
2829 
2830 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2831 
2832 	switch (io_type) {
2833 
2834 	case ZTEST_IO_WRITE_TAG:
2835 		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2836 		    offset, 0, 0, 0);
2837 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2838 		break;
2839 
2840 	case ZTEST_IO_WRITE_PATTERN:
2841 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2842 		if (ztest_random(2) == 0) {
2843 			/*
2844 			 * Induce fletcher2 collisions to ensure that
2845 			 * zio_ddt_collision() detects and resolves them
2846 			 * when using fletcher2-verify for deduplication.
2847 			 */
2848 			((uint64_t *)data)[0] ^= 1ULL << 63;
2849 			((uint64_t *)data)[4] ^= 1ULL << 63;
2850 		}
2851 		(void) ztest_write(zd, object, offset, blocksize, data);
2852 		break;
2853 
2854 	case ZTEST_IO_WRITE_ZEROES:
2855 		memset(data, 0, blocksize);
2856 		(void) ztest_write(zd, object, offset, blocksize, data);
2857 		break;
2858 
2859 	case ZTEST_IO_TRUNCATE:
2860 		(void) ztest_truncate(zd, object, offset, blocksize);
2861 		break;
2862 
2863 	case ZTEST_IO_SETATTR:
2864 		(void) ztest_setattr(zd, object);
2865 		break;
2866 	default:
2867 		break;
2868 
2869 	case ZTEST_IO_REWRITE:
2870 		(void) pthread_rwlock_rdlock(&ztest_name_lock);
2871 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2872 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2873 		    B_FALSE);
2874 		ASSERT(err == 0 || err == ENOSPC);
2875 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2876 		    ZFS_PROP_COMPRESSION,
2877 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2878 		    B_FALSE);
2879 		ASSERT(err == 0 || err == ENOSPC);
2880 		(void) pthread_rwlock_unlock(&ztest_name_lock);
2881 
2882 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2883 		    DMU_READ_NO_PREFETCH));
2884 
2885 		(void) ztest_write(zd, object, offset, blocksize, data);
2886 		break;
2887 	}
2888 
2889 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2890 
2891 	umem_free(data, blocksize);
2892 }
2893 
2894 /*
2895  * Initialize an object description template.
2896  */
2897 static void
2898 ztest_od_init(ztest_od_t *od, uint64_t id, const char *tag, uint64_t index,
2899     dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2900     uint64_t gen)
2901 {
2902 	od->od_dir = ZTEST_DIROBJ;
2903 	od->od_object = 0;
2904 
2905 	od->od_crtype = type;
2906 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2907 	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2908 	od->od_crgen = gen;
2909 
2910 	od->od_type = DMU_OT_NONE;
2911 	od->od_blocksize = 0;
2912 	od->od_gen = 0;
2913 
2914 	(void) snprintf(od->od_name, sizeof (od->od_name),
2915 	    "%s(%"PRId64")[%"PRIu64"]",
2916 	    tag, id, index);
2917 }
2918 
2919 /*
2920  * Lookup or create the objects for a test using the od template.
2921  * If the objects do not all exist, or if 'remove' is specified,
2922  * remove any existing objects and create new ones.  Otherwise,
2923  * use the existing objects.
2924  */
2925 static int
2926 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2927 {
2928 	int count = size / sizeof (*od);
2929 	int rv = 0;
2930 
2931 	mutex_enter(&zd->zd_dirobj_lock);
2932 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2933 	    (ztest_remove(zd, od, count) != 0 ||
2934 	    ztest_create(zd, od, count) != 0))
2935 		rv = -1;
2936 	zd->zd_od = od;
2937 	mutex_exit(&zd->zd_dirobj_lock);
2938 
2939 	return (rv);
2940 }
2941 
2942 void
2943 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2944 {
2945 	(void) id;
2946 	zilog_t *zilog = zd->zd_zilog;
2947 
2948 	(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
2949 
2950 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2951 
2952 	/*
2953 	 * Remember the committed values in zd, which is in parent/child
2954 	 * shared memory.  If we die, the next iteration of ztest_run()
2955 	 * will verify that the log really does contain this record.
2956 	 */
2957 	mutex_enter(&zilog->zl_lock);
2958 	ASSERT3P(zd->zd_shared, !=, NULL);
2959 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2960 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2961 	mutex_exit(&zilog->zl_lock);
2962 
2963 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
2964 }
2965 
2966 /*
2967  * This function is designed to simulate the operations that occur during a
2968  * mount/unmount operation.  We hold the dataset across these operations in an
2969  * attempt to expose any implicit assumptions about ZIL management.
2970  */
2971 void
2972 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2973 {
2974 	(void) id;
2975 	objset_t *os = zd->zd_os;
2976 
2977 	/*
2978 	 * We hold the ztest_vdev_lock so we don't cause problems with
2979 	 * other threads that wish to remove a log device, such as
2980 	 * ztest_device_removal().
2981 	 */
2982 	mutex_enter(&ztest_vdev_lock);
2983 
2984 	/*
2985 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2986 	 * updating the zil (i.e. adding in-memory log records) and the
2987 	 * zd_zilog_lock to block any I/O.
2988 	 */
2989 	mutex_enter(&zd->zd_dirobj_lock);
2990 	(void) pthread_rwlock_wrlock(&zd->zd_zilog_lock);
2991 
2992 	/* zfsvfs_teardown() */
2993 	zil_close(zd->zd_zilog);
2994 
2995 	/* zfsvfs_setup() */
2996 	VERIFY3P(zil_open(os, ztest_get_data, NULL), ==, zd->zd_zilog);
2997 	zil_replay(os, zd, ztest_replay_vector);
2998 
2999 	(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
3000 	mutex_exit(&zd->zd_dirobj_lock);
3001 	mutex_exit(&ztest_vdev_lock);
3002 }
3003 
3004 /*
3005  * Verify that we can't destroy an active pool, create an existing pool,
3006  * or create a pool with a bad vdev spec.
3007  */
3008 void
3009 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
3010 {
3011 	(void) zd, (void) id;
3012 	ztest_shared_opts_t *zo = &ztest_opts;
3013 	spa_t *spa;
3014 	nvlist_t *nvroot;
3015 
3016 	if (zo->zo_mmp_test)
3017 		return;
3018 
3019 	/*
3020 	 * Attempt to create using a bad file.
3021 	 */
3022 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
3023 	VERIFY3U(ENOENT, ==,
3024 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
3025 	fnvlist_free(nvroot);
3026 
3027 	/*
3028 	 * Attempt to create using a bad mirror.
3029 	 */
3030 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
3031 	VERIFY3U(ENOENT, ==,
3032 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
3033 	fnvlist_free(nvroot);
3034 
3035 	/*
3036 	 * Attempt to create an existing pool.  It shouldn't matter
3037 	 * what's in the nvroot; we should fail with EEXIST.
3038 	 */
3039 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
3040 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
3041 	VERIFY3U(EEXIST, ==,
3042 	    spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
3043 	fnvlist_free(nvroot);
3044 
3045 	/*
3046 	 * We open a reference to the spa and then we try to export it
3047 	 * expecting one of the following errors:
3048 	 *
3049 	 * EBUSY
3050 	 *	Because of the reference we just opened.
3051 	 *
3052 	 * ZFS_ERR_EXPORT_IN_PROGRESS
3053 	 *	For the case that there is another ztest thread doing
3054 	 *	an export concurrently.
3055 	 */
3056 	VERIFY0(spa_open(zo->zo_pool, &spa, FTAG));
3057 	int error = spa_destroy(zo->zo_pool);
3058 	if (error != EBUSY && error != ZFS_ERR_EXPORT_IN_PROGRESS) {
3059 		fatal(B_FALSE, "spa_destroy(%s) returned unexpected value %d",
3060 		    spa->spa_name, error);
3061 	}
3062 	spa_close(spa, FTAG);
3063 
3064 	(void) pthread_rwlock_unlock(&ztest_name_lock);
3065 }
3066 
3067 /*
3068  * Start and then stop the MMP threads to ensure the startup and shutdown code
3069  * works properly.  Actual protection and property-related code tested via ZTS.
3070  */
3071 void
3072 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
3073 {
3074 	(void) zd, (void) id;
3075 	ztest_shared_opts_t *zo = &ztest_opts;
3076 	spa_t *spa = ztest_spa;
3077 
3078 	if (zo->zo_mmp_test)
3079 		return;
3080 
3081 	/*
3082 	 * Since enabling MMP involves setting a property, it could not be done
3083 	 * while the pool is suspended.
3084 	 */
3085 	if (spa_suspended(spa))
3086 		return;
3087 
3088 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3089 	mutex_enter(&spa->spa_props_lock);
3090 
3091 	zfs_multihost_fail_intervals = 0;
3092 
3093 	if (!spa_multihost(spa)) {
3094 		spa->spa_multihost = B_TRUE;
3095 		mmp_thread_start(spa);
3096 	}
3097 
3098 	mutex_exit(&spa->spa_props_lock);
3099 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3100 
3101 	txg_wait_synced(spa_get_dsl(spa), 0);
3102 	mmp_signal_all_threads();
3103 	txg_wait_synced(spa_get_dsl(spa), 0);
3104 
3105 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3106 	mutex_enter(&spa->spa_props_lock);
3107 
3108 	if (spa_multihost(spa)) {
3109 		mmp_thread_stop(spa);
3110 		spa->spa_multihost = B_FALSE;
3111 	}
3112 
3113 	mutex_exit(&spa->spa_props_lock);
3114 	spa_config_exit(spa, SCL_CONFIG, FTAG);
3115 }
3116 
3117 static int
3118 ztest_get_raidz_children(spa_t *spa)
3119 {
3120 	(void) spa;
3121 	vdev_t *raidvd;
3122 
3123 	ASSERT(MUTEX_HELD(&ztest_vdev_lock));
3124 
3125 	if (ztest_opts.zo_raid_do_expand) {
3126 		raidvd = ztest_spa->spa_root_vdev->vdev_child[0];
3127 
3128 		ASSERT(raidvd->vdev_ops == &vdev_raidz_ops);
3129 
3130 		return (raidvd->vdev_children);
3131 	}
3132 
3133 	return (ztest_opts.zo_raid_children);
3134 }
3135 
3136 void
3137 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
3138 {
3139 	(void) zd, (void) id;
3140 	spa_t *spa;
3141 	uint64_t initial_version = SPA_VERSION_INITIAL;
3142 	uint64_t raidz_children, version, newversion;
3143 	nvlist_t *nvroot, *props;
3144 	char *name;
3145 
3146 	if (ztest_opts.zo_mmp_test)
3147 		return;
3148 
3149 	/* dRAID added after feature flags, skip upgrade test. */
3150 	if (strcmp(ztest_opts.zo_raid_type, VDEV_TYPE_DRAID) == 0)
3151 		return;
3152 
3153 	mutex_enter(&ztest_vdev_lock);
3154 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
3155 
3156 	/*
3157 	 * Clean up from previous runs.
3158 	 */
3159 	(void) spa_destroy(name);
3160 
3161 	raidz_children = ztest_get_raidz_children(ztest_spa);
3162 
3163 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
3164 	    NULL, raidz_children, ztest_opts.zo_mirrors, 1);
3165 
3166 	/*
3167 	 * If we're configuring a RAIDZ device then make sure that the
3168 	 * initial version is capable of supporting that feature.
3169 	 */
3170 	switch (ztest_opts.zo_raid_parity) {
3171 	case 0:
3172 	case 1:
3173 		initial_version = SPA_VERSION_INITIAL;
3174 		break;
3175 	case 2:
3176 		initial_version = SPA_VERSION_RAIDZ2;
3177 		break;
3178 	case 3:
3179 		initial_version = SPA_VERSION_RAIDZ3;
3180 		break;
3181 	}
3182 
3183 	/*
3184 	 * Create a pool with a spa version that can be upgraded. Pick
3185 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
3186 	 */
3187 	do {
3188 		version = ztest_random_spa_version(initial_version);
3189 	} while (version > SPA_VERSION_BEFORE_FEATURES);
3190 
3191 	props = fnvlist_alloc();
3192 	fnvlist_add_uint64(props,
3193 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
3194 	VERIFY0(spa_create(name, nvroot, props, NULL, NULL));
3195 	fnvlist_free(nvroot);
3196 	fnvlist_free(props);
3197 
3198 	VERIFY0(spa_open(name, &spa, FTAG));
3199 	VERIFY3U(spa_version(spa), ==, version);
3200 	newversion = ztest_random_spa_version(version + 1);
3201 
3202 	if (ztest_opts.zo_verbose >= 4) {
3203 		(void) printf("upgrading spa version from "
3204 		    "%"PRIu64" to %"PRIu64"\n",
3205 		    version, newversion);
3206 	}
3207 
3208 	spa_upgrade(spa, newversion);
3209 	VERIFY3U(spa_version(spa), >, version);
3210 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
3211 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
3212 	spa_close(spa, FTAG);
3213 
3214 	kmem_strfree(name);
3215 	mutex_exit(&ztest_vdev_lock);
3216 }
3217 
3218 static void
3219 ztest_spa_checkpoint(spa_t *spa)
3220 {
3221 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3222 
3223 	int error = spa_checkpoint(spa->spa_name);
3224 
3225 	switch (error) {
3226 	case 0:
3227 	case ZFS_ERR_DEVRM_IN_PROGRESS:
3228 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3229 	case ZFS_ERR_CHECKPOINT_EXISTS:
3230 	case ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS:
3231 		break;
3232 	case ENOSPC:
3233 		ztest_record_enospc(FTAG);
3234 		break;
3235 	default:
3236 		fatal(B_FALSE, "spa_checkpoint(%s) = %d", spa->spa_name, error);
3237 	}
3238 }
3239 
3240 static void
3241 ztest_spa_discard_checkpoint(spa_t *spa)
3242 {
3243 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
3244 
3245 	int error = spa_checkpoint_discard(spa->spa_name);
3246 
3247 	switch (error) {
3248 	case 0:
3249 	case ZFS_ERR_DISCARDING_CHECKPOINT:
3250 	case ZFS_ERR_NO_CHECKPOINT:
3251 		break;
3252 	default:
3253 		fatal(B_FALSE, "spa_discard_checkpoint(%s) = %d",
3254 		    spa->spa_name, error);
3255 	}
3256 
3257 }
3258 
3259 void
3260 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
3261 {
3262 	(void) zd, (void) id;
3263 	spa_t *spa = ztest_spa;
3264 
3265 	mutex_enter(&ztest_checkpoint_lock);
3266 	if (ztest_random(2) == 0) {
3267 		ztest_spa_checkpoint(spa);
3268 	} else {
3269 		ztest_spa_discard_checkpoint(spa);
3270 	}
3271 	mutex_exit(&ztest_checkpoint_lock);
3272 }
3273 
3274 
3275 static vdev_t *
3276 vdev_lookup_by_path(vdev_t *vd, const char *path)
3277 {
3278 	vdev_t *mvd;
3279 	int c;
3280 
3281 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
3282 		return (vd);
3283 
3284 	for (c = 0; c < vd->vdev_children; c++)
3285 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
3286 		    NULL)
3287 			return (mvd);
3288 
3289 	return (NULL);
3290 }
3291 
3292 static int
3293 spa_num_top_vdevs(spa_t *spa)
3294 {
3295 	vdev_t *rvd = spa->spa_root_vdev;
3296 	ASSERT3U(spa_config_held(spa, SCL_VDEV, RW_READER), ==, SCL_VDEV);
3297 	return (rvd->vdev_children);
3298 }
3299 
3300 /*
3301  * Verify that vdev_add() works as expected.
3302  */
3303 void
3304 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
3305 {
3306 	(void) zd, (void) id;
3307 	ztest_shared_t *zs = ztest_shared;
3308 	spa_t *spa = ztest_spa;
3309 	uint64_t leaves;
3310 	uint64_t guid;
3311 	uint64_t raidz_children;
3312 
3313 	nvlist_t *nvroot;
3314 	int error;
3315 
3316 	if (ztest_opts.zo_mmp_test)
3317 		return;
3318 
3319 	mutex_enter(&ztest_vdev_lock);
3320 	raidz_children = ztest_get_raidz_children(spa);
3321 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * raidz_children;
3322 
3323 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3324 
3325 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3326 
3327 	/*
3328 	 * If we have slogs then remove them 1/4 of the time.
3329 	 */
3330 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
3331 		metaslab_group_t *mg;
3332 
3333 		/*
3334 		 * find the first real slog in log allocation class
3335 		 */
3336 		mg =  spa_log_class(spa)->mc_allocator[0].mca_rotor;
3337 		while (!mg->mg_vd->vdev_islog)
3338 			mg = mg->mg_next;
3339 
3340 		guid = mg->mg_vd->vdev_guid;
3341 
3342 		spa_config_exit(spa, SCL_VDEV, FTAG);
3343 
3344 		/*
3345 		 * We have to grab the zs_name_lock as writer to
3346 		 * prevent a race between removing a slog (dmu_objset_find)
3347 		 * and destroying a dataset. Removing the slog will
3348 		 * grab a reference on the dataset which may cause
3349 		 * dsl_destroy_head() to fail with EBUSY thus
3350 		 * leaving the dataset in an inconsistent state.
3351 		 */
3352 		pthread_rwlock_wrlock(&ztest_name_lock);
3353 		error = spa_vdev_remove(spa, guid, B_FALSE);
3354 		pthread_rwlock_unlock(&ztest_name_lock);
3355 
3356 		switch (error) {
3357 		case 0:
3358 		case EEXIST:	/* Generic zil_reset() error */
3359 		case EBUSY:	/* Replay required */
3360 		case EACCES:	/* Crypto key not loaded */
3361 		case ZFS_ERR_CHECKPOINT_EXISTS:
3362 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3363 			break;
3364 		default:
3365 			fatal(B_FALSE, "spa_vdev_remove() = %d", error);
3366 		}
3367 	} else {
3368 		spa_config_exit(spa, SCL_VDEV, FTAG);
3369 
3370 		/*
3371 		 * Make 1/4 of the devices be log devices
3372 		 */
3373 		nvroot = make_vdev_root(NULL, NULL, NULL,
3374 		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
3375 		    "log" : NULL, raidz_children, zs->zs_mirrors,
3376 		    1);
3377 
3378 		error = spa_vdev_add(spa, nvroot);
3379 		fnvlist_free(nvroot);
3380 
3381 		switch (error) {
3382 		case 0:
3383 			break;
3384 		case ENOSPC:
3385 			ztest_record_enospc("spa_vdev_add");
3386 			break;
3387 		default:
3388 			fatal(B_FALSE, "spa_vdev_add() = %d", error);
3389 		}
3390 	}
3391 
3392 	mutex_exit(&ztest_vdev_lock);
3393 }
3394 
3395 void
3396 ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
3397 {
3398 	(void) zd, (void) id;
3399 	ztest_shared_t *zs = ztest_shared;
3400 	spa_t *spa = ztest_spa;
3401 	uint64_t leaves;
3402 	nvlist_t *nvroot;
3403 	uint64_t raidz_children;
3404 	const char *class = (ztest_random(2) == 0) ?
3405 	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
3406 	int error;
3407 
3408 	/*
3409 	 * By default add a special vdev 50% of the time
3410 	 */
3411 	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
3412 	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
3413 	    ztest_random(2) == 0)) {
3414 		return;
3415 	}
3416 
3417 	mutex_enter(&ztest_vdev_lock);
3418 
3419 	/* Only test with mirrors */
3420 	if (zs->zs_mirrors < 2) {
3421 		mutex_exit(&ztest_vdev_lock);
3422 		return;
3423 	}
3424 
3425 	/* requires feature@allocation_classes */
3426 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
3427 		mutex_exit(&ztest_vdev_lock);
3428 		return;
3429 	}
3430 
3431 	raidz_children = ztest_get_raidz_children(spa);
3432 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * raidz_children;
3433 
3434 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3435 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
3436 	spa_config_exit(spa, SCL_VDEV, FTAG);
3437 
3438 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
3439 	    class, raidz_children, zs->zs_mirrors, 1);
3440 
3441 	error = spa_vdev_add(spa, nvroot);
3442 	fnvlist_free(nvroot);
3443 
3444 	if (error == ENOSPC)
3445 		ztest_record_enospc("spa_vdev_add");
3446 	else if (error != 0)
3447 		fatal(B_FALSE, "spa_vdev_add() = %d", error);
3448 
3449 	/*
3450 	 * 50% of the time allow small blocks in the special class
3451 	 */
3452 	if (error == 0 &&
3453 	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
3454 		if (ztest_opts.zo_verbose >= 3)
3455 			(void) printf("Enabling special VDEV small blocks\n");
3456 		error = ztest_dsl_prop_set_uint64(zd->zd_name,
3457 		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
3458 		ASSERT(error == 0 || error == ENOSPC);
3459 	}
3460 
3461 	mutex_exit(&ztest_vdev_lock);
3462 
3463 	if (ztest_opts.zo_verbose >= 3) {
3464 		metaslab_class_t *mc;
3465 
3466 		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3467 			mc = spa_special_class(spa);
3468 		else
3469 			mc = spa_dedup_class(spa);
3470 		(void) printf("Added a %s mirrored vdev (of %d)\n",
3471 		    class, (int)mc->mc_groups);
3472 	}
3473 }
3474 
3475 /*
3476  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3477  */
3478 void
3479 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
3480 {
3481 	(void) zd, (void) id;
3482 	ztest_shared_t *zs = ztest_shared;
3483 	spa_t *spa = ztest_spa;
3484 	vdev_t *rvd = spa->spa_root_vdev;
3485 	spa_aux_vdev_t *sav;
3486 	const char *aux;
3487 	char *path;
3488 	uint64_t guid = 0;
3489 	int error, ignore_err = 0;
3490 
3491 	if (ztest_opts.zo_mmp_test)
3492 		return;
3493 
3494 	path = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3495 
3496 	if (ztest_random(2) == 0) {
3497 		sav = &spa->spa_spares;
3498 		aux = ZPOOL_CONFIG_SPARES;
3499 	} else {
3500 		sav = &spa->spa_l2cache;
3501 		aux = ZPOOL_CONFIG_L2CACHE;
3502 	}
3503 
3504 	mutex_enter(&ztest_vdev_lock);
3505 
3506 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3507 
3508 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
3509 		/*
3510 		 * Pick a random device to remove.
3511 		 */
3512 		vdev_t *svd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3513 
3514 		/* dRAID spares cannot be removed; try anyways to see ENOTSUP */
3515 		if (strstr(svd->vdev_path, VDEV_TYPE_DRAID) != NULL)
3516 			ignore_err = ENOTSUP;
3517 
3518 		guid = svd->vdev_guid;
3519 	} else {
3520 		/*
3521 		 * Find an unused device we can add.
3522 		 */
3523 		zs->zs_vdev_aux = 0;
3524 		for (;;) {
3525 			int c;
3526 			(void) snprintf(path, MAXPATHLEN, ztest_aux_template,
3527 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3528 			    zs->zs_vdev_aux);
3529 			for (c = 0; c < sav->sav_count; c++)
3530 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
3531 				    path) == 0)
3532 					break;
3533 			if (c == sav->sav_count &&
3534 			    vdev_lookup_by_path(rvd, path) == NULL)
3535 				break;
3536 			zs->zs_vdev_aux++;
3537 		}
3538 	}
3539 
3540 	spa_config_exit(spa, SCL_VDEV, FTAG);
3541 
3542 	if (guid == 0) {
3543 		/*
3544 		 * Add a new device.
3545 		 */
3546 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3547 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
3548 		error = spa_vdev_add(spa, nvroot);
3549 
3550 		switch (error) {
3551 		case 0:
3552 			break;
3553 		default:
3554 			fatal(B_FALSE, "spa_vdev_add(%p) = %d", nvroot, error);
3555 		}
3556 		fnvlist_free(nvroot);
3557 	} else {
3558 		/*
3559 		 * Remove an existing device.  Sometimes, dirty its
3560 		 * vdev state first to make sure we handle removal
3561 		 * of devices that have pending state changes.
3562 		 */
3563 		if (ztest_random(2) == 0)
3564 			(void) vdev_online(spa, guid, 0, NULL);
3565 
3566 		error = spa_vdev_remove(spa, guid, B_FALSE);
3567 
3568 		switch (error) {
3569 		case 0:
3570 		case EBUSY:
3571 		case ZFS_ERR_CHECKPOINT_EXISTS:
3572 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3573 			break;
3574 		default:
3575 			if (error != ignore_err)
3576 				fatal(B_FALSE,
3577 				    "spa_vdev_remove(%"PRIu64") = %d",
3578 				    guid, error);
3579 		}
3580 	}
3581 
3582 	mutex_exit(&ztest_vdev_lock);
3583 
3584 	umem_free(path, MAXPATHLEN);
3585 }
3586 
3587 /*
3588  * split a pool if it has mirror tlvdevs
3589  */
3590 void
3591 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3592 {
3593 	(void) zd, (void) id;
3594 	ztest_shared_t *zs = ztest_shared;
3595 	spa_t *spa = ztest_spa;
3596 	vdev_t *rvd = spa->spa_root_vdev;
3597 	nvlist_t *tree, **child, *config, *split, **schild;
3598 	uint_t c, children, schildren = 0, lastlogid = 0;
3599 	int error = 0;
3600 
3601 	if (ztest_opts.zo_mmp_test)
3602 		return;
3603 
3604 	mutex_enter(&ztest_vdev_lock);
3605 
3606 	/* ensure we have a usable config; mirrors of raidz aren't supported */
3607 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raid_children > 1) {
3608 		mutex_exit(&ztest_vdev_lock);
3609 		return;
3610 	}
3611 
3612 	/* clean up the old pool, if any */
3613 	(void) spa_destroy("splitp");
3614 
3615 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3616 
3617 	/* generate a config from the existing config */
3618 	mutex_enter(&spa->spa_props_lock);
3619 	tree = fnvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE);
3620 	mutex_exit(&spa->spa_props_lock);
3621 
3622 	VERIFY0(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN,
3623 	    &child, &children));
3624 
3625 	schild = umem_alloc(rvd->vdev_children * sizeof (nvlist_t *),
3626 	    UMEM_NOFAIL);
3627 	for (c = 0; c < children; c++) {
3628 		vdev_t *tvd = rvd->vdev_child[c];
3629 		nvlist_t **mchild;
3630 		uint_t mchildren;
3631 
3632 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3633 			schild[schildren] = fnvlist_alloc();
3634 			fnvlist_add_string(schild[schildren],
3635 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE);
3636 			fnvlist_add_uint64(schild[schildren],
3637 			    ZPOOL_CONFIG_IS_HOLE, 1);
3638 			if (lastlogid == 0)
3639 				lastlogid = schildren;
3640 			++schildren;
3641 			continue;
3642 		}
3643 		lastlogid = 0;
3644 		VERIFY0(nvlist_lookup_nvlist_array(child[c],
3645 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren));
3646 		schild[schildren++] = fnvlist_dup(mchild[0]);
3647 	}
3648 
3649 	/* OK, create a config that can be used to split */
3650 	split = fnvlist_alloc();
3651 	fnvlist_add_string(split, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
3652 	fnvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN,
3653 	    (const nvlist_t **)schild, lastlogid != 0 ? lastlogid : schildren);
3654 
3655 	config = fnvlist_alloc();
3656 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split);
3657 
3658 	for (c = 0; c < schildren; c++)
3659 		fnvlist_free(schild[c]);
3660 	umem_free(schild, rvd->vdev_children * sizeof (nvlist_t *));
3661 	fnvlist_free(split);
3662 
3663 	spa_config_exit(spa, SCL_VDEV, FTAG);
3664 
3665 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
3666 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3667 	(void) pthread_rwlock_unlock(&ztest_name_lock);
3668 
3669 	fnvlist_free(config);
3670 
3671 	if (error == 0) {
3672 		(void) printf("successful split - results:\n");
3673 		mutex_enter(&spa_namespace_lock);
3674 		show_pool_stats(spa);
3675 		show_pool_stats(spa_lookup("splitp"));
3676 		mutex_exit(&spa_namespace_lock);
3677 		++zs->zs_splits;
3678 		--zs->zs_mirrors;
3679 	}
3680 	mutex_exit(&ztest_vdev_lock);
3681 }
3682 
3683 /*
3684  * Verify that we can attach and detach devices.
3685  */
3686 void
3687 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3688 {
3689 	(void) zd, (void) id;
3690 	ztest_shared_t *zs = ztest_shared;
3691 	spa_t *spa = ztest_spa;
3692 	spa_aux_vdev_t *sav = &spa->spa_spares;
3693 	vdev_t *rvd = spa->spa_root_vdev;
3694 	vdev_t *oldvd, *newvd, *pvd;
3695 	nvlist_t *root;
3696 	uint64_t leaves;
3697 	uint64_t leaf, top;
3698 	uint64_t ashift = ztest_get_ashift();
3699 	uint64_t oldguid, pguid;
3700 	uint64_t oldsize, newsize;
3701 	uint64_t raidz_children;
3702 	char *oldpath, *newpath;
3703 	int replacing;
3704 	int oldvd_has_siblings = B_FALSE;
3705 	int newvd_is_spare = B_FALSE;
3706 	int newvd_is_dspare = B_FALSE;
3707 	int oldvd_is_log;
3708 	int oldvd_is_special;
3709 	int error, expected_error;
3710 
3711 	if (ztest_opts.zo_mmp_test)
3712 		return;
3713 
3714 	oldpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3715 	newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
3716 
3717 	mutex_enter(&ztest_vdev_lock);
3718 	raidz_children = ztest_get_raidz_children(spa);
3719 	leaves = MAX(zs->zs_mirrors, 1) * raidz_children;
3720 
3721 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3722 
3723 	/*
3724 	 * If a vdev is in the process of being removed, its removal may
3725 	 * finish while we are in progress, leading to an unexpected error
3726 	 * value.  Don't bother trying to attach while we are in the middle
3727 	 * of removal.
3728 	 */
3729 	if (ztest_device_removal_active) {
3730 		spa_config_exit(spa, SCL_ALL, FTAG);
3731 		goto out;
3732 	}
3733 
3734 	/*
3735 	 * RAIDZ leaf VDEV mirrors are not currently supported while a
3736 	 * RAIDZ expansion is in progress.
3737 	 */
3738 	if (ztest_opts.zo_raid_do_expand) {
3739 		spa_config_exit(spa, SCL_ALL, FTAG);
3740 		goto out;
3741 	}
3742 
3743 	/*
3744 	 * Decide whether to do an attach or a replace.
3745 	 */
3746 	replacing = ztest_random(2);
3747 
3748 	/*
3749 	 * Pick a random top-level vdev.
3750 	 */
3751 	top = ztest_random_vdev_top(spa, B_TRUE);
3752 
3753 	/*
3754 	 * Pick a random leaf within it.
3755 	 */
3756 	leaf = ztest_random(leaves);
3757 
3758 	/*
3759 	 * Locate this vdev.
3760 	 */
3761 	oldvd = rvd->vdev_child[top];
3762 
3763 	/* pick a child from the mirror */
3764 	if (zs->zs_mirrors >= 1) {
3765 		ASSERT3P(oldvd->vdev_ops, ==, &vdev_mirror_ops);
3766 		ASSERT3U(oldvd->vdev_children, >=, zs->zs_mirrors);
3767 		oldvd = oldvd->vdev_child[leaf / raidz_children];
3768 	}
3769 
3770 	/* pick a child out of the raidz group */
3771 	if (ztest_opts.zo_raid_children > 1) {
3772 		if (strcmp(oldvd->vdev_ops->vdev_op_type, "raidz") == 0)
3773 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_raidz_ops);
3774 		else
3775 			ASSERT3P(oldvd->vdev_ops, ==, &vdev_draid_ops);
3776 		oldvd = oldvd->vdev_child[leaf % raidz_children];
3777 	}
3778 
3779 	/*
3780 	 * If we're already doing an attach or replace, oldvd may be a
3781 	 * mirror vdev -- in which case, pick a random child.
3782 	 */
3783 	while (oldvd->vdev_children != 0) {
3784 		oldvd_has_siblings = B_TRUE;
3785 		ASSERT3U(oldvd->vdev_children, >=, 2);
3786 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3787 	}
3788 
3789 	oldguid = oldvd->vdev_guid;
3790 	oldsize = vdev_get_min_asize(oldvd);
3791 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3792 	oldvd_is_special =
3793 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_SPECIAL ||
3794 	    oldvd->vdev_top->vdev_alloc_bias == VDEV_BIAS_DEDUP;
3795 	(void) strlcpy(oldpath, oldvd->vdev_path, MAXPATHLEN);
3796 	pvd = oldvd->vdev_parent;
3797 	pguid = pvd->vdev_guid;
3798 
3799 	/*
3800 	 * If oldvd has siblings, then half of the time, detach it.  Prior
3801 	 * to the detach the pool is scrubbed in order to prevent creating
3802 	 * unrepairable blocks as a result of the data corruption injection.
3803 	 */
3804 	if (oldvd_has_siblings && ztest_random(2) == 0) {
3805 		spa_config_exit(spa, SCL_ALL, FTAG);
3806 
3807 		error = ztest_scrub_impl(spa);
3808 		if (error)
3809 			goto out;
3810 
3811 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3812 		if (error != 0 && error != ENODEV && error != EBUSY &&
3813 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3814 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3815 			fatal(B_FALSE, "detach (%s) returned %d",
3816 			    oldpath, error);
3817 		goto out;
3818 	}
3819 
3820 	/*
3821 	 * For the new vdev, choose with equal probability between the two
3822 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3823 	 */
3824 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3825 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3826 		newvd_is_spare = B_TRUE;
3827 
3828 		if (newvd->vdev_ops == &vdev_draid_spare_ops)
3829 			newvd_is_dspare = B_TRUE;
3830 
3831 		(void) strlcpy(newpath, newvd->vdev_path, MAXPATHLEN);
3832 	} else {
3833 		(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
3834 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3835 		    top * leaves + leaf);
3836 		if (ztest_random(2) == 0)
3837 			newpath[strlen(newpath) - 1] = 'b';
3838 		newvd = vdev_lookup_by_path(rvd, newpath);
3839 	}
3840 
3841 	if (newvd) {
3842 		/*
3843 		 * Reopen to ensure the vdev's asize field isn't stale.
3844 		 */
3845 		vdev_reopen(newvd);
3846 		newsize = vdev_get_min_asize(newvd);
3847 	} else {
3848 		/*
3849 		 * Make newsize a little bigger or smaller than oldsize.
3850 		 * If it's smaller, the attach should fail.
3851 		 * If it's larger, and we're doing a replace,
3852 		 * we should get dynamic LUN growth when we're done.
3853 		 */
3854 		newsize = 10 * oldsize / (9 + ztest_random(3));
3855 	}
3856 
3857 	/*
3858 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3859 	 * unless it's a replace; in that case any non-replacing parent is OK.
3860 	 *
3861 	 * If newvd is already part of the pool, it should fail with EBUSY.
3862 	 *
3863 	 * If newvd is too small, it should fail with EOVERFLOW.
3864 	 *
3865 	 * If newvd is a distributed spare and it's being attached to a
3866 	 * dRAID which is not its parent it should fail with EINVAL.
3867 	 */
3868 	if (pvd->vdev_ops != &vdev_mirror_ops &&
3869 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3870 	    pvd->vdev_ops == &vdev_replacing_ops ||
3871 	    pvd->vdev_ops == &vdev_spare_ops))
3872 		expected_error = ENOTSUP;
3873 	else if (newvd_is_spare &&
3874 	    (!replacing || oldvd_is_log || oldvd_is_special))
3875 		expected_error = ENOTSUP;
3876 	else if (newvd == oldvd)
3877 		expected_error = replacing ? 0 : EBUSY;
3878 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3879 		expected_error = EBUSY;
3880 	else if (!newvd_is_dspare && newsize < oldsize)
3881 		expected_error = EOVERFLOW;
3882 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3883 		expected_error = EDOM;
3884 	else if (newvd_is_dspare && pvd != vdev_draid_spare_get_parent(newvd))
3885 		expected_error = EINVAL;
3886 	else
3887 		expected_error = 0;
3888 
3889 	spa_config_exit(spa, SCL_ALL, FTAG);
3890 
3891 	/*
3892 	 * Build the nvlist describing newpath.
3893 	 */
3894 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3895 	    ashift, NULL, 0, 0, 1);
3896 
3897 	/*
3898 	 * When supported select either a healing or sequential resilver.
3899 	 */
3900 	boolean_t rebuilding = B_FALSE;
3901 	if (pvd->vdev_ops == &vdev_mirror_ops ||
3902 	    pvd->vdev_ops ==  &vdev_root_ops) {
3903 		rebuilding = !!ztest_random(2);
3904 	}
3905 
3906 	error = spa_vdev_attach(spa, oldguid, root, replacing, rebuilding);
3907 
3908 	fnvlist_free(root);
3909 
3910 	/*
3911 	 * If our parent was the replacing vdev, but the replace completed,
3912 	 * then instead of failing with ENOTSUP we may either succeed,
3913 	 * fail with ENODEV, or fail with EOVERFLOW.
3914 	 */
3915 	if (expected_error == ENOTSUP &&
3916 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3917 		expected_error = error;
3918 
3919 	/*
3920 	 * If someone grew the LUN, the replacement may be too small.
3921 	 */
3922 	if (error == EOVERFLOW || error == EBUSY)
3923 		expected_error = error;
3924 
3925 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3926 	    error == ZFS_ERR_DISCARDING_CHECKPOINT ||
3927 	    error == ZFS_ERR_RESILVER_IN_PROGRESS ||
3928 	    error == ZFS_ERR_REBUILD_IN_PROGRESS)
3929 		expected_error = error;
3930 
3931 	if (error != expected_error && expected_error != EBUSY) {
3932 		fatal(B_FALSE, "attach (%s %"PRIu64", %s %"PRIu64", %d) "
3933 		    "returned %d, expected %d",
3934 		    oldpath, oldsize, newpath,
3935 		    newsize, replacing, error, expected_error);
3936 	}
3937 out:
3938 	mutex_exit(&ztest_vdev_lock);
3939 
3940 	umem_free(oldpath, MAXPATHLEN);
3941 	umem_free(newpath, MAXPATHLEN);
3942 }
3943 
3944 static void
3945 raidz_scratch_verify(void)
3946 {
3947 	spa_t *spa;
3948 	uint64_t write_size, logical_size, offset;
3949 	raidz_reflow_scratch_state_t state;
3950 	vdev_raidz_expand_t *vre;
3951 	vdev_t *raidvd;
3952 
3953 	ASSERT(raidz_expand_pause_point == RAIDZ_EXPAND_PAUSE_NONE);
3954 
3955 	if (ztest_scratch_state->zs_raidz_scratch_verify_pause == 0)
3956 		return;
3957 
3958 	kernel_init(SPA_MODE_READ);
3959 
3960 	mutex_enter(&spa_namespace_lock);
3961 	spa = spa_lookup(ztest_opts.zo_pool);
3962 	ASSERT(spa);
3963 	spa->spa_import_flags |= ZFS_IMPORT_SKIP_MMP;
3964 	mutex_exit(&spa_namespace_lock);
3965 
3966 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
3967 
3968 	ASSERT3U(RRSS_GET_OFFSET(&spa->spa_uberblock), !=, UINT64_MAX);
3969 
3970 	mutex_enter(&ztest_vdev_lock);
3971 
3972 	spa_config_enter(spa, SCL_ALL, FTAG, RW_READER);
3973 
3974 	vre = spa->spa_raidz_expand;
3975 	if (vre == NULL)
3976 		goto out;
3977 
3978 	raidvd = vdev_lookup_top(spa, vre->vre_vdev_id);
3979 	offset = RRSS_GET_OFFSET(&spa->spa_uberblock);
3980 	state = RRSS_GET_STATE(&spa->spa_uberblock);
3981 	write_size = P2ALIGN(VDEV_BOOT_SIZE, 1 << raidvd->vdev_ashift);
3982 	logical_size = write_size * raidvd->vdev_children;
3983 
3984 	switch (state) {
3985 		/*
3986 		 * Initial state of reflow process.  RAIDZ expansion was
3987 		 * requested by user, but scratch object was not created.
3988 		 */
3989 		case RRSS_SCRATCH_NOT_IN_USE:
3990 			ASSERT3U(offset, ==, 0);
3991 			break;
3992 
3993 		/*
3994 		 * Scratch object was synced and stored in boot area.
3995 		 */
3996 		case RRSS_SCRATCH_VALID:
3997 
3998 		/*
3999 		 * Scratch object was synced back to raidz start offset,
4000 		 * raidz is ready for sector by sector reflow process.
4001 		 */
4002 		case RRSS_SCRATCH_INVALID_SYNCED:
4003 
4004 		/*
4005 		 * Scratch object was synced back to raidz start offset
4006 		 * on zpool importing, raidz is ready for sector by sector
4007 		 * reflow process.
4008 		 */
4009 		case RRSS_SCRATCH_INVALID_SYNCED_ON_IMPORT:
4010 			ASSERT3U(offset, ==, logical_size);
4011 			break;
4012 
4013 		/*
4014 		 * Sector by sector reflow process started.
4015 		 */
4016 		case RRSS_SCRATCH_INVALID_SYNCED_REFLOW:
4017 			ASSERT3U(offset, >=, logical_size);
4018 			break;
4019 	}
4020 
4021 out:
4022 	spa_config_exit(spa, SCL_ALL, FTAG);
4023 
4024 	mutex_exit(&ztest_vdev_lock);
4025 
4026 	ztest_scratch_state->zs_raidz_scratch_verify_pause = 0;
4027 
4028 	spa_close(spa, FTAG);
4029 	kernel_fini();
4030 }
4031 
4032 static void
4033 ztest_scratch_thread(void *arg)
4034 {
4035 	(void) arg;
4036 
4037 	/* wait up to 10 seconds */
4038 	for (int t = 100; t > 0; t -= 1) {
4039 		if (raidz_expand_pause_point == RAIDZ_EXPAND_PAUSE_NONE)
4040 			thread_exit();
4041 
4042 		(void) poll(NULL, 0, 100);
4043 	}
4044 
4045 	/* killed when the scratch area progress reached a certain point */
4046 	ztest_kill(ztest_shared);
4047 }
4048 
4049 /*
4050  * Verify that we can attach raidz device.
4051  */
4052 void
4053 ztest_vdev_raidz_attach(ztest_ds_t *zd, uint64_t id)
4054 {
4055 	(void) zd, (void) id;
4056 	ztest_shared_t *zs = ztest_shared;
4057 	spa_t *spa = ztest_spa;
4058 	uint64_t leaves, raidz_children, newsize, ashift = ztest_get_ashift();
4059 	kthread_t *scratch_thread = NULL;
4060 	vdev_t *newvd, *pvd;
4061 	nvlist_t *root;
4062 	char *newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
4063 	int error, expected_error = 0;
4064 
4065 	mutex_enter(&ztest_vdev_lock);
4066 
4067 	spa_config_enter(spa, SCL_ALL, FTAG, RW_READER);
4068 
4069 	/* Only allow attach when raid-kind = 'eraidz' */
4070 	if (!ztest_opts.zo_raid_do_expand) {
4071 		spa_config_exit(spa, SCL_ALL, FTAG);
4072 		goto out;
4073 	}
4074 
4075 	if (ztest_opts.zo_mmp_test) {
4076 		spa_config_exit(spa, SCL_ALL, FTAG);
4077 		goto out;
4078 	}
4079 
4080 	if (ztest_device_removal_active) {
4081 		spa_config_exit(spa, SCL_ALL, FTAG);
4082 		goto out;
4083 	}
4084 
4085 	pvd = vdev_lookup_top(spa, 0);
4086 
4087 	ASSERT(pvd->vdev_ops == &vdev_raidz_ops);
4088 
4089 	/*
4090 	 * Get size of a child of the raidz group,
4091 	 * make sure device is a bit bigger
4092 	 */
4093 	newvd = pvd->vdev_child[ztest_random(pvd->vdev_children)];
4094 	newsize = 10 * vdev_get_min_asize(newvd) / (9 + ztest_random(2));
4095 
4096 	/*
4097 	 * Get next attached leaf id
4098 	 */
4099 	raidz_children = ztest_get_raidz_children(spa);
4100 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * raidz_children;
4101 	zs->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
4102 
4103 	if (spa->spa_raidz_expand)
4104 		expected_error = ZFS_ERR_RAIDZ_EXPAND_IN_PROGRESS;
4105 
4106 	spa_config_exit(spa, SCL_ALL, FTAG);
4107 
4108 	/*
4109 	 * Path to vdev to be attached
4110 	 */
4111 	(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
4112 	    ztest_opts.zo_dir, ztest_opts.zo_pool, zs->zs_vdev_next_leaf);
4113 
4114 	/*
4115 	 * Build the nvlist describing newpath.
4116 	 */
4117 	root = make_vdev_root(newpath, NULL, NULL, newsize, ashift, NULL,
4118 	    0, 0, 1);
4119 
4120 	/*
4121 	 * 50% of the time, set raidz_expand_pause_point to cause
4122 	 * raidz_reflow_scratch_sync() to pause at a certain point and
4123 	 * then kill the test after 10 seconds so raidz_scratch_verify()
4124 	 * can confirm consistency when the pool is imported.
4125 	 */
4126 	if (ztest_random(2) == 0 && expected_error == 0) {
4127 		raidz_expand_pause_point =
4128 		    ztest_random(RAIDZ_EXPAND_PAUSE_SCRATCH_POST_REFLOW_2) + 1;
4129 		scratch_thread = thread_create(NULL, 0, ztest_scratch_thread,
4130 		    ztest_shared, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
4131 	}
4132 
4133 	error = spa_vdev_attach(spa, pvd->vdev_guid, root, B_FALSE, B_FALSE);
4134 
4135 	nvlist_free(root);
4136 
4137 	if (error == EOVERFLOW || error == ENXIO ||
4138 	    error == ZFS_ERR_CHECKPOINT_EXISTS ||
4139 	    error == ZFS_ERR_DISCARDING_CHECKPOINT)
4140 		expected_error = error;
4141 
4142 	if (error != 0 && error != expected_error) {
4143 		fatal(0, "raidz attach (%s %"PRIu64") returned %d, expected %d",
4144 		    newpath, newsize, error, expected_error);
4145 	}
4146 
4147 	if (raidz_expand_pause_point) {
4148 		if (error != 0) {
4149 			/*
4150 			 * Do not verify scratch object in case of error
4151 			 * returned by vdev attaching.
4152 			 */
4153 			raidz_expand_pause_point = RAIDZ_EXPAND_PAUSE_NONE;
4154 		}
4155 
4156 		VERIFY0(thread_join(scratch_thread));
4157 	}
4158 out:
4159 	mutex_exit(&ztest_vdev_lock);
4160 
4161 	umem_free(newpath, MAXPATHLEN);
4162 }
4163 
4164 void
4165 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
4166 {
4167 	(void) zd, (void) id;
4168 	spa_t *spa = ztest_spa;
4169 	vdev_t *vd;
4170 	uint64_t guid;
4171 	int error;
4172 
4173 	mutex_enter(&ztest_vdev_lock);
4174 
4175 	if (ztest_device_removal_active) {
4176 		mutex_exit(&ztest_vdev_lock);
4177 		return;
4178 	}
4179 
4180 	/*
4181 	 * Remove a random top-level vdev and wait for removal to finish.
4182 	 */
4183 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
4184 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
4185 	guid = vd->vdev_guid;
4186 	spa_config_exit(spa, SCL_VDEV, FTAG);
4187 
4188 	error = spa_vdev_remove(spa, guid, B_FALSE);
4189 	if (error == 0) {
4190 		ztest_device_removal_active = B_TRUE;
4191 		mutex_exit(&ztest_vdev_lock);
4192 
4193 		/*
4194 		 * spa->spa_vdev_removal is created in a sync task that
4195 		 * is initiated via dsl_sync_task_nowait(). Since the
4196 		 * task may not run before spa_vdev_remove() returns, we
4197 		 * must wait at least 1 txg to ensure that the removal
4198 		 * struct has been created.
4199 		 */
4200 		txg_wait_synced(spa_get_dsl(spa), 0);
4201 
4202 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
4203 			txg_wait_synced(spa_get_dsl(spa), 0);
4204 	} else {
4205 		mutex_exit(&ztest_vdev_lock);
4206 		return;
4207 	}
4208 
4209 	/*
4210 	 * The pool needs to be scrubbed after completing device removal.
4211 	 * Failure to do so may result in checksum errors due to the
4212 	 * strategy employed by ztest_fault_inject() when selecting which
4213 	 * offset are redundant and can be damaged.
4214 	 */
4215 	error = spa_scan(spa, POOL_SCAN_SCRUB);
4216 	if (error == 0) {
4217 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
4218 			txg_wait_synced(spa_get_dsl(spa), 0);
4219 	}
4220 
4221 	mutex_enter(&ztest_vdev_lock);
4222 	ztest_device_removal_active = B_FALSE;
4223 	mutex_exit(&ztest_vdev_lock);
4224 }
4225 
4226 /*
4227  * Callback function which expands the physical size of the vdev.
4228  */
4229 static vdev_t *
4230 grow_vdev(vdev_t *vd, void *arg)
4231 {
4232 	spa_t *spa __maybe_unused = vd->vdev_spa;
4233 	size_t *newsize = arg;
4234 	size_t fsize;
4235 	int fd;
4236 
4237 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
4238 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4239 
4240 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
4241 		return (vd);
4242 
4243 	fsize = lseek(fd, 0, SEEK_END);
4244 	VERIFY0(ftruncate(fd, *newsize));
4245 
4246 	if (ztest_opts.zo_verbose >= 6) {
4247 		(void) printf("%s grew from %lu to %lu bytes\n",
4248 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
4249 	}
4250 	(void) close(fd);
4251 	return (NULL);
4252 }
4253 
4254 /*
4255  * Callback function which expands a given vdev by calling vdev_online().
4256  */
4257 static vdev_t *
4258 online_vdev(vdev_t *vd, void *arg)
4259 {
4260 	(void) arg;
4261 	spa_t *spa = vd->vdev_spa;
4262 	vdev_t *tvd = vd->vdev_top;
4263 	uint64_t guid = vd->vdev_guid;
4264 	uint64_t generation = spa->spa_config_generation + 1;
4265 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
4266 	int error;
4267 
4268 	ASSERT3S(spa_config_held(spa, SCL_STATE, RW_READER), ==, SCL_STATE);
4269 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4270 
4271 	/* Calling vdev_online will initialize the new metaslabs */
4272 	spa_config_exit(spa, SCL_STATE, spa);
4273 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
4274 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4275 
4276 	/*
4277 	 * If vdev_online returned an error or the underlying vdev_open
4278 	 * failed then we abort the expand. The only way to know that
4279 	 * vdev_open fails is by checking the returned newstate.
4280 	 */
4281 	if (error || newstate != VDEV_STATE_HEALTHY) {
4282 		if (ztest_opts.zo_verbose >= 5) {
4283 			(void) printf("Unable to expand vdev, state %u, "
4284 			    "error %d\n", newstate, error);
4285 		}
4286 		return (vd);
4287 	}
4288 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
4289 
4290 	/*
4291 	 * Since we dropped the lock we need to ensure that we're
4292 	 * still talking to the original vdev. It's possible this
4293 	 * vdev may have been detached/replaced while we were
4294 	 * trying to online it.
4295 	 */
4296 	if (generation != spa->spa_config_generation) {
4297 		if (ztest_opts.zo_verbose >= 5) {
4298 			(void) printf("vdev configuration has changed, "
4299 			    "guid %"PRIu64", state %"PRIu64", "
4300 			    "expected gen %"PRIu64", got gen %"PRIu64"\n",
4301 			    guid,
4302 			    tvd->vdev_state,
4303 			    generation,
4304 			    spa->spa_config_generation);
4305 		}
4306 		return (vd);
4307 	}
4308 	return (NULL);
4309 }
4310 
4311 /*
4312  * Traverse the vdev tree calling the supplied function.
4313  * We continue to walk the tree until we either have walked all
4314  * children or we receive a non-NULL return from the callback.
4315  * If a NULL callback is passed, then we just return back the first
4316  * leaf vdev we encounter.
4317  */
4318 static vdev_t *
4319 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
4320 {
4321 	uint_t c;
4322 
4323 	if (vd->vdev_ops->vdev_op_leaf) {
4324 		if (func == NULL)
4325 			return (vd);
4326 		else
4327 			return (func(vd, arg));
4328 	}
4329 
4330 	for (c = 0; c < vd->vdev_children; c++) {
4331 		vdev_t *cvd = vd->vdev_child[c];
4332 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
4333 			return (cvd);
4334 	}
4335 	return (NULL);
4336 }
4337 
4338 /*
4339  * Verify that dynamic LUN growth works as expected.
4340  */
4341 void
4342 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
4343 {
4344 	(void) zd, (void) id;
4345 	spa_t *spa = ztest_spa;
4346 	vdev_t *vd, *tvd;
4347 	metaslab_class_t *mc;
4348 	metaslab_group_t *mg;
4349 	size_t psize, newsize;
4350 	uint64_t top;
4351 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
4352 
4353 	mutex_enter(&ztest_checkpoint_lock);
4354 	mutex_enter(&ztest_vdev_lock);
4355 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4356 
4357 	/*
4358 	 * If there is a vdev removal in progress, it could complete while
4359 	 * we are running, in which case we would not be able to verify
4360 	 * that the metaslab_class space increased (because it decreases
4361 	 * when the device removal completes).
4362 	 */
4363 	if (ztest_device_removal_active) {
4364 		spa_config_exit(spa, SCL_STATE, spa);
4365 		mutex_exit(&ztest_vdev_lock);
4366 		mutex_exit(&ztest_checkpoint_lock);
4367 		return;
4368 	}
4369 
4370 	/*
4371 	 * If we are under raidz expansion, the test can failed because the
4372 	 * metaslabs count will not increase immediately after the vdev is
4373 	 * expanded. It will happen only after raidz expansion completion.
4374 	 */
4375 	if (spa->spa_raidz_expand) {
4376 		spa_config_exit(spa, SCL_STATE, spa);
4377 		mutex_exit(&ztest_vdev_lock);
4378 		mutex_exit(&ztest_checkpoint_lock);
4379 		return;
4380 	}
4381 
4382 	top = ztest_random_vdev_top(spa, B_TRUE);
4383 
4384 	tvd = spa->spa_root_vdev->vdev_child[top];
4385 	mg = tvd->vdev_mg;
4386 	mc = mg->mg_class;
4387 	old_ms_count = tvd->vdev_ms_count;
4388 	old_class_space = metaslab_class_get_space(mc);
4389 
4390 	/*
4391 	 * Determine the size of the first leaf vdev associated with
4392 	 * our top-level device.
4393 	 */
4394 	vd = vdev_walk_tree(tvd, NULL, NULL);
4395 	ASSERT3P(vd, !=, NULL);
4396 	ASSERT(vd->vdev_ops->vdev_op_leaf);
4397 
4398 	psize = vd->vdev_psize;
4399 
4400 	/*
4401 	 * We only try to expand the vdev if it's healthy, less than 4x its
4402 	 * original size, and it has a valid psize.
4403 	 */
4404 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
4405 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
4406 		spa_config_exit(spa, SCL_STATE, spa);
4407 		mutex_exit(&ztest_vdev_lock);
4408 		mutex_exit(&ztest_checkpoint_lock);
4409 		return;
4410 	}
4411 	ASSERT3U(psize, >, 0);
4412 	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
4413 	ASSERT3U(newsize, >, psize);
4414 
4415 	if (ztest_opts.zo_verbose >= 6) {
4416 		(void) printf("Expanding LUN %s from %lu to %lu\n",
4417 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
4418 	}
4419 
4420 	/*
4421 	 * Growing the vdev is a two step process:
4422 	 *	1). expand the physical size (i.e. relabel)
4423 	 *	2). online the vdev to create the new metaslabs
4424 	 */
4425 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
4426 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
4427 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
4428 		if (ztest_opts.zo_verbose >= 5) {
4429 			(void) printf("Could not expand LUN because "
4430 			    "the vdev configuration changed.\n");
4431 		}
4432 		spa_config_exit(spa, SCL_STATE, spa);
4433 		mutex_exit(&ztest_vdev_lock);
4434 		mutex_exit(&ztest_checkpoint_lock);
4435 		return;
4436 	}
4437 
4438 	spa_config_exit(spa, SCL_STATE, spa);
4439 
4440 	/*
4441 	 * Expanding the LUN will update the config asynchronously,
4442 	 * thus we must wait for the async thread to complete any
4443 	 * pending tasks before proceeding.
4444 	 */
4445 	for (;;) {
4446 		boolean_t done;
4447 		mutex_enter(&spa->spa_async_lock);
4448 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
4449 		mutex_exit(&spa->spa_async_lock);
4450 		if (done)
4451 			break;
4452 		txg_wait_synced(spa_get_dsl(spa), 0);
4453 		(void) poll(NULL, 0, 100);
4454 	}
4455 
4456 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
4457 
4458 	tvd = spa->spa_root_vdev->vdev_child[top];
4459 	new_ms_count = tvd->vdev_ms_count;
4460 	new_class_space = metaslab_class_get_space(mc);
4461 
4462 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
4463 		if (ztest_opts.zo_verbose >= 5) {
4464 			(void) printf("Could not verify LUN expansion due to "
4465 			    "intervening vdev offline or remove.\n");
4466 		}
4467 		spa_config_exit(spa, SCL_STATE, spa);
4468 		mutex_exit(&ztest_vdev_lock);
4469 		mutex_exit(&ztest_checkpoint_lock);
4470 		return;
4471 	}
4472 
4473 	/*
4474 	 * Make sure we were able to grow the vdev.
4475 	 */
4476 	if (new_ms_count <= old_ms_count) {
4477 		fatal(B_FALSE,
4478 		    "LUN expansion failed: ms_count %"PRIu64" < %"PRIu64"\n",
4479 		    old_ms_count, new_ms_count);
4480 	}
4481 
4482 	/*
4483 	 * Make sure we were able to grow the pool.
4484 	 */
4485 	if (new_class_space <= old_class_space) {
4486 		fatal(B_FALSE,
4487 		    "LUN expansion failed: class_space %"PRIu64" < %"PRIu64"\n",
4488 		    old_class_space, new_class_space);
4489 	}
4490 
4491 	if (ztest_opts.zo_verbose >= 5) {
4492 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
4493 
4494 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
4495 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
4496 		(void) printf("%s grew from %s to %s\n",
4497 		    spa->spa_name, oldnumbuf, newnumbuf);
4498 	}
4499 
4500 	spa_config_exit(spa, SCL_STATE, spa);
4501 	mutex_exit(&ztest_vdev_lock);
4502 	mutex_exit(&ztest_checkpoint_lock);
4503 }
4504 
4505 /*
4506  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
4507  */
4508 static void
4509 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
4510 {
4511 	(void) arg, (void) cr;
4512 
4513 	/*
4514 	 * Create the objects common to all ztest datasets.
4515 	 */
4516 	VERIFY0(zap_create_claim(os, ZTEST_DIROBJ,
4517 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx));
4518 }
4519 
4520 static int
4521 ztest_dataset_create(char *dsname)
4522 {
4523 	int err;
4524 	uint64_t rand;
4525 	dsl_crypto_params_t *dcp = NULL;
4526 
4527 	/*
4528 	 * 50% of the time, we create encrypted datasets
4529 	 * using a random cipher suite and a hard-coded
4530 	 * wrapping key.
4531 	 */
4532 	rand = ztest_random(2);
4533 	if (rand != 0) {
4534 		nvlist_t *crypto_args = fnvlist_alloc();
4535 		nvlist_t *props = fnvlist_alloc();
4536 
4537 		/* slight bias towards the default cipher suite */
4538 		rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
4539 		if (rand < ZIO_CRYPT_AES_128_CCM)
4540 			rand = ZIO_CRYPT_ON;
4541 
4542 		fnvlist_add_uint64(props,
4543 		    zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
4544 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
4545 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
4546 
4547 		/*
4548 		 * These parameters aren't really used by the kernel. They
4549 		 * are simply stored so that userspace knows how to load
4550 		 * the wrapping key.
4551 		 */
4552 		fnvlist_add_uint64(props,
4553 		    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
4554 		fnvlist_add_string(props,
4555 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
4556 		fnvlist_add_uint64(props,
4557 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
4558 		fnvlist_add_uint64(props,
4559 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
4560 
4561 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
4562 		    crypto_args, &dcp));
4563 
4564 		/*
4565 		 * Cycle through all available encryption implementations
4566 		 * to verify interoperability.
4567 		 */
4568 		VERIFY0(gcm_impl_set("cycle"));
4569 		VERIFY0(aes_impl_set("cycle"));
4570 
4571 		fnvlist_free(crypto_args);
4572 		fnvlist_free(props);
4573 	}
4574 
4575 	err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
4576 	    ztest_objset_create_cb, NULL);
4577 	dsl_crypto_params_free(dcp, !!err);
4578 
4579 	rand = ztest_random(100);
4580 	if (err || rand < 80)
4581 		return (err);
4582 
4583 	if (ztest_opts.zo_verbose >= 5)
4584 		(void) printf("Setting dataset %s to sync always\n", dsname);
4585 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
4586 	    ZFS_SYNC_ALWAYS, B_FALSE));
4587 }
4588 
4589 static int
4590 ztest_objset_destroy_cb(const char *name, void *arg)
4591 {
4592 	(void) arg;
4593 	objset_t *os;
4594 	dmu_object_info_t doi;
4595 	int error;
4596 
4597 	/*
4598 	 * Verify that the dataset contains a directory object.
4599 	 */
4600 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
4601 	    B_TRUE, FTAG, &os));
4602 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
4603 	if (error != ENOENT) {
4604 		/* We could have crashed in the middle of destroying it */
4605 		ASSERT0(error);
4606 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
4607 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
4608 	}
4609 	dmu_objset_disown(os, B_TRUE, FTAG);
4610 
4611 	/*
4612 	 * Destroy the dataset.
4613 	 */
4614 	if (strchr(name, '@') != NULL) {
4615 		error = dsl_destroy_snapshot(name, B_TRUE);
4616 		if (error != ECHRNG) {
4617 			/*
4618 			 * The program was executed, but encountered a runtime
4619 			 * error, such as insufficient slop, or a hold on the
4620 			 * dataset.
4621 			 */
4622 			ASSERT0(error);
4623 		}
4624 	} else {
4625 		error = dsl_destroy_head(name);
4626 		if (error == ENOSPC) {
4627 			/* There could be checkpoint or insufficient slop */
4628 			ztest_record_enospc(FTAG);
4629 		} else if (error != EBUSY) {
4630 			/* There could be a hold on this dataset */
4631 			ASSERT0(error);
4632 		}
4633 	}
4634 	return (0);
4635 }
4636 
4637 static boolean_t
4638 ztest_snapshot_create(char *osname, uint64_t id)
4639 {
4640 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4641 	int error;
4642 
4643 	(void) snprintf(snapname, sizeof (snapname), "%"PRIu64"", id);
4644 
4645 	error = dmu_objset_snapshot_one(osname, snapname);
4646 	if (error == ENOSPC) {
4647 		ztest_record_enospc(FTAG);
4648 		return (B_FALSE);
4649 	}
4650 	if (error != 0 && error != EEXIST && error != ECHRNG) {
4651 		fatal(B_FALSE, "ztest_snapshot_create(%s@%s) = %d", osname,
4652 		    snapname, error);
4653 	}
4654 	return (B_TRUE);
4655 }
4656 
4657 static boolean_t
4658 ztest_snapshot_destroy(char *osname, uint64_t id)
4659 {
4660 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
4661 	int error;
4662 
4663 	(void) snprintf(snapname, sizeof (snapname), "%s@%"PRIu64"",
4664 	    osname, id);
4665 
4666 	error = dsl_destroy_snapshot(snapname, B_FALSE);
4667 	if (error != 0 && error != ENOENT && error != ECHRNG)
4668 		fatal(B_FALSE, "ztest_snapshot_destroy(%s) = %d",
4669 		    snapname, error);
4670 	return (B_TRUE);
4671 }
4672 
4673 void
4674 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
4675 {
4676 	(void) zd;
4677 	ztest_ds_t *zdtmp;
4678 	int iters;
4679 	int error;
4680 	objset_t *os, *os2;
4681 	char name[ZFS_MAX_DATASET_NAME_LEN];
4682 	zilog_t *zilog;
4683 	int i;
4684 
4685 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
4686 
4687 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4688 
4689 	(void) snprintf(name, sizeof (name), "%s/temp_%"PRIu64"",
4690 	    ztest_opts.zo_pool, id);
4691 
4692 	/*
4693 	 * If this dataset exists from a previous run, process its replay log
4694 	 * half of the time.  If we don't replay it, then dsl_destroy_head()
4695 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
4696 	 */
4697 	if (ztest_random(2) == 0 &&
4698 	    ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
4699 	    B_TRUE, FTAG, &os) == 0) {
4700 		ztest_zd_init(zdtmp, NULL, os);
4701 		zil_replay(os, zdtmp, ztest_replay_vector);
4702 		ztest_zd_fini(zdtmp);
4703 		dmu_objset_disown(os, B_TRUE, FTAG);
4704 	}
4705 
4706 	/*
4707 	 * There may be an old instance of the dataset we're about to
4708 	 * create lying around from a previous run.  If so, destroy it
4709 	 * and all of its snapshots.
4710 	 */
4711 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
4712 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
4713 
4714 	/*
4715 	 * Verify that the destroyed dataset is no longer in the namespace.
4716 	 * It may still be present if the destroy above fails with ENOSPC.
4717 	 */
4718 	error = ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, B_TRUE,
4719 	    FTAG, &os);
4720 	if (error == 0) {
4721 		dmu_objset_disown(os, B_TRUE, FTAG);
4722 		ztest_record_enospc(FTAG);
4723 		goto out;
4724 	}
4725 	VERIFY3U(ENOENT, ==, error);
4726 
4727 	/*
4728 	 * Verify that we can create a new dataset.
4729 	 */
4730 	error = ztest_dataset_create(name);
4731 	if (error) {
4732 		if (error == ENOSPC) {
4733 			ztest_record_enospc(FTAG);
4734 			goto out;
4735 		}
4736 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", name, error);
4737 	}
4738 
4739 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
4740 	    FTAG, &os));
4741 
4742 	ztest_zd_init(zdtmp, NULL, os);
4743 
4744 	/*
4745 	 * Open the intent log for it.
4746 	 */
4747 	zilog = zil_open(os, ztest_get_data, NULL);
4748 
4749 	/*
4750 	 * Put some objects in there, do a little I/O to them,
4751 	 * and randomly take a couple of snapshots along the way.
4752 	 */
4753 	iters = ztest_random(5);
4754 	for (i = 0; i < iters; i++) {
4755 		ztest_dmu_object_alloc_free(zdtmp, id);
4756 		if (ztest_random(iters) == 0)
4757 			(void) ztest_snapshot_create(name, i);
4758 	}
4759 
4760 	/*
4761 	 * Verify that we cannot create an existing dataset.
4762 	 */
4763 	VERIFY3U(EEXIST, ==,
4764 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
4765 
4766 	/*
4767 	 * Verify that we can hold an objset that is also owned.
4768 	 */
4769 	VERIFY0(dmu_objset_hold(name, FTAG, &os2));
4770 	dmu_objset_rele(os2, FTAG);
4771 
4772 	/*
4773 	 * Verify that we cannot own an objset that is already owned.
4774 	 */
4775 	VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
4776 	    B_FALSE, B_TRUE, FTAG, &os2));
4777 
4778 	zil_close(zilog);
4779 	dmu_objset_disown(os, B_TRUE, FTAG);
4780 	ztest_zd_fini(zdtmp);
4781 out:
4782 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4783 
4784 	umem_free(zdtmp, sizeof (ztest_ds_t));
4785 }
4786 
4787 /*
4788  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
4789  */
4790 void
4791 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
4792 {
4793 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4794 	(void) ztest_snapshot_destroy(zd->zd_name, id);
4795 	(void) ztest_snapshot_create(zd->zd_name, id);
4796 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4797 }
4798 
4799 /*
4800  * Cleanup non-standard snapshots and clones.
4801  */
4802 static void
4803 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
4804 {
4805 	char *snap1name;
4806 	char *clone1name;
4807 	char *snap2name;
4808 	char *clone2name;
4809 	char *snap3name;
4810 	int error;
4811 
4812 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4813 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4814 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4815 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4816 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4817 
4818 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4819 	    osname, id);
4820 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4821 	    osname, id);
4822 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4823 	    clone1name, id);
4824 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4825 	    osname, id);
4826 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4827 	    clone1name, id);
4828 
4829 	error = dsl_destroy_head(clone2name);
4830 	if (error && error != ENOENT)
4831 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone2name, error);
4832 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
4833 	if (error && error != ENOENT)
4834 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4835 		    snap3name, error);
4836 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
4837 	if (error && error != ENOENT)
4838 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4839 		    snap2name, error);
4840 	error = dsl_destroy_head(clone1name);
4841 	if (error && error != ENOENT)
4842 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clone1name, error);
4843 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
4844 	if (error && error != ENOENT)
4845 		fatal(B_FALSE, "dsl_destroy_snapshot(%s) = %d",
4846 		    snap1name, error);
4847 
4848 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4849 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4850 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4851 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4852 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4853 }
4854 
4855 /*
4856  * Verify dsl_dataset_promote handles EBUSY
4857  */
4858 void
4859 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
4860 {
4861 	objset_t *os;
4862 	char *snap1name;
4863 	char *clone1name;
4864 	char *snap2name;
4865 	char *clone2name;
4866 	char *snap3name;
4867 	char *osname = zd->zd_name;
4868 	int error;
4869 
4870 	snap1name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4871 	clone1name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4872 	snap2name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4873 	clone2name = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4874 	snap3name  = umem_alloc(ZFS_MAX_DATASET_NAME_LEN, UMEM_NOFAIL);
4875 
4876 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
4877 
4878 	ztest_dsl_dataset_cleanup(osname, id);
4879 
4880 	(void) snprintf(snap1name, ZFS_MAX_DATASET_NAME_LEN, "%s@s1_%"PRIu64"",
4881 	    osname, id);
4882 	(void) snprintf(clone1name, ZFS_MAX_DATASET_NAME_LEN, "%s/c1_%"PRIu64"",
4883 	    osname, id);
4884 	(void) snprintf(snap2name, ZFS_MAX_DATASET_NAME_LEN, "%s@s2_%"PRIu64"",
4885 	    clone1name, id);
4886 	(void) snprintf(clone2name, ZFS_MAX_DATASET_NAME_LEN, "%s/c2_%"PRIu64"",
4887 	    osname, id);
4888 	(void) snprintf(snap3name, ZFS_MAX_DATASET_NAME_LEN, "%s@s3_%"PRIu64"",
4889 	    clone1name, id);
4890 
4891 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
4892 	if (error && error != EEXIST) {
4893 		if (error == ENOSPC) {
4894 			ztest_record_enospc(FTAG);
4895 			goto out;
4896 		}
4897 		fatal(B_FALSE, "dmu_take_snapshot(%s) = %d", snap1name, error);
4898 	}
4899 
4900 	error = dmu_objset_clone(clone1name, snap1name);
4901 	if (error) {
4902 		if (error == ENOSPC) {
4903 			ztest_record_enospc(FTAG);
4904 			goto out;
4905 		}
4906 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone1name, error);
4907 	}
4908 
4909 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4910 	if (error && error != EEXIST) {
4911 		if (error == ENOSPC) {
4912 			ztest_record_enospc(FTAG);
4913 			goto out;
4914 		}
4915 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap2name, error);
4916 	}
4917 
4918 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4919 	if (error && error != EEXIST) {
4920 		if (error == ENOSPC) {
4921 			ztest_record_enospc(FTAG);
4922 			goto out;
4923 		}
4924 		fatal(B_FALSE, "dmu_open_snapshot(%s) = %d", snap3name, error);
4925 	}
4926 
4927 	error = dmu_objset_clone(clone2name, snap3name);
4928 	if (error) {
4929 		if (error == ENOSPC) {
4930 			ztest_record_enospc(FTAG);
4931 			goto out;
4932 		}
4933 		fatal(B_FALSE, "dmu_objset_create(%s) = %d", clone2name, error);
4934 	}
4935 
4936 	error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
4937 	    FTAG, &os);
4938 	if (error)
4939 		fatal(B_FALSE, "dmu_objset_own(%s) = %d", snap2name, error);
4940 	error = dsl_dataset_promote(clone2name, NULL);
4941 	if (error == ENOSPC) {
4942 		dmu_objset_disown(os, B_TRUE, FTAG);
4943 		ztest_record_enospc(FTAG);
4944 		goto out;
4945 	}
4946 	if (error != EBUSY)
4947 		fatal(B_FALSE, "dsl_dataset_promote(%s), %d, not EBUSY",
4948 		    clone2name, error);
4949 	dmu_objset_disown(os, B_TRUE, FTAG);
4950 
4951 out:
4952 	ztest_dsl_dataset_cleanup(osname, id);
4953 
4954 	(void) pthread_rwlock_unlock(&ztest_name_lock);
4955 
4956 	umem_free(snap1name, ZFS_MAX_DATASET_NAME_LEN);
4957 	umem_free(clone1name, ZFS_MAX_DATASET_NAME_LEN);
4958 	umem_free(snap2name, ZFS_MAX_DATASET_NAME_LEN);
4959 	umem_free(clone2name, ZFS_MAX_DATASET_NAME_LEN);
4960 	umem_free(snap3name, ZFS_MAX_DATASET_NAME_LEN);
4961 }
4962 
4963 #undef OD_ARRAY_SIZE
4964 #define	OD_ARRAY_SIZE	4
4965 
4966 /*
4967  * Verify that dmu_object_{alloc,free} work as expected.
4968  */
4969 void
4970 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4971 {
4972 	ztest_od_t *od;
4973 	int batchsize;
4974 	int size;
4975 	int b;
4976 
4977 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
4978 	od = umem_alloc(size, UMEM_NOFAIL);
4979 	batchsize = OD_ARRAY_SIZE;
4980 
4981 	for (b = 0; b < batchsize; b++)
4982 		ztest_od_init(od + b, id, FTAG, b, DMU_OT_UINT64_OTHER,
4983 		    0, 0, 0);
4984 
4985 	/*
4986 	 * Destroy the previous batch of objects, create a new batch,
4987 	 * and do some I/O on the new objects.
4988 	 */
4989 	if (ztest_object_init(zd, od, size, B_TRUE) != 0) {
4990 		zd->zd_od = NULL;
4991 		umem_free(od, size);
4992 		return;
4993 	}
4994 
4995 	while (ztest_random(4 * batchsize) != 0)
4996 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
4997 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4998 
4999 	umem_free(od, size);
5000 }
5001 
5002 /*
5003  * Rewind the global allocator to verify object allocation backfilling.
5004  */
5005 void
5006 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
5007 {
5008 	(void) id;
5009 	objset_t *os = zd->zd_os;
5010 	uint_t dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
5011 	uint64_t object;
5012 
5013 	/*
5014 	 * Rewind the global allocator randomly back to a lower object number
5015 	 * to force backfilling and reclamation of recently freed dnodes.
5016 	 */
5017 	mutex_enter(&os->os_obj_lock);
5018 	object = ztest_random(os->os_obj_next_chunk);
5019 	os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
5020 	mutex_exit(&os->os_obj_lock);
5021 }
5022 
5023 #undef OD_ARRAY_SIZE
5024 #define	OD_ARRAY_SIZE	2
5025 
5026 /*
5027  * Verify that dmu_{read,write} work as expected.
5028  */
5029 void
5030 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
5031 {
5032 	int size;
5033 	ztest_od_t *od;
5034 
5035 	objset_t *os = zd->zd_os;
5036 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
5037 	od = umem_alloc(size, UMEM_NOFAIL);
5038 	dmu_tx_t *tx;
5039 	int freeit, error;
5040 	uint64_t i, n, s, txg;
5041 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
5042 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
5043 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
5044 	uint64_t regions = 997;
5045 	uint64_t stride = 123456789ULL;
5046 	uint64_t width = 40;
5047 	int free_percent = 5;
5048 
5049 	/*
5050 	 * This test uses two objects, packobj and bigobj, that are always
5051 	 * updated together (i.e. in the same tx) so that their contents are
5052 	 * in sync and can be compared.  Their contents relate to each other
5053 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
5054 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
5055 	 * for any index n, there are three bufwads that should be identical:
5056 	 *
5057 	 *	packobj, at offset n * sizeof (bufwad_t)
5058 	 *	bigobj, at the head of the nth chunk
5059 	 *	bigobj, at the tail of the nth chunk
5060 	 *
5061 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
5062 	 * and it doesn't have any relation to the object blocksize.
5063 	 * The only requirement is that it can hold at least two bufwads.
5064 	 *
5065 	 * Normally, we write the bufwad to each of these locations.
5066 	 * However, free_percent of the time we instead write zeroes to
5067 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
5068 	 * bigobj to packobj, we can verify that the DMU is correctly
5069 	 * tracking which parts of an object are allocated and free,
5070 	 * and that the contents of the allocated blocks are correct.
5071 	 */
5072 
5073 	/*
5074 	 * Read the directory info.  If it's the first time, set things up.
5075 	 */
5076 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, chunksize);
5077 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
5078 	    chunksize);
5079 
5080 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
5081 		umem_free(od, size);
5082 		return;
5083 	}
5084 
5085 	bigobj = od[0].od_object;
5086 	packobj = od[1].od_object;
5087 	chunksize = od[0].od_gen;
5088 	ASSERT3U(chunksize, ==, od[1].od_gen);
5089 
5090 	/*
5091 	 * Prefetch a random chunk of the big object.
5092 	 * Our aim here is to get some async reads in flight
5093 	 * for blocks that we may free below; the DMU should
5094 	 * handle this race correctly.
5095 	 */
5096 	n = ztest_random(regions) * stride + ztest_random(width);
5097 	s = 1 + ztest_random(2 * width - 1);
5098 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
5099 	    ZIO_PRIORITY_SYNC_READ);
5100 
5101 	/*
5102 	 * Pick a random index and compute the offsets into packobj and bigobj.
5103 	 */
5104 	n = ztest_random(regions) * stride + ztest_random(width);
5105 	s = 1 + ztest_random(width - 1);
5106 
5107 	packoff = n * sizeof (bufwad_t);
5108 	packsize = s * sizeof (bufwad_t);
5109 
5110 	bigoff = n * chunksize;
5111 	bigsize = s * chunksize;
5112 
5113 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
5114 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
5115 
5116 	/*
5117 	 * free_percent of the time, free a range of bigobj rather than
5118 	 * overwriting it.
5119 	 */
5120 	freeit = (ztest_random(100) < free_percent);
5121 
5122 	/*
5123 	 * Read the current contents of our objects.
5124 	 */
5125 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
5126 	    DMU_READ_PREFETCH);
5127 	ASSERT0(error);
5128 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
5129 	    DMU_READ_PREFETCH);
5130 	ASSERT0(error);
5131 
5132 	/*
5133 	 * Get a tx for the mods to both packobj and bigobj.
5134 	 */
5135 	tx = dmu_tx_create(os);
5136 
5137 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
5138 
5139 	if (freeit)
5140 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
5141 	else
5142 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
5143 
5144 	/* This accounts for setting the checksum/compression. */
5145 	dmu_tx_hold_bonus(tx, bigobj);
5146 
5147 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5148 	if (txg == 0) {
5149 		umem_free(packbuf, packsize);
5150 		umem_free(bigbuf, bigsize);
5151 		umem_free(od, size);
5152 		return;
5153 	}
5154 
5155 	enum zio_checksum cksum;
5156 	do {
5157 		cksum = (enum zio_checksum)
5158 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
5159 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
5160 	dmu_object_set_checksum(os, bigobj, cksum, tx);
5161 
5162 	enum zio_compress comp;
5163 	do {
5164 		comp = (enum zio_compress)
5165 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
5166 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
5167 	dmu_object_set_compress(os, bigobj, comp, tx);
5168 
5169 	/*
5170 	 * For each index from n to n + s, verify that the existing bufwad
5171 	 * in packobj matches the bufwads at the head and tail of the
5172 	 * corresponding chunk in bigobj.  Then update all three bufwads
5173 	 * with the new values we want to write out.
5174 	 */
5175 	for (i = 0; i < s; i++) {
5176 		/* LINTED */
5177 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
5178 		/* LINTED */
5179 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
5180 		/* LINTED */
5181 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
5182 
5183 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
5184 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
5185 
5186 		if (pack->bw_txg > txg)
5187 			fatal(B_FALSE,
5188 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
5189 			    pack->bw_txg, txg);
5190 
5191 		if (pack->bw_data != 0 && pack->bw_index != n + i)
5192 			fatal(B_FALSE, "wrong index: "
5193 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
5194 			    pack->bw_index, n, i);
5195 
5196 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
5197 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
5198 			    pack, bigH);
5199 
5200 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
5201 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
5202 			    pack, bigT);
5203 
5204 		if (freeit) {
5205 			memset(pack, 0, sizeof (bufwad_t));
5206 		} else {
5207 			pack->bw_index = n + i;
5208 			pack->bw_txg = txg;
5209 			pack->bw_data = 1 + ztest_random(-2ULL);
5210 		}
5211 		*bigH = *pack;
5212 		*bigT = *pack;
5213 	}
5214 
5215 	/*
5216 	 * We've verified all the old bufwads, and made new ones.
5217 	 * Now write them out.
5218 	 */
5219 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
5220 
5221 	if (freeit) {
5222 		if (ztest_opts.zo_verbose >= 7) {
5223 			(void) printf("freeing offset %"PRIx64" size %"PRIx64""
5224 			    " txg %"PRIx64"\n",
5225 			    bigoff, bigsize, txg);
5226 		}
5227 		VERIFY0(dmu_free_range(os, bigobj, bigoff, bigsize, tx));
5228 	} else {
5229 		if (ztest_opts.zo_verbose >= 7) {
5230 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
5231 			    " txg %"PRIx64"\n",
5232 			    bigoff, bigsize, txg);
5233 		}
5234 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
5235 	}
5236 
5237 	dmu_tx_commit(tx);
5238 
5239 	/*
5240 	 * Sanity check the stuff we just wrote.
5241 	 */
5242 	{
5243 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
5244 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
5245 
5246 		VERIFY0(dmu_read(os, packobj, packoff,
5247 		    packsize, packcheck, DMU_READ_PREFETCH));
5248 		VERIFY0(dmu_read(os, bigobj, bigoff,
5249 		    bigsize, bigcheck, DMU_READ_PREFETCH));
5250 
5251 		ASSERT0(memcmp(packbuf, packcheck, packsize));
5252 		ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
5253 
5254 		umem_free(packcheck, packsize);
5255 		umem_free(bigcheck, bigsize);
5256 	}
5257 
5258 	umem_free(packbuf, packsize);
5259 	umem_free(bigbuf, bigsize);
5260 	umem_free(od, size);
5261 }
5262 
5263 static void
5264 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
5265     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
5266 {
5267 	uint64_t i;
5268 	bufwad_t *pack;
5269 	bufwad_t *bigH;
5270 	bufwad_t *bigT;
5271 
5272 	/*
5273 	 * For each index from n to n + s, verify that the existing bufwad
5274 	 * in packobj matches the bufwads at the head and tail of the
5275 	 * corresponding chunk in bigobj.  Then update all three bufwads
5276 	 * with the new values we want to write out.
5277 	 */
5278 	for (i = 0; i < s; i++) {
5279 		/* LINTED */
5280 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
5281 		/* LINTED */
5282 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
5283 		/* LINTED */
5284 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
5285 
5286 		ASSERT3U((uintptr_t)bigH - (uintptr_t)bigbuf, <, bigsize);
5287 		ASSERT3U((uintptr_t)bigT - (uintptr_t)bigbuf, <, bigsize);
5288 
5289 		if (pack->bw_txg > txg)
5290 			fatal(B_FALSE,
5291 			    "future leak: got %"PRIx64", open txg is %"PRIx64"",
5292 			    pack->bw_txg, txg);
5293 
5294 		if (pack->bw_data != 0 && pack->bw_index != n + i)
5295 			fatal(B_FALSE, "wrong index: "
5296 			    "got %"PRIx64", wanted %"PRIx64"+%"PRIx64"",
5297 			    pack->bw_index, n, i);
5298 
5299 		if (memcmp(pack, bigH, sizeof (bufwad_t)) != 0)
5300 			fatal(B_FALSE, "pack/bigH mismatch in %p/%p",
5301 			    pack, bigH);
5302 
5303 		if (memcmp(pack, bigT, sizeof (bufwad_t)) != 0)
5304 			fatal(B_FALSE, "pack/bigT mismatch in %p/%p",
5305 			    pack, bigT);
5306 
5307 		pack->bw_index = n + i;
5308 		pack->bw_txg = txg;
5309 		pack->bw_data = 1 + ztest_random(-2ULL);
5310 
5311 		*bigH = *pack;
5312 		*bigT = *pack;
5313 	}
5314 }
5315 
5316 #undef OD_ARRAY_SIZE
5317 #define	OD_ARRAY_SIZE	2
5318 
5319 void
5320 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
5321 {
5322 	objset_t *os = zd->zd_os;
5323 	ztest_od_t *od;
5324 	dmu_tx_t *tx;
5325 	uint64_t i;
5326 	int error;
5327 	int size;
5328 	uint64_t n, s, txg;
5329 	bufwad_t *packbuf, *bigbuf;
5330 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
5331 	uint64_t blocksize = ztest_random_blocksize();
5332 	uint64_t chunksize = blocksize;
5333 	uint64_t regions = 997;
5334 	uint64_t stride = 123456789ULL;
5335 	uint64_t width = 9;
5336 	dmu_buf_t *bonus_db;
5337 	arc_buf_t **bigbuf_arcbufs;
5338 	dmu_object_info_t doi;
5339 
5340 	size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
5341 	od = umem_alloc(size, UMEM_NOFAIL);
5342 
5343 	/*
5344 	 * This test uses two objects, packobj and bigobj, that are always
5345 	 * updated together (i.e. in the same tx) so that their contents are
5346 	 * in sync and can be compared.  Their contents relate to each other
5347 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
5348 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
5349 	 * for any index n, there are three bufwads that should be identical:
5350 	 *
5351 	 *	packobj, at offset n * sizeof (bufwad_t)
5352 	 *	bigobj, at the head of the nth chunk
5353 	 *	bigobj, at the tail of the nth chunk
5354 	 *
5355 	 * The chunk size is set equal to bigobj block size so that
5356 	 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
5357 	 */
5358 
5359 	/*
5360 	 * Read the directory info.  If it's the first time, set things up.
5361 	 */
5362 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5363 	ztest_od_init(od + 1, id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
5364 	    chunksize);
5365 
5366 
5367 	if (ztest_object_init(zd, od, size, B_FALSE) != 0) {
5368 		umem_free(od, size);
5369 		return;
5370 	}
5371 
5372 	bigobj = od[0].od_object;
5373 	packobj = od[1].od_object;
5374 	blocksize = od[0].od_blocksize;
5375 	chunksize = blocksize;
5376 	ASSERT3U(chunksize, ==, od[1].od_gen);
5377 
5378 	VERIFY0(dmu_object_info(os, bigobj, &doi));
5379 	VERIFY(ISP2(doi.doi_data_block_size));
5380 	VERIFY3U(chunksize, ==, doi.doi_data_block_size);
5381 	VERIFY3U(chunksize, >=, 2 * sizeof (bufwad_t));
5382 
5383 	/*
5384 	 * Pick a random index and compute the offsets into packobj and bigobj.
5385 	 */
5386 	n = ztest_random(regions) * stride + ztest_random(width);
5387 	s = 1 + ztest_random(width - 1);
5388 
5389 	packoff = n * sizeof (bufwad_t);
5390 	packsize = s * sizeof (bufwad_t);
5391 
5392 	bigoff = n * chunksize;
5393 	bigsize = s * chunksize;
5394 
5395 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
5396 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
5397 
5398 	VERIFY0(dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
5399 
5400 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
5401 
5402 	/*
5403 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
5404 	 * Iteration 1 test zcopy to already referenced dbufs.
5405 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
5406 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
5407 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
5408 	 * Iteration 5 test zcopy when it can't be done.
5409 	 * Iteration 6 one more zcopy write.
5410 	 */
5411 	for (i = 0; i < 7; i++) {
5412 		uint64_t j;
5413 		uint64_t off;
5414 
5415 		/*
5416 		 * In iteration 5 (i == 5) use arcbufs
5417 		 * that don't match bigobj blksz to test
5418 		 * dmu_assign_arcbuf_by_dbuf() when it can't directly
5419 		 * assign an arcbuf to a dbuf.
5420 		 */
5421 		for (j = 0; j < s; j++) {
5422 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5423 				bigbuf_arcbufs[j] =
5424 				    dmu_request_arcbuf(bonus_db, chunksize);
5425 			} else {
5426 				bigbuf_arcbufs[2 * j] =
5427 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5428 				bigbuf_arcbufs[2 * j + 1] =
5429 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
5430 			}
5431 		}
5432 
5433 		/*
5434 		 * Get a tx for the mods to both packobj and bigobj.
5435 		 */
5436 		tx = dmu_tx_create(os);
5437 
5438 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
5439 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
5440 
5441 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5442 		if (txg == 0) {
5443 			umem_free(packbuf, packsize);
5444 			umem_free(bigbuf, bigsize);
5445 			for (j = 0; j < s; j++) {
5446 				if (i != 5 ||
5447 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
5448 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
5449 				} else {
5450 					dmu_return_arcbuf(
5451 					    bigbuf_arcbufs[2 * j]);
5452 					dmu_return_arcbuf(
5453 					    bigbuf_arcbufs[2 * j + 1]);
5454 				}
5455 			}
5456 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5457 			umem_free(od, size);
5458 			dmu_buf_rele(bonus_db, FTAG);
5459 			return;
5460 		}
5461 
5462 		/*
5463 		 * 50% of the time don't read objects in the 1st iteration to
5464 		 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
5465 		 * no existing dbufs for the specified offsets.
5466 		 */
5467 		if (i != 0 || ztest_random(2) != 0) {
5468 			error = dmu_read(os, packobj, packoff,
5469 			    packsize, packbuf, DMU_READ_PREFETCH);
5470 			ASSERT0(error);
5471 			error = dmu_read(os, bigobj, bigoff, bigsize,
5472 			    bigbuf, DMU_READ_PREFETCH);
5473 			ASSERT0(error);
5474 		}
5475 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
5476 		    n, chunksize, txg);
5477 
5478 		/*
5479 		 * We've verified all the old bufwads, and made new ones.
5480 		 * Now write them out.
5481 		 */
5482 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
5483 		if (ztest_opts.zo_verbose >= 7) {
5484 			(void) printf("writing offset %"PRIx64" size %"PRIx64""
5485 			    " txg %"PRIx64"\n",
5486 			    bigoff, bigsize, txg);
5487 		}
5488 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
5489 			dmu_buf_t *dbt;
5490 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5491 				memcpy(bigbuf_arcbufs[j]->b_data,
5492 				    (caddr_t)bigbuf + (off - bigoff),
5493 				    chunksize);
5494 			} else {
5495 				memcpy(bigbuf_arcbufs[2 * j]->b_data,
5496 				    (caddr_t)bigbuf + (off - bigoff),
5497 				    chunksize / 2);
5498 				memcpy(bigbuf_arcbufs[2 * j + 1]->b_data,
5499 				    (caddr_t)bigbuf + (off - bigoff) +
5500 				    chunksize / 2,
5501 				    chunksize / 2);
5502 			}
5503 
5504 			if (i == 1) {
5505 				VERIFY(dmu_buf_hold(os, bigobj, off,
5506 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
5507 			}
5508 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
5509 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5510 				    off, bigbuf_arcbufs[j], tx));
5511 			} else {
5512 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5513 				    off, bigbuf_arcbufs[2 * j], tx));
5514 				VERIFY0(dmu_assign_arcbuf_by_dbuf(bonus_db,
5515 				    off + chunksize / 2,
5516 				    bigbuf_arcbufs[2 * j + 1], tx));
5517 			}
5518 			if (i == 1) {
5519 				dmu_buf_rele(dbt, FTAG);
5520 			}
5521 		}
5522 		dmu_tx_commit(tx);
5523 
5524 		/*
5525 		 * Sanity check the stuff we just wrote.
5526 		 */
5527 		{
5528 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
5529 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
5530 
5531 			VERIFY0(dmu_read(os, packobj, packoff,
5532 			    packsize, packcheck, DMU_READ_PREFETCH));
5533 			VERIFY0(dmu_read(os, bigobj, bigoff,
5534 			    bigsize, bigcheck, DMU_READ_PREFETCH));
5535 
5536 			ASSERT0(memcmp(packbuf, packcheck, packsize));
5537 			ASSERT0(memcmp(bigbuf, bigcheck, bigsize));
5538 
5539 			umem_free(packcheck, packsize);
5540 			umem_free(bigcheck, bigsize);
5541 		}
5542 		if (i == 2) {
5543 			txg_wait_open(dmu_objset_pool(os), 0, B_TRUE);
5544 		} else if (i == 3) {
5545 			txg_wait_synced(dmu_objset_pool(os), 0);
5546 		}
5547 	}
5548 
5549 	dmu_buf_rele(bonus_db, FTAG);
5550 	umem_free(packbuf, packsize);
5551 	umem_free(bigbuf, bigsize);
5552 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
5553 	umem_free(od, size);
5554 }
5555 
5556 void
5557 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
5558 {
5559 	(void) id;
5560 	ztest_od_t *od;
5561 
5562 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5563 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
5564 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5565 
5566 	/*
5567 	 * Have multiple threads write to large offsets in an object
5568 	 * to verify that parallel writes to an object -- even to the
5569 	 * same blocks within the object -- doesn't cause any trouble.
5570 	 */
5571 	ztest_od_init(od, ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5572 
5573 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0)
5574 		return;
5575 
5576 	while (ztest_random(10) != 0)
5577 		ztest_io(zd, od->od_object, offset);
5578 
5579 	umem_free(od, sizeof (ztest_od_t));
5580 }
5581 
5582 void
5583 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
5584 {
5585 	ztest_od_t *od;
5586 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
5587 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5588 	uint64_t count = ztest_random(20) + 1;
5589 	uint64_t blocksize = ztest_random_blocksize();
5590 	void *data;
5591 
5592 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5593 
5594 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0, 0);
5595 
5596 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5597 	    !ztest_random(2)) != 0) {
5598 		umem_free(od, sizeof (ztest_od_t));
5599 		return;
5600 	}
5601 
5602 	if (ztest_truncate(zd, od->od_object, offset, count * blocksize) != 0) {
5603 		umem_free(od, sizeof (ztest_od_t));
5604 		return;
5605 	}
5606 
5607 	ztest_prealloc(zd, od->od_object, offset, count * blocksize);
5608 
5609 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
5610 
5611 	while (ztest_random(count) != 0) {
5612 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
5613 		if (ztest_write(zd, od->od_object, randoff, blocksize,
5614 		    data) != 0)
5615 			break;
5616 		while (ztest_random(4) != 0)
5617 			ztest_io(zd, od->od_object, randoff);
5618 	}
5619 
5620 	umem_free(data, blocksize);
5621 	umem_free(od, sizeof (ztest_od_t));
5622 }
5623 
5624 /*
5625  * Verify that zap_{create,destroy,add,remove,update} work as expected.
5626  */
5627 #define	ZTEST_ZAP_MIN_INTS	1
5628 #define	ZTEST_ZAP_MAX_INTS	4
5629 #define	ZTEST_ZAP_MAX_PROPS	1000
5630 
5631 void
5632 ztest_zap(ztest_ds_t *zd, uint64_t id)
5633 {
5634 	objset_t *os = zd->zd_os;
5635 	ztest_od_t *od;
5636 	uint64_t object;
5637 	uint64_t txg, last_txg;
5638 	uint64_t value[ZTEST_ZAP_MAX_INTS];
5639 	uint64_t zl_ints, zl_intsize, prop;
5640 	int i, ints;
5641 	dmu_tx_t *tx;
5642 	char propname[100], txgname[100];
5643 	int error;
5644 	const char *const hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
5645 
5646 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5647 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5648 
5649 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5650 	    !ztest_random(2)) != 0)
5651 		goto out;
5652 
5653 	object = od->od_object;
5654 
5655 	/*
5656 	 * Generate a known hash collision, and verify that
5657 	 * we can lookup and remove both entries.
5658 	 */
5659 	tx = dmu_tx_create(os);
5660 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5661 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5662 	if (txg == 0)
5663 		goto out;
5664 	for (i = 0; i < 2; i++) {
5665 		value[i] = i;
5666 		VERIFY0(zap_add(os, object, hc[i], sizeof (uint64_t),
5667 		    1, &value[i], tx));
5668 	}
5669 	for (i = 0; i < 2; i++) {
5670 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
5671 		    sizeof (uint64_t), 1, &value[i], tx));
5672 		VERIFY0(
5673 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
5674 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5675 		ASSERT3U(zl_ints, ==, 1);
5676 	}
5677 	for (i = 0; i < 2; i++) {
5678 		VERIFY0(zap_remove(os, object, hc[i], tx));
5679 	}
5680 	dmu_tx_commit(tx);
5681 
5682 	/*
5683 	 * Generate a bunch of random entries.
5684 	 */
5685 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
5686 
5687 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5688 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5689 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5690 	memset(value, 0, sizeof (value));
5691 	last_txg = 0;
5692 
5693 	/*
5694 	 * If these zap entries already exist, validate their contents.
5695 	 */
5696 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5697 	if (error == 0) {
5698 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5699 		ASSERT3U(zl_ints, ==, 1);
5700 
5701 		VERIFY0(zap_lookup(os, object, txgname, zl_intsize,
5702 		    zl_ints, &last_txg));
5703 
5704 		VERIFY0(zap_length(os, object, propname, &zl_intsize,
5705 		    &zl_ints));
5706 
5707 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
5708 		ASSERT3U(zl_ints, ==, ints);
5709 
5710 		VERIFY0(zap_lookup(os, object, propname, zl_intsize,
5711 		    zl_ints, value));
5712 
5713 		for (i = 0; i < ints; i++) {
5714 			ASSERT3U(value[i], ==, last_txg + object + i);
5715 		}
5716 	} else {
5717 		ASSERT3U(error, ==, ENOENT);
5718 	}
5719 
5720 	/*
5721 	 * Atomically update two entries in our zap object.
5722 	 * The first is named txg_%llu, and contains the txg
5723 	 * in which the property was last updated.  The second
5724 	 * is named prop_%llu, and the nth element of its value
5725 	 * should be txg + object + n.
5726 	 */
5727 	tx = dmu_tx_create(os);
5728 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5729 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5730 	if (txg == 0)
5731 		goto out;
5732 
5733 	if (last_txg > txg)
5734 		fatal(B_FALSE, "zap future leak: old %"PRIu64" new %"PRIu64"",
5735 		    last_txg, txg);
5736 
5737 	for (i = 0; i < ints; i++)
5738 		value[i] = txg + object + i;
5739 
5740 	VERIFY0(zap_update(os, object, txgname, sizeof (uint64_t),
5741 	    1, &txg, tx));
5742 	VERIFY0(zap_update(os, object, propname, sizeof (uint64_t),
5743 	    ints, value, tx));
5744 
5745 	dmu_tx_commit(tx);
5746 
5747 	/*
5748 	 * Remove a random pair of entries.
5749 	 */
5750 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
5751 	(void) sprintf(propname, "prop_%"PRIu64"", prop);
5752 	(void) sprintf(txgname, "txg_%"PRIu64"", prop);
5753 
5754 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
5755 
5756 	if (error == ENOENT)
5757 		goto out;
5758 
5759 	ASSERT0(error);
5760 
5761 	tx = dmu_tx_create(os);
5762 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5763 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5764 	if (txg == 0)
5765 		goto out;
5766 	VERIFY0(zap_remove(os, object, txgname, tx));
5767 	VERIFY0(zap_remove(os, object, propname, tx));
5768 	dmu_tx_commit(tx);
5769 out:
5770 	umem_free(od, sizeof (ztest_od_t));
5771 }
5772 
5773 /*
5774  * Test case to test the upgrading of a microzap to fatzap.
5775  */
5776 void
5777 ztest_fzap(ztest_ds_t *zd, uint64_t id)
5778 {
5779 	objset_t *os = zd->zd_os;
5780 	ztest_od_t *od;
5781 	uint64_t object, txg, value;
5782 
5783 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5784 	ztest_od_init(od, id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
5785 
5786 	if (ztest_object_init(zd, od, sizeof (ztest_od_t),
5787 	    !ztest_random(2)) != 0)
5788 		goto out;
5789 	object = od->od_object;
5790 
5791 	/*
5792 	 * Add entries to this ZAP and make sure it spills over
5793 	 * and gets upgraded to a fatzap. Also, since we are adding
5794 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
5795 	 */
5796 	for (value = 0; value < 2050; value++) {
5797 		char name[ZFS_MAX_DATASET_NAME_LEN];
5798 		dmu_tx_t *tx;
5799 		int error;
5800 
5801 		(void) snprintf(name, sizeof (name), "fzap-%"PRIu64"-%"PRIu64"",
5802 		    id, value);
5803 
5804 		tx = dmu_tx_create(os);
5805 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
5806 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5807 		if (txg == 0)
5808 			goto out;
5809 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
5810 		    &value, tx);
5811 		ASSERT(error == 0 || error == EEXIST);
5812 		dmu_tx_commit(tx);
5813 	}
5814 out:
5815 	umem_free(od, sizeof (ztest_od_t));
5816 }
5817 
5818 void
5819 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
5820 {
5821 	(void) id;
5822 	objset_t *os = zd->zd_os;
5823 	ztest_od_t *od;
5824 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
5825 	dmu_tx_t *tx;
5826 	int i, namelen, error;
5827 	int micro = ztest_random(2);
5828 	char name[20], string_value[20];
5829 	void *data;
5830 
5831 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
5832 	ztest_od_init(od, ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0, 0);
5833 
5834 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
5835 		umem_free(od, sizeof (ztest_od_t));
5836 		return;
5837 	}
5838 
5839 	object = od->od_object;
5840 
5841 	/*
5842 	 * Generate a random name of the form 'xxx.....' where each
5843 	 * x is a random printable character and the dots are dots.
5844 	 * There are 94 such characters, and the name length goes from
5845 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
5846 	 */
5847 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
5848 
5849 	for (i = 0; i < 3; i++)
5850 		name[i] = '!' + ztest_random('~' - '!' + 1);
5851 	for (; i < namelen - 1; i++)
5852 		name[i] = '.';
5853 	name[i] = '\0';
5854 
5855 	if ((namelen & 1) || micro) {
5856 		wsize = sizeof (txg);
5857 		wc = 1;
5858 		data = &txg;
5859 	} else {
5860 		wsize = 1;
5861 		wc = namelen;
5862 		data = string_value;
5863 	}
5864 
5865 	count = -1ULL;
5866 	VERIFY0(zap_count(os, object, &count));
5867 	ASSERT3S(count, !=, -1ULL);
5868 
5869 	/*
5870 	 * Select an operation: length, lookup, add, update, remove.
5871 	 */
5872 	i = ztest_random(5);
5873 
5874 	if (i >= 2) {
5875 		tx = dmu_tx_create(os);
5876 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
5877 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
5878 		if (txg == 0) {
5879 			umem_free(od, sizeof (ztest_od_t));
5880 			return;
5881 		}
5882 		memcpy(string_value, name, namelen);
5883 	} else {
5884 		tx = NULL;
5885 		txg = 0;
5886 		memset(string_value, 0, namelen);
5887 	}
5888 
5889 	switch (i) {
5890 
5891 	case 0:
5892 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
5893 		if (error == 0) {
5894 			ASSERT3U(wsize, ==, zl_wsize);
5895 			ASSERT3U(wc, ==, zl_wc);
5896 		} else {
5897 			ASSERT3U(error, ==, ENOENT);
5898 		}
5899 		break;
5900 
5901 	case 1:
5902 		error = zap_lookup(os, object, name, wsize, wc, data);
5903 		if (error == 0) {
5904 			if (data == string_value &&
5905 			    memcmp(name, data, namelen) != 0)
5906 				fatal(B_FALSE, "name '%s' != val '%s' len %d",
5907 				    name, (char *)data, namelen);
5908 		} else {
5909 			ASSERT3U(error, ==, ENOENT);
5910 		}
5911 		break;
5912 
5913 	case 2:
5914 		error = zap_add(os, object, name, wsize, wc, data, tx);
5915 		ASSERT(error == 0 || error == EEXIST);
5916 		break;
5917 
5918 	case 3:
5919 		VERIFY0(zap_update(os, object, name, wsize, wc, data, tx));
5920 		break;
5921 
5922 	case 4:
5923 		error = zap_remove(os, object, name, tx);
5924 		ASSERT(error == 0 || error == ENOENT);
5925 		break;
5926 	}
5927 
5928 	if (tx != NULL)
5929 		dmu_tx_commit(tx);
5930 
5931 	umem_free(od, sizeof (ztest_od_t));
5932 }
5933 
5934 /*
5935  * Commit callback data.
5936  */
5937 typedef struct ztest_cb_data {
5938 	list_node_t		zcd_node;
5939 	uint64_t		zcd_txg;
5940 	int			zcd_expected_err;
5941 	boolean_t		zcd_added;
5942 	boolean_t		zcd_called;
5943 	spa_t			*zcd_spa;
5944 } ztest_cb_data_t;
5945 
5946 /* This is the actual commit callback function */
5947 static void
5948 ztest_commit_callback(void *arg, int error)
5949 {
5950 	ztest_cb_data_t *data = arg;
5951 	uint64_t synced_txg;
5952 
5953 	VERIFY3P(data, !=, NULL);
5954 	VERIFY3S(data->zcd_expected_err, ==, error);
5955 	VERIFY(!data->zcd_called);
5956 
5957 	synced_txg = spa_last_synced_txg(data->zcd_spa);
5958 	if (data->zcd_txg > synced_txg)
5959 		fatal(B_FALSE,
5960 		    "commit callback of txg %"PRIu64" called prematurely, "
5961 		    "last synced txg = %"PRIu64"\n",
5962 		    data->zcd_txg, synced_txg);
5963 
5964 	data->zcd_called = B_TRUE;
5965 
5966 	if (error == ECANCELED) {
5967 		ASSERT0(data->zcd_txg);
5968 		ASSERT(!data->zcd_added);
5969 
5970 		/*
5971 		 * The private callback data should be destroyed here, but
5972 		 * since we are going to check the zcd_called field after
5973 		 * dmu_tx_abort(), we will destroy it there.
5974 		 */
5975 		return;
5976 	}
5977 
5978 	ASSERT(data->zcd_added);
5979 	ASSERT3U(data->zcd_txg, !=, 0);
5980 
5981 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
5982 
5983 	/* See if this cb was called more quickly */
5984 	if ((synced_txg - data->zcd_txg) < zc_min_txg_delay)
5985 		zc_min_txg_delay = synced_txg - data->zcd_txg;
5986 
5987 	/* Remove our callback from the list */
5988 	list_remove(&zcl.zcl_callbacks, data);
5989 
5990 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
5991 
5992 	umem_free(data, sizeof (ztest_cb_data_t));
5993 }
5994 
5995 /* Allocate and initialize callback data structure */
5996 static ztest_cb_data_t *
5997 ztest_create_cb_data(objset_t *os, uint64_t txg)
5998 {
5999 	ztest_cb_data_t *cb_data;
6000 
6001 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
6002 
6003 	cb_data->zcd_txg = txg;
6004 	cb_data->zcd_spa = dmu_objset_spa(os);
6005 	list_link_init(&cb_data->zcd_node);
6006 
6007 	return (cb_data);
6008 }
6009 
6010 /*
6011  * Commit callback test.
6012  */
6013 void
6014 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
6015 {
6016 	objset_t *os = zd->zd_os;
6017 	ztest_od_t *od;
6018 	dmu_tx_t *tx;
6019 	ztest_cb_data_t *cb_data[3], *tmp_cb;
6020 	uint64_t old_txg, txg;
6021 	int i, error = 0;
6022 
6023 	od = umem_alloc(sizeof (ztest_od_t), UMEM_NOFAIL);
6024 	ztest_od_init(od, id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6025 
6026 	if (ztest_object_init(zd, od, sizeof (ztest_od_t), B_FALSE) != 0) {
6027 		umem_free(od, sizeof (ztest_od_t));
6028 		return;
6029 	}
6030 
6031 	tx = dmu_tx_create(os);
6032 
6033 	cb_data[0] = ztest_create_cb_data(os, 0);
6034 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
6035 
6036 	dmu_tx_hold_write(tx, od->od_object, 0, sizeof (uint64_t));
6037 
6038 	/* Every once in a while, abort the transaction on purpose */
6039 	if (ztest_random(100) == 0)
6040 		error = -1;
6041 
6042 	if (!error)
6043 		error = dmu_tx_assign(tx, TXG_NOWAIT);
6044 
6045 	txg = error ? 0 : dmu_tx_get_txg(tx);
6046 
6047 	cb_data[0]->zcd_txg = txg;
6048 	cb_data[1] = ztest_create_cb_data(os, txg);
6049 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
6050 
6051 	if (error) {
6052 		/*
6053 		 * It's not a strict requirement to call the registered
6054 		 * callbacks from inside dmu_tx_abort(), but that's what
6055 		 * it's supposed to happen in the current implementation
6056 		 * so we will check for that.
6057 		 */
6058 		for (i = 0; i < 2; i++) {
6059 			cb_data[i]->zcd_expected_err = ECANCELED;
6060 			VERIFY(!cb_data[i]->zcd_called);
6061 		}
6062 
6063 		dmu_tx_abort(tx);
6064 
6065 		for (i = 0; i < 2; i++) {
6066 			VERIFY(cb_data[i]->zcd_called);
6067 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
6068 		}
6069 
6070 		umem_free(od, sizeof (ztest_od_t));
6071 		return;
6072 	}
6073 
6074 	cb_data[2] = ztest_create_cb_data(os, txg);
6075 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
6076 
6077 	/*
6078 	 * Read existing data to make sure there isn't a future leak.
6079 	 */
6080 	VERIFY0(dmu_read(os, od->od_object, 0, sizeof (uint64_t),
6081 	    &old_txg, DMU_READ_PREFETCH));
6082 
6083 	if (old_txg > txg)
6084 		fatal(B_FALSE,
6085 		    "future leak: got %"PRIu64", open txg is %"PRIu64"",
6086 		    old_txg, txg);
6087 
6088 	dmu_write(os, od->od_object, 0, sizeof (uint64_t), &txg, tx);
6089 
6090 	(void) mutex_enter(&zcl.zcl_callbacks_lock);
6091 
6092 	/*
6093 	 * Since commit callbacks don't have any ordering requirement and since
6094 	 * it is theoretically possible for a commit callback to be called
6095 	 * after an arbitrary amount of time has elapsed since its txg has been
6096 	 * synced, it is difficult to reliably determine whether a commit
6097 	 * callback hasn't been called due to high load or due to a flawed
6098 	 * implementation.
6099 	 *
6100 	 * In practice, we will assume that if after a certain number of txgs a
6101 	 * commit callback hasn't been called, then most likely there's an
6102 	 * implementation bug..
6103 	 */
6104 	tmp_cb = list_head(&zcl.zcl_callbacks);
6105 	if (tmp_cb != NULL &&
6106 	    tmp_cb->zcd_txg + ZTEST_COMMIT_CB_THRESH < txg) {
6107 		fatal(B_FALSE,
6108 		    "Commit callback threshold exceeded, "
6109 		    "oldest txg: %"PRIu64", open txg: %"PRIu64"\n",
6110 		    tmp_cb->zcd_txg, txg);
6111 	}
6112 
6113 	/*
6114 	 * Let's find the place to insert our callbacks.
6115 	 *
6116 	 * Even though the list is ordered by txg, it is possible for the
6117 	 * insertion point to not be the end because our txg may already be
6118 	 * quiescing at this point and other callbacks in the open txg
6119 	 * (from other objsets) may have sneaked in.
6120 	 */
6121 	tmp_cb = list_tail(&zcl.zcl_callbacks);
6122 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
6123 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
6124 
6125 	/* Add the 3 callbacks to the list */
6126 	for (i = 0; i < 3; i++) {
6127 		if (tmp_cb == NULL)
6128 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
6129 		else
6130 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
6131 			    cb_data[i]);
6132 
6133 		cb_data[i]->zcd_added = B_TRUE;
6134 		VERIFY(!cb_data[i]->zcd_called);
6135 
6136 		tmp_cb = cb_data[i];
6137 	}
6138 
6139 	zc_cb_counter += 3;
6140 
6141 	(void) mutex_exit(&zcl.zcl_callbacks_lock);
6142 
6143 	dmu_tx_commit(tx);
6144 
6145 	umem_free(od, sizeof (ztest_od_t));
6146 }
6147 
6148 /*
6149  * Visit each object in the dataset. Verify that its properties
6150  * are consistent what was stored in the block tag when it was created,
6151  * and that its unused bonus buffer space has not been overwritten.
6152  */
6153 void
6154 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
6155 {
6156 	(void) id;
6157 	objset_t *os = zd->zd_os;
6158 	uint64_t obj;
6159 	int err = 0;
6160 
6161 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
6162 		ztest_block_tag_t *bt = NULL;
6163 		dmu_object_info_t doi;
6164 		dmu_buf_t *db;
6165 
6166 		ztest_object_lock(zd, obj, ZTRL_READER);
6167 		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0) {
6168 			ztest_object_unlock(zd, obj);
6169 			continue;
6170 		}
6171 
6172 		dmu_object_info_from_db(db, &doi);
6173 		if (doi.doi_bonus_size >= sizeof (*bt))
6174 			bt = ztest_bt_bonus(db);
6175 
6176 		if (bt && bt->bt_magic == BT_MAGIC) {
6177 			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
6178 			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
6179 			    bt->bt_crtxg);
6180 			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
6181 		}
6182 
6183 		dmu_buf_rele(db, FTAG);
6184 		ztest_object_unlock(zd, obj);
6185 	}
6186 }
6187 
6188 void
6189 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
6190 {
6191 	(void) id;
6192 	zfs_prop_t proplist[] = {
6193 		ZFS_PROP_CHECKSUM,
6194 		ZFS_PROP_COMPRESSION,
6195 		ZFS_PROP_COPIES,
6196 		ZFS_PROP_DEDUP
6197 	};
6198 
6199 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6200 
6201 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++) {
6202 		int error = ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
6203 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
6204 		ASSERT(error == 0 || error == ENOSPC);
6205 	}
6206 
6207 	int error = ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_RECORDSIZE,
6208 	    ztest_random_blocksize(), (int)ztest_random(2));
6209 	ASSERT(error == 0 || error == ENOSPC);
6210 
6211 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6212 }
6213 
6214 void
6215 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
6216 {
6217 	(void) zd, (void) id;
6218 	nvlist_t *props = NULL;
6219 
6220 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6221 
6222 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_AUTOTRIM, ztest_random(2));
6223 
6224 	VERIFY0(spa_prop_get(ztest_spa, &props));
6225 
6226 	if (ztest_opts.zo_verbose >= 6)
6227 		dump_nvlist(props, 4);
6228 
6229 	fnvlist_free(props);
6230 
6231 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6232 }
6233 
6234 static int
6235 user_release_one(const char *snapname, const char *holdname)
6236 {
6237 	nvlist_t *snaps, *holds;
6238 	int error;
6239 
6240 	snaps = fnvlist_alloc();
6241 	holds = fnvlist_alloc();
6242 	fnvlist_add_boolean(holds, holdname);
6243 	fnvlist_add_nvlist(snaps, snapname, holds);
6244 	fnvlist_free(holds);
6245 	error = dsl_dataset_user_release(snaps, NULL);
6246 	fnvlist_free(snaps);
6247 	return (error);
6248 }
6249 
6250 /*
6251  * Test snapshot hold/release and deferred destroy.
6252  */
6253 void
6254 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
6255 {
6256 	int error;
6257 	objset_t *os = zd->zd_os;
6258 	objset_t *origin;
6259 	char snapname[100];
6260 	char fullname[100];
6261 	char clonename[100];
6262 	char tag[100];
6263 	char osname[ZFS_MAX_DATASET_NAME_LEN];
6264 	nvlist_t *holds;
6265 
6266 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6267 
6268 	dmu_objset_name(os, osname);
6269 
6270 	(void) snprintf(snapname, sizeof (snapname), "sh1_%"PRIu64"", id);
6271 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
6272 	(void) snprintf(clonename, sizeof (clonename), "%s/ch1_%"PRIu64"",
6273 	    osname, id);
6274 	(void) snprintf(tag, sizeof (tag), "tag_%"PRIu64"", id);
6275 
6276 	/*
6277 	 * Clean up from any previous run.
6278 	 */
6279 	error = dsl_destroy_head(clonename);
6280 	if (error != ENOENT)
6281 		ASSERT0(error);
6282 	error = user_release_one(fullname, tag);
6283 	if (error != ESRCH && error != ENOENT)
6284 		ASSERT0(error);
6285 	error = dsl_destroy_snapshot(fullname, B_FALSE);
6286 	if (error != ENOENT)
6287 		ASSERT0(error);
6288 
6289 	/*
6290 	 * Create snapshot, clone it, mark snap for deferred destroy,
6291 	 * destroy clone, verify snap was also destroyed.
6292 	 */
6293 	error = dmu_objset_snapshot_one(osname, snapname);
6294 	if (error) {
6295 		if (error == ENOSPC) {
6296 			ztest_record_enospc("dmu_objset_snapshot");
6297 			goto out;
6298 		}
6299 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
6300 	}
6301 
6302 	error = dmu_objset_clone(clonename, fullname);
6303 	if (error) {
6304 		if (error == ENOSPC) {
6305 			ztest_record_enospc("dmu_objset_clone");
6306 			goto out;
6307 		}
6308 		fatal(B_FALSE, "dmu_objset_clone(%s) = %d", clonename, error);
6309 	}
6310 
6311 	error = dsl_destroy_snapshot(fullname, B_TRUE);
6312 	if (error) {
6313 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
6314 		    fullname, error);
6315 	}
6316 
6317 	error = dsl_destroy_head(clonename);
6318 	if (error)
6319 		fatal(B_FALSE, "dsl_destroy_head(%s) = %d", clonename, error);
6320 
6321 	error = dmu_objset_hold(fullname, FTAG, &origin);
6322 	if (error != ENOENT)
6323 		fatal(B_FALSE, "dmu_objset_hold(%s) = %d", fullname, error);
6324 
6325 	/*
6326 	 * Create snapshot, add temporary hold, verify that we can't
6327 	 * destroy a held snapshot, mark for deferred destroy,
6328 	 * release hold, verify snapshot was destroyed.
6329 	 */
6330 	error = dmu_objset_snapshot_one(osname, snapname);
6331 	if (error) {
6332 		if (error == ENOSPC) {
6333 			ztest_record_enospc("dmu_objset_snapshot");
6334 			goto out;
6335 		}
6336 		fatal(B_FALSE, "dmu_objset_snapshot(%s) = %d", fullname, error);
6337 	}
6338 
6339 	holds = fnvlist_alloc();
6340 	fnvlist_add_string(holds, fullname, tag);
6341 	error = dsl_dataset_user_hold(holds, 0, NULL);
6342 	fnvlist_free(holds);
6343 
6344 	if (error == ENOSPC) {
6345 		ztest_record_enospc("dsl_dataset_user_hold");
6346 		goto out;
6347 	} else if (error) {
6348 		fatal(B_FALSE, "dsl_dataset_user_hold(%s, %s) = %u",
6349 		    fullname, tag, error);
6350 	}
6351 
6352 	error = dsl_destroy_snapshot(fullname, B_FALSE);
6353 	if (error != EBUSY) {
6354 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
6355 		    fullname, error);
6356 	}
6357 
6358 	error = dsl_destroy_snapshot(fullname, B_TRUE);
6359 	if (error) {
6360 		fatal(B_FALSE, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
6361 		    fullname, error);
6362 	}
6363 
6364 	error = user_release_one(fullname, tag);
6365 	if (error)
6366 		fatal(B_FALSE, "user_release_one(%s, %s) = %d",
6367 		    fullname, tag, error);
6368 
6369 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
6370 
6371 out:
6372 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6373 }
6374 
6375 /*
6376  * Inject random faults into the on-disk data.
6377  */
6378 void
6379 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
6380 {
6381 	(void) zd, (void) id;
6382 	ztest_shared_t *zs = ztest_shared;
6383 	spa_t *spa = ztest_spa;
6384 	int fd;
6385 	uint64_t offset;
6386 	uint64_t leaves;
6387 	uint64_t bad = 0x1990c0ffeedecadeull;
6388 	uint64_t top, leaf;
6389 	uint64_t raidz_children;
6390 	char *path0;
6391 	char *pathrand;
6392 	size_t fsize;
6393 	int bshift = SPA_MAXBLOCKSHIFT + 2;
6394 	int iters = 1000;
6395 	int maxfaults;
6396 	int mirror_save;
6397 	vdev_t *vd0 = NULL;
6398 	uint64_t guid0 = 0;
6399 	boolean_t islog = B_FALSE;
6400 	boolean_t injected = B_FALSE;
6401 
6402 	path0 = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6403 	pathrand = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6404 
6405 	mutex_enter(&ztest_vdev_lock);
6406 
6407 	/*
6408 	 * Device removal is in progress, fault injection must be disabled
6409 	 * until it completes and the pool is scrubbed.  The fault injection
6410 	 * strategy for damaging blocks does not take in to account evacuated
6411 	 * blocks which may have already been damaged.
6412 	 */
6413 	if (ztest_device_removal_active)
6414 		goto out;
6415 
6416 	/*
6417 	 * The fault injection strategy for damaging blocks cannot be used
6418 	 * if raidz expansion is in progress. The leaves value
6419 	 * (attached raidz children) is variable and strategy for damaging
6420 	 * blocks will corrupt same data blocks on different child vdevs
6421 	 * because of the reflow process.
6422 	 */
6423 	if (spa->spa_raidz_expand != NULL)
6424 		goto out;
6425 
6426 	maxfaults = MAXFAULTS(zs);
6427 	raidz_children = ztest_get_raidz_children(spa);
6428 	leaves = MAX(zs->zs_mirrors, 1) * raidz_children;
6429 	mirror_save = zs->zs_mirrors;
6430 
6431 	ASSERT3U(leaves, >=, 1);
6432 
6433 	/*
6434 	 * While ztest is running the number of leaves will not change.  This
6435 	 * is critical for the fault injection logic as it determines where
6436 	 * errors can be safely injected such that they are always repairable.
6437 	 *
6438 	 * When restarting ztest a different number of leaves may be requested
6439 	 * which will shift the regions to be damaged.  This is fine as long
6440 	 * as the pool has been scrubbed prior to using the new mapping.
6441 	 * Failure to do can result in non-repairable damage being injected.
6442 	 */
6443 	if (ztest_pool_scrubbed == B_FALSE)
6444 		goto out;
6445 
6446 	/*
6447 	 * Grab the name lock as reader. There are some operations
6448 	 * which don't like to have their vdevs changed while
6449 	 * they are in progress (i.e. spa_change_guid). Those
6450 	 * operations will have grabbed the name lock as writer.
6451 	 */
6452 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
6453 
6454 	/*
6455 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
6456 	 */
6457 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6458 
6459 	if (ztest_random(2) == 0) {
6460 		/*
6461 		 * Inject errors on a normal data device or slog device.
6462 		 */
6463 		top = ztest_random_vdev_top(spa, B_TRUE);
6464 		leaf = ztest_random(leaves) + zs->zs_splits;
6465 
6466 		/*
6467 		 * Generate paths to the first leaf in this top-level vdev,
6468 		 * and to the random leaf we selected.  We'll induce transient
6469 		 * write failures and random online/offline activity on leaf 0,
6470 		 * and we'll write random garbage to the randomly chosen leaf.
6471 		 */
6472 		(void) snprintf(path0, MAXPATHLEN, ztest_dev_template,
6473 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6474 		    top * leaves + zs->zs_splits);
6475 		(void) snprintf(pathrand, MAXPATHLEN, ztest_dev_template,
6476 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
6477 		    top * leaves + leaf);
6478 
6479 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
6480 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
6481 			islog = B_TRUE;
6482 
6483 		/*
6484 		 * If the top-level vdev needs to be resilvered
6485 		 * then we only allow faults on the device that is
6486 		 * resilvering.
6487 		 */
6488 		if (vd0 != NULL && maxfaults != 1 &&
6489 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
6490 		    vd0->vdev_resilver_txg != 0)) {
6491 			/*
6492 			 * Make vd0 explicitly claim to be unreadable,
6493 			 * or unwritable, or reach behind its back
6494 			 * and close the underlying fd.  We can do this if
6495 			 * maxfaults == 0 because we'll fail and reexecute,
6496 			 * and we can do it if maxfaults >= 2 because we'll
6497 			 * have enough redundancy.  If maxfaults == 1, the
6498 			 * combination of this with injection of random data
6499 			 * corruption below exceeds the pool's fault tolerance.
6500 			 */
6501 			vdev_file_t *vf = vd0->vdev_tsd;
6502 
6503 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
6504 			    (long long)vd0->vdev_id, (int)maxfaults);
6505 
6506 			if (vf != NULL && ztest_random(3) == 0) {
6507 				(void) close(vf->vf_file->f_fd);
6508 				vf->vf_file->f_fd = -1;
6509 			} else if (ztest_random(2) == 0) {
6510 				vd0->vdev_cant_read = B_TRUE;
6511 			} else {
6512 				vd0->vdev_cant_write = B_TRUE;
6513 			}
6514 			guid0 = vd0->vdev_guid;
6515 		}
6516 	} else {
6517 		/*
6518 		 * Inject errors on an l2cache device.
6519 		 */
6520 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
6521 
6522 		if (sav->sav_count == 0) {
6523 			spa_config_exit(spa, SCL_STATE, FTAG);
6524 			(void) pthread_rwlock_unlock(&ztest_name_lock);
6525 			goto out;
6526 		}
6527 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
6528 		guid0 = vd0->vdev_guid;
6529 		(void) strlcpy(path0, vd0->vdev_path, MAXPATHLEN);
6530 		(void) strlcpy(pathrand, vd0->vdev_path, MAXPATHLEN);
6531 
6532 		leaf = 0;
6533 		leaves = 1;
6534 		maxfaults = INT_MAX;	/* no limit on cache devices */
6535 	}
6536 
6537 	spa_config_exit(spa, SCL_STATE, FTAG);
6538 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6539 
6540 	/*
6541 	 * If we can tolerate two or more faults, or we're dealing
6542 	 * with a slog, randomly online/offline vd0.
6543 	 */
6544 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
6545 		if (ztest_random(10) < 6) {
6546 			int flags = (ztest_random(2) == 0 ?
6547 			    ZFS_OFFLINE_TEMPORARY : 0);
6548 
6549 			/*
6550 			 * We have to grab the zs_name_lock as writer to
6551 			 * prevent a race between offlining a slog and
6552 			 * destroying a dataset. Offlining the slog will
6553 			 * grab a reference on the dataset which may cause
6554 			 * dsl_destroy_head() to fail with EBUSY thus
6555 			 * leaving the dataset in an inconsistent state.
6556 			 */
6557 			if (islog)
6558 				(void) pthread_rwlock_wrlock(&ztest_name_lock);
6559 
6560 			VERIFY3U(vdev_offline(spa, guid0, flags), !=, EBUSY);
6561 
6562 			if (islog)
6563 				(void) pthread_rwlock_unlock(&ztest_name_lock);
6564 		} else {
6565 			/*
6566 			 * Ideally we would like to be able to randomly
6567 			 * call vdev_[on|off]line without holding locks
6568 			 * to force unpredictable failures but the side
6569 			 * effects of vdev_[on|off]line prevent us from
6570 			 * doing so.
6571 			 */
6572 			(void) vdev_online(spa, guid0, 0, NULL);
6573 		}
6574 	}
6575 
6576 	if (maxfaults == 0)
6577 		goto out;
6578 
6579 	/*
6580 	 * We have at least single-fault tolerance, so inject data corruption.
6581 	 */
6582 	fd = open(pathrand, O_RDWR);
6583 
6584 	if (fd == -1) /* we hit a gap in the device namespace */
6585 		goto out;
6586 
6587 	fsize = lseek(fd, 0, SEEK_END);
6588 
6589 	while (--iters != 0) {
6590 		/*
6591 		 * The offset must be chosen carefully to ensure that
6592 		 * we do not inject a given logical block with errors
6593 		 * on two different leaf devices, because ZFS can not
6594 		 * tolerate that (if maxfaults==1).
6595 		 *
6596 		 * To achieve this we divide each leaf device into
6597 		 * chunks of size (# leaves * SPA_MAXBLOCKSIZE * 4).
6598 		 * Each chunk is further divided into error-injection
6599 		 * ranges (can accept errors) and clear ranges (we do
6600 		 * not inject errors in those). Each error-injection
6601 		 * range can accept errors only for a single leaf vdev.
6602 		 * Error-injection ranges are separated by clear ranges.
6603 		 *
6604 		 * For example, with 3 leaves, each chunk looks like:
6605 		 *    0 to  32M: injection range for leaf 0
6606 		 *  32M to  64M: clear range - no injection allowed
6607 		 *  64M to  96M: injection range for leaf 1
6608 		 *  96M to 128M: clear range - no injection allowed
6609 		 * 128M to 160M: injection range for leaf 2
6610 		 * 160M to 192M: clear range - no injection allowed
6611 		 *
6612 		 * Each clear range must be large enough such that a
6613 		 * single block cannot straddle it. This way a block
6614 		 * can't be a target in two different injection ranges
6615 		 * (on different leaf vdevs).
6616 		 */
6617 		offset = ztest_random(fsize / (leaves << bshift)) *
6618 		    (leaves << bshift) + (leaf << bshift) +
6619 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
6620 
6621 		/*
6622 		 * Only allow damage to the labels at one end of the vdev.
6623 		 *
6624 		 * If all labels are damaged, the device will be totally
6625 		 * inaccessible, which will result in loss of data,
6626 		 * because we also damage (parts of) the other side of
6627 		 * the mirror/raidz.
6628 		 *
6629 		 * Additionally, we will always have both an even and an
6630 		 * odd label, so that we can handle crashes in the
6631 		 * middle of vdev_config_sync().
6632 		 */
6633 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
6634 			continue;
6635 
6636 		/*
6637 		 * The two end labels are stored at the "end" of the disk, but
6638 		 * the end of the disk (vdev_psize) is aligned to
6639 		 * sizeof (vdev_label_t).
6640 		 */
6641 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
6642 		if ((leaf & 1) == 1 &&
6643 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
6644 			continue;
6645 
6646 		if (mirror_save != zs->zs_mirrors) {
6647 			(void) close(fd);
6648 			goto out;
6649 		}
6650 
6651 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
6652 			fatal(B_TRUE,
6653 			    "can't inject bad word at 0x%"PRIx64" in %s",
6654 			    offset, pathrand);
6655 
6656 		if (ztest_opts.zo_verbose >= 7)
6657 			(void) printf("injected bad word into %s,"
6658 			    " offset 0x%"PRIx64"\n", pathrand, offset);
6659 
6660 		injected = B_TRUE;
6661 	}
6662 
6663 	(void) close(fd);
6664 out:
6665 	mutex_exit(&ztest_vdev_lock);
6666 
6667 	if (injected && ztest_opts.zo_raid_do_expand) {
6668 		int error = spa_scan(spa, POOL_SCAN_SCRUB);
6669 		if (error == 0) {
6670 			while (dsl_scan_scrubbing(spa_get_dsl(spa)))
6671 				txg_wait_synced(spa_get_dsl(spa), 0);
6672 		}
6673 	}
6674 
6675 	umem_free(path0, MAXPATHLEN);
6676 	umem_free(pathrand, MAXPATHLEN);
6677 }
6678 
6679 /*
6680  * By design ztest will never inject uncorrectable damage in to the pool.
6681  * Issue a scrub, wait for it to complete, and verify there is never any
6682  * persistent damage.
6683  *
6684  * Only after a full scrub has been completed is it safe to start injecting
6685  * data corruption.  See the comment in zfs_fault_inject().
6686  */
6687 static int
6688 ztest_scrub_impl(spa_t *spa)
6689 {
6690 	int error = spa_scan(spa, POOL_SCAN_SCRUB);
6691 	if (error)
6692 		return (error);
6693 
6694 	while (dsl_scan_scrubbing(spa_get_dsl(spa)))
6695 		txg_wait_synced(spa_get_dsl(spa), 0);
6696 
6697 	if (spa_approx_errlog_size(spa) > 0)
6698 		return (ECKSUM);
6699 
6700 	ztest_pool_scrubbed = B_TRUE;
6701 
6702 	return (0);
6703 }
6704 
6705 /*
6706  * Scrub the pool.
6707  */
6708 void
6709 ztest_scrub(ztest_ds_t *zd, uint64_t id)
6710 {
6711 	(void) zd, (void) id;
6712 	spa_t *spa = ztest_spa;
6713 	int error;
6714 
6715 	/*
6716 	 * Scrub in progress by device removal.
6717 	 */
6718 	if (ztest_device_removal_active)
6719 		return;
6720 
6721 	/*
6722 	 * Start a scrub, wait a moment, then force a restart.
6723 	 */
6724 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
6725 	(void) poll(NULL, 0, 100);
6726 
6727 	error = ztest_scrub_impl(spa);
6728 	if (error == EBUSY)
6729 		error = 0;
6730 	ASSERT0(error);
6731 }
6732 
6733 /*
6734  * Change the guid for the pool.
6735  */
6736 void
6737 ztest_reguid(ztest_ds_t *zd, uint64_t id)
6738 {
6739 	(void) zd, (void) id;
6740 	spa_t *spa = ztest_spa;
6741 	uint64_t orig, load;
6742 	int error;
6743 	ztest_shared_t *zs = ztest_shared;
6744 
6745 	if (ztest_opts.zo_mmp_test)
6746 		return;
6747 
6748 	orig = spa_guid(spa);
6749 	load = spa_load_guid(spa);
6750 
6751 	(void) pthread_rwlock_wrlock(&ztest_name_lock);
6752 	error = spa_change_guid(spa);
6753 	zs->zs_guid = spa_guid(spa);
6754 	(void) pthread_rwlock_unlock(&ztest_name_lock);
6755 
6756 	if (error != 0)
6757 		return;
6758 
6759 	if (ztest_opts.zo_verbose >= 4) {
6760 		(void) printf("Changed guid old %"PRIu64" -> %"PRIu64"\n",
6761 		    orig, spa_guid(spa));
6762 	}
6763 
6764 	VERIFY3U(orig, !=, spa_guid(spa));
6765 	VERIFY3U(load, ==, spa_load_guid(spa));
6766 }
6767 
6768 void
6769 ztest_blake3(ztest_ds_t *zd, uint64_t id)
6770 {
6771 	(void) zd, (void) id;
6772 	hrtime_t end = gethrtime() + NANOSEC;
6773 	zio_cksum_salt_t salt;
6774 	void *salt_ptr = &salt.zcs_bytes;
6775 	struct abd *abd_data, *abd_meta;
6776 	void *buf, *templ;
6777 	int i, *ptr;
6778 	uint32_t size;
6779 	BLAKE3_CTX ctx;
6780 	const zfs_impl_t *blake3 = zfs_impl_get_ops("blake3");
6781 
6782 	size = ztest_random_blocksize();
6783 	buf = umem_alloc(size, UMEM_NOFAIL);
6784 	abd_data = abd_alloc(size, B_FALSE);
6785 	abd_meta = abd_alloc(size, B_TRUE);
6786 
6787 	for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6788 		*ptr = ztest_random(UINT_MAX);
6789 	memset(salt_ptr, 'A', 32);
6790 
6791 	abd_copy_from_buf_off(abd_data, buf, 0, size);
6792 	abd_copy_from_buf_off(abd_meta, buf, 0, size);
6793 
6794 	while (gethrtime() <= end) {
6795 		int run_count = 100;
6796 		zio_cksum_t zc_ref1, zc_ref2;
6797 		zio_cksum_t zc_res1, zc_res2;
6798 
6799 		void *ref1 = &zc_ref1;
6800 		void *ref2 = &zc_ref2;
6801 		void *res1 = &zc_res1;
6802 		void *res2 = &zc_res2;
6803 
6804 		/* BLAKE3_KEY_LEN = 32 */
6805 		VERIFY0(blake3->setname("generic"));
6806 		templ = abd_checksum_blake3_tmpl_init(&salt);
6807 		Blake3_InitKeyed(&ctx, salt_ptr);
6808 		Blake3_Update(&ctx, buf, size);
6809 		Blake3_Final(&ctx, ref1);
6810 		zc_ref2 = zc_ref1;
6811 		ZIO_CHECKSUM_BSWAP(&zc_ref2);
6812 		abd_checksum_blake3_tmpl_free(templ);
6813 
6814 		VERIFY0(blake3->setname("cycle"));
6815 		while (run_count-- > 0) {
6816 
6817 			/* Test current implementation */
6818 			Blake3_InitKeyed(&ctx, salt_ptr);
6819 			Blake3_Update(&ctx, buf, size);
6820 			Blake3_Final(&ctx, res1);
6821 			zc_res2 = zc_res1;
6822 			ZIO_CHECKSUM_BSWAP(&zc_res2);
6823 
6824 			VERIFY0(memcmp(ref1, res1, 32));
6825 			VERIFY0(memcmp(ref2, res2, 32));
6826 
6827 			/* Test ABD - data */
6828 			templ = abd_checksum_blake3_tmpl_init(&salt);
6829 			abd_checksum_blake3_native(abd_data, size,
6830 			    templ, &zc_res1);
6831 			abd_checksum_blake3_byteswap(abd_data, size,
6832 			    templ, &zc_res2);
6833 
6834 			VERIFY0(memcmp(ref1, res1, 32));
6835 			VERIFY0(memcmp(ref2, res2, 32));
6836 
6837 			/* Test ABD - metadata */
6838 			abd_checksum_blake3_native(abd_meta, size,
6839 			    templ, &zc_res1);
6840 			abd_checksum_blake3_byteswap(abd_meta, size,
6841 			    templ, &zc_res2);
6842 			abd_checksum_blake3_tmpl_free(templ);
6843 
6844 			VERIFY0(memcmp(ref1, res1, 32));
6845 			VERIFY0(memcmp(ref2, res2, 32));
6846 
6847 		}
6848 	}
6849 
6850 	abd_free(abd_data);
6851 	abd_free(abd_meta);
6852 	umem_free(buf, size);
6853 }
6854 
6855 void
6856 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
6857 {
6858 	(void) zd, (void) id;
6859 	hrtime_t end = gethrtime() + NANOSEC;
6860 
6861 	while (gethrtime() <= end) {
6862 		int run_count = 100;
6863 		void *buf;
6864 		struct abd *abd_data, *abd_meta;
6865 		uint32_t size;
6866 		int *ptr;
6867 		int i;
6868 		zio_cksum_t zc_ref;
6869 		zio_cksum_t zc_ref_byteswap;
6870 
6871 		size = ztest_random_blocksize();
6872 
6873 		buf = umem_alloc(size, UMEM_NOFAIL);
6874 		abd_data = abd_alloc(size, B_FALSE);
6875 		abd_meta = abd_alloc(size, B_TRUE);
6876 
6877 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6878 			*ptr = ztest_random(UINT_MAX);
6879 
6880 		abd_copy_from_buf_off(abd_data, buf, 0, size);
6881 		abd_copy_from_buf_off(abd_meta, buf, 0, size);
6882 
6883 		VERIFY0(fletcher_4_impl_set("scalar"));
6884 		fletcher_4_native(buf, size, NULL, &zc_ref);
6885 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
6886 
6887 		VERIFY0(fletcher_4_impl_set("cycle"));
6888 		while (run_count-- > 0) {
6889 			zio_cksum_t zc;
6890 			zio_cksum_t zc_byteswap;
6891 
6892 			fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
6893 			fletcher_4_native(buf, size, NULL, &zc);
6894 
6895 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6896 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6897 			    sizeof (zc_byteswap)));
6898 
6899 			/* Test ABD - data */
6900 			abd_fletcher_4_byteswap(abd_data, size, NULL,
6901 			    &zc_byteswap);
6902 			abd_fletcher_4_native(abd_data, size, NULL, &zc);
6903 
6904 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6905 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6906 			    sizeof (zc_byteswap)));
6907 
6908 			/* Test ABD - metadata */
6909 			abd_fletcher_4_byteswap(abd_meta, size, NULL,
6910 			    &zc_byteswap);
6911 			abd_fletcher_4_native(abd_meta, size, NULL, &zc);
6912 
6913 			VERIFY0(memcmp(&zc, &zc_ref, sizeof (zc)));
6914 			VERIFY0(memcmp(&zc_byteswap, &zc_ref_byteswap,
6915 			    sizeof (zc_byteswap)));
6916 
6917 		}
6918 
6919 		umem_free(buf, size);
6920 		abd_free(abd_data);
6921 		abd_free(abd_meta);
6922 	}
6923 }
6924 
6925 void
6926 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
6927 {
6928 	(void) zd, (void) id;
6929 	void *buf;
6930 	size_t size;
6931 	int *ptr;
6932 	int i;
6933 	zio_cksum_t zc_ref;
6934 	zio_cksum_t zc_ref_bswap;
6935 
6936 	hrtime_t end = gethrtime() + NANOSEC;
6937 
6938 	while (gethrtime() <= end) {
6939 		int run_count = 100;
6940 
6941 		size = ztest_random_blocksize();
6942 		buf = umem_alloc(size, UMEM_NOFAIL);
6943 
6944 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
6945 			*ptr = ztest_random(UINT_MAX);
6946 
6947 		VERIFY0(fletcher_4_impl_set("scalar"));
6948 		fletcher_4_native(buf, size, NULL, &zc_ref);
6949 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
6950 
6951 		VERIFY0(fletcher_4_impl_set("cycle"));
6952 
6953 		while (run_count-- > 0) {
6954 			zio_cksum_t zc;
6955 			zio_cksum_t zc_bswap;
6956 			size_t pos = 0;
6957 
6958 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6959 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6960 
6961 			while (pos < size) {
6962 				size_t inc = 64 * ztest_random(size / 67);
6963 				/* sometimes add few bytes to test non-simd */
6964 				if (ztest_random(100) < 10)
6965 					inc += P2ALIGN(ztest_random(64),
6966 					    sizeof (uint32_t));
6967 
6968 				if (inc > (size - pos))
6969 					inc = size - pos;
6970 
6971 				fletcher_4_incremental_native(buf + pos, inc,
6972 				    &zc);
6973 				fletcher_4_incremental_byteswap(buf + pos, inc,
6974 				    &zc_bswap);
6975 
6976 				pos += inc;
6977 			}
6978 
6979 			VERIFY3U(pos, ==, size);
6980 
6981 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6982 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6983 
6984 			/*
6985 			 * verify if incremental on the whole buffer is
6986 			 * equivalent to non-incremental version
6987 			 */
6988 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
6989 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
6990 
6991 			fletcher_4_incremental_native(buf, size, &zc);
6992 			fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
6993 
6994 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
6995 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
6996 		}
6997 
6998 		umem_free(buf, size);
6999 	}
7000 }
7001 
7002 static int
7003 ztest_set_global_vars(void)
7004 {
7005 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
7006 		char *kv = ztest_opts.zo_gvars[i];
7007 		VERIFY3U(strlen(kv), <=, ZO_GVARS_MAX_ARGLEN);
7008 		VERIFY3U(strlen(kv), >, 0);
7009 		int err = set_global_var(kv);
7010 		if (ztest_opts.zo_verbose > 0) {
7011 			(void) printf("setting global var %s ... %s\n", kv,
7012 			    err ? "failed" : "ok");
7013 		}
7014 		if (err != 0) {
7015 			(void) fprintf(stderr,
7016 			    "failed to set global var '%s'\n", kv);
7017 			return (err);
7018 		}
7019 	}
7020 	return (0);
7021 }
7022 
7023 static char **
7024 ztest_global_vars_to_zdb_args(void)
7025 {
7026 	char **args = calloc(2*ztest_opts.zo_gvars_count + 1, sizeof (char *));
7027 	char **cur = args;
7028 	if (args == NULL)
7029 		return (NULL);
7030 	for (size_t i = 0; i < ztest_opts.zo_gvars_count; i++) {
7031 		*cur++ = (char *)"-o";
7032 		*cur++ = ztest_opts.zo_gvars[i];
7033 	}
7034 	ASSERT3P(cur, ==, &args[2*ztest_opts.zo_gvars_count]);
7035 	*cur = NULL;
7036 	return (args);
7037 }
7038 
7039 /* The end of strings is indicated by a NULL element */
7040 static char *
7041 join_strings(char **strings, const char *sep)
7042 {
7043 	size_t totallen = 0;
7044 	for (char **sp = strings; *sp != NULL; sp++) {
7045 		totallen += strlen(*sp);
7046 		totallen += strlen(sep);
7047 	}
7048 	if (totallen > 0) {
7049 		ASSERT(totallen >= strlen(sep));
7050 		totallen -= strlen(sep);
7051 	}
7052 
7053 	size_t buflen = totallen + 1;
7054 	char *o = umem_alloc(buflen, UMEM_NOFAIL); /* trailing 0 byte */
7055 	o[0] = '\0';
7056 	for (char **sp = strings; *sp != NULL; sp++) {
7057 		size_t would;
7058 		would = strlcat(o, *sp, buflen);
7059 		VERIFY3U(would, <, buflen);
7060 		if (*(sp+1) == NULL) {
7061 			break;
7062 		}
7063 		would = strlcat(o, sep, buflen);
7064 		VERIFY3U(would, <, buflen);
7065 	}
7066 	ASSERT3S(strlen(o), ==, totallen);
7067 	return (o);
7068 }
7069 
7070 static int
7071 ztest_check_path(char *path)
7072 {
7073 	struct stat s;
7074 	/* return true on success */
7075 	return (!stat(path, &s));
7076 }
7077 
7078 static void
7079 ztest_get_zdb_bin(char *bin, int len)
7080 {
7081 	char *zdb_path;
7082 	/*
7083 	 * Try to use $ZDB and in-tree zdb path. If not successful, just
7084 	 * let popen to search through PATH.
7085 	 */
7086 	if ((zdb_path = getenv("ZDB"))) {
7087 		strlcpy(bin, zdb_path, len); /* In env */
7088 		if (!ztest_check_path(bin)) {
7089 			ztest_dump_core = 0;
7090 			fatal(B_TRUE, "invalid ZDB '%s'", bin);
7091 		}
7092 		return;
7093 	}
7094 
7095 	VERIFY3P(realpath(getexecname(), bin), !=, NULL);
7096 	if (strstr(bin, ".libs/ztest")) {
7097 		strstr(bin, ".libs/ztest")[0] = '\0'; /* In-tree */
7098 		strcat(bin, "zdb");
7099 		if (ztest_check_path(bin))
7100 			return;
7101 	}
7102 	strcpy(bin, "zdb");
7103 }
7104 
7105 static vdev_t *
7106 ztest_random_concrete_vdev_leaf(vdev_t *vd)
7107 {
7108 	if (vd == NULL)
7109 		return (NULL);
7110 
7111 	if (vd->vdev_children == 0)
7112 		return (vd);
7113 
7114 	vdev_t *eligible[vd->vdev_children];
7115 	int eligible_idx = 0, i;
7116 	for (i = 0; i < vd->vdev_children; i++) {
7117 		vdev_t *cvd = vd->vdev_child[i];
7118 		if (cvd->vdev_top->vdev_removing)
7119 			continue;
7120 		if (cvd->vdev_children > 0 ||
7121 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
7122 			eligible[eligible_idx++] = cvd;
7123 		}
7124 	}
7125 	VERIFY3S(eligible_idx, >, 0);
7126 
7127 	uint64_t child_no = ztest_random(eligible_idx);
7128 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
7129 }
7130 
7131 void
7132 ztest_initialize(ztest_ds_t *zd, uint64_t id)
7133 {
7134 	(void) zd, (void) id;
7135 	spa_t *spa = ztest_spa;
7136 	int error = 0;
7137 
7138 	mutex_enter(&ztest_vdev_lock);
7139 
7140 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
7141 
7142 	/* Random leaf vdev */
7143 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
7144 	if (rand_vd == NULL) {
7145 		spa_config_exit(spa, SCL_VDEV, FTAG);
7146 		mutex_exit(&ztest_vdev_lock);
7147 		return;
7148 	}
7149 
7150 	/*
7151 	 * The random vdev we've selected may change as soon as we
7152 	 * drop the spa_config_lock. We create local copies of things
7153 	 * we're interested in.
7154 	 */
7155 	uint64_t guid = rand_vd->vdev_guid;
7156 	char *path = strdup(rand_vd->vdev_path);
7157 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
7158 
7159 	zfs_dbgmsg("vd %px, guid %llu", rand_vd, (u_longlong_t)guid);
7160 	spa_config_exit(spa, SCL_VDEV, FTAG);
7161 
7162 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
7163 
7164 	nvlist_t *vdev_guids = fnvlist_alloc();
7165 	nvlist_t *vdev_errlist = fnvlist_alloc();
7166 	fnvlist_add_uint64(vdev_guids, path, guid);
7167 	error = spa_vdev_initialize(spa, vdev_guids, cmd, vdev_errlist);
7168 	fnvlist_free(vdev_guids);
7169 	fnvlist_free(vdev_errlist);
7170 
7171 	switch (cmd) {
7172 	case POOL_INITIALIZE_CANCEL:
7173 		if (ztest_opts.zo_verbose >= 4) {
7174 			(void) printf("Cancel initialize %s", path);
7175 			if (!active)
7176 				(void) printf(" failed (no initialize active)");
7177 			(void) printf("\n");
7178 		}
7179 		break;
7180 	case POOL_INITIALIZE_START:
7181 		if (ztest_opts.zo_verbose >= 4) {
7182 			(void) printf("Start initialize %s", path);
7183 			if (active && error == 0)
7184 				(void) printf(" failed (already active)");
7185 			else if (error != 0)
7186 				(void) printf(" failed (error %d)", error);
7187 			(void) printf("\n");
7188 		}
7189 		break;
7190 	case POOL_INITIALIZE_SUSPEND:
7191 		if (ztest_opts.zo_verbose >= 4) {
7192 			(void) printf("Suspend initialize %s", path);
7193 			if (!active)
7194 				(void) printf(" failed (no initialize active)");
7195 			(void) printf("\n");
7196 		}
7197 		break;
7198 	}
7199 	free(path);
7200 	mutex_exit(&ztest_vdev_lock);
7201 }
7202 
7203 void
7204 ztest_trim(ztest_ds_t *zd, uint64_t id)
7205 {
7206 	(void) zd, (void) id;
7207 	spa_t *spa = ztest_spa;
7208 	int error = 0;
7209 
7210 	mutex_enter(&ztest_vdev_lock);
7211 
7212 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
7213 
7214 	/* Random leaf vdev */
7215 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
7216 	if (rand_vd == NULL) {
7217 		spa_config_exit(spa, SCL_VDEV, FTAG);
7218 		mutex_exit(&ztest_vdev_lock);
7219 		return;
7220 	}
7221 
7222 	/*
7223 	 * The random vdev we've selected may change as soon as we
7224 	 * drop the spa_config_lock. We create local copies of things
7225 	 * we're interested in.
7226 	 */
7227 	uint64_t guid = rand_vd->vdev_guid;
7228 	char *path = strdup(rand_vd->vdev_path);
7229 	boolean_t active = rand_vd->vdev_trim_thread != NULL;
7230 
7231 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, (u_longlong_t)guid);
7232 	spa_config_exit(spa, SCL_VDEV, FTAG);
7233 
7234 	uint64_t cmd = ztest_random(POOL_TRIM_FUNCS);
7235 	uint64_t rate = 1 << ztest_random(30);
7236 	boolean_t partial = (ztest_random(5) > 0);
7237 	boolean_t secure = (ztest_random(5) > 0);
7238 
7239 	nvlist_t *vdev_guids = fnvlist_alloc();
7240 	nvlist_t *vdev_errlist = fnvlist_alloc();
7241 	fnvlist_add_uint64(vdev_guids, path, guid);
7242 	error = spa_vdev_trim(spa, vdev_guids, cmd, rate, partial,
7243 	    secure, vdev_errlist);
7244 	fnvlist_free(vdev_guids);
7245 	fnvlist_free(vdev_errlist);
7246 
7247 	switch (cmd) {
7248 	case POOL_TRIM_CANCEL:
7249 		if (ztest_opts.zo_verbose >= 4) {
7250 			(void) printf("Cancel TRIM %s", path);
7251 			if (!active)
7252 				(void) printf(" failed (no TRIM active)");
7253 			(void) printf("\n");
7254 		}
7255 		break;
7256 	case POOL_TRIM_START:
7257 		if (ztest_opts.zo_verbose >= 4) {
7258 			(void) printf("Start TRIM %s", path);
7259 			if (active && error == 0)
7260 				(void) printf(" failed (already active)");
7261 			else if (error != 0)
7262 				(void) printf(" failed (error %d)", error);
7263 			(void) printf("\n");
7264 		}
7265 		break;
7266 	case POOL_TRIM_SUSPEND:
7267 		if (ztest_opts.zo_verbose >= 4) {
7268 			(void) printf("Suspend TRIM %s", path);
7269 			if (!active)
7270 				(void) printf(" failed (no TRIM active)");
7271 			(void) printf("\n");
7272 		}
7273 		break;
7274 	}
7275 	free(path);
7276 	mutex_exit(&ztest_vdev_lock);
7277 }
7278 
7279 /*
7280  * Verify pool integrity by running zdb.
7281  */
7282 static void
7283 ztest_run_zdb(uint64_t guid)
7284 {
7285 	int status;
7286 	char *bin;
7287 	char *zdb;
7288 	char *zbuf;
7289 	const int len = MAXPATHLEN + MAXNAMELEN + 20;
7290 	FILE *fp;
7291 
7292 	bin = umem_alloc(len, UMEM_NOFAIL);
7293 	zdb = umem_alloc(len, UMEM_NOFAIL);
7294 	zbuf = umem_alloc(1024, UMEM_NOFAIL);
7295 
7296 	ztest_get_zdb_bin(bin, len);
7297 
7298 	char **set_gvars_args = ztest_global_vars_to_zdb_args();
7299 	if (set_gvars_args == NULL) {
7300 		fatal(B_FALSE, "Failed to allocate memory in "
7301 		    "ztest_global_vars_to_zdb_args(). Cannot run zdb.\n");
7302 	}
7303 	char *set_gvars_args_joined = join_strings(set_gvars_args, " ");
7304 	free(set_gvars_args);
7305 
7306 	size_t would = snprintf(zdb, len,
7307 	    "%s -bcc%s%s -G -d -Y -e -y %s -p %s %"PRIu64,
7308 	    bin,
7309 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
7310 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
7311 	    set_gvars_args_joined,
7312 	    ztest_opts.zo_dir,
7313 	    guid);
7314 	ASSERT3U(would, <, len);
7315 
7316 	umem_free(set_gvars_args_joined, strlen(set_gvars_args_joined) + 1);
7317 
7318 	if (ztest_opts.zo_verbose >= 5)
7319 		(void) printf("Executing %s\n", zdb);
7320 
7321 	fp = popen(zdb, "r");
7322 
7323 	while (fgets(zbuf, 1024, fp) != NULL)
7324 		if (ztest_opts.zo_verbose >= 3)
7325 			(void) printf("%s", zbuf);
7326 
7327 	status = pclose(fp);
7328 
7329 	if (status == 0)
7330 		goto out;
7331 
7332 	ztest_dump_core = 0;
7333 	if (WIFEXITED(status))
7334 		fatal(B_FALSE, "'%s' exit code %d", zdb, WEXITSTATUS(status));
7335 	else
7336 		fatal(B_FALSE, "'%s' died with signal %d",
7337 		    zdb, WTERMSIG(status));
7338 out:
7339 	umem_free(bin, len);
7340 	umem_free(zdb, len);
7341 	umem_free(zbuf, 1024);
7342 }
7343 
7344 static void
7345 ztest_walk_pool_directory(const char *header)
7346 {
7347 	spa_t *spa = NULL;
7348 
7349 	if (ztest_opts.zo_verbose >= 6)
7350 		(void) puts(header);
7351 
7352 	mutex_enter(&spa_namespace_lock);
7353 	while ((spa = spa_next(spa)) != NULL)
7354 		if (ztest_opts.zo_verbose >= 6)
7355 			(void) printf("\t%s\n", spa_name(spa));
7356 	mutex_exit(&spa_namespace_lock);
7357 }
7358 
7359 static void
7360 ztest_spa_import_export(char *oldname, char *newname)
7361 {
7362 	nvlist_t *config, *newconfig;
7363 	uint64_t pool_guid;
7364 	spa_t *spa;
7365 	int error;
7366 
7367 	if (ztest_opts.zo_verbose >= 4) {
7368 		(void) printf("import/export: old = %s, new = %s\n",
7369 		    oldname, newname);
7370 	}
7371 
7372 	/*
7373 	 * Clean up from previous runs.
7374 	 */
7375 	(void) spa_destroy(newname);
7376 
7377 	/*
7378 	 * Get the pool's configuration and guid.
7379 	 */
7380 	VERIFY0(spa_open(oldname, &spa, FTAG));
7381 
7382 	/*
7383 	 * Kick off a scrub to tickle scrub/export races.
7384 	 */
7385 	if (ztest_random(2) == 0)
7386 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
7387 
7388 	pool_guid = spa_guid(spa);
7389 	spa_close(spa, FTAG);
7390 
7391 	ztest_walk_pool_directory("pools before export");
7392 
7393 	/*
7394 	 * Export it.
7395 	 */
7396 	VERIFY0(spa_export(oldname, &config, B_FALSE, B_FALSE));
7397 
7398 	ztest_walk_pool_directory("pools after export");
7399 
7400 	/*
7401 	 * Try to import it.
7402 	 */
7403 	newconfig = spa_tryimport(config);
7404 	ASSERT3P(newconfig, !=, NULL);
7405 	fnvlist_free(newconfig);
7406 
7407 	/*
7408 	 * Import it under the new name.
7409 	 */
7410 	error = spa_import(newname, config, NULL, 0);
7411 	if (error != 0) {
7412 		dump_nvlist(config, 0);
7413 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
7414 		    oldname, newname, error);
7415 	}
7416 
7417 	ztest_walk_pool_directory("pools after import");
7418 
7419 	/*
7420 	 * Try to import it again -- should fail with EEXIST.
7421 	 */
7422 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
7423 
7424 	/*
7425 	 * Try to import it under a different name -- should fail with EEXIST.
7426 	 */
7427 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
7428 
7429 	/*
7430 	 * Verify that the pool is no longer visible under the old name.
7431 	 */
7432 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
7433 
7434 	/*
7435 	 * Verify that we can open and close the pool using the new name.
7436 	 */
7437 	VERIFY0(spa_open(newname, &spa, FTAG));
7438 	ASSERT3U(pool_guid, ==, spa_guid(spa));
7439 	spa_close(spa, FTAG);
7440 
7441 	fnvlist_free(config);
7442 }
7443 
7444 static void
7445 ztest_resume(spa_t *spa)
7446 {
7447 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
7448 		(void) printf("resuming from suspended state\n");
7449 	spa_vdev_state_enter(spa, SCL_NONE);
7450 	vdev_clear(spa, NULL);
7451 	(void) spa_vdev_state_exit(spa, NULL, 0);
7452 	(void) zio_resume(spa);
7453 }
7454 
7455 static __attribute__((noreturn)) void
7456 ztest_resume_thread(void *arg)
7457 {
7458 	spa_t *spa = arg;
7459 
7460 	while (!ztest_exiting) {
7461 		if (spa_suspended(spa))
7462 			ztest_resume(spa);
7463 		(void) poll(NULL, 0, 100);
7464 
7465 		/*
7466 		 * Periodically change the zfs_compressed_arc_enabled setting.
7467 		 */
7468 		if (ztest_random(10) == 0)
7469 			zfs_compressed_arc_enabled = ztest_random(2);
7470 
7471 		/*
7472 		 * Periodically change the zfs_abd_scatter_enabled setting.
7473 		 */
7474 		if (ztest_random(10) == 0)
7475 			zfs_abd_scatter_enabled = ztest_random(2);
7476 	}
7477 
7478 	thread_exit();
7479 }
7480 
7481 static __attribute__((noreturn)) void
7482 ztest_deadman_thread(void *arg)
7483 {
7484 	ztest_shared_t *zs = arg;
7485 	spa_t *spa = ztest_spa;
7486 	hrtime_t delay, overdue, last_run = gethrtime();
7487 
7488 	delay = (zs->zs_thread_stop - zs->zs_thread_start) +
7489 	    MSEC2NSEC(zfs_deadman_synctime_ms);
7490 
7491 	while (!ztest_exiting) {
7492 		/*
7493 		 * Wait for the delay timer while checking occasionally
7494 		 * if we should stop.
7495 		 */
7496 		if (gethrtime() < last_run + delay) {
7497 			(void) poll(NULL, 0, 1000);
7498 			continue;
7499 		}
7500 
7501 		/*
7502 		 * If the pool is suspended then fail immediately. Otherwise,
7503 		 * check to see if the pool is making any progress. If
7504 		 * vdev_deadman() discovers that there hasn't been any recent
7505 		 * I/Os then it will end up aborting the tests.
7506 		 */
7507 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
7508 			fatal(B_FALSE,
7509 			    "aborting test after %llu seconds because "
7510 			    "pool has transitioned to a suspended state.",
7511 			    (u_longlong_t)zfs_deadman_synctime_ms / 1000);
7512 		}
7513 		vdev_deadman(spa->spa_root_vdev, FTAG);
7514 
7515 		/*
7516 		 * If the process doesn't complete within a grace period of
7517 		 * zfs_deadman_synctime_ms over the expected finish time,
7518 		 * then it may be hung and is terminated.
7519 		 */
7520 		overdue = zs->zs_proc_stop + MSEC2NSEC(zfs_deadman_synctime_ms);
7521 		if (gethrtime() > overdue) {
7522 			fatal(B_FALSE,
7523 			    "aborting test after %llu seconds because "
7524 			    "the process is overdue for termination.",
7525 			    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7526 		}
7527 
7528 		(void) printf("ztest has been running for %lld seconds\n",
7529 		    (gethrtime() - zs->zs_proc_start) / NANOSEC);
7530 
7531 		last_run = gethrtime();
7532 		delay = MSEC2NSEC(zfs_deadman_checktime_ms);
7533 	}
7534 
7535 	thread_exit();
7536 }
7537 
7538 static void
7539 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
7540 {
7541 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
7542 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
7543 	hrtime_t functime = gethrtime();
7544 	int i;
7545 
7546 	for (i = 0; i < zi->zi_iters; i++)
7547 		zi->zi_func(zd, id);
7548 
7549 	functime = gethrtime() - functime;
7550 
7551 	atomic_add_64(&zc->zc_count, 1);
7552 	atomic_add_64(&zc->zc_time, functime);
7553 
7554 	if (ztest_opts.zo_verbose >= 4)
7555 		(void) printf("%6.2f sec in %s\n",
7556 		    (double)functime / NANOSEC, zi->zi_funcname);
7557 }
7558 
7559 typedef struct ztest_raidz_expand_io {
7560 	uint64_t	rzx_id;
7561 	uint64_t	rzx_amount;
7562 	uint64_t	rzx_bufsize;
7563 	const void	*rzx_buffer;
7564 	uint64_t	rzx_alloc_max;
7565 	spa_t		*rzx_spa;
7566 } ztest_expand_io_t;
7567 
7568 #undef OD_ARRAY_SIZE
7569 #define	OD_ARRAY_SIZE	10
7570 
7571 /*
7572  * Write a request amount of data to some dataset objects.
7573  * There will be ztest_opts.zo_threads count of these running in parallel.
7574  */
7575 static __attribute__((noreturn)) void
7576 ztest_rzx_thread(void *arg)
7577 {
7578 	ztest_expand_io_t *info = (ztest_expand_io_t *)arg;
7579 	ztest_od_t *od;
7580 	int batchsize;
7581 	int od_size;
7582 	ztest_ds_t *zd = &ztest_ds[info->rzx_id % ztest_opts.zo_datasets];
7583 	spa_t *spa = info->rzx_spa;
7584 
7585 	od_size = sizeof (ztest_od_t) * OD_ARRAY_SIZE;
7586 	od = umem_alloc(od_size, UMEM_NOFAIL);
7587 	batchsize = OD_ARRAY_SIZE;
7588 
7589 	/* Create objects to write to */
7590 	for (int b = 0; b < batchsize; b++) {
7591 		ztest_od_init(od + b, info->rzx_id, FTAG, b,
7592 		    DMU_OT_UINT64_OTHER, 0, 0, 0);
7593 	}
7594 	if (ztest_object_init(zd, od, od_size, B_FALSE) != 0) {
7595 		umem_free(od, od_size);
7596 		thread_exit();
7597 	}
7598 
7599 	for (uint64_t offset = 0, written = 0; written < info->rzx_amount;
7600 	    offset += info->rzx_bufsize) {
7601 		/* write to 10 objects */
7602 		for (int i = 0; i < batchsize && written < info->rzx_amount;
7603 		    i++) {
7604 			(void) pthread_rwlock_rdlock(&zd->zd_zilog_lock);
7605 			ztest_write(zd, od[i].od_object, offset,
7606 			    info->rzx_bufsize, info->rzx_buffer);
7607 			(void) pthread_rwlock_unlock(&zd->zd_zilog_lock);
7608 			written += info->rzx_bufsize;
7609 		}
7610 		txg_wait_synced(spa_get_dsl(spa), 0);
7611 		/* due to inflation, we'll typically bail here */
7612 		if (metaslab_class_get_alloc(spa_normal_class(spa)) >
7613 		    info->rzx_alloc_max) {
7614 			break;
7615 		}
7616 	}
7617 
7618 	/* Remove a few objects to leave some holes in allocation space */
7619 	mutex_enter(&zd->zd_dirobj_lock);
7620 	(void) ztest_remove(zd, od, 2);
7621 	mutex_exit(&zd->zd_dirobj_lock);
7622 
7623 	umem_free(od, od_size);
7624 
7625 	thread_exit();
7626 }
7627 
7628 static __attribute__((noreturn)) void
7629 ztest_thread(void *arg)
7630 {
7631 	int rand;
7632 	uint64_t id = (uintptr_t)arg;
7633 	ztest_shared_t *zs = ztest_shared;
7634 	uint64_t call_next;
7635 	hrtime_t now;
7636 	ztest_info_t *zi;
7637 	ztest_shared_callstate_t *zc;
7638 
7639 	while ((now = gethrtime()) < zs->zs_thread_stop) {
7640 		/*
7641 		 * See if it's time to force a crash.
7642 		 */
7643 		if (now > zs->zs_thread_kill &&
7644 		    raidz_expand_pause_point == RAIDZ_EXPAND_PAUSE_NONE) {
7645 			ztest_kill(zs);
7646 		}
7647 
7648 		/*
7649 		 * If we're getting ENOSPC with some regularity, stop.
7650 		 */
7651 		if (zs->zs_enospc_count > 10)
7652 			break;
7653 
7654 		/*
7655 		 * Pick a random function to execute.
7656 		 */
7657 		rand = ztest_random(ZTEST_FUNCS);
7658 		zi = &ztest_info[rand];
7659 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
7660 		call_next = zc->zc_next;
7661 
7662 		if (now >= call_next &&
7663 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
7664 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
7665 			ztest_execute(rand, zi, id);
7666 		}
7667 	}
7668 
7669 	thread_exit();
7670 }
7671 
7672 static void
7673 ztest_dataset_name(char *dsname, const char *pool, int d)
7674 {
7675 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
7676 }
7677 
7678 static void
7679 ztest_dataset_destroy(int d)
7680 {
7681 	char name[ZFS_MAX_DATASET_NAME_LEN];
7682 	int t;
7683 
7684 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7685 
7686 	if (ztest_opts.zo_verbose >= 3)
7687 		(void) printf("Destroying %s to free up space\n", name);
7688 
7689 	/*
7690 	 * Cleanup any non-standard clones and snapshots.  In general,
7691 	 * ztest thread t operates on dataset (t % zopt_datasets),
7692 	 * so there may be more than one thing to clean up.
7693 	 */
7694 	for (t = d; t < ztest_opts.zo_threads;
7695 	    t += ztest_opts.zo_datasets)
7696 		ztest_dsl_dataset_cleanup(name, t);
7697 
7698 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
7699 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
7700 }
7701 
7702 static void
7703 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
7704 {
7705 	uint64_t usedobjs, dirobjs, scratch;
7706 
7707 	/*
7708 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
7709 	 * Therefore, the number of objects in use should equal the
7710 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
7711 	 * If not, we have an object leak.
7712 	 *
7713 	 * Note that we can only check this in ztest_dataset_open(),
7714 	 * when the open-context and syncing-context values agree.
7715 	 * That's because zap_count() returns the open-context value,
7716 	 * while dmu_objset_space() returns the rootbp fill count.
7717 	 */
7718 	VERIFY0(zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
7719 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
7720 	ASSERT3U(dirobjs + 1, ==, usedobjs);
7721 }
7722 
7723 static int
7724 ztest_dataset_open(int d)
7725 {
7726 	ztest_ds_t *zd = &ztest_ds[d];
7727 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
7728 	objset_t *os;
7729 	zilog_t *zilog;
7730 	char name[ZFS_MAX_DATASET_NAME_LEN];
7731 	int error;
7732 
7733 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
7734 
7735 	(void) pthread_rwlock_rdlock(&ztest_name_lock);
7736 
7737 	error = ztest_dataset_create(name);
7738 	if (error == ENOSPC) {
7739 		(void) pthread_rwlock_unlock(&ztest_name_lock);
7740 		ztest_record_enospc(FTAG);
7741 		return (error);
7742 	}
7743 	ASSERT(error == 0 || error == EEXIST);
7744 
7745 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
7746 	    B_TRUE, zd, &os));
7747 	(void) pthread_rwlock_unlock(&ztest_name_lock);
7748 
7749 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
7750 
7751 	zilog = zd->zd_zilog;
7752 
7753 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
7754 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
7755 		fatal(B_FALSE, "missing log records: "
7756 		    "claimed %"PRIu64" < committed %"PRIu64"",
7757 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
7758 
7759 	ztest_dataset_dirobj_verify(zd);
7760 
7761 	zil_replay(os, zd, ztest_replay_vector);
7762 
7763 	ztest_dataset_dirobj_verify(zd);
7764 
7765 	if (ztest_opts.zo_verbose >= 6)
7766 		(void) printf("%s replay %"PRIu64" blocks, "
7767 		    "%"PRIu64" records, seq %"PRIu64"\n",
7768 		    zd->zd_name,
7769 		    zilog->zl_parse_blk_count,
7770 		    zilog->zl_parse_lr_count,
7771 		    zilog->zl_replaying_seq);
7772 
7773 	zilog = zil_open(os, ztest_get_data, NULL);
7774 
7775 	if (zilog->zl_replaying_seq != 0 &&
7776 	    zilog->zl_replaying_seq < committed_seq)
7777 		fatal(B_FALSE, "missing log records: "
7778 		    "replayed %"PRIu64" < committed %"PRIu64"",
7779 		    zilog->zl_replaying_seq, committed_seq);
7780 
7781 	return (0);
7782 }
7783 
7784 static void
7785 ztest_dataset_close(int d)
7786 {
7787 	ztest_ds_t *zd = &ztest_ds[d];
7788 
7789 	zil_close(zd->zd_zilog);
7790 	dmu_objset_disown(zd->zd_os, B_TRUE, zd);
7791 
7792 	ztest_zd_fini(zd);
7793 }
7794 
7795 static int
7796 ztest_replay_zil_cb(const char *name, void *arg)
7797 {
7798 	(void) arg;
7799 	objset_t *os;
7800 	ztest_ds_t *zdtmp;
7801 
7802 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_ANY, B_TRUE,
7803 	    B_TRUE, FTAG, &os));
7804 
7805 	zdtmp = umem_alloc(sizeof (ztest_ds_t), UMEM_NOFAIL);
7806 
7807 	ztest_zd_init(zdtmp, NULL, os);
7808 	zil_replay(os, zdtmp, ztest_replay_vector);
7809 	ztest_zd_fini(zdtmp);
7810 
7811 	if (dmu_objset_zil(os)->zl_parse_lr_count != 0 &&
7812 	    ztest_opts.zo_verbose >= 6) {
7813 		zilog_t *zilog = dmu_objset_zil(os);
7814 
7815 		(void) printf("%s replay %"PRIu64" blocks, "
7816 		    "%"PRIu64" records, seq %"PRIu64"\n",
7817 		    name,
7818 		    zilog->zl_parse_blk_count,
7819 		    zilog->zl_parse_lr_count,
7820 		    zilog->zl_replaying_seq);
7821 	}
7822 
7823 	umem_free(zdtmp, sizeof (ztest_ds_t));
7824 
7825 	dmu_objset_disown(os, B_TRUE, FTAG);
7826 	return (0);
7827 }
7828 
7829 static void
7830 ztest_freeze(void)
7831 {
7832 	ztest_ds_t *zd = &ztest_ds[0];
7833 	spa_t *spa;
7834 	int numloops = 0;
7835 
7836 	/* freeze not supported during RAIDZ expansion */
7837 	if (ztest_opts.zo_raid_do_expand)
7838 		return;
7839 
7840 	if (ztest_opts.zo_verbose >= 3)
7841 		(void) printf("testing spa_freeze()...\n");
7842 
7843 	raidz_scratch_verify();
7844 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7845 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7846 	VERIFY0(ztest_dataset_open(0));
7847 	ztest_spa = spa;
7848 
7849 	/*
7850 	 * Force the first log block to be transactionally allocated.
7851 	 * We have to do this before we freeze the pool -- otherwise
7852 	 * the log chain won't be anchored.
7853 	 */
7854 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
7855 		ztest_dmu_object_alloc_free(zd, 0);
7856 		zil_commit(zd->zd_zilog, 0);
7857 	}
7858 
7859 	txg_wait_synced(spa_get_dsl(spa), 0);
7860 
7861 	/*
7862 	 * Freeze the pool.  This stops spa_sync() from doing anything,
7863 	 * so that the only way to record changes from now on is the ZIL.
7864 	 */
7865 	spa_freeze(spa);
7866 
7867 	/*
7868 	 * Because it is hard to predict how much space a write will actually
7869 	 * require beforehand, we leave ourselves some fudge space to write over
7870 	 * capacity.
7871 	 */
7872 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
7873 
7874 	/*
7875 	 * Run tests that generate log records but don't alter the pool config
7876 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
7877 	 * We do a txg_wait_synced() after each iteration to force the txg
7878 	 * to increase well beyond the last synced value in the uberblock.
7879 	 * The ZIL should be OK with that.
7880 	 *
7881 	 * Run a random number of times less than zo_maxloops and ensure we do
7882 	 * not run out of space on the pool.
7883 	 */
7884 	while (ztest_random(10) != 0 &&
7885 	    numloops++ < ztest_opts.zo_maxloops &&
7886 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
7887 		ztest_od_t od;
7888 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
7889 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
7890 		ztest_io(zd, od.od_object,
7891 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
7892 		txg_wait_synced(spa_get_dsl(spa), 0);
7893 	}
7894 
7895 	/*
7896 	 * Commit all of the changes we just generated.
7897 	 */
7898 	zil_commit(zd->zd_zilog, 0);
7899 	txg_wait_synced(spa_get_dsl(spa), 0);
7900 
7901 	/*
7902 	 * Close our dataset and close the pool.
7903 	 */
7904 	ztest_dataset_close(0);
7905 	spa_close(spa, FTAG);
7906 	kernel_fini();
7907 
7908 	/*
7909 	 * Open and close the pool and dataset to induce log replay.
7910 	 */
7911 	raidz_scratch_verify();
7912 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7913 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7914 	ASSERT3U(spa_freeze_txg(spa), ==, UINT64_MAX);
7915 	VERIFY0(ztest_dataset_open(0));
7916 	ztest_spa = spa;
7917 	txg_wait_synced(spa_get_dsl(spa), 0);
7918 	ztest_dataset_close(0);
7919 	ztest_reguid(NULL, 0);
7920 
7921 	spa_close(spa, FTAG);
7922 	kernel_fini();
7923 }
7924 
7925 static void
7926 ztest_import_impl(void)
7927 {
7928 	importargs_t args = { 0 };
7929 	nvlist_t *cfg = NULL;
7930 	int nsearch = 1;
7931 	char *searchdirs[nsearch];
7932 	int flags = ZFS_IMPORT_MISSING_LOG;
7933 
7934 	searchdirs[0] = ztest_opts.zo_dir;
7935 	args.paths = nsearch;
7936 	args.path = searchdirs;
7937 	args.can_be_active = B_FALSE;
7938 
7939 	libpc_handle_t lpch = {
7940 		.lpc_lib_handle = NULL,
7941 		.lpc_ops = &libzpool_config_ops,
7942 		.lpc_printerr = B_TRUE
7943 	};
7944 	VERIFY0(zpool_find_config(&lpch, ztest_opts.zo_pool, &cfg, &args));
7945 	VERIFY0(spa_import(ztest_opts.zo_pool, cfg, NULL, flags));
7946 	fnvlist_free(cfg);
7947 }
7948 
7949 /*
7950  * Import a storage pool with the given name.
7951  */
7952 static void
7953 ztest_import(ztest_shared_t *zs)
7954 {
7955 	spa_t *spa;
7956 
7957 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
7958 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
7959 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
7960 
7961 	raidz_scratch_verify();
7962 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
7963 
7964 	ztest_import_impl();
7965 
7966 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
7967 	zs->zs_metaslab_sz =
7968 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7969 	zs->zs_guid = spa_guid(spa);
7970 	spa_close(spa, FTAG);
7971 
7972 	kernel_fini();
7973 
7974 	if (!ztest_opts.zo_mmp_test) {
7975 		ztest_run_zdb(zs->zs_guid);
7976 		ztest_freeze();
7977 		ztest_run_zdb(zs->zs_guid);
7978 	}
7979 
7980 	(void) pthread_rwlock_destroy(&ztest_name_lock);
7981 	mutex_destroy(&ztest_vdev_lock);
7982 	mutex_destroy(&ztest_checkpoint_lock);
7983 }
7984 
7985 /*
7986  * After the expansion was killed, check that the pool is healthy
7987  */
7988 static void
7989 ztest_raidz_expand_check(spa_t *spa)
7990 {
7991 	ASSERT3U(ztest_opts.zo_raidz_expand_test, ==, RAIDZ_EXPAND_KILLED);
7992 	/*
7993 	 * Set pool check done flag, main program will run a zdb check
7994 	 * of the pool when we exit.
7995 	 */
7996 	ztest_shared_opts->zo_raidz_expand_test = RAIDZ_EXPAND_CHECKED;
7997 
7998 	/* Wait for reflow to finish */
7999 	if (ztest_opts.zo_verbose >= 1) {
8000 		(void) printf("\nwaiting for reflow to finish ...\n");
8001 	}
8002 	pool_raidz_expand_stat_t rzx_stats;
8003 	pool_raidz_expand_stat_t *pres = &rzx_stats;
8004 	do {
8005 		txg_wait_synced(spa_get_dsl(spa), 0);
8006 		(void) poll(NULL, 0, 500); /* wait 1/2 second */
8007 
8008 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8009 		(void) spa_raidz_expand_get_stats(spa, pres);
8010 		spa_config_exit(spa, SCL_CONFIG, FTAG);
8011 	} while (pres->pres_state != DSS_FINISHED &&
8012 	    pres->pres_reflowed < pres->pres_to_reflow);
8013 
8014 	if (ztest_opts.zo_verbose >= 1) {
8015 		(void) printf("verifying an interrupted raidz "
8016 		    "expansion using a pool scrub ...\n");
8017 	}
8018 	/* Will fail here if there is non-recoverable corruption detected */
8019 	VERIFY0(ztest_scrub_impl(spa));
8020 	if (ztest_opts.zo_verbose >= 1) {
8021 		(void) printf("raidz expansion scrub check complete\n");
8022 	}
8023 }
8024 
8025 /*
8026  * Start a raidz expansion test.  We run some I/O on the pool for a while
8027  * to get some data in the pool.  Then we grow the raidz and
8028  * kill the test at the requested offset into the reflow, verifying that
8029  * doing such does not lead to pool corruption.
8030  */
8031 static void
8032 ztest_raidz_expand_run(ztest_shared_t *zs, spa_t *spa)
8033 {
8034 	nvlist_t *root;
8035 	pool_raidz_expand_stat_t rzx_stats;
8036 	pool_raidz_expand_stat_t *pres = &rzx_stats;
8037 	kthread_t **run_threads;
8038 	vdev_t *cvd, *rzvd = spa->spa_root_vdev->vdev_child[0];
8039 	int total_disks = rzvd->vdev_children;
8040 	int data_disks = total_disks - vdev_get_nparity(rzvd);
8041 	uint64_t alloc_goal;
8042 	uint64_t csize;
8043 	int error, t;
8044 	int threads = ztest_opts.zo_threads;
8045 	ztest_expand_io_t *thread_args;
8046 
8047 	ASSERT3U(ztest_opts.zo_raidz_expand_test, !=, RAIDZ_EXPAND_NONE);
8048 	ASSERT3U(rzvd->vdev_ops, ==, &vdev_raidz_ops);
8049 	ztest_opts.zo_raidz_expand_test = RAIDZ_EXPAND_STARTED;
8050 
8051 	/* Setup a 1 MiB buffer of random data */
8052 	uint64_t bufsize = 1024 * 1024;
8053 	void *buffer = umem_alloc(bufsize, UMEM_NOFAIL);
8054 
8055 	if (read(ztest_fd_rand, buffer, bufsize) != bufsize) {
8056 		fatal(B_TRUE, "short read from /dev/urandom");
8057 	}
8058 	/*
8059 	 * Put some data in the pool and then attach a vdev to initiate
8060 	 * reflow.
8061 	 */
8062 	run_threads = umem_zalloc(threads * sizeof (kthread_t *), UMEM_NOFAIL);
8063 	thread_args = umem_zalloc(threads * sizeof (ztest_expand_io_t),
8064 	    UMEM_NOFAIL);
8065 	/* Aim for roughly 25% of allocatable space up to 1GB */
8066 	alloc_goal = (vdev_get_min_asize(rzvd) * data_disks) / total_disks;
8067 	alloc_goal = MIN(alloc_goal >> 2, 1024*1024*1024);
8068 	if (ztest_opts.zo_verbose >= 1) {
8069 		(void) printf("adding data to pool '%s', goal %llu bytes\n",
8070 		    ztest_opts.zo_pool, (u_longlong_t)alloc_goal);
8071 	}
8072 
8073 	/*
8074 	 * Kick off all the I/O generators that run in parallel.
8075 	 */
8076 	for (t = 0; t < threads; t++) {
8077 		if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
8078 			umem_free(run_threads, threads * sizeof (kthread_t *));
8079 			umem_free(buffer, bufsize);
8080 			return;
8081 		}
8082 		thread_args[t].rzx_id = t;
8083 		thread_args[t].rzx_amount = alloc_goal / threads;
8084 		thread_args[t].rzx_bufsize = bufsize;
8085 		thread_args[t].rzx_buffer = buffer;
8086 		thread_args[t].rzx_alloc_max = alloc_goal;
8087 		thread_args[t].rzx_spa = spa;
8088 		run_threads[t] = thread_create(NULL, 0, ztest_rzx_thread,
8089 		    &thread_args[t], 0, NULL, TS_RUN | TS_JOINABLE,
8090 		    defclsyspri);
8091 	}
8092 
8093 	/*
8094 	 * Wait for all of the writers to complete.
8095 	 */
8096 	for (t = 0; t < threads; t++)
8097 		VERIFY0(thread_join(run_threads[t]));
8098 
8099 	/*
8100 	 * Close all datasets. This must be done after all the threads
8101 	 * are joined so we can be sure none of the datasets are in-use
8102 	 * by any of the threads.
8103 	 */
8104 	for (t = 0; t < ztest_opts.zo_threads; t++) {
8105 		if (t < ztest_opts.zo_datasets)
8106 			ztest_dataset_close(t);
8107 	}
8108 
8109 	txg_wait_synced(spa_get_dsl(spa), 0);
8110 
8111 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
8112 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
8113 
8114 	umem_free(buffer, bufsize);
8115 	umem_free(run_threads, threads * sizeof (kthread_t *));
8116 	umem_free(thread_args, threads * sizeof (ztest_expand_io_t));
8117 
8118 	/* Set our reflow target to 25%, 50% or 75% of allocated size */
8119 	uint_t multiple = ztest_random(3) + 1;
8120 	uint64_t reflow_max = (rzvd->vdev_stat.vs_alloc * multiple) / 4;
8121 	raidz_expand_max_reflow_bytes = reflow_max;
8122 
8123 	if (ztest_opts.zo_verbose >= 1) {
8124 		(void) printf("running raidz expansion test, killing when "
8125 		    "reflow reaches %llu bytes (%u/4 of allocated space)\n",
8126 		    (u_longlong_t)reflow_max, multiple);
8127 	}
8128 
8129 	/* XXX - do we want some I/O load during the reflow? */
8130 
8131 	/*
8132 	 * Use a disk size that is larger than existing ones
8133 	 */
8134 	cvd = rzvd->vdev_child[0];
8135 	csize = vdev_get_min_asize(cvd);
8136 	csize += csize / 10;
8137 	/*
8138 	 * Path to vdev to be attached
8139 	 */
8140 	char *newpath = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
8141 	(void) snprintf(newpath, MAXPATHLEN, ztest_dev_template,
8142 	    ztest_opts.zo_dir, ztest_opts.zo_pool, rzvd->vdev_children);
8143 	/*
8144 	 * Build the nvlist describing newpath.
8145 	 */
8146 	root = make_vdev_root(newpath, NULL, NULL, csize, ztest_get_ashift(),
8147 	    NULL, 0, 0, 1);
8148 	/*
8149 	 * Expand the raidz vdev by attaching the new disk
8150 	 */
8151 	if (ztest_opts.zo_verbose >= 1) {
8152 		(void) printf("expanding raidz: %d wide to %d wide with '%s'\n",
8153 		    (int)rzvd->vdev_children, (int)rzvd->vdev_children + 1,
8154 		    newpath);
8155 	}
8156 	error = spa_vdev_attach(spa, rzvd->vdev_guid, root, B_FALSE, B_FALSE);
8157 	nvlist_free(root);
8158 	if (error != 0) {
8159 		fatal(0, "raidz expand: attach (%s %llu) returned %d",
8160 		    newpath, (long long)csize, error);
8161 	}
8162 
8163 	/*
8164 	 * Wait for reflow to begin
8165 	 */
8166 	while (spa->spa_raidz_expand == NULL) {
8167 		txg_wait_synced(spa_get_dsl(spa), 0);
8168 		(void) poll(NULL, 0, 100); /* wait 1/10 second */
8169 	}
8170 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8171 	(void) spa_raidz_expand_get_stats(spa, pres);
8172 	spa_config_exit(spa, SCL_CONFIG, FTAG);
8173 	while (pres->pres_state != DSS_SCANNING) {
8174 		txg_wait_synced(spa_get_dsl(spa), 0);
8175 		(void) poll(NULL, 0, 100); /* wait 1/10 second */
8176 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8177 		(void) spa_raidz_expand_get_stats(spa, pres);
8178 		spa_config_exit(spa, SCL_CONFIG, FTAG);
8179 	}
8180 
8181 	ASSERT3U(pres->pres_state, ==, DSS_SCANNING);
8182 	ASSERT3U(pres->pres_to_reflow, !=, 0);
8183 	/*
8184 	 * Set so when we are killed we go to raidz checking rather than
8185 	 * restarting test.
8186 	 */
8187 	ztest_shared_opts->zo_raidz_expand_test = RAIDZ_EXPAND_KILLED;
8188 	if (ztest_opts.zo_verbose >= 1) {
8189 		(void) printf("raidz expansion reflow started, waiting for "
8190 		    "%llu bytes to be copied\n", (u_longlong_t)reflow_max);
8191 	}
8192 
8193 	/*
8194 	 * Wait for reflow maximum to be reached and then kill the test
8195 	 */
8196 	while (pres->pres_reflowed < reflow_max) {
8197 		txg_wait_synced(spa_get_dsl(spa), 0);
8198 		(void) poll(NULL, 0, 100); /* wait 1/10 second */
8199 		spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
8200 		(void) spa_raidz_expand_get_stats(spa, pres);
8201 		spa_config_exit(spa, SCL_CONFIG, FTAG);
8202 	}
8203 
8204 	/* Reset the reflow pause before killing */
8205 	raidz_expand_max_reflow_bytes = 0;
8206 
8207 	if (ztest_opts.zo_verbose >= 1) {
8208 		(void) printf("killing raidz expansion test after reflow "
8209 		    "reached %llu bytes\n", (u_longlong_t)pres->pres_reflowed);
8210 	}
8211 
8212 	/*
8213 	 * Kill ourself to simulate a panic during a reflow.  Our parent will
8214 	 * restart the test and the changed flag value will drive the test
8215 	 * through the scrub/check code to verify the pool is not corrupted.
8216 	 */
8217 	ztest_kill(zs);
8218 }
8219 
8220 static void
8221 ztest_generic_run(ztest_shared_t *zs, spa_t *spa)
8222 {
8223 	kthread_t **run_threads;
8224 	int t;
8225 
8226 	run_threads = umem_zalloc(ztest_opts.zo_threads * sizeof (kthread_t *),
8227 	    UMEM_NOFAIL);
8228 
8229 	/*
8230 	 * Kick off all the tests that run in parallel.
8231 	 */
8232 	for (t = 0; t < ztest_opts.zo_threads; t++) {
8233 		if (t < ztest_opts.zo_datasets && ztest_dataset_open(t) != 0) {
8234 			umem_free(run_threads, ztest_opts.zo_threads *
8235 			    sizeof (kthread_t *));
8236 			return;
8237 		}
8238 
8239 		run_threads[t] = thread_create(NULL, 0, ztest_thread,
8240 		    (void *)(uintptr_t)t, 0, NULL, TS_RUN | TS_JOINABLE,
8241 		    defclsyspri);
8242 	}
8243 
8244 	/*
8245 	 * Wait for all of the tests to complete.
8246 	 */
8247 	for (t = 0; t < ztest_opts.zo_threads; t++)
8248 		VERIFY0(thread_join(run_threads[t]));
8249 
8250 	/*
8251 	 * Close all datasets. This must be done after all the threads
8252 	 * are joined so we can be sure none of the datasets are in-use
8253 	 * by any of the threads.
8254 	 */
8255 	for (t = 0; t < ztest_opts.zo_threads; t++) {
8256 		if (t < ztest_opts.zo_datasets)
8257 			ztest_dataset_close(t);
8258 	}
8259 
8260 	txg_wait_synced(spa_get_dsl(spa), 0);
8261 
8262 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
8263 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
8264 
8265 	umem_free(run_threads, ztest_opts.zo_threads * sizeof (kthread_t *));
8266 }
8267 
8268 /*
8269  * Setup our test context and kick off threads to run tests on all datasets
8270  * in parallel.
8271  */
8272 static void
8273 ztest_run(ztest_shared_t *zs)
8274 {
8275 	spa_t *spa;
8276 	objset_t *os;
8277 	kthread_t *resume_thread, *deadman_thread;
8278 	uint64_t object;
8279 	int error;
8280 	int t, d;
8281 
8282 	ztest_exiting = B_FALSE;
8283 
8284 	/*
8285 	 * Initialize parent/child shared state.
8286 	 */
8287 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
8288 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
8289 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
8290 
8291 	zs->zs_thread_start = gethrtime();
8292 	zs->zs_thread_stop =
8293 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
8294 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
8295 	zs->zs_thread_kill = zs->zs_thread_stop;
8296 	if (ztest_random(100) < ztest_opts.zo_killrate) {
8297 		zs->zs_thread_kill -=
8298 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
8299 	}
8300 
8301 	mutex_init(&zcl.zcl_callbacks_lock, NULL, MUTEX_DEFAULT, NULL);
8302 
8303 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
8304 	    offsetof(ztest_cb_data_t, zcd_node));
8305 
8306 	/*
8307 	 * Open our pool.  It may need to be imported first depending on
8308 	 * what tests were running when the previous pass was terminated.
8309 	 */
8310 	raidz_scratch_verify();
8311 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
8312 	error = spa_open(ztest_opts.zo_pool, &spa, FTAG);
8313 	if (error) {
8314 		VERIFY3S(error, ==, ENOENT);
8315 		ztest_import_impl();
8316 		VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
8317 		zs->zs_metaslab_sz =
8318 		    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
8319 	}
8320 
8321 	metaslab_preload_limit = ztest_random(20) + 1;
8322 	ztest_spa = spa;
8323 
8324 	/*
8325 	 * XXX - BUGBUG raidz expansion do not run this for generic for now
8326 	 */
8327 	if (ztest_opts.zo_raidz_expand_test != RAIDZ_EXPAND_NONE)
8328 		VERIFY0(vdev_raidz_impl_set("cycle"));
8329 
8330 	dmu_objset_stats_t dds;
8331 	VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
8332 	    DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
8333 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
8334 	dmu_objset_fast_stat(os, &dds);
8335 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
8336 	dmu_objset_disown(os, B_TRUE, FTAG);
8337 
8338 	/* Give the dedicated raidz expansion test more grace time */
8339 	if (ztest_opts.zo_raidz_expand_test != RAIDZ_EXPAND_NONE)
8340 		zfs_deadman_synctime_ms *= 2;
8341 
8342 	/*
8343 	 * Create a thread to periodically resume suspended I/O.
8344 	 */
8345 	resume_thread = thread_create(NULL, 0, ztest_resume_thread,
8346 	    spa, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
8347 
8348 	/*
8349 	 * Create a deadman thread and set to panic if we hang.
8350 	 */
8351 	deadman_thread = thread_create(NULL, 0, ztest_deadman_thread,
8352 	    zs, 0, NULL, TS_RUN | TS_JOINABLE, defclsyspri);
8353 
8354 	spa->spa_deadman_failmode = ZIO_FAILURE_MODE_PANIC;
8355 
8356 	/*
8357 	 * Verify that we can safely inquire about any object,
8358 	 * whether it's allocated or not.  To make it interesting,
8359 	 * we probe a 5-wide window around each power of two.
8360 	 * This hits all edge cases, including zero and the max.
8361 	 */
8362 	for (t = 0; t < 64; t++) {
8363 		for (d = -5; d <= 5; d++) {
8364 			error = dmu_object_info(spa->spa_meta_objset,
8365 			    (1ULL << t) + d, NULL);
8366 			ASSERT(error == 0 || error == ENOENT ||
8367 			    error == EINVAL);
8368 		}
8369 	}
8370 
8371 	/*
8372 	 * If we got any ENOSPC errors on the previous run, destroy something.
8373 	 */
8374 	if (zs->zs_enospc_count != 0) {
8375 		/* Not expecting ENOSPC errors during raidz expansion tests */
8376 		ASSERT3U(ztest_opts.zo_raidz_expand_test, ==,
8377 		    RAIDZ_EXPAND_NONE);
8378 
8379 		int d = ztest_random(ztest_opts.zo_datasets);
8380 		ztest_dataset_destroy(d);
8381 	}
8382 	zs->zs_enospc_count = 0;
8383 
8384 	/*
8385 	 * If we were in the middle of ztest_device_removal() and were killed
8386 	 * we need to ensure the removal and scrub complete before running
8387 	 * any tests that check ztest_device_removal_active. The removal will
8388 	 * be restarted automatically when the spa is opened, but we need to
8389 	 * initiate the scrub manually if it is not already in progress. Note
8390 	 * that we always run the scrub whenever an indirect vdev exists
8391 	 * because we have no way of knowing for sure if ztest_device_removal()
8392 	 * fully completed its scrub before the pool was reimported.
8393 	 *
8394 	 * Does not apply for the RAIDZ expansion specific test runs
8395 	 */
8396 	if (ztest_opts.zo_raidz_expand_test == RAIDZ_EXPAND_NONE &&
8397 	    (spa->spa_removing_phys.sr_state == DSS_SCANNING ||
8398 	    spa->spa_removing_phys.sr_prev_indirect_vdev != -1)) {
8399 		while (spa->spa_removing_phys.sr_state == DSS_SCANNING)
8400 			txg_wait_synced(spa_get_dsl(spa), 0);
8401 
8402 		error = ztest_scrub_impl(spa);
8403 		if (error == EBUSY)
8404 			error = 0;
8405 		ASSERT0(error);
8406 	}
8407 
8408 	if (ztest_opts.zo_verbose >= 4)
8409 		(void) printf("starting main threads...\n");
8410 
8411 	/*
8412 	 * Replay all logs of all datasets in the pool. This is primarily for
8413 	 * temporary datasets which wouldn't otherwise get replayed, which
8414 	 * can trigger failures when attempting to offline a SLOG in
8415 	 * ztest_fault_inject().
8416 	 */
8417 	(void) dmu_objset_find(ztest_opts.zo_pool, ztest_replay_zil_cb,
8418 	    NULL, DS_FIND_CHILDREN);
8419 
8420 	if (ztest_opts.zo_raidz_expand_test == RAIDZ_EXPAND_REQUESTED)
8421 		ztest_raidz_expand_run(zs, spa);
8422 	else if (ztest_opts.zo_raidz_expand_test == RAIDZ_EXPAND_KILLED)
8423 		ztest_raidz_expand_check(spa);
8424 	else
8425 		ztest_generic_run(zs, spa);
8426 
8427 	/* Kill the resume and deadman threads */
8428 	ztest_exiting = B_TRUE;
8429 	VERIFY0(thread_join(resume_thread));
8430 	VERIFY0(thread_join(deadman_thread));
8431 	ztest_resume(spa);
8432 
8433 	/*
8434 	 * Right before closing the pool, kick off a bunch of async I/O;
8435 	 * spa_close() should wait for it to complete.
8436 	 */
8437 	for (object = 1; object < 50; object++) {
8438 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
8439 		    ZIO_PRIORITY_SYNC_READ);
8440 	}
8441 
8442 	/* Verify that at least one commit cb was called in a timely fashion */
8443 	if (zc_cb_counter >= ZTEST_COMMIT_CB_MIN_REG)
8444 		VERIFY0(zc_min_txg_delay);
8445 
8446 	spa_close(spa, FTAG);
8447 
8448 	/*
8449 	 * Verify that we can loop over all pools.
8450 	 */
8451 	mutex_enter(&spa_namespace_lock);
8452 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
8453 		if (ztest_opts.zo_verbose > 3)
8454 			(void) printf("spa_next: found %s\n", spa_name(spa));
8455 	mutex_exit(&spa_namespace_lock);
8456 
8457 	/*
8458 	 * Verify that we can export the pool and reimport it under a
8459 	 * different name.
8460 	 */
8461 	if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
8462 		char name[ZFS_MAX_DATASET_NAME_LEN];
8463 		(void) snprintf(name, sizeof (name), "%s_import",
8464 		    ztest_opts.zo_pool);
8465 		ztest_spa_import_export(ztest_opts.zo_pool, name);
8466 		ztest_spa_import_export(name, ztest_opts.zo_pool);
8467 	}
8468 
8469 	kernel_fini();
8470 
8471 	list_destroy(&zcl.zcl_callbacks);
8472 	mutex_destroy(&zcl.zcl_callbacks_lock);
8473 	(void) pthread_rwlock_destroy(&ztest_name_lock);
8474 	mutex_destroy(&ztest_vdev_lock);
8475 	mutex_destroy(&ztest_checkpoint_lock);
8476 }
8477 
8478 static void
8479 print_time(hrtime_t t, char *timebuf)
8480 {
8481 	hrtime_t s = t / NANOSEC;
8482 	hrtime_t m = s / 60;
8483 	hrtime_t h = m / 60;
8484 	hrtime_t d = h / 24;
8485 
8486 	s -= m * 60;
8487 	m -= h * 60;
8488 	h -= d * 24;
8489 
8490 	timebuf[0] = '\0';
8491 
8492 	if (d)
8493 		(void) sprintf(timebuf,
8494 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
8495 	else if (h)
8496 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
8497 	else if (m)
8498 		(void) sprintf(timebuf, "%llum%02llus", m, s);
8499 	else
8500 		(void) sprintf(timebuf, "%llus", s);
8501 }
8502 
8503 static nvlist_t *
8504 make_random_props(void)
8505 {
8506 	nvlist_t *props;
8507 
8508 	props = fnvlist_alloc();
8509 
8510 	if (ztest_random(2) == 0)
8511 		return (props);
8512 
8513 	fnvlist_add_uint64(props,
8514 	    zpool_prop_to_name(ZPOOL_PROP_AUTOREPLACE), 1);
8515 
8516 	return (props);
8517 }
8518 
8519 /*
8520  * Create a storage pool with the given name and initial vdev size.
8521  * Then test spa_freeze() functionality.
8522  */
8523 static void
8524 ztest_init(ztest_shared_t *zs)
8525 {
8526 	spa_t *spa;
8527 	nvlist_t *nvroot, *props;
8528 	int i;
8529 
8530 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
8531 	mutex_init(&ztest_checkpoint_lock, NULL, MUTEX_DEFAULT, NULL);
8532 	VERIFY0(pthread_rwlock_init(&ztest_name_lock, NULL));
8533 
8534 	raidz_scratch_verify();
8535 	kernel_init(SPA_MODE_READ | SPA_MODE_WRITE);
8536 
8537 	/*
8538 	 * Create the storage pool.
8539 	 */
8540 	(void) spa_destroy(ztest_opts.zo_pool);
8541 	ztest_shared->zs_vdev_next_leaf = 0;
8542 	zs->zs_splits = 0;
8543 	zs->zs_mirrors = ztest_opts.zo_mirrors;
8544 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
8545 	    NULL, ztest_opts.zo_raid_children, zs->zs_mirrors, 1);
8546 	props = make_random_props();
8547 
8548 	/*
8549 	 * We don't expect the pool to suspend unless maxfaults == 0,
8550 	 * in which case ztest_fault_inject() temporarily takes away
8551 	 * the only valid replica.
8552 	 */
8553 	fnvlist_add_uint64(props,
8554 	    zpool_prop_to_name(ZPOOL_PROP_FAILUREMODE),
8555 	    MAXFAULTS(zs) ? ZIO_FAILURE_MODE_PANIC : ZIO_FAILURE_MODE_WAIT);
8556 
8557 	for (i = 0; i < SPA_FEATURES; i++) {
8558 		char *buf;
8559 
8560 		if (!spa_feature_table[i].fi_zfs_mod_supported)
8561 			continue;
8562 
8563 		/*
8564 		 * 75% chance of using the log space map feature. We want ztest
8565 		 * to exercise both the code paths that use the log space map
8566 		 * feature and the ones that don't.
8567 		 */
8568 		if (i == SPA_FEATURE_LOG_SPACEMAP && ztest_random(4) == 0)
8569 			continue;
8570 
8571 		VERIFY3S(-1, !=, asprintf(&buf, "feature@%s",
8572 		    spa_feature_table[i].fi_uname));
8573 		fnvlist_add_uint64(props, buf, 0);
8574 		free(buf);
8575 	}
8576 
8577 	VERIFY0(spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
8578 	fnvlist_free(nvroot);
8579 	fnvlist_free(props);
8580 
8581 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
8582 	zs->zs_metaslab_sz =
8583 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
8584 	zs->zs_guid = spa_guid(spa);
8585 	spa_close(spa, FTAG);
8586 
8587 	kernel_fini();
8588 
8589 	if (!ztest_opts.zo_mmp_test) {
8590 		ztest_run_zdb(zs->zs_guid);
8591 		ztest_freeze();
8592 		ztest_run_zdb(zs->zs_guid);
8593 	}
8594 
8595 	(void) pthread_rwlock_destroy(&ztest_name_lock);
8596 	mutex_destroy(&ztest_vdev_lock);
8597 	mutex_destroy(&ztest_checkpoint_lock);
8598 }
8599 
8600 static void
8601 setup_data_fd(void)
8602 {
8603 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
8604 
8605 	ztest_fd_data = mkstemp(ztest_name_data);
8606 	ASSERT3S(ztest_fd_data, >=, 0);
8607 	(void) unlink(ztest_name_data);
8608 }
8609 
8610 static int
8611 shared_data_size(ztest_shared_hdr_t *hdr)
8612 {
8613 	int size;
8614 
8615 	size = hdr->zh_hdr_size;
8616 	size += hdr->zh_opts_size;
8617 	size += hdr->zh_size;
8618 	size += hdr->zh_stats_size * hdr->zh_stats_count;
8619 	size += hdr->zh_ds_size * hdr->zh_ds_count;
8620 	size += hdr->zh_scratch_state_size;
8621 
8622 	return (size);
8623 }
8624 
8625 static void
8626 setup_hdr(void)
8627 {
8628 	int size;
8629 	ztest_shared_hdr_t *hdr;
8630 
8631 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
8632 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
8633 	ASSERT3P(hdr, !=, MAP_FAILED);
8634 
8635 	VERIFY0(ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
8636 
8637 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
8638 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
8639 	hdr->zh_size = sizeof (ztest_shared_t);
8640 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
8641 	hdr->zh_stats_count = ZTEST_FUNCS;
8642 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
8643 	hdr->zh_ds_count = ztest_opts.zo_datasets;
8644 	hdr->zh_scratch_state_size = sizeof (ztest_shared_scratch_state_t);
8645 
8646 	size = shared_data_size(hdr);
8647 	VERIFY0(ftruncate(ztest_fd_data, size));
8648 
8649 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
8650 }
8651 
8652 static void
8653 setup_data(void)
8654 {
8655 	int size, offset;
8656 	ztest_shared_hdr_t *hdr;
8657 	uint8_t *buf;
8658 
8659 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
8660 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
8661 	ASSERT3P(hdr, !=, MAP_FAILED);
8662 
8663 	size = shared_data_size(hdr);
8664 
8665 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
8666 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
8667 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
8668 	ASSERT3P(hdr, !=, MAP_FAILED);
8669 	buf = (uint8_t *)hdr;
8670 
8671 	offset = hdr->zh_hdr_size;
8672 	ztest_shared_opts = (void *)&buf[offset];
8673 	offset += hdr->zh_opts_size;
8674 	ztest_shared = (void *)&buf[offset];
8675 	offset += hdr->zh_size;
8676 	ztest_shared_callstate = (void *)&buf[offset];
8677 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
8678 	ztest_shared_ds = (void *)&buf[offset];
8679 	offset += hdr->zh_ds_size * hdr->zh_ds_count;
8680 	ztest_scratch_state = (void *)&buf[offset];
8681 }
8682 
8683 static boolean_t
8684 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
8685 {
8686 	pid_t pid;
8687 	int status;
8688 	char *cmdbuf = NULL;
8689 
8690 	pid = fork();
8691 
8692 	if (cmd == NULL) {
8693 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
8694 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
8695 		cmd = cmdbuf;
8696 	}
8697 
8698 	if (pid == -1)
8699 		fatal(B_TRUE, "fork failed");
8700 
8701 	if (pid == 0) {	/* child */
8702 		char fd_data_str[12];
8703 
8704 		VERIFY3S(11, >=,
8705 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
8706 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
8707 
8708 		if (libpath != NULL) {
8709 			const char *curlp = getenv("LD_LIBRARY_PATH");
8710 			if (curlp == NULL)
8711 				VERIFY0(setenv("LD_LIBRARY_PATH", libpath, 1));
8712 			else {
8713 				char *newlp = NULL;
8714 				VERIFY3S(-1, !=,
8715 				    asprintf(&newlp, "%s:%s", libpath, curlp));
8716 				VERIFY0(setenv("LD_LIBRARY_PATH", newlp, 1));
8717 				free(newlp);
8718 			}
8719 		}
8720 		(void) execl(cmd, cmd, (char *)NULL);
8721 		ztest_dump_core = B_FALSE;
8722 		fatal(B_TRUE, "exec failed: %s", cmd);
8723 	}
8724 
8725 	if (cmdbuf != NULL) {
8726 		umem_free(cmdbuf, MAXPATHLEN);
8727 		cmd = NULL;
8728 	}
8729 
8730 	while (waitpid(pid, &status, 0) != pid)
8731 		continue;
8732 	if (statusp != NULL)
8733 		*statusp = status;
8734 
8735 	if (WIFEXITED(status)) {
8736 		if (WEXITSTATUS(status) != 0) {
8737 			(void) fprintf(stderr, "child exited with code %d\n",
8738 			    WEXITSTATUS(status));
8739 			exit(2);
8740 		}
8741 		return (B_FALSE);
8742 	} else if (WIFSIGNALED(status)) {
8743 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
8744 			(void) fprintf(stderr, "child died with signal %d\n",
8745 			    WTERMSIG(status));
8746 			exit(3);
8747 		}
8748 		return (B_TRUE);
8749 	} else {
8750 		(void) fprintf(stderr, "something strange happened to child\n");
8751 		exit(4);
8752 	}
8753 }
8754 
8755 static void
8756 ztest_run_init(void)
8757 {
8758 	int i;
8759 
8760 	ztest_shared_t *zs = ztest_shared;
8761 
8762 	/*
8763 	 * Blow away any existing copy of zpool.cache
8764 	 */
8765 	(void) remove(spa_config_path);
8766 
8767 	if (ztest_opts.zo_init == 0) {
8768 		if (ztest_opts.zo_verbose >= 1)
8769 			(void) printf("Importing pool %s\n",
8770 			    ztest_opts.zo_pool);
8771 		ztest_import(zs);
8772 		return;
8773 	}
8774 
8775 	/*
8776 	 * Create and initialize our storage pool.
8777 	 */
8778 	for (i = 1; i <= ztest_opts.zo_init; i++) {
8779 		memset(zs, 0, sizeof (*zs));
8780 		if (ztest_opts.zo_verbose >= 3 &&
8781 		    ztest_opts.zo_init != 1) {
8782 			(void) printf("ztest_init(), pass %d\n", i);
8783 		}
8784 		ztest_init(zs);
8785 	}
8786 }
8787 
8788 int
8789 main(int argc, char **argv)
8790 {
8791 	int kills = 0;
8792 	int iters = 0;
8793 	int older = 0;
8794 	int newer = 0;
8795 	ztest_shared_t *zs;
8796 	ztest_info_t *zi;
8797 	ztest_shared_callstate_t *zc;
8798 	char timebuf[100];
8799 	char numbuf[NN_NUMBUF_SZ];
8800 	char *cmd;
8801 	boolean_t hasalt;
8802 	int f, err;
8803 	char *fd_data_str = getenv("ZTEST_FD_DATA");
8804 	struct sigaction action;
8805 
8806 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
8807 
8808 	dprintf_setup(&argc, argv);
8809 	zfs_deadman_synctime_ms = 300000;
8810 	zfs_deadman_checktime_ms = 30000;
8811 	/*
8812 	 * As two-word space map entries may not come up often (especially
8813 	 * if pool and vdev sizes are small) we want to force at least some
8814 	 * of them so the feature get tested.
8815 	 */
8816 	zfs_force_some_double_word_sm_entries = B_TRUE;
8817 
8818 	/*
8819 	 * Verify that even extensively damaged split blocks with many
8820 	 * segments can be reconstructed in a reasonable amount of time
8821 	 * when reconstruction is known to be possible.
8822 	 *
8823 	 * Note: the lower this value is, the more damage we inflict, and
8824 	 * the more time ztest spends in recovering that damage. We chose
8825 	 * to induce damage 1/100th of the time so recovery is tested but
8826 	 * not so frequently that ztest doesn't get to test other code paths.
8827 	 */
8828 	zfs_reconstruct_indirect_damage_fraction = 100;
8829 
8830 	action.sa_handler = sig_handler;
8831 	sigemptyset(&action.sa_mask);
8832 	action.sa_flags = 0;
8833 
8834 	if (sigaction(SIGSEGV, &action, NULL) < 0) {
8835 		(void) fprintf(stderr, "ztest: cannot catch SIGSEGV: %s.\n",
8836 		    strerror(errno));
8837 		exit(EXIT_FAILURE);
8838 	}
8839 
8840 	if (sigaction(SIGABRT, &action, NULL) < 0) {
8841 		(void) fprintf(stderr, "ztest: cannot catch SIGABRT: %s.\n",
8842 		    strerror(errno));
8843 		exit(EXIT_FAILURE);
8844 	}
8845 
8846 	/*
8847 	 * Force random_get_bytes() to use /dev/urandom in order to prevent
8848 	 * ztest from needlessly depleting the system entropy pool.
8849 	 */
8850 	random_path = "/dev/urandom";
8851 	ztest_fd_rand = open(random_path, O_RDONLY | O_CLOEXEC);
8852 	ASSERT3S(ztest_fd_rand, >=, 0);
8853 
8854 	if (!fd_data_str) {
8855 		process_options(argc, argv);
8856 
8857 		setup_data_fd();
8858 		setup_hdr();
8859 		setup_data();
8860 		memcpy(ztest_shared_opts, &ztest_opts,
8861 		    sizeof (*ztest_shared_opts));
8862 	} else {
8863 		ztest_fd_data = atoi(fd_data_str);
8864 		setup_data();
8865 		memcpy(&ztest_opts, ztest_shared_opts, sizeof (ztest_opts));
8866 	}
8867 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
8868 
8869 	err = ztest_set_global_vars();
8870 	if (err != 0 && !fd_data_str) {
8871 		/* error message done by ztest_set_global_vars */
8872 		exit(EXIT_FAILURE);
8873 	} else {
8874 		/* children should not be spawned if setting gvars fails */
8875 		VERIFY3S(err, ==, 0);
8876 	}
8877 
8878 	/* Override location of zpool.cache */
8879 	VERIFY3S(asprintf((char **)&spa_config_path, "%s/zpool.cache",
8880 	    ztest_opts.zo_dir), !=, -1);
8881 
8882 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
8883 	    UMEM_NOFAIL);
8884 	zs = ztest_shared;
8885 
8886 	if (fd_data_str) {
8887 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
8888 		metaslab_df_alloc_threshold =
8889 		    zs->zs_metaslab_df_alloc_threshold;
8890 
8891 		if (zs->zs_do_init)
8892 			ztest_run_init();
8893 		else
8894 			ztest_run(zs);
8895 		exit(0);
8896 	}
8897 
8898 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
8899 
8900 	if (ztest_opts.zo_verbose >= 1) {
8901 		(void) printf("%"PRIu64" vdevs, %d datasets, %d threads, "
8902 		    "%d %s disks, parity %d, %"PRIu64" seconds...\n\n",
8903 		    ztest_opts.zo_vdevs,
8904 		    ztest_opts.zo_datasets,
8905 		    ztest_opts.zo_threads,
8906 		    ztest_opts.zo_raid_children,
8907 		    ztest_opts.zo_raid_type,
8908 		    ztest_opts.zo_raid_parity,
8909 		    ztest_opts.zo_time);
8910 	}
8911 
8912 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
8913 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
8914 
8915 	zs->zs_do_init = B_TRUE;
8916 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
8917 		if (ztest_opts.zo_verbose >= 1) {
8918 			(void) printf("Executing older ztest for "
8919 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
8920 		}
8921 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
8922 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
8923 	} else {
8924 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
8925 	}
8926 	zs->zs_do_init = B_FALSE;
8927 
8928 	zs->zs_proc_start = gethrtime();
8929 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
8930 
8931 	for (f = 0; f < ZTEST_FUNCS; f++) {
8932 		zi = &ztest_info[f];
8933 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
8934 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
8935 			zc->zc_next = UINT64_MAX;
8936 		else
8937 			zc->zc_next = zs->zs_proc_start +
8938 			    ztest_random(2 * zi->zi_interval[0] + 1);
8939 	}
8940 
8941 	/*
8942 	 * Run the tests in a loop.  These tests include fault injection
8943 	 * to verify that self-healing data works, and forced crashes
8944 	 * to verify that we never lose on-disk consistency.
8945 	 */
8946 	while (gethrtime() < zs->zs_proc_stop) {
8947 		int status;
8948 		boolean_t killed;
8949 
8950 		/*
8951 		 * Initialize the workload counters for each function.
8952 		 */
8953 		for (f = 0; f < ZTEST_FUNCS; f++) {
8954 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
8955 			zc->zc_count = 0;
8956 			zc->zc_time = 0;
8957 		}
8958 
8959 		/* Set the allocation switch size */
8960 		zs->zs_metaslab_df_alloc_threshold =
8961 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
8962 
8963 		if (!hasalt || ztest_random(2) == 0) {
8964 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8965 				(void) printf("Executing newer ztest: %s\n",
8966 				    cmd);
8967 			}
8968 			newer++;
8969 			killed = exec_child(cmd, NULL, B_TRUE, &status);
8970 		} else {
8971 			if (hasalt && ztest_opts.zo_verbose >= 1) {
8972 				(void) printf("Executing older ztest: %s\n",
8973 				    ztest_opts.zo_alt_ztest);
8974 			}
8975 			older++;
8976 			killed = exec_child(ztest_opts.zo_alt_ztest,
8977 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
8978 		}
8979 
8980 		if (killed)
8981 			kills++;
8982 		iters++;
8983 
8984 		if (ztest_opts.zo_verbose >= 1) {
8985 			hrtime_t now = gethrtime();
8986 
8987 			now = MIN(now, zs->zs_proc_stop);
8988 			print_time(zs->zs_proc_stop - now, timebuf);
8989 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
8990 
8991 			(void) printf("Pass %3d, %8s, %3"PRIu64" ENOSPC, "
8992 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
8993 			    iters,
8994 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
8995 			    zs->zs_enospc_count,
8996 			    100.0 * zs->zs_alloc / zs->zs_space,
8997 			    numbuf,
8998 			    100.0 * (now - zs->zs_proc_start) /
8999 			    (ztest_opts.zo_time * NANOSEC), timebuf);
9000 		}
9001 
9002 		if (ztest_opts.zo_verbose >= 2) {
9003 			(void) printf("\nWorkload summary:\n\n");
9004 			(void) printf("%7s %9s   %s\n",
9005 			    "Calls", "Time", "Function");
9006 			(void) printf("%7s %9s   %s\n",
9007 			    "-----", "----", "--------");
9008 			for (f = 0; f < ZTEST_FUNCS; f++) {
9009 				zi = &ztest_info[f];
9010 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
9011 				print_time(zc->zc_time, timebuf);
9012 				(void) printf("%7"PRIu64" %9s   %s\n",
9013 				    zc->zc_count, timebuf,
9014 				    zi->zi_funcname);
9015 			}
9016 			(void) printf("\n");
9017 		}
9018 
9019 		if (!ztest_opts.zo_mmp_test)
9020 			ztest_run_zdb(zs->zs_guid);
9021 		if (ztest_shared_opts->zo_raidz_expand_test ==
9022 		    RAIDZ_EXPAND_CHECKED)
9023 			break; /* raidz expand test complete */
9024 	}
9025 
9026 	if (ztest_opts.zo_verbose >= 1) {
9027 		if (hasalt) {
9028 			(void) printf("%d runs of older ztest: %s\n", older,
9029 			    ztest_opts.zo_alt_ztest);
9030 			(void) printf("%d runs of newer ztest: %s\n", newer,
9031 			    cmd);
9032 		}
9033 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
9034 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
9035 	}
9036 
9037 	umem_free(cmd, MAXNAMELEN);
9038 
9039 	return (0);
9040 }
9041