xref: /illumos-gate/usr/src/cmd/ztest/ztest.c (revision 02ac56e010f18fc0c5aafe47377586d8ba8c897c)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Joyent, Inc.
28  * Copyright (c) 2017, Intel Corporation.
29  * Copyright 2017 RackTop Systems.
30  */
31 
32 /*
33  * The objective of this program is to provide a DMU/ZAP/SPA stress test
34  * that runs entirely in userland, is easy to use, and easy to extend.
35  *
36  * The overall design of the ztest program is as follows:
37  *
38  * (1) For each major functional area (e.g. adding vdevs to a pool,
39  *     creating and destroying datasets, reading and writing objects, etc)
40  *     we have a simple routine to test that functionality.  These
41  *     individual routines do not have to do anything "stressful".
42  *
43  * (2) We turn these simple functionality tests into a stress test by
44  *     running them all in parallel, with as many threads as desired,
45  *     and spread across as many datasets, objects, and vdevs as desired.
46  *
47  * (3) While all this is happening, we inject faults into the pool to
48  *     verify that self-healing data really works.
49  *
50  * (4) Every time we open a dataset, we change its checksum and compression
51  *     functions.  Thus even individual objects vary from block to block
52  *     in which checksum they use and whether they're compressed.
53  *
54  * (5) To verify that we never lose on-disk consistency after a crash,
55  *     we run the entire test in a child of the main process.
56  *     At random times, the child self-immolates with a SIGKILL.
57  *     This is the software equivalent of pulling the power cord.
58  *     The parent then runs the test again, using the existing
59  *     storage pool, as many times as desired. If backwards compatibility
60  *     testing is enabled ztest will sometimes run the "older" version
61  *     of ztest after a SIGKILL.
62  *
63  * (6) To verify that we don't have future leaks or temporal incursions,
64  *     many of the functional tests record the transaction group number
65  *     as part of their data.  When reading old data, they verify that
66  *     the transaction group number is less than the current, open txg.
67  *     If you add a new test, please do this if applicable.
68  *
69  * When run with no arguments, ztest runs for about five minutes and
70  * produces no output if successful.  To get a little bit of information,
71  * specify -V.  To get more information, specify -VV, and so on.
72  *
73  * To turn this into an overnight stress test, use -T to specify run time.
74  *
75  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
76  * to increase the pool capacity, fanout, and overall stress level.
77  *
78  * Use the -k option to set the desired frequency of kills.
79  *
80  * When ztest invokes itself it passes all relevant information through a
81  * temporary file which is mmap-ed in the child process. This allows shared
82  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
83  * stored at offset 0 of this file and contains information on the size and
84  * number of shared structures in the file. The information stored in this file
85  * must remain backwards compatible with older versions of ztest so that
86  * ztest can invoke them during backwards compatibility testing (-B).
87  */
88 
89 #include <sys/zfs_context.h>
90 #include <sys/spa.h>
91 #include <sys/dmu.h>
92 #include <sys/txg.h>
93 #include <sys/dbuf.h>
94 #include <sys/zap.h>
95 #include <sys/dmu_objset.h>
96 #include <sys/poll.h>
97 #include <sys/stat.h>
98 #include <sys/time.h>
99 #include <sys/wait.h>
100 #include <sys/mman.h>
101 #include <sys/resource.h>
102 #include <sys/zio.h>
103 #include <sys/zil.h>
104 #include <sys/zil_impl.h>
105 #include <sys/vdev_impl.h>
106 #include <sys/vdev_file.h>
107 #include <sys/vdev_initialize.h>
108 #include <sys/vdev_trim.h>
109 #include <sys/spa_impl.h>
110 #include <sys/metaslab_impl.h>
111 #include <sys/dsl_prop.h>
112 #include <sys/dsl_dataset.h>
113 #include <sys/dsl_destroy.h>
114 #include <sys/dsl_scan.h>
115 #include <sys/zio_checksum.h>
116 #include <sys/refcount.h>
117 #include <sys/zfeature.h>
118 #include <sys/dsl_userhold.h>
119 #include <sys/abd.h>
120 #include <stdio.h>
121 #include <stdio_ext.h>
122 #include <stdlib.h>
123 #include <unistd.h>
124 #include <signal.h>
125 #include <umem.h>
126 #include <dlfcn.h>
127 #include <ctype.h>
128 #include <math.h>
129 #include <sys/fs/zfs.h>
130 #include <zfs_fletcher.h>
131 #include <libnvpair.h>
132 #include <libzutil.h>
133 #include <libcmdutils.h>
134 
135 static int ztest_fd_data = -1;
136 static int ztest_fd_rand = -1;
137 
138 typedef struct ztest_shared_hdr {
139 	uint64_t	zh_hdr_size;
140 	uint64_t	zh_opts_size;
141 	uint64_t	zh_size;
142 	uint64_t	zh_stats_size;
143 	uint64_t	zh_stats_count;
144 	uint64_t	zh_ds_size;
145 	uint64_t	zh_ds_count;
146 } ztest_shared_hdr_t;
147 
148 static ztest_shared_hdr_t *ztest_shared_hdr;
149 
150 enum ztest_class_state {
151 	ZTEST_VDEV_CLASS_OFF,
152 	ZTEST_VDEV_CLASS_ON,
153 	ZTEST_VDEV_CLASS_RND
154 };
155 
156 typedef struct ztest_shared_opts {
157 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
158 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
159 	char zo_alt_ztest[MAXNAMELEN];
160 	char zo_alt_libpath[MAXNAMELEN];
161 	uint64_t zo_vdevs;
162 	uint64_t zo_vdevtime;
163 	size_t zo_vdev_size;
164 	int zo_ashift;
165 	int zo_mirrors;
166 	int zo_raidz;
167 	int zo_raidz_parity;
168 	int zo_datasets;
169 	int zo_threads;
170 	uint64_t zo_passtime;
171 	uint64_t zo_killrate;
172 	int zo_verbose;
173 	int zo_init;
174 	uint64_t zo_time;
175 	uint64_t zo_maxloops;
176 	uint64_t zo_metaslab_force_ganging;
177 	int zo_mmp_test;
178 	int zo_special_vdevs;
179 } ztest_shared_opts_t;
180 
181 static const ztest_shared_opts_t ztest_opts_defaults = {
182 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
183 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
184 	.zo_alt_ztest = { '\0' },
185 	.zo_alt_libpath = { '\0' },
186 	.zo_vdevs = 5,
187 	.zo_ashift = SPA_MINBLOCKSHIFT,
188 	.zo_mirrors = 2,
189 	.zo_raidz = 4,
190 	.zo_raidz_parity = 1,
191 	.zo_vdev_size = SPA_MINDEVSIZE * 4,	/* 256m default size */
192 	.zo_datasets = 7,
193 	.zo_threads = 23,
194 	.zo_passtime = 60,		/* 60 seconds */
195 	.zo_killrate = 70,		/* 70% kill rate */
196 	.zo_verbose = 0,
197 	.zo_mmp_test = 0,
198 	.zo_init = 1,
199 	.zo_time = 300,			/* 5 minutes */
200 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
201 	.zo_metaslab_force_ganging = 32 << 10,
202 	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
203 };
204 
205 extern uint64_t metaslab_force_ganging;
206 extern uint64_t metaslab_df_alloc_threshold;
207 extern uint64_t zfs_deadman_synctime_ms;
208 extern int metaslab_preload_limit;
209 extern boolean_t zfs_compressed_arc_enabled;
210 extern boolean_t zfs_abd_scatter_enabled;
211 extern int dmu_object_alloc_chunk_shift;
212 extern boolean_t zfs_force_some_double_word_sm_entries;
213 extern unsigned long zfs_reconstruct_indirect_damage_fraction;
214 
215 static ztest_shared_opts_t *ztest_shared_opts;
216 static ztest_shared_opts_t ztest_opts;
217 static char *ztest_wkeydata = "abcdefghijklmnopqrstuvwxyz012345";
218 
219 typedef struct ztest_shared_ds {
220 	uint64_t	zd_seq;
221 } ztest_shared_ds_t;
222 
223 static ztest_shared_ds_t *ztest_shared_ds;
224 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
225 
226 #define	BT_MAGIC	0x123456789abcdefULL
227 #define	MAXFAULTS() \
228 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
229 
230 enum ztest_io_type {
231 	ZTEST_IO_WRITE_TAG,
232 	ZTEST_IO_WRITE_PATTERN,
233 	ZTEST_IO_WRITE_ZEROES,
234 	ZTEST_IO_TRUNCATE,
235 	ZTEST_IO_SETATTR,
236 	ZTEST_IO_REWRITE,
237 	ZTEST_IO_TYPES
238 };
239 
240 typedef struct ztest_block_tag {
241 	uint64_t	bt_magic;
242 	uint64_t	bt_objset;
243 	uint64_t	bt_object;
244 	uint64_t	bt_dnodesize;
245 	uint64_t	bt_offset;
246 	uint64_t	bt_gen;
247 	uint64_t	bt_txg;
248 	uint64_t	bt_crtxg;
249 } ztest_block_tag_t;
250 
251 typedef struct bufwad {
252 	uint64_t	bw_index;
253 	uint64_t	bw_txg;
254 	uint64_t	bw_data;
255 } bufwad_t;
256 
257 /*
258  * It would be better to use a rangelock_t per object.  Unfortunately
259  * the rangelock_t is not a drop-in replacement for rl_t, because we
260  * still need to map from object ID to rangelock_t.
261  */
262 typedef enum {
263 	RL_READER,
264 	RL_WRITER,
265 	RL_APPEND
266 } rl_type_t;
267 
268 typedef struct rll {
269 	void		*rll_writer;
270 	int		rll_readers;
271 	kmutex_t	rll_lock;
272 	kcondvar_t	rll_cv;
273 } rll_t;
274 
275 typedef struct rl {
276 	uint64_t	rl_object;
277 	uint64_t	rl_offset;
278 	uint64_t	rl_size;
279 	rll_t		*rl_lock;
280 } rl_t;
281 
282 #define	ZTEST_RANGE_LOCKS	64
283 #define	ZTEST_OBJECT_LOCKS	64
284 
285 /*
286  * Object descriptor.  Used as a template for object lookup/create/remove.
287  */
288 typedef struct ztest_od {
289 	uint64_t	od_dir;
290 	uint64_t	od_object;
291 	dmu_object_type_t od_type;
292 	dmu_object_type_t od_crtype;
293 	uint64_t	od_blocksize;
294 	uint64_t	od_crblocksize;
295 	uint64_t	od_crdnodesize;
296 	uint64_t	od_gen;
297 	uint64_t	od_crgen;
298 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
299 } ztest_od_t;
300 
301 /*
302  * Per-dataset state.
303  */
304 typedef struct ztest_ds {
305 	ztest_shared_ds_t *zd_shared;
306 	objset_t	*zd_os;
307 	krwlock_t	zd_zilog_lock;
308 	zilog_t		*zd_zilog;
309 	ztest_od_t	*zd_od;		/* debugging aid */
310 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
311 	kmutex_t	zd_dirobj_lock;
312 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
313 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
314 } ztest_ds_t;
315 
316 /*
317  * Per-iteration state.
318  */
319 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
320 
321 typedef struct ztest_info {
322 	ztest_func_t	*zi_func;	/* test function */
323 	uint64_t	zi_iters;	/* iterations per execution */
324 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
325 } ztest_info_t;
326 
327 typedef struct ztest_shared_callstate {
328 	uint64_t	zc_count;	/* per-pass count */
329 	uint64_t	zc_time;	/* per-pass time */
330 	uint64_t	zc_next;	/* next time to call this function */
331 } ztest_shared_callstate_t;
332 
333 static ztest_shared_callstate_t *ztest_shared_callstate;
334 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
335 
336 /*
337  * Note: these aren't static because we want dladdr() to work.
338  */
339 ztest_func_t ztest_dmu_read_write;
340 ztest_func_t ztest_dmu_write_parallel;
341 ztest_func_t ztest_dmu_object_alloc_free;
342 ztest_func_t ztest_dmu_object_next_chunk;
343 ztest_func_t ztest_dmu_commit_callbacks;
344 ztest_func_t ztest_zap;
345 ztest_func_t ztest_zap_parallel;
346 ztest_func_t ztest_zil_commit;
347 ztest_func_t ztest_zil_remount;
348 ztest_func_t ztest_dmu_read_write_zcopy;
349 ztest_func_t ztest_dmu_objset_create_destroy;
350 ztest_func_t ztest_dmu_prealloc;
351 ztest_func_t ztest_fzap;
352 ztest_func_t ztest_dmu_snapshot_create_destroy;
353 ztest_func_t ztest_dsl_prop_get_set;
354 ztest_func_t ztest_spa_prop_get_set;
355 ztest_func_t ztest_spa_create_destroy;
356 ztest_func_t ztest_fault_inject;
357 ztest_func_t ztest_ddt_repair;
358 ztest_func_t ztest_dmu_snapshot_hold;
359 ztest_func_t ztest_mmp_enable_disable;
360 ztest_func_t ztest_scrub;
361 ztest_func_t ztest_dsl_dataset_promote_busy;
362 ztest_func_t ztest_vdev_attach_detach;
363 ztest_func_t ztest_vdev_LUN_growth;
364 ztest_func_t ztest_vdev_add_remove;
365 ztest_func_t ztest_vdev_class_add;
366 ztest_func_t ztest_vdev_aux_add_remove;
367 ztest_func_t ztest_split_pool;
368 ztest_func_t ztest_reguid;
369 ztest_func_t ztest_spa_upgrade;
370 ztest_func_t ztest_fletcher;
371 ztest_func_t ztest_fletcher_incr;
372 ztest_func_t ztest_device_removal;
373 ztest_func_t ztest_remap_blocks;
374 ztest_func_t ztest_spa_checkpoint_create_discard;
375 ztest_func_t ztest_initialize;
376 ztest_func_t ztest_trim;
377 ztest_func_t ztest_verify_dnode_bt;
378 
379 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
380 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
381 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
382 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
383 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
384 
385 ztest_info_t ztest_info[] = {
386 	{ ztest_dmu_read_write,			1,	&zopt_always	},
387 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
388 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
389 	{ ztest_dmu_object_next_chunk,		1,	&zopt_sometimes	},
390 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
391 	{ ztest_zap,				30,	&zopt_always	},
392 	{ ztest_zap_parallel,			100,	&zopt_always	},
393 	{ ztest_split_pool,			1,	&zopt_always	},
394 	{ ztest_zil_commit,			1,	&zopt_incessant	},
395 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
396 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
397 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
398 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
399 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
400 #if 0
401 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
402 #endif
403 	{ ztest_fzap,				1,	&zopt_sometimes	},
404 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
405 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
406 	{ ztest_fault_inject,			1,	&zopt_incessant	},
407 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
408 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
409 	{ ztest_mmp_enable_disable,		1,	&zopt_sometimes	},
410 	{ ztest_reguid,				1,	&zopt_rarely	},
411 	{ ztest_scrub,				1,	&zopt_often	},
412 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
413 	{ ztest_fletcher,			1,	&zopt_rarely	},
414 	{ ztest_fletcher_incr,			1,	&zopt_rarely	},
415 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
416 	{ ztest_vdev_attach_detach,		1,	&zopt_incessant	},
417 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
418 	{ ztest_vdev_add_remove,		1,
419 	    &ztest_opts.zo_vdevtime				},
420 	{ ztest_vdev_class_add,			1,
421 	    &ztest_opts.zo_vdevtime				},
422 	{ ztest_vdev_aux_add_remove,		1,
423 	    &ztest_opts.zo_vdevtime				},
424 	{ ztest_device_removal,			1,	&zopt_sometimes	},
425 	{ ztest_remap_blocks,			1,	&zopt_sometimes },
426 	{ ztest_spa_checkpoint_create_discard,	1,	&zopt_rarely	},
427 	{ ztest_initialize,			1,	&zopt_sometimes },
428 	{ ztest_trim,				1,	&zopt_sometimes },
429 	{ ztest_verify_dnode_bt,		1,	&zopt_sometimes }
430 };
431 
432 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
433 
434 /*
435  * The following struct is used to hold a list of uncalled commit callbacks.
436  * The callbacks are ordered by txg number.
437  */
438 typedef struct ztest_cb_list {
439 	kmutex_t zcl_callbacks_lock;
440 	list_t	zcl_callbacks;
441 } ztest_cb_list_t;
442 
443 /*
444  * Stuff we need to share writably between parent and child.
445  */
446 typedef struct ztest_shared {
447 	boolean_t	zs_do_init;
448 	hrtime_t	zs_proc_start;
449 	hrtime_t	zs_proc_stop;
450 	hrtime_t	zs_thread_start;
451 	hrtime_t	zs_thread_stop;
452 	hrtime_t	zs_thread_kill;
453 	uint64_t	zs_enospc_count;
454 	uint64_t	zs_vdev_next_leaf;
455 	uint64_t	zs_vdev_aux;
456 	uint64_t	zs_alloc;
457 	uint64_t	zs_space;
458 	uint64_t	zs_splits;
459 	uint64_t	zs_mirrors;
460 	uint64_t	zs_metaslab_sz;
461 	uint64_t	zs_metaslab_df_alloc_threshold;
462 	uint64_t	zs_guid;
463 } ztest_shared_t;
464 
465 #define	ID_PARALLEL	-1ULL
466 
467 static char ztest_dev_template[] = "%s/%s.%llua";
468 static char ztest_aux_template[] = "%s/%s.%s.%llu";
469 ztest_shared_t *ztest_shared;
470 
471 static spa_t *ztest_spa = NULL;
472 static ztest_ds_t *ztest_ds;
473 
474 static kmutex_t ztest_vdev_lock;
475 static boolean_t ztest_device_removal_active = B_FALSE;
476 static kmutex_t ztest_checkpoint_lock;
477 
478 /*
479  * The ztest_name_lock protects the pool and dataset namespace used by
480  * the individual tests. To modify the namespace, consumers must grab
481  * this lock as writer. Grabbing the lock as reader will ensure that the
482  * namespace does not change while the lock is held.
483  */
484 static krwlock_t ztest_name_lock;
485 
486 static boolean_t ztest_dump_core = B_TRUE;
487 static boolean_t ztest_exiting;
488 
489 /* Global commit callback list */
490 static ztest_cb_list_t zcl;
491 
492 enum ztest_object {
493 	ZTEST_META_DNODE = 0,
494 	ZTEST_DIROBJ,
495 	ZTEST_OBJECTS
496 };
497 
498 static void usage(boolean_t) __NORETURN;
499 
500 /*
501  * These libumem hooks provide a reasonable set of defaults for the allocator's
502  * debugging facilities.
503  */
504 const char *
505 _umem_debug_init()
506 {
507 	return ("default,verbose"); /* $UMEM_DEBUG setting */
508 }
509 
510 const char *
511 _umem_logging_init(void)
512 {
513 	return ("fail,contents"); /* $UMEM_LOGGING setting */
514 }
515 
516 #define	FATAL_MSG_SZ	1024
517 
518 char *fatal_msg;
519 
520 static void
521 fatal(int do_perror, char *message, ...)
522 {
523 	va_list args;
524 	int save_errno = errno;
525 	char buf[FATAL_MSG_SZ];
526 
527 	(void) fflush(stdout);
528 
529 	va_start(args, message);
530 	(void) sprintf(buf, "ztest: ");
531 	/* LINTED */
532 	(void) vsprintf(buf + strlen(buf), message, args);
533 	va_end(args);
534 	if (do_perror) {
535 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
536 		    ": %s", strerror(save_errno));
537 	}
538 	(void) fprintf(stderr, "%s\n", buf);
539 	fatal_msg = buf;			/* to ease debugging */
540 	if (ztest_dump_core)
541 		abort();
542 	exit(3);
543 }
544 
545 static int
546 str2shift(const char *buf)
547 {
548 	const char *ends = "BKMGTPEZ";
549 	int i;
550 
551 	if (buf[0] == '\0')
552 		return (0);
553 	for (i = 0; i < strlen(ends); i++) {
554 		if (toupper(buf[0]) == ends[i])
555 			break;
556 	}
557 	if (i == strlen(ends)) {
558 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
559 		    buf);
560 		usage(B_FALSE);
561 	}
562 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
563 		return (10*i);
564 	}
565 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
566 	usage(B_FALSE);
567 	/* NOTREACHED */
568 }
569 
570 static uint64_t
571 nicenumtoull(const char *buf)
572 {
573 	char *end;
574 	uint64_t val;
575 
576 	val = strtoull(buf, &end, 0);
577 	if (end == buf) {
578 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
579 		usage(B_FALSE);
580 	} else if (end[0] == '.') {
581 		double fval = strtod(buf, &end);
582 		fval *= pow(2, str2shift(end));
583 		if (fval > UINT64_MAX) {
584 			(void) fprintf(stderr, "ztest: value too large: %s\n",
585 			    buf);
586 			usage(B_FALSE);
587 		}
588 		val = (uint64_t)fval;
589 	} else {
590 		int shift = str2shift(end);
591 		if (shift >= 64 || (val << shift) >> shift != val) {
592 			(void) fprintf(stderr, "ztest: value too large: %s\n",
593 			    buf);
594 			usage(B_FALSE);
595 		}
596 		val <<= shift;
597 	}
598 	return (val);
599 }
600 
601 static void
602 usage(boolean_t requested)
603 {
604 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
605 
606 	char nice_vdev_size[NN_NUMBUF_SZ];
607 	char nice_force_ganging[NN_NUMBUF_SZ];
608 	FILE *fp = requested ? stdout : stderr;
609 
610 	nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
611 	nicenum(zo->zo_metaslab_force_ganging, nice_force_ganging,
612 	    sizeof (nice_force_ganging));
613 
614 	(void) fprintf(fp, "Usage: %s\n"
615 	    "\t[-v vdevs (default: %llu)]\n"
616 	    "\t[-s size_of_each_vdev (default: %s)]\n"
617 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
618 	    "\t[-m mirror_copies (default: %d)]\n"
619 	    "\t[-r raidz_disks (default: %d)]\n"
620 	    "\t[-R raidz_parity (default: %d)]\n"
621 	    "\t[-d datasets (default: %d)]\n"
622 	    "\t[-t threads (default: %d)]\n"
623 	    "\t[-g gang_block_threshold (default: %s)]\n"
624 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
625 	    "\t[-k kill_percentage (default: %llu%%)]\n"
626 	    "\t[-p pool_name (default: %s)]\n"
627 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
628 	    "\t[-M] Multi-host simulate pool imported on remote host\n"
629 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
630 	    "\t[-E] use existing pool instead of creating new one\n"
631 	    "\t[-T time (default: %llu sec)] total run time\n"
632 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
633 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
634 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
635 	    "\t[-C vdev class state (default: random)] special=on|off|random\n"
636 	    "\t[-o variable=value] ... set global variable to an unsigned\n"
637 	    "\t    32-bit integer value\n"
638 	    "\t[-h] (print help)\n"
639 	    "",
640 	    zo->zo_pool,
641 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
642 	    nice_vdev_size,				/* -s */
643 	    zo->zo_ashift,				/* -a */
644 	    zo->zo_mirrors,				/* -m */
645 	    zo->zo_raidz,				/* -r */
646 	    zo->zo_raidz_parity,			/* -R */
647 	    zo->zo_datasets,				/* -d */
648 	    zo->zo_threads,				/* -t */
649 	    nice_force_ganging,				/* -g */
650 	    zo->zo_init,				/* -i */
651 	    (u_longlong_t)zo->zo_killrate,		/* -k */
652 	    zo->zo_pool,				/* -p */
653 	    zo->zo_dir,					/* -f */
654 	    (u_longlong_t)zo->zo_time,			/* -T */
655 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
656 	    (u_longlong_t)zo->zo_passtime);
657 	exit(requested ? 0 : 1);
658 }
659 
660 
661 static void
662 ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
663 {
664 	char name[32];
665 	char *value;
666 	int state = ZTEST_VDEV_CLASS_RND;
667 
668 	(void) strlcpy(name, input, sizeof (name));
669 
670 	value = strchr(name, '=');
671 	if (value == NULL) {
672 		(void) fprintf(stderr, "missing value in property=value "
673 		    "'-C' argument (%s)\n", input);
674 		usage(B_FALSE);
675 	}
676 	*(value) = '\0';
677 	value++;
678 
679 	if (strcmp(value, "on") == 0) {
680 		state = ZTEST_VDEV_CLASS_ON;
681 	} else if (strcmp(value, "off") == 0) {
682 		state = ZTEST_VDEV_CLASS_OFF;
683 	} else if (strcmp(value, "random") == 0) {
684 		state = ZTEST_VDEV_CLASS_RND;
685 	} else {
686 		(void) fprintf(stderr, "invalid property value '%s'\n", value);
687 		usage(B_FALSE);
688 	}
689 
690 	if (strcmp(name, "special") == 0) {
691 		zo->zo_special_vdevs = state;
692 	} else {
693 		(void) fprintf(stderr, "invalid property name '%s'\n", name);
694 		usage(B_FALSE);
695 	}
696 	if (zo->zo_verbose >= 3)
697 		(void) printf("%s vdev state is '%s'\n", name, value);
698 }
699 
700 static void
701 process_options(int argc, char **argv)
702 {
703 	char *path;
704 	ztest_shared_opts_t *zo = &ztest_opts;
705 
706 	int opt;
707 	uint64_t value;
708 	char altdir[MAXNAMELEN] = { 0 };
709 
710 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
711 
712 	while ((opt = getopt(argc, argv,
713 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:C:o:")) != EOF) {
714 		value = 0;
715 		switch (opt) {
716 		case 'v':
717 		case 's':
718 		case 'a':
719 		case 'm':
720 		case 'r':
721 		case 'R':
722 		case 'd':
723 		case 't':
724 		case 'g':
725 		case 'i':
726 		case 'k':
727 		case 'T':
728 		case 'P':
729 		case 'F':
730 			value = nicenumtoull(optarg);
731 		}
732 		switch (opt) {
733 		case 'v':
734 			zo->zo_vdevs = value;
735 			break;
736 		case 's':
737 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
738 			break;
739 		case 'a':
740 			zo->zo_ashift = value;
741 			break;
742 		case 'm':
743 			zo->zo_mirrors = value;
744 			break;
745 		case 'r':
746 			zo->zo_raidz = MAX(1, value);
747 			break;
748 		case 'R':
749 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
750 			break;
751 		case 'd':
752 			zo->zo_datasets = MAX(1, value);
753 			break;
754 		case 't':
755 			zo->zo_threads = MAX(1, value);
756 			break;
757 		case 'g':
758 			zo->zo_metaslab_force_ganging =
759 			    MAX(SPA_MINBLOCKSIZE << 1, value);
760 			break;
761 		case 'i':
762 			zo->zo_init = value;
763 			break;
764 		case 'k':
765 			zo->zo_killrate = value;
766 			break;
767 		case 'p':
768 			(void) strlcpy(zo->zo_pool, optarg,
769 			    sizeof (zo->zo_pool));
770 			break;
771 		case 'f':
772 			path = realpath(optarg, NULL);
773 			if (path == NULL) {
774 				(void) fprintf(stderr, "error: %s: %s\n",
775 				    optarg, strerror(errno));
776 				usage(B_FALSE);
777 			} else {
778 				(void) strlcpy(zo->zo_dir, path,
779 				    sizeof (zo->zo_dir));
780 			}
781 			break;
782 		case 'M':
783 			zo->zo_mmp_test = 1;
784 			break;
785 		case 'V':
786 			zo->zo_verbose++;
787 			break;
788 		case 'E':
789 			zo->zo_init = 0;
790 			break;
791 		case 'T':
792 			zo->zo_time = value;
793 			break;
794 		case 'P':
795 			zo->zo_passtime = MAX(1, value);
796 			break;
797 		case 'F':
798 			zo->zo_maxloops = MAX(1, value);
799 			break;
800 		case 'B':
801 			(void) strlcpy(altdir, optarg, sizeof (altdir));
802 			break;
803 		case 'C':
804 			ztest_parse_name_value(optarg, zo);
805 			break;
806 		case 'o':
807 			if (set_global_var(optarg) != 0)
808 				usage(B_FALSE);
809 			break;
810 		case 'h':
811 			usage(B_TRUE);
812 			break;
813 		case '?':
814 		default:
815 			usage(B_FALSE);
816 			break;
817 		}
818 	}
819 
820 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
821 
822 	zo->zo_vdevtime =
823 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
824 	    UINT64_MAX >> 2);
825 
826 	if (strlen(altdir) > 0) {
827 		char *cmd;
828 		char *realaltdir;
829 		char *bin;
830 		char *ztest;
831 		char *isa;
832 		int isalen;
833 
834 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
835 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
836 
837 		VERIFY(NULL != realpath(getexecname(), cmd));
838 		if (0 != access(altdir, F_OK)) {
839 			ztest_dump_core = B_FALSE;
840 			fatal(B_TRUE, "invalid alternate ztest path: %s",
841 			    altdir);
842 		}
843 		VERIFY(NULL != realpath(altdir, realaltdir));
844 
845 		/*
846 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
847 		 * We want to extract <isa> to determine if we should use
848 		 * 32 or 64 bit binaries.
849 		 */
850 		bin = strstr(cmd, "/usr/bin/");
851 		ztest = strstr(bin, "/ztest");
852 		isa = bin + 9;
853 		isalen = ztest - isa;
854 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
855 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
856 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
857 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
858 
859 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
860 			ztest_dump_core = B_FALSE;
861 			fatal(B_TRUE, "invalid alternate ztest: %s",
862 			    zo->zo_alt_ztest);
863 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
864 			ztest_dump_core = B_FALSE;
865 			fatal(B_TRUE, "invalid alternate lib directory %s",
866 			    zo->zo_alt_libpath);
867 		}
868 
869 		umem_free(cmd, MAXPATHLEN);
870 		umem_free(realaltdir, MAXPATHLEN);
871 	}
872 }
873 
874 static void
875 ztest_kill(ztest_shared_t *zs)
876 {
877 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
878 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
879 
880 	/*
881 	 * Before we kill off ztest, make sure that the config is updated.
882 	 * See comment above spa_write_cachefile().
883 	 */
884 	mutex_enter(&spa_namespace_lock);
885 	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
886 	mutex_exit(&spa_namespace_lock);
887 
888 	zfs_dbgmsg_print(FTAG);
889 	(void) kill(getpid(), SIGKILL);
890 }
891 
892 static uint64_t
893 ztest_random(uint64_t range)
894 {
895 	uint64_t r;
896 
897 	ASSERT3S(ztest_fd_rand, >=, 0);
898 
899 	if (range == 0)
900 		return (0);
901 
902 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
903 		fatal(1, "short read from /dev/urandom");
904 
905 	return (r % range);
906 }
907 
908 /* ARGSUSED */
909 static void
910 ztest_record_enospc(const char *s)
911 {
912 	ztest_shared->zs_enospc_count++;
913 }
914 
915 static uint64_t
916 ztest_get_ashift(void)
917 {
918 	if (ztest_opts.zo_ashift == 0)
919 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
920 	return (ztest_opts.zo_ashift);
921 }
922 
923 static nvlist_t *
924 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
925 {
926 	char pathbuf[MAXPATHLEN];
927 	uint64_t vdev;
928 	nvlist_t *file;
929 
930 	if (ashift == 0)
931 		ashift = ztest_get_ashift();
932 
933 	if (path == NULL) {
934 		path = pathbuf;
935 
936 		if (aux != NULL) {
937 			vdev = ztest_shared->zs_vdev_aux;
938 			(void) snprintf(path, sizeof (pathbuf),
939 			    ztest_aux_template, ztest_opts.zo_dir,
940 			    pool == NULL ? ztest_opts.zo_pool : pool,
941 			    aux, vdev);
942 		} else {
943 			vdev = ztest_shared->zs_vdev_next_leaf++;
944 			(void) snprintf(path, sizeof (pathbuf),
945 			    ztest_dev_template, ztest_opts.zo_dir,
946 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
947 		}
948 	}
949 
950 	if (size != 0) {
951 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
952 		if (fd == -1)
953 			fatal(1, "can't open %s", path);
954 		if (ftruncate(fd, size) != 0)
955 			fatal(1, "can't ftruncate %s", path);
956 		(void) close(fd);
957 	}
958 
959 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
960 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
961 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
962 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
963 
964 	return (file);
965 }
966 
967 static nvlist_t *
968 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
969     uint64_t ashift, int r)
970 {
971 	nvlist_t *raidz, **child;
972 	int c;
973 
974 	if (r < 2)
975 		return (make_vdev_file(path, aux, pool, size, ashift));
976 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
977 
978 	for (c = 0; c < r; c++)
979 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
980 
981 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
982 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
983 	    VDEV_TYPE_RAIDZ) == 0);
984 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
985 	    ztest_opts.zo_raidz_parity) == 0);
986 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
987 	    child, r) == 0);
988 
989 	for (c = 0; c < r; c++)
990 		nvlist_free(child[c]);
991 
992 	umem_free(child, r * sizeof (nvlist_t *));
993 
994 	return (raidz);
995 }
996 
997 static nvlist_t *
998 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
999     uint64_t ashift, int r, int m)
1000 {
1001 	nvlist_t *mirror, **child;
1002 	int c;
1003 
1004 	if (m < 1)
1005 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
1006 
1007 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1008 
1009 	for (c = 0; c < m; c++)
1010 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
1011 
1012 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
1013 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
1014 	    VDEV_TYPE_MIRROR) == 0);
1015 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1016 	    child, m) == 0);
1017 
1018 	for (c = 0; c < m; c++)
1019 		nvlist_free(child[c]);
1020 
1021 	umem_free(child, m * sizeof (nvlist_t *));
1022 
1023 	return (mirror);
1024 }
1025 
1026 static nvlist_t *
1027 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
1028     const char *class, int r, int m, int t)
1029 {
1030 	nvlist_t *root, **child;
1031 	int c;
1032 	boolean_t log;
1033 
1034 	ASSERT(t > 0);
1035 
1036 	log = (class != NULL && strcmp(class, "log") == 0);
1037 
1038 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1039 
1040 	for (c = 0; c < t; c++) {
1041 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1042 		    r, m);
1043 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
1044 		    log) == 0);
1045 
1046 		if (class != NULL && class[0] != '\0') {
1047 			ASSERT(m > 1 || log);   /* expecting a mirror */
1048 			VERIFY(nvlist_add_string(child[c],
1049 			    ZPOOL_CONFIG_ALLOCATION_BIAS, class) == 0);
1050 		}
1051 	}
1052 
1053 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
1054 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
1055 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1056 	    child, t) == 0);
1057 
1058 	for (c = 0; c < t; c++)
1059 		nvlist_free(child[c]);
1060 
1061 	umem_free(child, t * sizeof (nvlist_t *));
1062 
1063 	return (root);
1064 }
1065 
1066 /*
1067  * Find a random spa version. Returns back a random spa version in the
1068  * range [initial_version, SPA_VERSION_FEATURES].
1069  */
1070 static uint64_t
1071 ztest_random_spa_version(uint64_t initial_version)
1072 {
1073 	uint64_t version = initial_version;
1074 
1075 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1076 		version = version +
1077 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1078 	}
1079 
1080 	if (version > SPA_VERSION_BEFORE_FEATURES)
1081 		version = SPA_VERSION_FEATURES;
1082 
1083 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1084 	return (version);
1085 }
1086 
1087 static int
1088 ztest_random_blocksize(void)
1089 {
1090 	uint64_t block_shift;
1091 
1092 	ASSERT(ztest_spa->spa_max_ashift != 0);
1093 
1094 	/*
1095 	 * Choose a block size >= the ashift.
1096 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1097 	 */
1098 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1099 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1100 		maxbs = 20;
1101 	block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1102 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1103 }
1104 
1105 static int
1106 ztest_random_dnodesize(void)
1107 {
1108 	int slots;
1109 	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1110 
1111 	if (max_slots == DNODE_MIN_SLOTS)
1112 		return (DNODE_MIN_SIZE);
1113 
1114 	/*
1115 	 * Weight the random distribution more heavily toward smaller
1116 	 * dnode sizes since that is more likely to reflect real-world
1117 	 * usage.
1118 	 */
1119 	ASSERT3U(max_slots, >, 4);
1120 	switch (ztest_random(10)) {
1121 	case 0:
1122 		slots = 5 + ztest_random(max_slots - 4);
1123 		break;
1124 	case 1 ... 4:
1125 		slots = 2 + ztest_random(3);
1126 		break;
1127 	default:
1128 		slots = 1;
1129 		break;
1130 	}
1131 
1132 	return (slots << DNODE_SHIFT);
1133 }
1134 
1135 static int
1136 ztest_random_ibshift(void)
1137 {
1138 	return (DN_MIN_INDBLKSHIFT +
1139 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1140 }
1141 
1142 static uint64_t
1143 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1144 {
1145 	uint64_t top;
1146 	vdev_t *rvd = spa->spa_root_vdev;
1147 	vdev_t *tvd;
1148 
1149 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1150 
1151 	do {
1152 		top = ztest_random(rvd->vdev_children);
1153 		tvd = rvd->vdev_child[top];
1154 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1155 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1156 
1157 	return (top);
1158 }
1159 
1160 static uint64_t
1161 ztest_random_dsl_prop(zfs_prop_t prop)
1162 {
1163 	uint64_t value;
1164 
1165 	do {
1166 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1167 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1168 
1169 	return (value);
1170 }
1171 
1172 static int
1173 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1174     boolean_t inherit)
1175 {
1176 	const char *propname = zfs_prop_to_name(prop);
1177 	const char *valname;
1178 	char setpoint[MAXPATHLEN];
1179 	uint64_t curval;
1180 	int error;
1181 
1182 	error = dsl_prop_set_int(osname, propname,
1183 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1184 
1185 	if (error == ENOSPC) {
1186 		ztest_record_enospc(FTAG);
1187 		return (error);
1188 	}
1189 	ASSERT0(error);
1190 
1191 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1192 
1193 	if (ztest_opts.zo_verbose >= 6) {
1194 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1195 		(void) printf("%s %s = %s at '%s'\n",
1196 		    osname, propname, valname, setpoint);
1197 	}
1198 
1199 	return (error);
1200 }
1201 
1202 static int
1203 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1204 {
1205 	spa_t *spa = ztest_spa;
1206 	nvlist_t *props = NULL;
1207 	int error;
1208 
1209 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1210 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1211 
1212 	error = spa_prop_set(spa, props);
1213 
1214 	nvlist_free(props);
1215 
1216 	if (error == ENOSPC) {
1217 		ztest_record_enospc(FTAG);
1218 		return (error);
1219 	}
1220 	ASSERT0(error);
1221 
1222 	return (error);
1223 }
1224 
1225 static int
1226 ztest_dmu_objset_own(const char *name, dmu_objset_type_t type,
1227     boolean_t readonly, boolean_t decrypt, void *tag, objset_t **osp)
1228 {
1229 	int err;
1230 
1231 	err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1232 	if (decrypt && err == EACCES) {
1233 		char ddname[ZFS_MAX_DATASET_NAME_LEN];
1234 		dsl_crypto_params_t *dcp;
1235 		nvlist_t *crypto_args = fnvlist_alloc();
1236 		char *cp = NULL;
1237 
1238 		/* spa_keystore_load_wkey() expects a dsl dir name */
1239 		(void) strcpy(ddname, name);
1240 		cp = strchr(ddname, '@');
1241 		if (cp != NULL)
1242 			*cp = '\0';
1243 
1244 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
1245 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
1246 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, NULL,
1247 		    crypto_args, &dcp));
1248 		err = spa_keystore_load_wkey(ddname, dcp, B_FALSE);
1249 		dsl_crypto_params_free(dcp, B_FALSE);
1250 		fnvlist_free(crypto_args);
1251 
1252 		if (err != 0)
1253 			return (err);
1254 
1255 		err = dmu_objset_own(name, type, readonly, decrypt, tag, osp);
1256 	}
1257 
1258 	return (err);
1259 }
1260 
1261 static void
1262 ztest_rll_init(rll_t *rll)
1263 {
1264 	rll->rll_writer = NULL;
1265 	rll->rll_readers = 0;
1266 	mutex_init(&rll->rll_lock, NULL, USYNC_THREAD, NULL);
1267 	cv_init(&rll->rll_cv, NULL, USYNC_THREAD, NULL);
1268 }
1269 
1270 static void
1271 ztest_rll_destroy(rll_t *rll)
1272 {
1273 	ASSERT(rll->rll_writer == NULL);
1274 	ASSERT(rll->rll_readers == 0);
1275 	mutex_destroy(&rll->rll_lock);
1276 	cv_destroy(&rll->rll_cv);
1277 }
1278 
1279 static void
1280 ztest_rll_lock(rll_t *rll, rl_type_t type)
1281 {
1282 	mutex_enter(&rll->rll_lock);
1283 
1284 	if (type == RL_READER) {
1285 		while (rll->rll_writer != NULL)
1286 			cv_wait(&rll->rll_cv, &rll->rll_lock);
1287 		rll->rll_readers++;
1288 	} else {
1289 		while (rll->rll_writer != NULL || rll->rll_readers)
1290 			cv_wait(&rll->rll_cv, &rll->rll_lock);
1291 		rll->rll_writer = curthread;
1292 	}
1293 
1294 	mutex_exit(&rll->rll_lock);
1295 }
1296 
1297 static void
1298 ztest_rll_unlock(rll_t *rll)
1299 {
1300 	mutex_enter(&rll->rll_lock);
1301 
1302 	if (rll->rll_writer) {
1303 		ASSERT(rll->rll_readers == 0);
1304 		rll->rll_writer = NULL;
1305 	} else {
1306 		ASSERT(rll->rll_readers != 0);
1307 		ASSERT(rll->rll_writer == NULL);
1308 		rll->rll_readers--;
1309 	}
1310 
1311 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1312 		cv_broadcast(&rll->rll_cv);
1313 
1314 	mutex_exit(&rll->rll_lock);
1315 }
1316 
1317 static void
1318 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1319 {
1320 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1321 
1322 	ztest_rll_lock(rll, type);
1323 }
1324 
1325 static void
1326 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1327 {
1328 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1329 
1330 	ztest_rll_unlock(rll);
1331 }
1332 
1333 static rl_t *
1334 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1335     uint64_t size, rl_type_t type)
1336 {
1337 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1338 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1339 	rl_t *rl;
1340 
1341 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1342 	rl->rl_object = object;
1343 	rl->rl_offset = offset;
1344 	rl->rl_size = size;
1345 	rl->rl_lock = rll;
1346 
1347 	ztest_rll_lock(rll, type);
1348 
1349 	return (rl);
1350 }
1351 
1352 static void
1353 ztest_range_unlock(rl_t *rl)
1354 {
1355 	rll_t *rll = rl->rl_lock;
1356 
1357 	ztest_rll_unlock(rll);
1358 
1359 	umem_free(rl, sizeof (*rl));
1360 }
1361 
1362 static void
1363 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1364 {
1365 	zd->zd_os = os;
1366 	zd->zd_zilog = dmu_objset_zil(os);
1367 	zd->zd_shared = szd;
1368 	dmu_objset_name(os, zd->zd_name);
1369 
1370 	if (zd->zd_shared != NULL)
1371 		zd->zd_shared->zd_seq = 0;
1372 
1373 	rw_init(&zd->zd_zilog_lock, NULL, USYNC_THREAD, NULL);
1374 	mutex_init(&zd->zd_dirobj_lock, NULL, USYNC_THREAD, NULL);
1375 
1376 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1377 		ztest_rll_init(&zd->zd_object_lock[l]);
1378 
1379 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1380 		ztest_rll_init(&zd->zd_range_lock[l]);
1381 }
1382 
1383 static void
1384 ztest_zd_fini(ztest_ds_t *zd)
1385 {
1386 	mutex_destroy(&zd->zd_dirobj_lock);
1387 
1388 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1389 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1390 
1391 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1392 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1393 }
1394 
1395 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1396 
1397 static uint64_t
1398 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1399 {
1400 	uint64_t txg;
1401 	int error;
1402 
1403 	/*
1404 	 * Attempt to assign tx to some transaction group.
1405 	 */
1406 	error = dmu_tx_assign(tx, txg_how);
1407 	if (error) {
1408 		if (error == ERESTART) {
1409 			ASSERT(txg_how == TXG_NOWAIT);
1410 			dmu_tx_wait(tx);
1411 		} else {
1412 			ASSERT3U(error, ==, ENOSPC);
1413 			ztest_record_enospc(tag);
1414 		}
1415 		dmu_tx_abort(tx);
1416 		return (0);
1417 	}
1418 	txg = dmu_tx_get_txg(tx);
1419 	ASSERT(txg != 0);
1420 	return (txg);
1421 }
1422 
1423 static void
1424 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1425 {
1426 	uint64_t *ip = buf;
1427 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1428 
1429 	while (ip < ip_end)
1430 		*ip++ = value;
1431 }
1432 
1433 static boolean_t
1434 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1435 {
1436 	uint64_t *ip = buf;
1437 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1438 	uint64_t diff = 0;
1439 
1440 	while (ip < ip_end)
1441 		diff |= (value - *ip++);
1442 
1443 	return (diff == 0);
1444 }
1445 
1446 static void
1447 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1448     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1449     uint64_t crtxg)
1450 {
1451 	bt->bt_magic = BT_MAGIC;
1452 	bt->bt_objset = dmu_objset_id(os);
1453 	bt->bt_object = object;
1454 	bt->bt_dnodesize = dnodesize;
1455 	bt->bt_offset = offset;
1456 	bt->bt_gen = gen;
1457 	bt->bt_txg = txg;
1458 	bt->bt_crtxg = crtxg;
1459 }
1460 
1461 static void
1462 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1463     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1464     uint64_t crtxg)
1465 {
1466 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1467 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1468 	ASSERT3U(bt->bt_object, ==, object);
1469 	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1470 	ASSERT3U(bt->bt_offset, ==, offset);
1471 	ASSERT3U(bt->bt_gen, <=, gen);
1472 	ASSERT3U(bt->bt_txg, <=, txg);
1473 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1474 }
1475 
1476 static ztest_block_tag_t *
1477 ztest_bt_bonus(dmu_buf_t *db)
1478 {
1479 	dmu_object_info_t doi;
1480 	ztest_block_tag_t *bt;
1481 
1482 	dmu_object_info_from_db(db, &doi);
1483 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1484 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1485 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1486 
1487 	return (bt);
1488 }
1489 
1490 /*
1491  * Generate a token to fill up unused bonus buffer space.  Try to make
1492  * it unique to the object, generation, and offset to verify that data
1493  * is not getting overwritten by data from other dnodes.
1494  */
1495 #define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset)	\
1496 	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1497 
1498 /*
1499  * Fill up the unused bonus buffer region before the block tag with a
1500  * verifiable pattern. Filling the whole bonus area with non-zero data
1501  * helps ensure that all dnode traversal code properly skips the
1502  * interior regions of large dnodes.
1503  */
1504 void
1505 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1506     objset_t *os, uint64_t gen)
1507 {
1508 	uint64_t *bonusp;
1509 
1510 	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1511 
1512 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1513 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1514 		    gen, bonusp - (uint64_t *)db->db_data);
1515 		*bonusp = token;
1516 	}
1517 }
1518 
1519 /*
1520  * Verify that the unused area of a bonus buffer is filled with the
1521  * expected tokens.
1522  */
1523 void
1524 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1525     objset_t *os, uint64_t gen)
1526 {
1527 	uint64_t *bonusp;
1528 
1529 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1530 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1531 		    gen, bonusp - (uint64_t *)db->db_data);
1532 		VERIFY3U(*bonusp, ==, token);
1533 	}
1534 }
1535 
1536 /*
1537  * ZIL logging ops
1538  */
1539 
1540 #define	lrz_type	lr_mode
1541 #define	lrz_blocksize	lr_uid
1542 #define	lrz_ibshift	lr_gid
1543 #define	lrz_bonustype	lr_rdev
1544 #define	lrz_dnodesize	lr_crtime[1]
1545 
1546 static void
1547 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1548 {
1549 	char *name = (void *)(lr + 1);		/* name follows lr */
1550 	size_t namesize = strlen(name) + 1;
1551 	itx_t *itx;
1552 
1553 	if (zil_replaying(zd->zd_zilog, tx))
1554 		return;
1555 
1556 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1557 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1558 	    sizeof (*lr) + namesize - sizeof (lr_t));
1559 
1560 	zil_itx_assign(zd->zd_zilog, itx, tx);
1561 }
1562 
1563 static void
1564 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1565 {
1566 	char *name = (void *)(lr + 1);		/* name follows lr */
1567 	size_t namesize = strlen(name) + 1;
1568 	itx_t *itx;
1569 
1570 	if (zil_replaying(zd->zd_zilog, tx))
1571 		return;
1572 
1573 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1574 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1575 	    sizeof (*lr) + namesize - sizeof (lr_t));
1576 
1577 	itx->itx_oid = object;
1578 	zil_itx_assign(zd->zd_zilog, itx, tx);
1579 }
1580 
1581 static void
1582 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1583 {
1584 	itx_t *itx;
1585 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1586 
1587 	if (zil_replaying(zd->zd_zilog, tx))
1588 		return;
1589 
1590 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1591 		write_state = WR_INDIRECT;
1592 
1593 	itx = zil_itx_create(TX_WRITE,
1594 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1595 
1596 	if (write_state == WR_COPIED &&
1597 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1598 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1599 		zil_itx_destroy(itx);
1600 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1601 		write_state = WR_NEED_COPY;
1602 	}
1603 	itx->itx_private = zd;
1604 	itx->itx_wr_state = write_state;
1605 	itx->itx_sync = (ztest_random(8) == 0);
1606 
1607 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1608 	    sizeof (*lr) - sizeof (lr_t));
1609 
1610 	zil_itx_assign(zd->zd_zilog, itx, tx);
1611 }
1612 
1613 static void
1614 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1615 {
1616 	itx_t *itx;
1617 
1618 	if (zil_replaying(zd->zd_zilog, tx))
1619 		return;
1620 
1621 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1622 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1623 	    sizeof (*lr) - sizeof (lr_t));
1624 
1625 	itx->itx_sync = B_FALSE;
1626 	zil_itx_assign(zd->zd_zilog, itx, tx);
1627 }
1628 
1629 static void
1630 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1631 {
1632 	itx_t *itx;
1633 
1634 	if (zil_replaying(zd->zd_zilog, tx))
1635 		return;
1636 
1637 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1638 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1639 	    sizeof (*lr) - sizeof (lr_t));
1640 
1641 	itx->itx_sync = B_FALSE;
1642 	zil_itx_assign(zd->zd_zilog, itx, tx);
1643 }
1644 
1645 /*
1646  * ZIL replay ops
1647  */
1648 static int
1649 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1650 {
1651 	ztest_ds_t *zd = arg1;
1652 	lr_create_t *lr = arg2;
1653 	char *name = (void *)(lr + 1);		/* name follows lr */
1654 	objset_t *os = zd->zd_os;
1655 	ztest_block_tag_t *bbt;
1656 	dmu_buf_t *db;
1657 	dmu_tx_t *tx;
1658 	uint64_t txg;
1659 	int error = 0;
1660 	int bonuslen;
1661 
1662 	if (byteswap)
1663 		byteswap_uint64_array(lr, sizeof (*lr));
1664 
1665 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1666 	ASSERT(name[0] != '\0');
1667 
1668 	tx = dmu_tx_create(os);
1669 
1670 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1671 
1672 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1673 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1674 	} else {
1675 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1676 	}
1677 
1678 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1679 	if (txg == 0)
1680 		return (ENOSPC);
1681 
1682 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1683 	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1684 
1685 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1686 		if (lr->lr_foid == 0) {
1687 			lr->lr_foid = zap_create_dnsize(os,
1688 			    lr->lrz_type, lr->lrz_bonustype,
1689 			    bonuslen, lr->lrz_dnodesize, tx);
1690 		} else {
1691 			error = zap_create_claim_dnsize(os, lr->lr_foid,
1692 			    lr->lrz_type, lr->lrz_bonustype,
1693 			    bonuslen, lr->lrz_dnodesize, tx);
1694 		}
1695 	} else {
1696 		if (lr->lr_foid == 0) {
1697 			lr->lr_foid = dmu_object_alloc_dnsize(os,
1698 			    lr->lrz_type, 0, lr->lrz_bonustype,
1699 			    bonuslen, lr->lrz_dnodesize, tx);
1700 		} else {
1701 			error = dmu_object_claim_dnsize(os, lr->lr_foid,
1702 			    lr->lrz_type, 0, lr->lrz_bonustype,
1703 			    bonuslen, lr->lrz_dnodesize, tx);
1704 		}
1705 	}
1706 
1707 	if (error) {
1708 		ASSERT3U(error, ==, EEXIST);
1709 		ASSERT(zd->zd_zilog->zl_replay);
1710 		dmu_tx_commit(tx);
1711 		return (error);
1712 	}
1713 
1714 	ASSERT(lr->lr_foid != 0);
1715 
1716 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1717 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1718 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1719 
1720 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1721 	bbt = ztest_bt_bonus(db);
1722 	dmu_buf_will_dirty(db, tx);
1723 	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
1724 	    lr->lr_gen, txg, txg);
1725 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
1726 	dmu_buf_rele(db, FTAG);
1727 
1728 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1729 	    &lr->lr_foid, tx));
1730 
1731 	(void) ztest_log_create(zd, tx, lr);
1732 
1733 	dmu_tx_commit(tx);
1734 
1735 	return (0);
1736 }
1737 
1738 static int
1739 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1740 {
1741 	ztest_ds_t *zd = arg1;
1742 	lr_remove_t *lr = arg2;
1743 	char *name = (void *)(lr + 1);		/* name follows lr */
1744 	objset_t *os = zd->zd_os;
1745 	dmu_object_info_t doi;
1746 	dmu_tx_t *tx;
1747 	uint64_t object, txg;
1748 
1749 	if (byteswap)
1750 		byteswap_uint64_array(lr, sizeof (*lr));
1751 
1752 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1753 	ASSERT(name[0] != '\0');
1754 
1755 	VERIFY3U(0, ==,
1756 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1757 	ASSERT(object != 0);
1758 
1759 	ztest_object_lock(zd, object, RL_WRITER);
1760 
1761 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1762 
1763 	tx = dmu_tx_create(os);
1764 
1765 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1766 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1767 
1768 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1769 	if (txg == 0) {
1770 		ztest_object_unlock(zd, object);
1771 		return (ENOSPC);
1772 	}
1773 
1774 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1775 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1776 	} else {
1777 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1778 	}
1779 
1780 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1781 
1782 	(void) ztest_log_remove(zd, tx, lr, object);
1783 
1784 	dmu_tx_commit(tx);
1785 
1786 	ztest_object_unlock(zd, object);
1787 
1788 	return (0);
1789 }
1790 
1791 static int
1792 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1793 {
1794 	ztest_ds_t *zd = arg1;
1795 	lr_write_t *lr = arg2;
1796 	objset_t *os = zd->zd_os;
1797 	void *data = lr + 1;			/* data follows lr */
1798 	uint64_t offset, length;
1799 	ztest_block_tag_t *bt = data;
1800 	ztest_block_tag_t *bbt;
1801 	uint64_t gen, txg, lrtxg, crtxg;
1802 	dmu_object_info_t doi;
1803 	dmu_tx_t *tx;
1804 	dmu_buf_t *db;
1805 	arc_buf_t *abuf = NULL;
1806 	rl_t *rl;
1807 
1808 	if (byteswap)
1809 		byteswap_uint64_array(lr, sizeof (*lr));
1810 
1811 	offset = lr->lr_offset;
1812 	length = lr->lr_length;
1813 
1814 	/* If it's a dmu_sync() block, write the whole block */
1815 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1816 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1817 		if (length < blocksize) {
1818 			offset -= offset % blocksize;
1819 			length = blocksize;
1820 		}
1821 	}
1822 
1823 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1824 		byteswap_uint64_array(bt, sizeof (*bt));
1825 
1826 	if (bt->bt_magic != BT_MAGIC)
1827 		bt = NULL;
1828 
1829 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1830 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1831 
1832 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1833 
1834 	dmu_object_info_from_db(db, &doi);
1835 
1836 	bbt = ztest_bt_bonus(db);
1837 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1838 	gen = bbt->bt_gen;
1839 	crtxg = bbt->bt_crtxg;
1840 	lrtxg = lr->lr_common.lrc_txg;
1841 
1842 	tx = dmu_tx_create(os);
1843 
1844 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1845 
1846 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1847 	    P2PHASE(offset, length) == 0)
1848 		abuf = dmu_request_arcbuf(db, length);
1849 
1850 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1851 	if (txg == 0) {
1852 		if (abuf != NULL)
1853 			dmu_return_arcbuf(abuf);
1854 		dmu_buf_rele(db, FTAG);
1855 		ztest_range_unlock(rl);
1856 		ztest_object_unlock(zd, lr->lr_foid);
1857 		return (ENOSPC);
1858 	}
1859 
1860 	if (bt != NULL) {
1861 		/*
1862 		 * Usually, verify the old data before writing new data --
1863 		 * but not always, because we also want to verify correct
1864 		 * behavior when the data was not recently read into cache.
1865 		 */
1866 		ASSERT(offset % doi.doi_data_block_size == 0);
1867 		if (ztest_random(4) != 0) {
1868 			int prefetch = ztest_random(2) ?
1869 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1870 			ztest_block_tag_t rbt;
1871 
1872 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1873 			    sizeof (rbt), &rbt, prefetch) == 0);
1874 			if (rbt.bt_magic == BT_MAGIC) {
1875 				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
1876 				    offset, gen, txg, crtxg);
1877 			}
1878 		}
1879 
1880 		/*
1881 		 * Writes can appear to be newer than the bonus buffer because
1882 		 * the ztest_get_data() callback does a dmu_read() of the
1883 		 * open-context data, which may be different than the data
1884 		 * as it was when the write was generated.
1885 		 */
1886 		if (zd->zd_zilog->zl_replay) {
1887 			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
1888 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1889 			    bt->bt_crtxg);
1890 		}
1891 
1892 		/*
1893 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1894 		 * so that all of the usual ASSERTs will work.
1895 		 */
1896 		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
1897 		    crtxg);
1898 	}
1899 
1900 	if (abuf == NULL) {
1901 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1902 	} else {
1903 		bcopy(data, abuf->b_data, length);
1904 		dmu_assign_arcbuf_by_dbuf(db, offset, abuf, tx);
1905 	}
1906 
1907 	(void) ztest_log_write(zd, tx, lr);
1908 
1909 	dmu_buf_rele(db, FTAG);
1910 
1911 	dmu_tx_commit(tx);
1912 
1913 	ztest_range_unlock(rl);
1914 	ztest_object_unlock(zd, lr->lr_foid);
1915 
1916 	return (0);
1917 }
1918 
1919 static int
1920 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
1921 {
1922 	ztest_ds_t *zd = arg1;
1923 	lr_truncate_t *lr = arg2;
1924 	objset_t *os = zd->zd_os;
1925 	dmu_tx_t *tx;
1926 	uint64_t txg;
1927 	rl_t *rl;
1928 
1929 	if (byteswap)
1930 		byteswap_uint64_array(lr, sizeof (*lr));
1931 
1932 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1933 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1934 	    RL_WRITER);
1935 
1936 	tx = dmu_tx_create(os);
1937 
1938 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1939 
1940 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1941 	if (txg == 0) {
1942 		ztest_range_unlock(rl);
1943 		ztest_object_unlock(zd, lr->lr_foid);
1944 		return (ENOSPC);
1945 	}
1946 
1947 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1948 	    lr->lr_length, tx) == 0);
1949 
1950 	(void) ztest_log_truncate(zd, tx, lr);
1951 
1952 	dmu_tx_commit(tx);
1953 
1954 	ztest_range_unlock(rl);
1955 	ztest_object_unlock(zd, lr->lr_foid);
1956 
1957 	return (0);
1958 }
1959 
1960 static int
1961 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
1962 {
1963 	ztest_ds_t *zd = arg1;
1964 	lr_setattr_t *lr = arg2;
1965 	objset_t *os = zd->zd_os;
1966 	dmu_tx_t *tx;
1967 	dmu_buf_t *db;
1968 	ztest_block_tag_t *bbt;
1969 	uint64_t txg, lrtxg, crtxg, dnodesize;
1970 
1971 	if (byteswap)
1972 		byteswap_uint64_array(lr, sizeof (*lr));
1973 
1974 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1975 
1976 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1977 
1978 	tx = dmu_tx_create(os);
1979 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1980 
1981 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1982 	if (txg == 0) {
1983 		dmu_buf_rele(db, FTAG);
1984 		ztest_object_unlock(zd, lr->lr_foid);
1985 		return (ENOSPC);
1986 	}
1987 
1988 	bbt = ztest_bt_bonus(db);
1989 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1990 	crtxg = bbt->bt_crtxg;
1991 	lrtxg = lr->lr_common.lrc_txg;
1992 	dnodesize = bbt->bt_dnodesize;
1993 
1994 	if (zd->zd_zilog->zl_replay) {
1995 		ASSERT(lr->lr_size != 0);
1996 		ASSERT(lr->lr_mode != 0);
1997 		ASSERT(lrtxg != 0);
1998 	} else {
1999 		/*
2000 		 * Randomly change the size and increment the generation.
2001 		 */
2002 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
2003 		    sizeof (*bbt);
2004 		lr->lr_mode = bbt->bt_gen + 1;
2005 		ASSERT(lrtxg == 0);
2006 	}
2007 
2008 	/*
2009 	 * Verify that the current bonus buffer is not newer than our txg.
2010 	 */
2011 	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2012 	    MAX(txg, lrtxg), crtxg);
2013 
2014 	dmu_buf_will_dirty(db, tx);
2015 
2016 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
2017 	ASSERT3U(lr->lr_size, <=, db->db_size);
2018 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
2019 	bbt = ztest_bt_bonus(db);
2020 
2021 	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
2022 	    txg, crtxg);
2023 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
2024 
2025 	dmu_buf_rele(db, FTAG);
2026 
2027 	(void) ztest_log_setattr(zd, tx, lr);
2028 
2029 	dmu_tx_commit(tx);
2030 
2031 	ztest_object_unlock(zd, lr->lr_foid);
2032 
2033 	return (0);
2034 }
2035 
2036 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
2037 	NULL,			/* 0 no such transaction type */
2038 	ztest_replay_create,	/* TX_CREATE */
2039 	NULL,			/* TX_MKDIR */
2040 	NULL,			/* TX_MKXATTR */
2041 	NULL,			/* TX_SYMLINK */
2042 	ztest_replay_remove,	/* TX_REMOVE */
2043 	NULL,			/* TX_RMDIR */
2044 	NULL,			/* TX_LINK */
2045 	NULL,			/* TX_RENAME */
2046 	ztest_replay_write,	/* TX_WRITE */
2047 	ztest_replay_truncate,	/* TX_TRUNCATE */
2048 	ztest_replay_setattr,	/* TX_SETATTR */
2049 	NULL,			/* TX_ACL */
2050 	NULL,			/* TX_CREATE_ACL */
2051 	NULL,			/* TX_CREATE_ATTR */
2052 	NULL,			/* TX_CREATE_ACL_ATTR */
2053 	NULL,			/* TX_MKDIR_ACL */
2054 	NULL,			/* TX_MKDIR_ATTR */
2055 	NULL,			/* TX_MKDIR_ACL_ATTR */
2056 	NULL,			/* TX_WRITE2 */
2057 };
2058 
2059 /*
2060  * ZIL get_data callbacks
2061  */
2062 
2063 /* ARGSUSED */
2064 static void
2065 ztest_get_done(zgd_t *zgd, int error)
2066 {
2067 	ztest_ds_t *zd = zgd->zgd_private;
2068 	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
2069 
2070 	if (zgd->zgd_db)
2071 		dmu_buf_rele(zgd->zgd_db, zgd);
2072 
2073 	ztest_range_unlock((rl_t *)zgd->zgd_lr);
2074 	ztest_object_unlock(zd, object);
2075 
2076 	umem_free(zgd, sizeof (*zgd));
2077 }
2078 
2079 static int
2080 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
2081     zio_t *zio)
2082 {
2083 	ztest_ds_t *zd = arg;
2084 	objset_t *os = zd->zd_os;
2085 	uint64_t object = lr->lr_foid;
2086 	uint64_t offset = lr->lr_offset;
2087 	uint64_t size = lr->lr_length;
2088 	uint64_t txg = lr->lr_common.lrc_txg;
2089 	uint64_t crtxg;
2090 	dmu_object_info_t doi;
2091 	dmu_buf_t *db;
2092 	zgd_t *zgd;
2093 	int error;
2094 
2095 	ASSERT3P(lwb, !=, NULL);
2096 	ASSERT3P(zio, !=, NULL);
2097 	ASSERT3U(size, !=, 0);
2098 
2099 	ztest_object_lock(zd, object, RL_READER);
2100 	error = dmu_bonus_hold(os, object, FTAG, &db);
2101 	if (error) {
2102 		ztest_object_unlock(zd, object);
2103 		return (error);
2104 	}
2105 
2106 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
2107 
2108 	if (crtxg == 0 || crtxg > txg) {
2109 		dmu_buf_rele(db, FTAG);
2110 		ztest_object_unlock(zd, object);
2111 		return (ENOENT);
2112 	}
2113 
2114 	dmu_object_info_from_db(db, &doi);
2115 	dmu_buf_rele(db, FTAG);
2116 	db = NULL;
2117 
2118 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2119 	zgd->zgd_lwb = lwb;
2120 	zgd->zgd_private = zd;
2121 
2122 	if (buf != NULL) {	/* immediate write */
2123 		zgd->zgd_lr = (struct locked_range *)ztest_range_lock(zd,
2124 		    object, offset, size, RL_READER);
2125 
2126 		error = dmu_read(os, object, offset, size, buf,
2127 		    DMU_READ_NO_PREFETCH);
2128 		ASSERT(error == 0);
2129 	} else {
2130 		size = doi.doi_data_block_size;
2131 		if (ISP2(size)) {
2132 			offset = P2ALIGN(offset, size);
2133 		} else {
2134 			ASSERT(offset < size);
2135 			offset = 0;
2136 		}
2137 
2138 		zgd->zgd_lr = (struct locked_range *)ztest_range_lock(zd,
2139 		    object, offset, size, RL_READER);
2140 
2141 		error = dmu_buf_hold(os, object, offset, zgd, &db,
2142 		    DMU_READ_NO_PREFETCH);
2143 
2144 		if (error == 0) {
2145 			blkptr_t *bp = &lr->lr_blkptr;
2146 
2147 			zgd->zgd_db = db;
2148 			zgd->zgd_bp = bp;
2149 
2150 			ASSERT(db->db_offset == offset);
2151 			ASSERT(db->db_size == size);
2152 
2153 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2154 			    ztest_get_done, zgd);
2155 
2156 			if (error == 0)
2157 				return (0);
2158 		}
2159 	}
2160 
2161 	ztest_get_done(zgd, error);
2162 
2163 	return (error);
2164 }
2165 
2166 static void *
2167 ztest_lr_alloc(size_t lrsize, char *name)
2168 {
2169 	char *lr;
2170 	size_t namesize = name ? strlen(name) + 1 : 0;
2171 
2172 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2173 
2174 	if (name)
2175 		bcopy(name, lr + lrsize, namesize);
2176 
2177 	return (lr);
2178 }
2179 
2180 void
2181 ztest_lr_free(void *lr, size_t lrsize, char *name)
2182 {
2183 	size_t namesize = name ? strlen(name) + 1 : 0;
2184 
2185 	umem_free(lr, lrsize + namesize);
2186 }
2187 
2188 /*
2189  * Lookup a bunch of objects.  Returns the number of objects not found.
2190  */
2191 static int
2192 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2193 {
2194 	int missing = 0;
2195 	int error;
2196 
2197 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2198 
2199 	for (int i = 0; i < count; i++, od++) {
2200 		od->od_object = 0;
2201 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2202 		    sizeof (uint64_t), 1, &od->od_object);
2203 		if (error) {
2204 			ASSERT(error == ENOENT);
2205 			ASSERT(od->od_object == 0);
2206 			missing++;
2207 		} else {
2208 			dmu_buf_t *db;
2209 			ztest_block_tag_t *bbt;
2210 			dmu_object_info_t doi;
2211 
2212 			ASSERT(od->od_object != 0);
2213 			ASSERT(missing == 0);	/* there should be no gaps */
2214 
2215 			ztest_object_lock(zd, od->od_object, RL_READER);
2216 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2217 			    od->od_object, FTAG, &db));
2218 			dmu_object_info_from_db(db, &doi);
2219 			bbt = ztest_bt_bonus(db);
2220 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2221 			od->od_type = doi.doi_type;
2222 			od->od_blocksize = doi.doi_data_block_size;
2223 			od->od_gen = bbt->bt_gen;
2224 			dmu_buf_rele(db, FTAG);
2225 			ztest_object_unlock(zd, od->od_object);
2226 		}
2227 	}
2228 
2229 	return (missing);
2230 }
2231 
2232 static int
2233 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2234 {
2235 	int missing = 0;
2236 
2237 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2238 
2239 	for (int i = 0; i < count; i++, od++) {
2240 		if (missing) {
2241 			od->od_object = 0;
2242 			missing++;
2243 			continue;
2244 		}
2245 
2246 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2247 
2248 		lr->lr_doid = od->od_dir;
2249 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2250 		lr->lrz_type = od->od_crtype;
2251 		lr->lrz_blocksize = od->od_crblocksize;
2252 		lr->lrz_ibshift = ztest_random_ibshift();
2253 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2254 		lr->lrz_dnodesize = od->od_crdnodesize;
2255 		lr->lr_gen = od->od_crgen;
2256 		lr->lr_crtime[0] = time(NULL);
2257 
2258 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2259 			ASSERT(missing == 0);
2260 			od->od_object = 0;
2261 			missing++;
2262 		} else {
2263 			od->od_object = lr->lr_foid;
2264 			od->od_type = od->od_crtype;
2265 			od->od_blocksize = od->od_crblocksize;
2266 			od->od_gen = od->od_crgen;
2267 			ASSERT(od->od_object != 0);
2268 		}
2269 
2270 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2271 	}
2272 
2273 	return (missing);
2274 }
2275 
2276 static int
2277 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2278 {
2279 	int missing = 0;
2280 	int error;
2281 
2282 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2283 
2284 	od += count - 1;
2285 
2286 	for (int i = count - 1; i >= 0; i--, od--) {
2287 		if (missing) {
2288 			missing++;
2289 			continue;
2290 		}
2291 
2292 		/*
2293 		 * No object was found.
2294 		 */
2295 		if (od->od_object == 0)
2296 			continue;
2297 
2298 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2299 
2300 		lr->lr_doid = od->od_dir;
2301 
2302 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2303 			ASSERT3U(error, ==, ENOSPC);
2304 			missing++;
2305 		} else {
2306 			od->od_object = 0;
2307 		}
2308 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2309 	}
2310 
2311 	return (missing);
2312 }
2313 
2314 static int
2315 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2316     void *data)
2317 {
2318 	lr_write_t *lr;
2319 	int error;
2320 
2321 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2322 
2323 	lr->lr_foid = object;
2324 	lr->lr_offset = offset;
2325 	lr->lr_length = size;
2326 	lr->lr_blkoff = 0;
2327 	BP_ZERO(&lr->lr_blkptr);
2328 
2329 	bcopy(data, lr + 1, size);
2330 
2331 	error = ztest_replay_write(zd, lr, B_FALSE);
2332 
2333 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2334 
2335 	return (error);
2336 }
2337 
2338 static int
2339 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2340 {
2341 	lr_truncate_t *lr;
2342 	int error;
2343 
2344 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2345 
2346 	lr->lr_foid = object;
2347 	lr->lr_offset = offset;
2348 	lr->lr_length = size;
2349 
2350 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2351 
2352 	ztest_lr_free(lr, sizeof (*lr), NULL);
2353 
2354 	return (error);
2355 }
2356 
2357 static int
2358 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2359 {
2360 	lr_setattr_t *lr;
2361 	int error;
2362 
2363 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2364 
2365 	lr->lr_foid = object;
2366 	lr->lr_size = 0;
2367 	lr->lr_mode = 0;
2368 
2369 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2370 
2371 	ztest_lr_free(lr, sizeof (*lr), NULL);
2372 
2373 	return (error);
2374 }
2375 
2376 static void
2377 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2378 {
2379 	objset_t *os = zd->zd_os;
2380 	dmu_tx_t *tx;
2381 	uint64_t txg;
2382 	rl_t *rl;
2383 
2384 	txg_wait_synced(dmu_objset_pool(os), 0);
2385 
2386 	ztest_object_lock(zd, object, RL_READER);
2387 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2388 
2389 	tx = dmu_tx_create(os);
2390 
2391 	dmu_tx_hold_write(tx, object, offset, size);
2392 
2393 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2394 
2395 	if (txg != 0) {
2396 		dmu_prealloc(os, object, offset, size, tx);
2397 		dmu_tx_commit(tx);
2398 		txg_wait_synced(dmu_objset_pool(os), txg);
2399 	} else {
2400 		(void) dmu_free_long_range(os, object, offset, size);
2401 	}
2402 
2403 	ztest_range_unlock(rl);
2404 	ztest_object_unlock(zd, object);
2405 }
2406 
2407 static void
2408 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2409 {
2410 	int err;
2411 	ztest_block_tag_t wbt;
2412 	dmu_object_info_t doi;
2413 	enum ztest_io_type io_type;
2414 	uint64_t blocksize;
2415 	void *data;
2416 
2417 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2418 	blocksize = doi.doi_data_block_size;
2419 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2420 
2421 	/*
2422 	 * Pick an i/o type at random, biased toward writing block tags.
2423 	 */
2424 	io_type = ztest_random(ZTEST_IO_TYPES);
2425 	if (ztest_random(2) == 0)
2426 		io_type = ZTEST_IO_WRITE_TAG;
2427 
2428 	rw_enter(&zd->zd_zilog_lock, RW_READER);
2429 
2430 	switch (io_type) {
2431 
2432 	case ZTEST_IO_WRITE_TAG:
2433 		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2434 		    offset, 0, 0, 0);
2435 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2436 		break;
2437 
2438 	case ZTEST_IO_WRITE_PATTERN:
2439 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2440 		if (ztest_random(2) == 0) {
2441 			/*
2442 			 * Induce fletcher2 collisions to ensure that
2443 			 * zio_ddt_collision() detects and resolves them
2444 			 * when using fletcher2-verify for deduplication.
2445 			 */
2446 			((uint64_t *)data)[0] ^= 1ULL << 63;
2447 			((uint64_t *)data)[4] ^= 1ULL << 63;
2448 		}
2449 		(void) ztest_write(zd, object, offset, blocksize, data);
2450 		break;
2451 
2452 	case ZTEST_IO_WRITE_ZEROES:
2453 		bzero(data, blocksize);
2454 		(void) ztest_write(zd, object, offset, blocksize, data);
2455 		break;
2456 
2457 	case ZTEST_IO_TRUNCATE:
2458 		(void) ztest_truncate(zd, object, offset, blocksize);
2459 		break;
2460 
2461 	case ZTEST_IO_SETATTR:
2462 		(void) ztest_setattr(zd, object);
2463 		break;
2464 
2465 	case ZTEST_IO_REWRITE:
2466 		rw_enter(&ztest_name_lock, RW_READER);
2467 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2468 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2469 		    B_FALSE);
2470 		VERIFY(err == 0 || err == ENOSPC);
2471 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2472 		    ZFS_PROP_COMPRESSION,
2473 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2474 		    B_FALSE);
2475 		VERIFY(err == 0 || err == ENOSPC);
2476 		rw_exit(&ztest_name_lock);
2477 
2478 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2479 		    DMU_READ_NO_PREFETCH));
2480 
2481 		(void) ztest_write(zd, object, offset, blocksize, data);
2482 		break;
2483 	}
2484 
2485 	rw_exit(&zd->zd_zilog_lock);
2486 
2487 	umem_free(data, blocksize);
2488 }
2489 
2490 /*
2491  * Initialize an object description template.
2492  */
2493 static void
2494 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2495     dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2496     uint64_t gen)
2497 {
2498 	od->od_dir = ZTEST_DIROBJ;
2499 	od->od_object = 0;
2500 
2501 	od->od_crtype = type;
2502 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2503 	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2504 	od->od_crgen = gen;
2505 
2506 	od->od_type = DMU_OT_NONE;
2507 	od->od_blocksize = 0;
2508 	od->od_gen = 0;
2509 
2510 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2511 	    tag, (int64_t)id, index);
2512 }
2513 
2514 /*
2515  * Lookup or create the objects for a test using the od template.
2516  * If the objects do not all exist, or if 'remove' is specified,
2517  * remove any existing objects and create new ones.  Otherwise,
2518  * use the existing objects.
2519  */
2520 static int
2521 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2522 {
2523 	int count = size / sizeof (*od);
2524 	int rv = 0;
2525 
2526 	mutex_enter(&zd->zd_dirobj_lock);
2527 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2528 	    (ztest_remove(zd, od, count) != 0 ||
2529 	    ztest_create(zd, od, count) != 0))
2530 		rv = -1;
2531 	zd->zd_od = od;
2532 	mutex_exit(&zd->zd_dirobj_lock);
2533 
2534 	return (rv);
2535 }
2536 
2537 /* ARGSUSED */
2538 void
2539 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2540 {
2541 	zilog_t *zilog = zd->zd_zilog;
2542 
2543 	rw_enter(&zd->zd_zilog_lock, RW_READER);
2544 
2545 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2546 
2547 	/*
2548 	 * Remember the committed values in zd, which is in parent/child
2549 	 * shared memory.  If we die, the next iteration of ztest_run()
2550 	 * will verify that the log really does contain this record.
2551 	 */
2552 	mutex_enter(&zilog->zl_lock);
2553 	ASSERT(zd->zd_shared != NULL);
2554 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2555 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2556 	mutex_exit(&zilog->zl_lock);
2557 
2558 	rw_exit(&zd->zd_zilog_lock);
2559 }
2560 
2561 /*
2562  * This function is designed to simulate the operations that occur during a
2563  * mount/unmount operation.  We hold the dataset across these operations in an
2564  * attempt to expose any implicit assumptions about ZIL management.
2565  */
2566 /* ARGSUSED */
2567 void
2568 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2569 {
2570 	objset_t *os = zd->zd_os;
2571 
2572 	/*
2573 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2574 	 * updating the zil (i.e. adding in-memory log records) and the
2575 	 * zd_zilog_lock to block any I/O.
2576 	 */
2577 	mutex_enter(&zd->zd_dirobj_lock);
2578 	rw_enter(&zd->zd_zilog_lock, RW_WRITER);
2579 
2580 	/* zfsvfs_teardown() */
2581 	zil_close(zd->zd_zilog);
2582 
2583 	/* zfsvfs_setup() */
2584 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2585 	zil_replay(os, zd, ztest_replay_vector);
2586 
2587 	rw_exit(&zd->zd_zilog_lock);
2588 	mutex_exit(&zd->zd_dirobj_lock);
2589 }
2590 
2591 /*
2592  * Verify that we can't destroy an active pool, create an existing pool,
2593  * or create a pool with a bad vdev spec.
2594  */
2595 /* ARGSUSED */
2596 void
2597 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2598 {
2599 	ztest_shared_opts_t *zo = &ztest_opts;
2600 	spa_t *spa;
2601 	nvlist_t *nvroot;
2602 
2603 	if (zo->zo_mmp_test)
2604 		return;
2605 
2606 	/*
2607 	 * Attempt to create using a bad file.
2608 	 */
2609 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2610 	VERIFY3U(ENOENT, ==,
2611 	    spa_create("ztest_bad_file", nvroot, NULL, NULL, NULL));
2612 	nvlist_free(nvroot);
2613 
2614 	/*
2615 	 * Attempt to create using a bad mirror.
2616 	 */
2617 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
2618 	VERIFY3U(ENOENT, ==,
2619 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL, NULL));
2620 	nvlist_free(nvroot);
2621 
2622 	/*
2623 	 * Attempt to create an existing pool.  It shouldn't matter
2624 	 * what's in the nvroot; we should fail with EEXIST.
2625 	 */
2626 	rw_enter(&ztest_name_lock, RW_READER);
2627 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2628 	VERIFY3U(EEXIST, ==,
2629 	    spa_create(zo->zo_pool, nvroot, NULL, NULL, NULL));
2630 	nvlist_free(nvroot);
2631 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2632 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2633 	spa_close(spa, FTAG);
2634 
2635 	rw_exit(&ztest_name_lock);
2636 }
2637 
2638 /*
2639  * Start and then stop the MMP threads to ensure the startup and shutdown code
2640  * works properly.  Actual protection and property-related code tested via ZTS.
2641  */
2642 /* ARGSUSED */
2643 void
2644 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2645 {
2646 	ztest_shared_opts_t *zo = &ztest_opts;
2647 	spa_t *spa = ztest_spa;
2648 
2649 	if (zo->zo_mmp_test)
2650 		return;
2651 
2652 	/*
2653 	 * Since enabling MMP involves setting a property, it could not be done
2654 	 * while the pool is suspended.
2655 	 */
2656 	if (spa_suspended(spa))
2657 		return;
2658 
2659 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2660 	mutex_enter(&spa->spa_props_lock);
2661 
2662 	zfs_multihost_fail_intervals = 0;
2663 
2664 	if (!spa_multihost(spa)) {
2665 		spa->spa_multihost = B_TRUE;
2666 		mmp_thread_start(spa);
2667 	}
2668 
2669 	mutex_exit(&spa->spa_props_lock);
2670 	spa_config_exit(spa, SCL_CONFIG, FTAG);
2671 
2672 	txg_wait_synced(spa_get_dsl(spa), 0);
2673 	mmp_signal_all_threads();
2674 	txg_wait_synced(spa_get_dsl(spa), 0);
2675 
2676 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2677 	mutex_enter(&spa->spa_props_lock);
2678 
2679 	if (spa_multihost(spa)) {
2680 		mmp_thread_stop(spa);
2681 		spa->spa_multihost = B_FALSE;
2682 	}
2683 
2684 	mutex_exit(&spa->spa_props_lock);
2685 	spa_config_exit(spa, SCL_CONFIG, FTAG);
2686 }
2687 
2688 /* ARGSUSED */
2689 void
2690 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2691 {
2692 	spa_t *spa;
2693 	uint64_t initial_version = SPA_VERSION_INITIAL;
2694 	uint64_t version, newversion;
2695 	nvlist_t *nvroot, *props;
2696 	char *name;
2697 
2698 	if (ztest_opts.zo_mmp_test)
2699 		return;
2700 
2701 	mutex_enter(&ztest_vdev_lock);
2702 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2703 
2704 	/*
2705 	 * Clean up from previous runs.
2706 	 */
2707 	(void) spa_destroy(name);
2708 
2709 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2710 	    NULL, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2711 
2712 	/*
2713 	 * If we're configuring a RAIDZ device then make sure that the
2714 	 * the initial version is capable of supporting that feature.
2715 	 */
2716 	switch (ztest_opts.zo_raidz_parity) {
2717 	case 0:
2718 	case 1:
2719 		initial_version = SPA_VERSION_INITIAL;
2720 		break;
2721 	case 2:
2722 		initial_version = SPA_VERSION_RAIDZ2;
2723 		break;
2724 	case 3:
2725 		initial_version = SPA_VERSION_RAIDZ3;
2726 		break;
2727 	}
2728 
2729 	/*
2730 	 * Create a pool with a spa version that can be upgraded. Pick
2731 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2732 	 */
2733 	do {
2734 		version = ztest_random_spa_version(initial_version);
2735 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2736 
2737 	props = fnvlist_alloc();
2738 	fnvlist_add_uint64(props,
2739 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2740 	VERIFY0(spa_create(name, nvroot, props, NULL, NULL));
2741 	fnvlist_free(nvroot);
2742 	fnvlist_free(props);
2743 
2744 	VERIFY0(spa_open(name, &spa, FTAG));
2745 	VERIFY3U(spa_version(spa), ==, version);
2746 	newversion = ztest_random_spa_version(version + 1);
2747 
2748 	if (ztest_opts.zo_verbose >= 4) {
2749 		(void) printf("upgrading spa version from %llu to %llu\n",
2750 		    (u_longlong_t)version, (u_longlong_t)newversion);
2751 	}
2752 
2753 	spa_upgrade(spa, newversion);
2754 	VERIFY3U(spa_version(spa), >, version);
2755 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2756 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2757 	spa_close(spa, FTAG);
2758 
2759 	strfree(name);
2760 	mutex_exit(&ztest_vdev_lock);
2761 }
2762 
2763 static void
2764 ztest_spa_checkpoint(spa_t *spa)
2765 {
2766 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2767 
2768 	int error = spa_checkpoint(spa->spa_name);
2769 
2770 	switch (error) {
2771 	case 0:
2772 	case ZFS_ERR_DEVRM_IN_PROGRESS:
2773 	case ZFS_ERR_DISCARDING_CHECKPOINT:
2774 	case ZFS_ERR_CHECKPOINT_EXISTS:
2775 		break;
2776 	case ENOSPC:
2777 		ztest_record_enospc(FTAG);
2778 		break;
2779 	default:
2780 		fatal(0, "spa_checkpoint(%s) = %d", spa->spa_name, error);
2781 	}
2782 }
2783 
2784 static void
2785 ztest_spa_discard_checkpoint(spa_t *spa)
2786 {
2787 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2788 
2789 	int error = spa_checkpoint_discard(spa->spa_name);
2790 
2791 	switch (error) {
2792 	case 0:
2793 	case ZFS_ERR_DISCARDING_CHECKPOINT:
2794 	case ZFS_ERR_NO_CHECKPOINT:
2795 		break;
2796 	default:
2797 		fatal(0, "spa_discard_checkpoint(%s) = %d",
2798 		    spa->spa_name, error);
2799 	}
2800 
2801 }
2802 
2803 /* ARGSUSED */
2804 void
2805 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
2806 {
2807 	spa_t *spa = ztest_spa;
2808 
2809 	mutex_enter(&ztest_checkpoint_lock);
2810 	if (ztest_random(2) == 0) {
2811 		ztest_spa_checkpoint(spa);
2812 	} else {
2813 		ztest_spa_discard_checkpoint(spa);
2814 	}
2815 	mutex_exit(&ztest_checkpoint_lock);
2816 }
2817 
2818 
2819 static vdev_t *
2820 vdev_lookup_by_path(vdev_t *vd, const char *path)
2821 {
2822 	vdev_t *mvd;
2823 
2824 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2825 		return (vd);
2826 
2827 	for (int c = 0; c < vd->vdev_children; c++)
2828 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2829 		    NULL)
2830 			return (mvd);
2831 
2832 	return (NULL);
2833 }
2834 
2835 static int
2836 spa_num_top_vdevs(spa_t *spa)
2837 {
2838 	vdev_t *rvd = spa->spa_root_vdev;
2839 	ASSERT3U(spa_config_held(spa, SCL_VDEV, RW_READER), ==, SCL_VDEV);
2840 	return (rvd->vdev_children);
2841 }
2842 
2843 /*
2844  * Verify that vdev_add() works as expected.
2845  */
2846 /* ARGSUSED */
2847 void
2848 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2849 {
2850 	ztest_shared_t *zs = ztest_shared;
2851 	spa_t *spa = ztest_spa;
2852 	uint64_t leaves;
2853 	uint64_t guid;
2854 	nvlist_t *nvroot;
2855 	int error;
2856 
2857 	if (ztest_opts.zo_mmp_test)
2858 		return;
2859 
2860 	mutex_enter(&ztest_vdev_lock);
2861 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2862 
2863 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2864 
2865 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
2866 
2867 	/*
2868 	 * If we have slogs then remove them 1/4 of the time.
2869 	 */
2870 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2871 		metaslab_group_t *mg;
2872 
2873 		/*
2874 		 * find the first real slog in log allocation class
2875 		 */
2876 		mg =  spa_log_class(spa)->mc_rotor;
2877 		while (!mg->mg_vd->vdev_islog)
2878 			mg = mg->mg_next;
2879 
2880 		guid = mg->mg_vd->vdev_guid;
2881 
2882 		spa_config_exit(spa, SCL_VDEV, FTAG);
2883 
2884 		/*
2885 		 * We have to grab the zs_name_lock as writer to
2886 		 * prevent a race between removing a slog (dmu_objset_find)
2887 		 * and destroying a dataset. Removing the slog will
2888 		 * grab a reference on the dataset which may cause
2889 		 * dmu_objset_destroy() to fail with EBUSY thus
2890 		 * leaving the dataset in an inconsistent state.
2891 		 */
2892 		rw_enter(&ztest_name_lock, RW_WRITER);
2893 		error = spa_vdev_remove(spa, guid, B_FALSE);
2894 		rw_exit(&ztest_name_lock);
2895 
2896 		switch (error) {
2897 		case 0:
2898 		case EEXIST:
2899 		case ZFS_ERR_CHECKPOINT_EXISTS:
2900 		case ZFS_ERR_DISCARDING_CHECKPOINT:
2901 			break;
2902 		default:
2903 			fatal(0, "spa_vdev_remove() = %d", error);
2904 		}
2905 	} else {
2906 		spa_config_exit(spa, SCL_VDEV, FTAG);
2907 
2908 		/*
2909 		 * Make 1/4 of the devices be log devices
2910 		 */
2911 		nvroot = make_vdev_root(NULL, NULL, NULL,
2912 		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
2913 		    "log" : NULL, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
2914 
2915 		error = spa_vdev_add(spa, nvroot);
2916 		nvlist_free(nvroot);
2917 
2918 		switch (error) {
2919 		case 0:
2920 			break;
2921 		case ENOSPC:
2922 			ztest_record_enospc("spa_vdev_add");
2923 			break;
2924 		default:
2925 			fatal(0, "spa_vdev_add() = %d", error);
2926 		}
2927 	}
2928 
2929 	mutex_exit(&ztest_vdev_lock);
2930 }
2931 
2932 /* ARGSUSED */
2933 void
2934 ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
2935 {
2936 	ztest_shared_t *zs = ztest_shared;
2937 	spa_t *spa = ztest_spa;
2938 	uint64_t leaves;
2939 	nvlist_t *nvroot;
2940 	const char *class = (ztest_random(2) == 0) ?
2941 	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
2942 	int error;
2943 
2944 	/*
2945 	 * By default add a special vdev 50% of the time
2946 	 */
2947 	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
2948 	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
2949 	    ztest_random(2) == 0)) {
2950 		return;
2951 	}
2952 
2953 	mutex_enter(&ztest_vdev_lock);
2954 
2955 	/* Only test with mirrors */
2956 	if (zs->zs_mirrors < 2) {
2957 		mutex_exit(&ztest_vdev_lock);
2958 		return;
2959 	}
2960 
2961 	/* requires feature@allocation_classes */
2962 	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
2963 		mutex_exit(&ztest_vdev_lock);
2964 		return;
2965 	}
2966 
2967 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2968 
2969 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2970 	ztest_shared->zs_vdev_next_leaf = spa_num_top_vdevs(spa) * leaves;
2971 	spa_config_exit(spa, SCL_VDEV, FTAG);
2972 
2973 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
2974 	    class, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
2975 
2976 	error = spa_vdev_add(spa, nvroot);
2977 	nvlist_free(nvroot);
2978 
2979 	if (error == ENOSPC)
2980 		ztest_record_enospc("spa_vdev_add");
2981 	else if (error != 0)
2982 		fatal(0, "spa_vdev_add() = %d", error);
2983 
2984 	/*
2985 	 * 50% of the time allow small blocks in the special class
2986 	 */
2987 	if (error == 0 &&
2988 	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
2989 		if (ztest_opts.zo_verbose >= 3)
2990 			(void) printf("Enabling special VDEV small blocks\n");
2991 		(void) ztest_dsl_prop_set_uint64(zd->zd_name,
2992 		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
2993 	}
2994 
2995 	mutex_exit(&ztest_vdev_lock);
2996 
2997 	if (ztest_opts.zo_verbose >= 3) {
2998 		metaslab_class_t *mc;
2999 
3000 		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
3001 			mc = spa_special_class(spa);
3002 		else
3003 			mc = spa_dedup_class(spa);
3004 		(void) printf("Added a %s mirrored vdev (of %d)\n",
3005 		    class, (int)mc->mc_groups);
3006 	}
3007 }
3008 
3009 /*
3010  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
3011  */
3012 /* ARGSUSED */
3013 void
3014 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
3015 {
3016 	ztest_shared_t *zs = ztest_shared;
3017 	spa_t *spa = ztest_spa;
3018 	vdev_t *rvd = spa->spa_root_vdev;
3019 	spa_aux_vdev_t *sav;
3020 	char *aux;
3021 	uint64_t guid = 0;
3022 	int error;
3023 
3024 	if (ztest_opts.zo_mmp_test)
3025 		return;
3026 
3027 	if (ztest_random(2) == 0) {
3028 		sav = &spa->spa_spares;
3029 		aux = ZPOOL_CONFIG_SPARES;
3030 	} else {
3031 		sav = &spa->spa_l2cache;
3032 		aux = ZPOOL_CONFIG_L2CACHE;
3033 	}
3034 
3035 	mutex_enter(&ztest_vdev_lock);
3036 
3037 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3038 
3039 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
3040 		/*
3041 		 * Pick a random device to remove.
3042 		 */
3043 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
3044 	} else {
3045 		/*
3046 		 * Find an unused device we can add.
3047 		 */
3048 		zs->zs_vdev_aux = 0;
3049 		for (;;) {
3050 			char path[MAXPATHLEN];
3051 			int c;
3052 			(void) snprintf(path, sizeof (path), ztest_aux_template,
3053 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3054 			    zs->zs_vdev_aux);
3055 			for (c = 0; c < sav->sav_count; c++)
3056 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
3057 				    path) == 0)
3058 					break;
3059 			if (c == sav->sav_count &&
3060 			    vdev_lookup_by_path(rvd, path) == NULL)
3061 				break;
3062 			zs->zs_vdev_aux++;
3063 		}
3064 	}
3065 
3066 	spa_config_exit(spa, SCL_VDEV, FTAG);
3067 
3068 	if (guid == 0) {
3069 		/*
3070 		 * Add a new device.
3071 		 */
3072 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3073 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
3074 		error = spa_vdev_add(spa, nvroot);
3075 
3076 		switch (error) {
3077 		case 0:
3078 			break;
3079 		default:
3080 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
3081 		}
3082 		nvlist_free(nvroot);
3083 	} else {
3084 		/*
3085 		 * Remove an existing device.  Sometimes, dirty its
3086 		 * vdev state first to make sure we handle removal
3087 		 * of devices that have pending state changes.
3088 		 */
3089 		if (ztest_random(2) == 0)
3090 			(void) vdev_online(spa, guid, 0, NULL);
3091 
3092 		error = spa_vdev_remove(spa, guid, B_FALSE);
3093 
3094 		switch (error) {
3095 		case 0:
3096 		case EBUSY:
3097 		case ZFS_ERR_CHECKPOINT_EXISTS:
3098 		case ZFS_ERR_DISCARDING_CHECKPOINT:
3099 			break;
3100 		default:
3101 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
3102 		}
3103 	}
3104 
3105 	mutex_exit(&ztest_vdev_lock);
3106 }
3107 
3108 /*
3109  * split a pool if it has mirror tlvdevs
3110  */
3111 /* ARGSUSED */
3112 void
3113 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3114 {
3115 	ztest_shared_t *zs = ztest_shared;
3116 	spa_t *spa = ztest_spa;
3117 	vdev_t *rvd = spa->spa_root_vdev;
3118 	nvlist_t *tree, **child, *config, *split, **schild;
3119 	uint_t c, children, schildren = 0, lastlogid = 0;
3120 	int error = 0;
3121 
3122 	if (ztest_opts.zo_mmp_test)
3123 		return;
3124 
3125 	mutex_enter(&ztest_vdev_lock);
3126 
3127 	/* ensure we have a useable config; mirrors of raidz aren't supported */
3128 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
3129 		mutex_exit(&ztest_vdev_lock);
3130 		return;
3131 	}
3132 
3133 	/* clean up the old pool, if any */
3134 	(void) spa_destroy("splitp");
3135 
3136 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3137 
3138 	/* generate a config from the existing config */
3139 	mutex_enter(&spa->spa_props_lock);
3140 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
3141 	    &tree) == 0);
3142 	mutex_exit(&spa->spa_props_lock);
3143 
3144 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3145 	    &children) == 0);
3146 
3147 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
3148 	for (c = 0; c < children; c++) {
3149 		vdev_t *tvd = rvd->vdev_child[c];
3150 		nvlist_t **mchild;
3151 		uint_t mchildren;
3152 
3153 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3154 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
3155 			    0) == 0);
3156 			VERIFY(nvlist_add_string(schild[schildren],
3157 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
3158 			VERIFY(nvlist_add_uint64(schild[schildren],
3159 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
3160 			if (lastlogid == 0)
3161 				lastlogid = schildren;
3162 			++schildren;
3163 			continue;
3164 		}
3165 		lastlogid = 0;
3166 		VERIFY(nvlist_lookup_nvlist_array(child[c],
3167 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3168 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
3169 	}
3170 
3171 	/* OK, create a config that can be used to split */
3172 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
3173 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
3174 	    VDEV_TYPE_ROOT) == 0);
3175 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
3176 	    lastlogid != 0 ? lastlogid : schildren) == 0);
3177 
3178 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
3179 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
3180 
3181 	for (c = 0; c < schildren; c++)
3182 		nvlist_free(schild[c]);
3183 	free(schild);
3184 	nvlist_free(split);
3185 
3186 	spa_config_exit(spa, SCL_VDEV, FTAG);
3187 
3188 	rw_enter(&ztest_name_lock, RW_WRITER);
3189 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3190 	rw_exit(&ztest_name_lock);
3191 
3192 	nvlist_free(config);
3193 
3194 	if (error == 0) {
3195 		(void) printf("successful split - results:\n");
3196 		mutex_enter(&spa_namespace_lock);
3197 		show_pool_stats(spa);
3198 		show_pool_stats(spa_lookup("splitp"));
3199 		mutex_exit(&spa_namespace_lock);
3200 		++zs->zs_splits;
3201 		--zs->zs_mirrors;
3202 	}
3203 	mutex_exit(&ztest_vdev_lock);
3204 }
3205 
3206 /*
3207  * Verify that we can attach and detach devices.
3208  */
3209 /* ARGSUSED */
3210 void
3211 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3212 {
3213 	ztest_shared_t *zs = ztest_shared;
3214 	spa_t *spa = ztest_spa;
3215 	spa_aux_vdev_t *sav = &spa->spa_spares;
3216 	vdev_t *rvd = spa->spa_root_vdev;
3217 	vdev_t *oldvd, *newvd, *pvd;
3218 	nvlist_t *root;
3219 	uint64_t leaves;
3220 	uint64_t leaf, top;
3221 	uint64_t ashift = ztest_get_ashift();
3222 	uint64_t oldguid, pguid;
3223 	uint64_t oldsize, newsize;
3224 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
3225 	int replacing;
3226 	int oldvd_has_siblings = B_FALSE;
3227 	int newvd_is_spare = B_FALSE;
3228 	int oldvd_is_log;
3229 	int error, expected_error;
3230 
3231 	if (ztest_opts.zo_mmp_test)
3232 		return;
3233 
3234 	mutex_enter(&ztest_vdev_lock);
3235 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
3236 
3237 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3238 
3239 	/*
3240 	 * If a vdev is in the process of being removed, its removal may
3241 	 * finish while we are in progress, leading to an unexpected error
3242 	 * value.  Don't bother trying to attach while we are in the middle
3243 	 * of removal.
3244 	 */
3245 	if (ztest_device_removal_active) {
3246 		spa_config_exit(spa, SCL_ALL, FTAG);
3247 		mutex_exit(&ztest_vdev_lock);
3248 		return;
3249 	}
3250 
3251 	/*
3252 	 * Decide whether to do an attach or a replace.
3253 	 */
3254 	replacing = ztest_random(2);
3255 
3256 	/*
3257 	 * Pick a random top-level vdev.
3258 	 */
3259 	top = ztest_random_vdev_top(spa, B_TRUE);
3260 
3261 	/*
3262 	 * Pick a random leaf within it.
3263 	 */
3264 	leaf = ztest_random(leaves);
3265 
3266 	/*
3267 	 * Locate this vdev.
3268 	 */
3269 	oldvd = rvd->vdev_child[top];
3270 
3271 	/* pick a child from the mirror */
3272 	if (zs->zs_mirrors >= 1) {
3273 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
3274 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
3275 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
3276 	}
3277 
3278 	/* pick a child out of the raidz group */
3279 	if (ztest_opts.zo_raidz > 1) {
3280 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
3281 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
3282 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
3283 	}
3284 
3285 	/*
3286 	 * If we're already doing an attach or replace, oldvd may be a
3287 	 * mirror vdev -- in which case, pick a random child.
3288 	 */
3289 	while (oldvd->vdev_children != 0) {
3290 		oldvd_has_siblings = B_TRUE;
3291 		ASSERT(oldvd->vdev_children >= 2);
3292 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3293 	}
3294 
3295 	oldguid = oldvd->vdev_guid;
3296 	oldsize = vdev_get_min_asize(oldvd);
3297 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3298 	(void) strcpy(oldpath, oldvd->vdev_path);
3299 	pvd = oldvd->vdev_parent;
3300 	pguid = pvd->vdev_guid;
3301 
3302 	/*
3303 	 * If oldvd has siblings, then half of the time, detach it.
3304 	 */
3305 	if (oldvd_has_siblings && ztest_random(2) == 0) {
3306 		spa_config_exit(spa, SCL_ALL, FTAG);
3307 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3308 		if (error != 0 && error != ENODEV && error != EBUSY &&
3309 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3310 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3311 			fatal(0, "detach (%s) returned %d", oldpath, error);
3312 		mutex_exit(&ztest_vdev_lock);
3313 		return;
3314 	}
3315 
3316 	/*
3317 	 * For the new vdev, choose with equal probability between the two
3318 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3319 	 */
3320 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3321 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3322 		newvd_is_spare = B_TRUE;
3323 		(void) strcpy(newpath, newvd->vdev_path);
3324 	} else {
3325 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
3326 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3327 		    top * leaves + leaf);
3328 		if (ztest_random(2) == 0)
3329 			newpath[strlen(newpath) - 1] = 'b';
3330 		newvd = vdev_lookup_by_path(rvd, newpath);
3331 	}
3332 
3333 	if (newvd) {
3334 		/*
3335 		 * Reopen to ensure the vdev's asize field isn't stale.
3336 		 */
3337 		vdev_reopen(newvd);
3338 		newsize = vdev_get_min_asize(newvd);
3339 	} else {
3340 		/*
3341 		 * Make newsize a little bigger or smaller than oldsize.
3342 		 * If it's smaller, the attach should fail.
3343 		 * If it's larger, and we're doing a replace,
3344 		 * we should get dynamic LUN growth when we're done.
3345 		 */
3346 		newsize = 10 * oldsize / (9 + ztest_random(3));
3347 	}
3348 
3349 	/*
3350 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3351 	 * unless it's a replace; in that case any non-replacing parent is OK.
3352 	 *
3353 	 * If newvd is already part of the pool, it should fail with EBUSY.
3354 	 *
3355 	 * If newvd is too small, it should fail with EOVERFLOW.
3356 	 */
3357 	if (pvd->vdev_ops != &vdev_mirror_ops &&
3358 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3359 	    pvd->vdev_ops == &vdev_replacing_ops ||
3360 	    pvd->vdev_ops == &vdev_spare_ops))
3361 		expected_error = ENOTSUP;
3362 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
3363 		expected_error = ENOTSUP;
3364 	else if (newvd == oldvd)
3365 		expected_error = replacing ? 0 : EBUSY;
3366 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3367 		expected_error = EBUSY;
3368 	else if (newsize < oldsize)
3369 		expected_error = EOVERFLOW;
3370 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3371 		expected_error = EDOM;
3372 	else
3373 		expected_error = 0;
3374 
3375 	spa_config_exit(spa, SCL_ALL, FTAG);
3376 
3377 	/*
3378 	 * Build the nvlist describing newpath.
3379 	 */
3380 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3381 	    ashift, NULL, 0, 0, 1);
3382 
3383 	error = spa_vdev_attach(spa, oldguid, root, replacing);
3384 
3385 	nvlist_free(root);
3386 
3387 	/*
3388 	 * If our parent was the replacing vdev, but the replace completed,
3389 	 * then instead of failing with ENOTSUP we may either succeed,
3390 	 * fail with ENODEV, or fail with EOVERFLOW.
3391 	 */
3392 	if (expected_error == ENOTSUP &&
3393 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3394 		expected_error = error;
3395 
3396 	/*
3397 	 * If someone grew the LUN, the replacement may be too small.
3398 	 */
3399 	if (error == EOVERFLOW || error == EBUSY)
3400 		expected_error = error;
3401 
3402 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3403 	    error == ZFS_ERR_DISCARDING_CHECKPOINT)
3404 		expected_error = error;
3405 
3406 	/* XXX workaround 6690467 */
3407 	if (error != expected_error && expected_error != EBUSY) {
3408 		fatal(0, "attach (%s %llu, %s %llu, %d) "
3409 		    "returned %d, expected %d",
3410 		    oldpath, oldsize, newpath,
3411 		    newsize, replacing, error, expected_error);
3412 	}
3413 
3414 	mutex_exit(&ztest_vdev_lock);
3415 }
3416 
3417 /* ARGSUSED */
3418 void
3419 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3420 {
3421 	spa_t *spa = ztest_spa;
3422 	vdev_t *vd;
3423 	uint64_t guid;
3424 	int error;
3425 
3426 	mutex_enter(&ztest_vdev_lock);
3427 
3428 	if (ztest_device_removal_active) {
3429 		mutex_exit(&ztest_vdev_lock);
3430 		return;
3431 	}
3432 
3433 	/*
3434 	 * Remove a random top-level vdev and wait for removal to finish.
3435 	 */
3436 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3437 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3438 	guid = vd->vdev_guid;
3439 	spa_config_exit(spa, SCL_VDEV, FTAG);
3440 
3441 	error = spa_vdev_remove(spa, guid, B_FALSE);
3442 	if (error == 0) {
3443 		ztest_device_removal_active = B_TRUE;
3444 		mutex_exit(&ztest_vdev_lock);
3445 
3446 		while (spa->spa_vdev_removal != NULL)
3447 			txg_wait_synced(spa_get_dsl(spa), 0);
3448 	} else {
3449 		mutex_exit(&ztest_vdev_lock);
3450 		return;
3451 	}
3452 
3453 	/*
3454 	 * The pool needs to be scrubbed after completing device removal.
3455 	 * Failure to do so may result in checksum errors due to the
3456 	 * strategy employed by ztest_fault_inject() when selecting which
3457 	 * offset are redundant and can be damaged.
3458 	 */
3459 	error = spa_scan(spa, POOL_SCAN_SCRUB);
3460 	if (error == 0) {
3461 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3462 			txg_wait_synced(spa_get_dsl(spa), 0);
3463 	}
3464 
3465 	mutex_enter(&ztest_vdev_lock);
3466 	ztest_device_removal_active = B_FALSE;
3467 	mutex_exit(&ztest_vdev_lock);
3468 }
3469 
3470 /*
3471  * Callback function which expands the physical size of the vdev.
3472  */
3473 vdev_t *
3474 grow_vdev(vdev_t *vd, void *arg)
3475 {
3476 	spa_t *spa = vd->vdev_spa;
3477 	size_t *newsize = arg;
3478 	size_t fsize;
3479 	int fd;
3480 
3481 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3482 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3483 
3484 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3485 		return (vd);
3486 
3487 	fsize = lseek(fd, 0, SEEK_END);
3488 	(void) ftruncate(fd, *newsize);
3489 
3490 	if (ztest_opts.zo_verbose >= 6) {
3491 		(void) printf("%s grew from %lu to %lu bytes\n",
3492 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3493 	}
3494 	(void) close(fd);
3495 	return (NULL);
3496 }
3497 
3498 /*
3499  * Callback function which expands a given vdev by calling vdev_online().
3500  */
3501 /* ARGSUSED */
3502 vdev_t *
3503 online_vdev(vdev_t *vd, void *arg)
3504 {
3505 	spa_t *spa = vd->vdev_spa;
3506 	vdev_t *tvd = vd->vdev_top;
3507 	uint64_t guid = vd->vdev_guid;
3508 	uint64_t generation = spa->spa_config_generation + 1;
3509 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3510 	int error;
3511 
3512 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3513 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3514 
3515 	/* Calling vdev_online will initialize the new metaslabs */
3516 	spa_config_exit(spa, SCL_STATE, spa);
3517 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3518 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3519 
3520 	/*
3521 	 * If vdev_online returned an error or the underlying vdev_open
3522 	 * failed then we abort the expand. The only way to know that
3523 	 * vdev_open fails is by checking the returned newstate.
3524 	 */
3525 	if (error || newstate != VDEV_STATE_HEALTHY) {
3526 		if (ztest_opts.zo_verbose >= 5) {
3527 			(void) printf("Unable to expand vdev, state %llu, "
3528 			    "error %d\n", (u_longlong_t)newstate, error);
3529 		}
3530 		return (vd);
3531 	}
3532 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3533 
3534 	/*
3535 	 * Since we dropped the lock we need to ensure that we're
3536 	 * still talking to the original vdev. It's possible this
3537 	 * vdev may have been detached/replaced while we were
3538 	 * trying to online it.
3539 	 */
3540 	if (generation != spa->spa_config_generation) {
3541 		if (ztest_opts.zo_verbose >= 5) {
3542 			(void) printf("vdev configuration has changed, "
3543 			    "guid %llu, state %llu, expected gen %llu, "
3544 			    "got gen %llu\n",
3545 			    (u_longlong_t)guid,
3546 			    (u_longlong_t)tvd->vdev_state,
3547 			    (u_longlong_t)generation,
3548 			    (u_longlong_t)spa->spa_config_generation);
3549 		}
3550 		return (vd);
3551 	}
3552 	return (NULL);
3553 }
3554 
3555 /*
3556  * Traverse the vdev tree calling the supplied function.
3557  * We continue to walk the tree until we either have walked all
3558  * children or we receive a non-NULL return from the callback.
3559  * If a NULL callback is passed, then we just return back the first
3560  * leaf vdev we encounter.
3561  */
3562 vdev_t *
3563 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3564 {
3565 	if (vd->vdev_ops->vdev_op_leaf) {
3566 		if (func == NULL)
3567 			return (vd);
3568 		else
3569 			return (func(vd, arg));
3570 	}
3571 
3572 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3573 		vdev_t *cvd = vd->vdev_child[c];
3574 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3575 			return (cvd);
3576 	}
3577 	return (NULL);
3578 }
3579 
3580 /*
3581  * Verify that dynamic LUN growth works as expected.
3582  */
3583 /* ARGSUSED */
3584 void
3585 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3586 {
3587 	spa_t *spa = ztest_spa;
3588 	vdev_t *vd, *tvd;
3589 	metaslab_class_t *mc;
3590 	metaslab_group_t *mg;
3591 	size_t psize, newsize;
3592 	uint64_t top;
3593 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3594 
3595 	mutex_enter(&ztest_checkpoint_lock);
3596 	mutex_enter(&ztest_vdev_lock);
3597 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3598 
3599 	/*
3600 	 * If there is a vdev removal in progress, it could complete while
3601 	 * we are running, in which case we would not be able to verify
3602 	 * that the metaslab_class space increased (because it decreases
3603 	 * when the device removal completes).
3604 	 */
3605 	if (ztest_device_removal_active) {
3606 		spa_config_exit(spa, SCL_STATE, spa);
3607 		mutex_exit(&ztest_vdev_lock);
3608 		mutex_exit(&ztest_checkpoint_lock);
3609 		return;
3610 	}
3611 
3612 	top = ztest_random_vdev_top(spa, B_TRUE);
3613 
3614 	tvd = spa->spa_root_vdev->vdev_child[top];
3615 	mg = tvd->vdev_mg;
3616 	mc = mg->mg_class;
3617 	old_ms_count = tvd->vdev_ms_count;
3618 	old_class_space = metaslab_class_get_space(mc);
3619 
3620 	/*
3621 	 * Determine the size of the first leaf vdev associated with
3622 	 * our top-level device.
3623 	 */
3624 	vd = vdev_walk_tree(tvd, NULL, NULL);
3625 	ASSERT3P(vd, !=, NULL);
3626 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3627 
3628 	psize = vd->vdev_psize;
3629 
3630 	/*
3631 	 * We only try to expand the vdev if it's healthy, less than 4x its
3632 	 * original size, and it has a valid psize.
3633 	 */
3634 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3635 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3636 		spa_config_exit(spa, SCL_STATE, spa);
3637 		mutex_exit(&ztest_vdev_lock);
3638 		mutex_exit(&ztest_checkpoint_lock);
3639 		return;
3640 	}
3641 	ASSERT(psize > 0);
3642 	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
3643 	ASSERT3U(newsize, >, psize);
3644 
3645 	if (ztest_opts.zo_verbose >= 6) {
3646 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3647 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3648 	}
3649 
3650 	/*
3651 	 * Growing the vdev is a two step process:
3652 	 *	1). expand the physical size (i.e. relabel)
3653 	 *	2). online the vdev to create the new metaslabs
3654 	 */
3655 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3656 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3657 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3658 		if (ztest_opts.zo_verbose >= 5) {
3659 			(void) printf("Could not expand LUN because "
3660 			    "the vdev configuration changed.\n");
3661 		}
3662 		spa_config_exit(spa, SCL_STATE, spa);
3663 		mutex_exit(&ztest_vdev_lock);
3664 		mutex_exit(&ztest_checkpoint_lock);
3665 		return;
3666 	}
3667 
3668 	spa_config_exit(spa, SCL_STATE, spa);
3669 
3670 	/*
3671 	 * Expanding the LUN will update the config asynchronously,
3672 	 * thus we must wait for the async thread to complete any
3673 	 * pending tasks before proceeding.
3674 	 */
3675 	for (;;) {
3676 		boolean_t done;
3677 		mutex_enter(&spa->spa_async_lock);
3678 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3679 		mutex_exit(&spa->spa_async_lock);
3680 		if (done)
3681 			break;
3682 		txg_wait_synced(spa_get_dsl(spa), 0);
3683 		(void) poll(NULL, 0, 100);
3684 	}
3685 
3686 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3687 
3688 	tvd = spa->spa_root_vdev->vdev_child[top];
3689 	new_ms_count = tvd->vdev_ms_count;
3690 	new_class_space = metaslab_class_get_space(mc);
3691 
3692 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3693 		if (ztest_opts.zo_verbose >= 5) {
3694 			(void) printf("Could not verify LUN expansion due to "
3695 			    "intervening vdev offline or remove.\n");
3696 		}
3697 		spa_config_exit(spa, SCL_STATE, spa);
3698 		mutex_exit(&ztest_vdev_lock);
3699 		mutex_exit(&ztest_checkpoint_lock);
3700 		return;
3701 	}
3702 
3703 	/*
3704 	 * Make sure we were able to grow the vdev.
3705 	 */
3706 	if (new_ms_count <= old_ms_count) {
3707 		fatal(0, "LUN expansion failed: ms_count %llu < %llu\n",
3708 		    old_ms_count, new_ms_count);
3709 	}
3710 
3711 	/*
3712 	 * Make sure we were able to grow the pool.
3713 	 */
3714 	if (new_class_space <= old_class_space) {
3715 		fatal(0, "LUN expansion failed: class_space %llu < %llu\n",
3716 		    old_class_space, new_class_space);
3717 	}
3718 
3719 	if (ztest_opts.zo_verbose >= 5) {
3720 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3721 
3722 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3723 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3724 		(void) printf("%s grew from %s to %s\n",
3725 		    spa->spa_name, oldnumbuf, newnumbuf);
3726 	}
3727 
3728 	spa_config_exit(spa, SCL_STATE, spa);
3729 	mutex_exit(&ztest_vdev_lock);
3730 	mutex_exit(&ztest_checkpoint_lock);
3731 }
3732 
3733 /*
3734  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3735  */
3736 /* ARGSUSED */
3737 static void
3738 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3739 {
3740 	/*
3741 	 * Create the objects common to all ztest datasets.
3742 	 */
3743 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3744 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3745 }
3746 
3747 static int
3748 ztest_dataset_create(char *dsname)
3749 {
3750 	int err;
3751 	uint64_t rand;
3752 	dsl_crypto_params_t *dcp = NULL;
3753 
3754 	/*
3755 	 * 50% of the time, we create encrypted datasets
3756 	 * using a random cipher suite and a hard-coded
3757 	 * wrapping key.
3758 	 */
3759 #ifdef WITHCRYPTO
3760 	/*
3761 	 * Until the crypto framework is compiled in userland, the ztest using
3762 	 * crypto will not work.
3763 	 */
3764 	rand = ztest_random(2);
3765 #else
3766 	rand = 0;
3767 #endif
3768 	if (rand != 0) {
3769 		nvlist_t *crypto_args = fnvlist_alloc();
3770 		nvlist_t *props = fnvlist_alloc();
3771 
3772 		/* slight bias towards the default cipher suite */
3773 		rand = ztest_random(ZIO_CRYPT_FUNCTIONS);
3774 		if (rand < ZIO_CRYPT_AES_128_CCM)
3775 			rand = ZIO_CRYPT_ON;
3776 
3777 		fnvlist_add_uint64(props,
3778 		    zfs_prop_to_name(ZFS_PROP_ENCRYPTION), rand);
3779 		fnvlist_add_uint8_array(crypto_args, "wkeydata",
3780 		    (uint8_t *)ztest_wkeydata, WRAPPING_KEY_LEN);
3781 
3782 		/*
3783 		 * These parameters aren't really used by the kernel. They
3784 		 * are simply stored so that userspace knows how to load
3785 		 * the wrapping key.
3786 		 */
3787 		fnvlist_add_uint64(props,
3788 		    zfs_prop_to_name(ZFS_PROP_KEYFORMAT), ZFS_KEYFORMAT_RAW);
3789 		fnvlist_add_string(props,
3790 		    zfs_prop_to_name(ZFS_PROP_KEYLOCATION), "prompt");
3791 		fnvlist_add_uint64(props,
3792 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_SALT), 0ULL);
3793 		fnvlist_add_uint64(props,
3794 		    zfs_prop_to_name(ZFS_PROP_PBKDF2_ITERS), 0ULL);
3795 
3796 		VERIFY0(dsl_crypto_params_create_nvlist(DCP_CMD_NONE, props,
3797 		    crypto_args, &dcp));
3798 
3799 		fnvlist_free(crypto_args);
3800 		fnvlist_free(props);
3801 	}
3802 
3803 	err = dmu_objset_create(dsname, DMU_OST_OTHER, 0, dcp,
3804 	    ztest_objset_create_cb, NULL);
3805 	dsl_crypto_params_free(dcp, !!err);
3806 
3807 	rand = ztest_random(100);
3808 	if (err || rand < 80)
3809 		return (err);
3810 
3811 	if (ztest_opts.zo_verbose >= 6)
3812 		(void) printf("Setting dataset %s to sync always\n", dsname);
3813 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3814 	    ZFS_SYNC_ALWAYS, B_FALSE));
3815 }
3816 
3817 /* ARGSUSED */
3818 static int
3819 ztest_objset_destroy_cb(const char *name, void *arg)
3820 {
3821 	objset_t *os;
3822 	dmu_object_info_t doi;
3823 	int error;
3824 
3825 	/*
3826 	 * Verify that the dataset contains a directory object.
3827 	 */
3828 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3829 	    B_TRUE, FTAG, &os));
3830 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3831 	if (error != ENOENT) {
3832 		/* We could have crashed in the middle of destroying it */
3833 		ASSERT0(error);
3834 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3835 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3836 	}
3837 	dmu_objset_disown(os, B_TRUE, FTAG);
3838 
3839 	/*
3840 	 * Destroy the dataset.
3841 	 */
3842 	if (strchr(name, '@') != NULL) {
3843 		VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3844 	} else {
3845 		error = dsl_destroy_head(name);
3846 		/* There could be a hold on this dataset */
3847 		if (error != EBUSY)
3848 			ASSERT0(error);
3849 	}
3850 	return (0);
3851 }
3852 
3853 static boolean_t
3854 ztest_snapshot_create(char *osname, uint64_t id)
3855 {
3856 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3857 	int error;
3858 
3859 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3860 
3861 	error = dmu_objset_snapshot_one(osname, snapname);
3862 	if (error == ENOSPC) {
3863 		ztest_record_enospc(FTAG);
3864 		return (B_FALSE);
3865 	}
3866 	if (error != 0 && error != EEXIST) {
3867 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3868 		    snapname, error);
3869 	}
3870 	return (B_TRUE);
3871 }
3872 
3873 static boolean_t
3874 ztest_snapshot_destroy(char *osname, uint64_t id)
3875 {
3876 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3877 	int error;
3878 
3879 	(void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3880 	    (u_longlong_t)id);
3881 
3882 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3883 	if (error != 0 && error != ENOENT)
3884 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3885 	return (B_TRUE);
3886 }
3887 
3888 /* ARGSUSED */
3889 void
3890 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3891 {
3892 	ztest_ds_t zdtmp;
3893 	int iters;
3894 	int error;
3895 	objset_t *os, *os2;
3896 	char name[ZFS_MAX_DATASET_NAME_LEN];
3897 	zilog_t *zilog;
3898 
3899 	rw_enter(&ztest_name_lock, RW_READER);
3900 
3901 	(void) snprintf(name, sizeof (name), "%s/temp_%llu",
3902 	    ztest_opts.zo_pool, (u_longlong_t)id);
3903 
3904 	/*
3905 	 * If this dataset exists from a previous run, process its replay log
3906 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3907 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3908 	 */
3909 	if (ztest_random(2) == 0 &&
3910 	    ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
3911 	    B_TRUE, FTAG, &os) == 0) {
3912 		ztest_zd_init(&zdtmp, NULL, os);
3913 		zil_replay(os, &zdtmp, ztest_replay_vector);
3914 		ztest_zd_fini(&zdtmp);
3915 		dmu_objset_disown(os, B_TRUE, FTAG);
3916 	}
3917 
3918 	/*
3919 	 * There may be an old instance of the dataset we're about to
3920 	 * create lying around from a previous run.  If so, destroy it
3921 	 * and all of its snapshots.
3922 	 */
3923 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3924 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3925 
3926 	/*
3927 	 * Verify that the destroyed dataset is no longer in the namespace.
3928 	 */
3929 	VERIFY3U(ENOENT, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3930 	    B_TRUE, FTAG, &os));
3931 
3932 	/*
3933 	 * Verify that we can create a new dataset.
3934 	 */
3935 	error = ztest_dataset_create(name);
3936 	if (error) {
3937 		if (error == ENOSPC) {
3938 			ztest_record_enospc(FTAG);
3939 			rw_exit(&ztest_name_lock);
3940 			return;
3941 		}
3942 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3943 	}
3944 
3945 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, B_TRUE,
3946 	    FTAG, &os));
3947 
3948 	ztest_zd_init(&zdtmp, NULL, os);
3949 
3950 	/*
3951 	 * Open the intent log for it.
3952 	 */
3953 	zilog = zil_open(os, ztest_get_data);
3954 
3955 	/*
3956 	 * Put some objects in there, do a little I/O to them,
3957 	 * and randomly take a couple of snapshots along the way.
3958 	 */
3959 	iters = ztest_random(5);
3960 	for (int i = 0; i < iters; i++) {
3961 		ztest_dmu_object_alloc_free(&zdtmp, id);
3962 		if (ztest_random(iters) == 0)
3963 			(void) ztest_snapshot_create(name, i);
3964 	}
3965 
3966 	/*
3967 	 * Verify that we cannot create an existing dataset.
3968 	 */
3969 	VERIFY3U(EEXIST, ==,
3970 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL, NULL));
3971 
3972 	/*
3973 	 * Verify that we can hold an objset that is also owned.
3974 	 */
3975 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3976 	dmu_objset_rele(os2, FTAG);
3977 
3978 	/*
3979 	 * Verify that we cannot own an objset that is already owned.
3980 	 */
3981 	VERIFY3U(EBUSY, ==, ztest_dmu_objset_own(name, DMU_OST_OTHER,
3982 	    B_FALSE, B_TRUE, FTAG, &os2));
3983 
3984 	zil_close(zilog);
3985 	dmu_objset_disown(os, B_TRUE, FTAG);
3986 	ztest_zd_fini(&zdtmp);
3987 
3988 	rw_exit(&ztest_name_lock);
3989 }
3990 
3991 /*
3992  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3993  */
3994 void
3995 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3996 {
3997 	rw_enter(&ztest_name_lock, RW_READER);
3998 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3999 	(void) ztest_snapshot_create(zd->zd_name, id);
4000 	rw_exit(&ztest_name_lock);
4001 }
4002 
4003 /*
4004  * Cleanup non-standard snapshots and clones.
4005  */
4006 void
4007 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
4008 {
4009 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
4010 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
4011 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
4012 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
4013 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
4014 	int error;
4015 
4016 	(void) snprintf(snap1name, sizeof (snap1name),
4017 	    "%s@s1_%llu", osname, id);
4018 	(void) snprintf(clone1name, sizeof (clone1name),
4019 	    "%s/c1_%llu", osname, id);
4020 	(void) snprintf(snap2name, sizeof (snap2name),
4021 	    "%s@s2_%llu", clone1name, id);
4022 	(void) snprintf(clone2name, sizeof (clone2name),
4023 	    "%s/c2_%llu", osname, id);
4024 	(void) snprintf(snap3name, sizeof (snap3name),
4025 	    "%s@s3_%llu", clone1name, id);
4026 
4027 	error = dsl_destroy_head(clone2name);
4028 	if (error && error != ENOENT)
4029 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
4030 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
4031 	if (error && error != ENOENT)
4032 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
4033 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
4034 	if (error && error != ENOENT)
4035 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
4036 	error = dsl_destroy_head(clone1name);
4037 	if (error && error != ENOENT)
4038 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
4039 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
4040 	if (error && error != ENOENT)
4041 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
4042 }
4043 
4044 /*
4045  * Verify dsl_dataset_promote handles EBUSY
4046  */
4047 void
4048 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
4049 {
4050 	objset_t *os;
4051 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
4052 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
4053 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
4054 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
4055 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
4056 	char *osname = zd->zd_name;
4057 	int error;
4058 
4059 	rw_enter(&ztest_name_lock, RW_READER);
4060 
4061 	ztest_dsl_dataset_cleanup(osname, id);
4062 
4063 	(void) snprintf(snap1name, sizeof (snap1name),
4064 	    "%s@s1_%llu", osname, id);
4065 	(void) snprintf(clone1name, sizeof (clone1name),
4066 	    "%s/c1_%llu", osname, id);
4067 	(void) snprintf(snap2name, sizeof (snap2name),
4068 	    "%s@s2_%llu", clone1name, id);
4069 	(void) snprintf(clone2name, sizeof (clone2name),
4070 	    "%s/c2_%llu", osname, id);
4071 	(void) snprintf(snap3name, sizeof (snap3name),
4072 	    "%s@s3_%llu", clone1name, id);
4073 
4074 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
4075 	if (error && error != EEXIST) {
4076 		if (error == ENOSPC) {
4077 			ztest_record_enospc(FTAG);
4078 			goto out;
4079 		}
4080 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
4081 	}
4082 
4083 	error = dmu_objset_clone(clone1name, snap1name);
4084 	if (error) {
4085 		if (error == ENOSPC) {
4086 			ztest_record_enospc(FTAG);
4087 			goto out;
4088 		}
4089 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
4090 	}
4091 
4092 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4093 	if (error && error != EEXIST) {
4094 		if (error == ENOSPC) {
4095 			ztest_record_enospc(FTAG);
4096 			goto out;
4097 		}
4098 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
4099 	}
4100 
4101 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4102 	if (error && error != EEXIST) {
4103 		if (error == ENOSPC) {
4104 			ztest_record_enospc(FTAG);
4105 			goto out;
4106 		}
4107 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
4108 	}
4109 
4110 	error = dmu_objset_clone(clone2name, snap3name);
4111 	if (error) {
4112 		if (error == ENOSPC) {
4113 			ztest_record_enospc(FTAG);
4114 			goto out;
4115 		}
4116 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
4117 	}
4118 
4119 	error = ztest_dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, B_TRUE,
4120 	    FTAG, &os);
4121 	if (error)
4122 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
4123 	error = dsl_dataset_promote(clone2name, NULL);
4124 	if (error == ENOSPC) {
4125 		dmu_objset_disown(os, B_TRUE, FTAG);
4126 		ztest_record_enospc(FTAG);
4127 		goto out;
4128 	}
4129 	if (error != EBUSY)
4130 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
4131 		    error);
4132 	dmu_objset_disown(os, B_TRUE, FTAG);
4133 
4134 out:
4135 	ztest_dsl_dataset_cleanup(osname, id);
4136 
4137 	rw_exit(&ztest_name_lock);
4138 }
4139 
4140 /*
4141  * Verify that dmu_object_{alloc,free} work as expected.
4142  */
4143 void
4144 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4145 {
4146 	ztest_od_t od[4];
4147 	int batchsize = sizeof (od) / sizeof (od[0]);
4148 
4149 	for (int b = 0; b < batchsize; b++) {
4150 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER,
4151 		    0, 0, 0);
4152 	}
4153 
4154 	/*
4155 	 * Destroy the previous batch of objects, create a new batch,
4156 	 * and do some I/O on the new objects.
4157 	 */
4158 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
4159 		return;
4160 
4161 	while (ztest_random(4 * batchsize) != 0)
4162 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
4163 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4164 }
4165 
4166 /*
4167  * Rewind the global allocator to verify object allocation backfilling.
4168  */
4169 void
4170 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
4171 {
4172 	objset_t *os = zd->zd_os;
4173 	int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
4174 	uint64_t object;
4175 
4176 	/*
4177 	 * Rewind the global allocator randomly back to a lower object number
4178 	 * to force backfilling and reclamation of recently freed dnodes.
4179 	 */
4180 	mutex_enter(&os->os_obj_lock);
4181 	object = ztest_random(os->os_obj_next_chunk);
4182 	os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
4183 	mutex_exit(&os->os_obj_lock);
4184 }
4185 
4186 /*
4187  * Verify that dmu_{read,write} work as expected.
4188  */
4189 void
4190 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4191 {
4192 	objset_t *os = zd->zd_os;
4193 	ztest_od_t od[2];
4194 	dmu_tx_t *tx;
4195 	int i, freeit, error;
4196 	uint64_t n, s, txg;
4197 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4198 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4199 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4200 	uint64_t regions = 997;
4201 	uint64_t stride = 123456789ULL;
4202 	uint64_t width = 40;
4203 	int free_percent = 5;
4204 
4205 	/*
4206 	 * This test uses two objects, packobj and bigobj, that are always
4207 	 * updated together (i.e. in the same tx) so that their contents are
4208 	 * in sync and can be compared.  Their contents relate to each other
4209 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4210 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4211 	 * for any index n, there are three bufwads that should be identical:
4212 	 *
4213 	 *	packobj, at offset n * sizeof (bufwad_t)
4214 	 *	bigobj, at the head of the nth chunk
4215 	 *	bigobj, at the tail of the nth chunk
4216 	 *
4217 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
4218 	 * and it doesn't have any relation to the object blocksize.
4219 	 * The only requirement is that it can hold at least two bufwads.
4220 	 *
4221 	 * Normally, we write the bufwad to each of these locations.
4222 	 * However, free_percent of the time we instead write zeroes to
4223 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
4224 	 * bigobj to packobj, we can verify that the DMU is correctly
4225 	 * tracking which parts of an object are allocated and free,
4226 	 * and that the contents of the allocated blocks are correct.
4227 	 */
4228 
4229 	/*
4230 	 * Read the directory info.  If it's the first time, set things up.
4231 	 */
4232 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0,
4233 	    chunksize);
4234 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4235 	    chunksize);
4236 
4237 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4238 		return;
4239 
4240 	bigobj = od[0].od_object;
4241 	packobj = od[1].od_object;
4242 	chunksize = od[0].od_gen;
4243 	ASSERT(chunksize == od[1].od_gen);
4244 
4245 	/*
4246 	 * Prefetch a random chunk of the big object.
4247 	 * Our aim here is to get some async reads in flight
4248 	 * for blocks that we may free below; the DMU should
4249 	 * handle this race correctly.
4250 	 */
4251 	n = ztest_random(regions) * stride + ztest_random(width);
4252 	s = 1 + ztest_random(2 * width - 1);
4253 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4254 	    ZIO_PRIORITY_SYNC_READ);
4255 
4256 	/*
4257 	 * Pick a random index and compute the offsets into packobj and bigobj.
4258 	 */
4259 	n = ztest_random(regions) * stride + ztest_random(width);
4260 	s = 1 + ztest_random(width - 1);
4261 
4262 	packoff = n * sizeof (bufwad_t);
4263 	packsize = s * sizeof (bufwad_t);
4264 
4265 	bigoff = n * chunksize;
4266 	bigsize = s * chunksize;
4267 
4268 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4269 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4270 
4271 	/*
4272 	 * free_percent of the time, free a range of bigobj rather than
4273 	 * overwriting it.
4274 	 */
4275 	freeit = (ztest_random(100) < free_percent);
4276 
4277 	/*
4278 	 * Read the current contents of our objects.
4279 	 */
4280 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
4281 	    DMU_READ_PREFETCH);
4282 	ASSERT0(error);
4283 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4284 	    DMU_READ_PREFETCH);
4285 	ASSERT0(error);
4286 
4287 	/*
4288 	 * Get a tx for the mods to both packobj and bigobj.
4289 	 */
4290 	tx = dmu_tx_create(os);
4291 
4292 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
4293 
4294 	if (freeit)
4295 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4296 	else
4297 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4298 
4299 	/* This accounts for setting the checksum/compression. */
4300 	dmu_tx_hold_bonus(tx, bigobj);
4301 
4302 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4303 	if (txg == 0) {
4304 		umem_free(packbuf, packsize);
4305 		umem_free(bigbuf, bigsize);
4306 		return;
4307 	}
4308 
4309 	enum zio_checksum cksum;
4310 	do {
4311 		cksum = (enum zio_checksum)
4312 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4313 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4314 	dmu_object_set_checksum(os, bigobj, cksum, tx);
4315 
4316 	enum zio_compress comp;
4317 	do {
4318 		comp = (enum zio_compress)
4319 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4320 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4321 	dmu_object_set_compress(os, bigobj, comp, tx);
4322 
4323 	/*
4324 	 * For each index from n to n + s, verify that the existing bufwad
4325 	 * in packobj matches the bufwads at the head and tail of the
4326 	 * corresponding chunk in bigobj.  Then update all three bufwads
4327 	 * with the new values we want to write out.
4328 	 */
4329 	for (i = 0; i < s; i++) {
4330 		/* LINTED */
4331 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4332 		/* LINTED */
4333 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4334 		/* LINTED */
4335 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4336 
4337 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4338 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4339 
4340 		if (pack->bw_txg > txg)
4341 			fatal(0, "future leak: got %llx, open txg is %llx",
4342 			    pack->bw_txg, txg);
4343 
4344 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4345 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4346 			    pack->bw_index, n, i);
4347 
4348 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4349 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4350 
4351 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4352 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4353 
4354 		if (freeit) {
4355 			bzero(pack, sizeof (bufwad_t));
4356 		} else {
4357 			pack->bw_index = n + i;
4358 			pack->bw_txg = txg;
4359 			pack->bw_data = 1 + ztest_random(-2ULL);
4360 		}
4361 		*bigH = *pack;
4362 		*bigT = *pack;
4363 	}
4364 
4365 	/*
4366 	 * We've verified all the old bufwads, and made new ones.
4367 	 * Now write them out.
4368 	 */
4369 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4370 
4371 	if (freeit) {
4372 		if (ztest_opts.zo_verbose >= 7) {
4373 			(void) printf("freeing offset %llx size %llx"
4374 			    " txg %llx\n",
4375 			    (u_longlong_t)bigoff,
4376 			    (u_longlong_t)bigsize,
4377 			    (u_longlong_t)txg);
4378 		}
4379 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4380 	} else {
4381 		if (ztest_opts.zo_verbose >= 7) {
4382 			(void) printf("writing offset %llx size %llx"
4383 			    " txg %llx\n",
4384 			    (u_longlong_t)bigoff,
4385 			    (u_longlong_t)bigsize,
4386 			    (u_longlong_t)txg);
4387 		}
4388 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4389 	}
4390 
4391 	dmu_tx_commit(tx);
4392 
4393 	/*
4394 	 * Sanity check the stuff we just wrote.
4395 	 */
4396 	{
4397 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4398 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4399 
4400 		VERIFY(0 == dmu_read(os, packobj, packoff,
4401 		    packsize, packcheck, DMU_READ_PREFETCH));
4402 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
4403 		    bigsize, bigcheck, DMU_READ_PREFETCH));
4404 
4405 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4406 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4407 
4408 		umem_free(packcheck, packsize);
4409 		umem_free(bigcheck, bigsize);
4410 	}
4411 
4412 	umem_free(packbuf, packsize);
4413 	umem_free(bigbuf, bigsize);
4414 }
4415 
4416 void
4417 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4418     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4419 {
4420 	uint64_t i;
4421 	bufwad_t *pack;
4422 	bufwad_t *bigH;
4423 	bufwad_t *bigT;
4424 
4425 	/*
4426 	 * For each index from n to n + s, verify that the existing bufwad
4427 	 * in packobj matches the bufwads at the head and tail of the
4428 	 * corresponding chunk in bigobj.  Then update all three bufwads
4429 	 * with the new values we want to write out.
4430 	 */
4431 	for (i = 0; i < s; i++) {
4432 		/* LINTED */
4433 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4434 		/* LINTED */
4435 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4436 		/* LINTED */
4437 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4438 
4439 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4440 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4441 
4442 		if (pack->bw_txg > txg)
4443 			fatal(0, "future leak: got %llx, open txg is %llx",
4444 			    pack->bw_txg, txg);
4445 
4446 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4447 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4448 			    pack->bw_index, n, i);
4449 
4450 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4451 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4452 
4453 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4454 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4455 
4456 		pack->bw_index = n + i;
4457 		pack->bw_txg = txg;
4458 		pack->bw_data = 1 + ztest_random(-2ULL);
4459 
4460 		*bigH = *pack;
4461 		*bigT = *pack;
4462 	}
4463 }
4464 
4465 void
4466 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4467 {
4468 	objset_t *os = zd->zd_os;
4469 	ztest_od_t od[2];
4470 	dmu_tx_t *tx;
4471 	uint64_t i;
4472 	int error;
4473 	uint64_t n, s, txg;
4474 	bufwad_t *packbuf, *bigbuf;
4475 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4476 	uint64_t blocksize = ztest_random_blocksize();
4477 	uint64_t chunksize = blocksize;
4478 	uint64_t regions = 997;
4479 	uint64_t stride = 123456789ULL;
4480 	uint64_t width = 9;
4481 	dmu_buf_t *bonus_db;
4482 	arc_buf_t **bigbuf_arcbufs;
4483 	dmu_object_info_t doi;
4484 
4485 	/*
4486 	 * This test uses two objects, packobj and bigobj, that are always
4487 	 * updated together (i.e. in the same tx) so that their contents are
4488 	 * in sync and can be compared.  Their contents relate to each other
4489 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4490 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4491 	 * for any index n, there are three bufwads that should be identical:
4492 	 *
4493 	 *	packobj, at offset n * sizeof (bufwad_t)
4494 	 *	bigobj, at the head of the nth chunk
4495 	 *	bigobj, at the tail of the nth chunk
4496 	 *
4497 	 * The chunk size is set equal to bigobj block size so that
4498 	 * dmu_assign_arcbuf_by_dbuf() can be tested for object updates.
4499 	 */
4500 
4501 	/*
4502 	 * Read the directory info.  If it's the first time, set things up.
4503 	 */
4504 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
4505 	    0, 0);
4506 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4507 	    chunksize);
4508 
4509 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4510 		return;
4511 
4512 	bigobj = od[0].od_object;
4513 	packobj = od[1].od_object;
4514 	blocksize = od[0].od_blocksize;
4515 	chunksize = blocksize;
4516 	ASSERT(chunksize == od[1].od_gen);
4517 
4518 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4519 	VERIFY(ISP2(doi.doi_data_block_size));
4520 	VERIFY(chunksize == doi.doi_data_block_size);
4521 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4522 
4523 	/*
4524 	 * Pick a random index and compute the offsets into packobj and bigobj.
4525 	 */
4526 	n = ztest_random(regions) * stride + ztest_random(width);
4527 	s = 1 + ztest_random(width - 1);
4528 
4529 	packoff = n * sizeof (bufwad_t);
4530 	packsize = s * sizeof (bufwad_t);
4531 
4532 	bigoff = n * chunksize;
4533 	bigsize = s * chunksize;
4534 
4535 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4536 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4537 
4538 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4539 
4540 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4541 
4542 	/*
4543 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4544 	 * Iteration 1 test zcopy to already referenced dbufs.
4545 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4546 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4547 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
4548 	 * Iteration 5 test zcopy when it can't be done.
4549 	 * Iteration 6 one more zcopy write.
4550 	 */
4551 	for (i = 0; i < 7; i++) {
4552 		uint64_t j;
4553 		uint64_t off;
4554 
4555 		/*
4556 		 * In iteration 5 (i == 5) use arcbufs
4557 		 * that don't match bigobj blksz to test
4558 		 * dmu_assign_arcbuf_by_dbuf() when it can't directly
4559 		 * assign an arcbuf to a dbuf.
4560 		 */
4561 		for (j = 0; j < s; j++) {
4562 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4563 				bigbuf_arcbufs[j] =
4564 				    dmu_request_arcbuf(bonus_db, chunksize);
4565 			} else {
4566 				bigbuf_arcbufs[2 * j] =
4567 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
4568 				bigbuf_arcbufs[2 * j + 1] =
4569 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
4570 			}
4571 		}
4572 
4573 		/*
4574 		 * Get a tx for the mods to both packobj and bigobj.
4575 		 */
4576 		tx = dmu_tx_create(os);
4577 
4578 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
4579 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4580 
4581 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4582 		if (txg == 0) {
4583 			umem_free(packbuf, packsize);
4584 			umem_free(bigbuf, bigsize);
4585 			for (j = 0; j < s; j++) {
4586 				if (i != 5 ||
4587 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
4588 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
4589 				} else {
4590 					dmu_return_arcbuf(
4591 					    bigbuf_arcbufs[2 * j]);
4592 					dmu_return_arcbuf(
4593 					    bigbuf_arcbufs[2 * j + 1]);
4594 				}
4595 			}
4596 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4597 			dmu_buf_rele(bonus_db, FTAG);
4598 			return;
4599 		}
4600 
4601 		/*
4602 		 * 50% of the time don't read objects in the 1st iteration to
4603 		 * test dmu_assign_arcbuf_by_dbuf() for the case when there are
4604 		 * no existing dbufs for the specified offsets.
4605 		 */
4606 		if (i != 0 || ztest_random(2) != 0) {
4607 			error = dmu_read(os, packobj, packoff,
4608 			    packsize, packbuf, DMU_READ_PREFETCH);
4609 			ASSERT0(error);
4610 			error = dmu_read(os, bigobj, bigoff, bigsize,
4611 			    bigbuf, DMU_READ_PREFETCH);
4612 			ASSERT0(error);
4613 		}
4614 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4615 		    n, chunksize, txg);
4616 
4617 		/*
4618 		 * We've verified all the old bufwads, and made new ones.
4619 		 * Now write them out.
4620 		 */
4621 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4622 		if (ztest_opts.zo_verbose >= 7) {
4623 			(void) printf("writing offset %llx size %llx"
4624 			    " txg %llx\n",
4625 			    (u_longlong_t)bigoff,
4626 			    (u_longlong_t)bigsize,
4627 			    (u_longlong_t)txg);
4628 		}
4629 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4630 			dmu_buf_t *dbt;
4631 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4632 				bcopy((caddr_t)bigbuf + (off - bigoff),
4633 				    bigbuf_arcbufs[j]->b_data, chunksize);
4634 			} else {
4635 				bcopy((caddr_t)bigbuf + (off - bigoff),
4636 				    bigbuf_arcbufs[2 * j]->b_data,
4637 				    chunksize / 2);
4638 				bcopy((caddr_t)bigbuf + (off - bigoff) +
4639 				    chunksize / 2,
4640 				    bigbuf_arcbufs[2 * j + 1]->b_data,
4641 				    chunksize / 2);
4642 			}
4643 
4644 			if (i == 1) {
4645 				VERIFY(dmu_buf_hold(os, bigobj, off,
4646 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4647 			}
4648 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4649 				dmu_assign_arcbuf_by_dbuf(bonus_db, off,
4650 				    bigbuf_arcbufs[j], tx);
4651 			} else {
4652 				dmu_assign_arcbuf_by_dbuf(bonus_db, off,
4653 				    bigbuf_arcbufs[2 * j], tx);
4654 				dmu_assign_arcbuf_by_dbuf(bonus_db,
4655 				    off + chunksize / 2,
4656 				    bigbuf_arcbufs[2 * j + 1], tx);
4657 			}
4658 			if (i == 1) {
4659 				dmu_buf_rele(dbt, FTAG);
4660 			}
4661 		}
4662 		dmu_tx_commit(tx);
4663 
4664 		/*
4665 		 * Sanity check the stuff we just wrote.
4666 		 */
4667 		{
4668 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4669 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4670 
4671 			VERIFY(0 == dmu_read(os, packobj, packoff,
4672 			    packsize, packcheck, DMU_READ_PREFETCH));
4673 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4674 			    bigsize, bigcheck, DMU_READ_PREFETCH));
4675 
4676 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4677 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4678 
4679 			umem_free(packcheck, packsize);
4680 			umem_free(bigcheck, bigsize);
4681 		}
4682 		if (i == 2) {
4683 			txg_wait_open(dmu_objset_pool(os), 0, B_TRUE);
4684 		} else if (i == 3) {
4685 			txg_wait_synced(dmu_objset_pool(os), 0);
4686 		}
4687 	}
4688 
4689 	dmu_buf_rele(bonus_db, FTAG);
4690 	umem_free(packbuf, packsize);
4691 	umem_free(bigbuf, bigsize);
4692 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4693 }
4694 
4695 /* ARGSUSED */
4696 void
4697 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4698 {
4699 	ztest_od_t od[1];
4700 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4701 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4702 
4703 	/*
4704 	 * Have multiple threads write to large offsets in an object
4705 	 * to verify that parallel writes to an object -- even to the
4706 	 * same blocks within the object -- doesn't cause any trouble.
4707 	 */
4708 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER,
4709 	    0, 0, 0);
4710 
4711 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4712 		return;
4713 
4714 	while (ztest_random(10) != 0)
4715 		ztest_io(zd, od[0].od_object, offset);
4716 }
4717 
4718 void
4719 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4720 {
4721 	ztest_od_t od[1];
4722 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4723 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4724 	uint64_t count = ztest_random(20) + 1;
4725 	uint64_t blocksize = ztest_random_blocksize();
4726 	void *data;
4727 
4728 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
4729 	    0, 0);
4730 
4731 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4732 		return;
4733 
4734 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4735 		return;
4736 
4737 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4738 
4739 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4740 
4741 	while (ztest_random(count) != 0) {
4742 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4743 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4744 		    data) != 0)
4745 			break;
4746 		while (ztest_random(4) != 0)
4747 			ztest_io(zd, od[0].od_object, randoff);
4748 	}
4749 
4750 	umem_free(data, blocksize);
4751 }
4752 
4753 /*
4754  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4755  */
4756 #define	ZTEST_ZAP_MIN_INTS	1
4757 #define	ZTEST_ZAP_MAX_INTS	4
4758 #define	ZTEST_ZAP_MAX_PROPS	1000
4759 
4760 void
4761 ztest_zap(ztest_ds_t *zd, uint64_t id)
4762 {
4763 	objset_t *os = zd->zd_os;
4764 	ztest_od_t od[1];
4765 	uint64_t object;
4766 	uint64_t txg, last_txg;
4767 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4768 	uint64_t zl_ints, zl_intsize, prop;
4769 	int i, ints;
4770 	dmu_tx_t *tx;
4771 	char propname[100], txgname[100];
4772 	int error;
4773 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4774 
4775 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4776 
4777 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4778 		return;
4779 
4780 	object = od[0].od_object;
4781 
4782 	/*
4783 	 * Generate a known hash collision, and verify that
4784 	 * we can lookup and remove both entries.
4785 	 */
4786 	tx = dmu_tx_create(os);
4787 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4788 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4789 	if (txg == 0)
4790 		return;
4791 	for (i = 0; i < 2; i++) {
4792 		value[i] = i;
4793 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4794 		    1, &value[i], tx));
4795 	}
4796 	for (i = 0; i < 2; i++) {
4797 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4798 		    sizeof (uint64_t), 1, &value[i], tx));
4799 		VERIFY3U(0, ==,
4800 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4801 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4802 		ASSERT3U(zl_ints, ==, 1);
4803 	}
4804 	for (i = 0; i < 2; i++) {
4805 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4806 	}
4807 	dmu_tx_commit(tx);
4808 
4809 	/*
4810 	 * Generate a buch of random entries.
4811 	 */
4812 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4813 
4814 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4815 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4816 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4817 	bzero(value, sizeof (value));
4818 	last_txg = 0;
4819 
4820 	/*
4821 	 * If these zap entries already exist, validate their contents.
4822 	 */
4823 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4824 	if (error == 0) {
4825 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4826 		ASSERT3U(zl_ints, ==, 1);
4827 
4828 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4829 		    zl_ints, &last_txg) == 0);
4830 
4831 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4832 		    &zl_ints) == 0);
4833 
4834 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4835 		ASSERT3U(zl_ints, ==, ints);
4836 
4837 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4838 		    zl_ints, value) == 0);
4839 
4840 		for (i = 0; i < ints; i++) {
4841 			ASSERT3U(value[i], ==, last_txg + object + i);
4842 		}
4843 	} else {
4844 		ASSERT3U(error, ==, ENOENT);
4845 	}
4846 
4847 	/*
4848 	 * Atomically update two entries in our zap object.
4849 	 * The first is named txg_%llu, and contains the txg
4850 	 * in which the property was last updated.  The second
4851 	 * is named prop_%llu, and the nth element of its value
4852 	 * should be txg + object + n.
4853 	 */
4854 	tx = dmu_tx_create(os);
4855 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4856 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4857 	if (txg == 0)
4858 		return;
4859 
4860 	if (last_txg > txg)
4861 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4862 
4863 	for (i = 0; i < ints; i++)
4864 		value[i] = txg + object + i;
4865 
4866 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4867 	    1, &txg, tx));
4868 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4869 	    ints, value, tx));
4870 
4871 	dmu_tx_commit(tx);
4872 
4873 	/*
4874 	 * Remove a random pair of entries.
4875 	 */
4876 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4877 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4878 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4879 
4880 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4881 
4882 	if (error == ENOENT)
4883 		return;
4884 
4885 	ASSERT0(error);
4886 
4887 	tx = dmu_tx_create(os);
4888 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4889 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4890 	if (txg == 0)
4891 		return;
4892 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4893 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4894 	dmu_tx_commit(tx);
4895 }
4896 
4897 /*
4898  * Testcase to test the upgrading of a microzap to fatzap.
4899  */
4900 void
4901 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4902 {
4903 	objset_t *os = zd->zd_os;
4904 	ztest_od_t od[1];
4905 	uint64_t object, txg;
4906 
4907 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4908 
4909 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4910 		return;
4911 
4912 	object = od[0].od_object;
4913 
4914 	/*
4915 	 * Add entries to this ZAP and make sure it spills over
4916 	 * and gets upgraded to a fatzap. Also, since we are adding
4917 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4918 	 */
4919 	for (int i = 0; i < 2050; i++) {
4920 		char name[ZFS_MAX_DATASET_NAME_LEN];
4921 		uint64_t value = i;
4922 		dmu_tx_t *tx;
4923 		int error;
4924 
4925 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4926 		    id, value);
4927 
4928 		tx = dmu_tx_create(os);
4929 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4930 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4931 		if (txg == 0)
4932 			return;
4933 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4934 		    &value, tx);
4935 		ASSERT(error == 0 || error == EEXIST);
4936 		dmu_tx_commit(tx);
4937 	}
4938 }
4939 
4940 /* ARGSUSED */
4941 void
4942 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4943 {
4944 	objset_t *os = zd->zd_os;
4945 	ztest_od_t od[1];
4946 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4947 	dmu_tx_t *tx;
4948 	int i, namelen, error;
4949 	int micro = ztest_random(2);
4950 	char name[20], string_value[20];
4951 	void *data;
4952 
4953 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER,
4954 	    0, 0, 0);
4955 
4956 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4957 		return;
4958 
4959 	object = od[0].od_object;
4960 
4961 	/*
4962 	 * Generate a random name of the form 'xxx.....' where each
4963 	 * x is a random printable character and the dots are dots.
4964 	 * There are 94 such characters, and the name length goes from
4965 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4966 	 */
4967 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4968 
4969 	for (i = 0; i < 3; i++)
4970 		name[i] = '!' + ztest_random('~' - '!' + 1);
4971 	for (; i < namelen - 1; i++)
4972 		name[i] = '.';
4973 	name[i] = '\0';
4974 
4975 	if ((namelen & 1) || micro) {
4976 		wsize = sizeof (txg);
4977 		wc = 1;
4978 		data = &txg;
4979 	} else {
4980 		wsize = 1;
4981 		wc = namelen;
4982 		data = string_value;
4983 	}
4984 
4985 	count = -1ULL;
4986 	VERIFY0(zap_count(os, object, &count));
4987 	ASSERT(count != -1ULL);
4988 
4989 	/*
4990 	 * Select an operation: length, lookup, add, update, remove.
4991 	 */
4992 	i = ztest_random(5);
4993 
4994 	if (i >= 2) {
4995 		tx = dmu_tx_create(os);
4996 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4997 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4998 		if (txg == 0)
4999 			return;
5000 		bcopy(name, string_value, namelen);
5001 	} else {
5002 		tx = NULL;
5003 		txg = 0;
5004 		bzero(string_value, namelen);
5005 	}
5006 
5007 	switch (i) {
5008 
5009 	case 0:
5010 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
5011 		if (error == 0) {
5012 			ASSERT3U(wsize, ==, zl_wsize);
5013 			ASSERT3U(wc, ==, zl_wc);
5014 		} else {
5015 			ASSERT3U(error, ==, ENOENT);
5016 		}
5017 		break;
5018 
5019 	case 1:
5020 		error = zap_lookup(os, object, name, wsize, wc, data);
5021 		if (error == 0) {
5022 			if (data == string_value &&
5023 			    bcmp(name, data, namelen) != 0)
5024 				fatal(0, "name '%s' != val '%s' len %d",
5025 				    name, data, namelen);
5026 		} else {
5027 			ASSERT3U(error, ==, ENOENT);
5028 		}
5029 		break;
5030 
5031 	case 2:
5032 		error = zap_add(os, object, name, wsize, wc, data, tx);
5033 		ASSERT(error == 0 || error == EEXIST);
5034 		break;
5035 
5036 	case 3:
5037 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
5038 		break;
5039 
5040 	case 4:
5041 		error = zap_remove(os, object, name, tx);
5042 		ASSERT(error == 0 || error == ENOENT);
5043 		break;
5044 	}
5045 
5046 	if (tx != NULL)
5047 		dmu_tx_commit(tx);
5048 }
5049 
5050 /*
5051  * Commit callback data.
5052  */
5053 typedef struct ztest_cb_data {
5054 	list_node_t		zcd_node;
5055 	uint64_t		zcd_txg;
5056 	int			zcd_expected_err;
5057 	boolean_t		zcd_added;
5058 	boolean_t		zcd_called;
5059 	spa_t			*zcd_spa;
5060 } ztest_cb_data_t;
5061 
5062 /* This is the actual commit callback function */
5063 static void
5064 ztest_commit_callback(void *arg, int error)
5065 {
5066 	ztest_cb_data_t *data = arg;
5067 	uint64_t synced_txg;
5068 
5069 	VERIFY(data != NULL);
5070 	VERIFY3S(data->zcd_expected_err, ==, error);
5071 	VERIFY(!data->zcd_called);
5072 
5073 	synced_txg = spa_last_synced_txg(data->zcd_spa);
5074 	if (data->zcd_txg > synced_txg)
5075 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
5076 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
5077 		    synced_txg);
5078 
5079 	data->zcd_called = B_TRUE;
5080 
5081 	if (error == ECANCELED) {
5082 		ASSERT0(data->zcd_txg);
5083 		ASSERT(!data->zcd_added);
5084 
5085 		/*
5086 		 * The private callback data should be destroyed here, but
5087 		 * since we are going to check the zcd_called field after
5088 		 * dmu_tx_abort(), we will destroy it there.
5089 		 */
5090 		return;
5091 	}
5092 
5093 	/* Was this callback added to the global callback list? */
5094 	if (!data->zcd_added)
5095 		goto out;
5096 
5097 	ASSERT3U(data->zcd_txg, !=, 0);
5098 
5099 	/* Remove our callback from the list */
5100 	mutex_enter(&zcl.zcl_callbacks_lock);
5101 	list_remove(&zcl.zcl_callbacks, data);
5102 	mutex_exit(&zcl.zcl_callbacks_lock);
5103 
5104 out:
5105 	umem_free(data, sizeof (ztest_cb_data_t));
5106 }
5107 
5108 /* Allocate and initialize callback data structure */
5109 static ztest_cb_data_t *
5110 ztest_create_cb_data(objset_t *os, uint64_t txg)
5111 {
5112 	ztest_cb_data_t *cb_data;
5113 
5114 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
5115 
5116 	cb_data->zcd_txg = txg;
5117 	cb_data->zcd_spa = dmu_objset_spa(os);
5118 
5119 	return (cb_data);
5120 }
5121 
5122 /*
5123  * If a number of txgs equal to this threshold have been created after a commit
5124  * callback has been registered but not called, then we assume there is an
5125  * implementation bug.
5126  */
5127 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
5128 
5129 /*
5130  * Commit callback test.
5131  */
5132 void
5133 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5134 {
5135 	objset_t *os = zd->zd_os;
5136 	ztest_od_t od[1];
5137 	dmu_tx_t *tx;
5138 	ztest_cb_data_t *cb_data[3], *tmp_cb;
5139 	uint64_t old_txg, txg;
5140 	int i, error;
5141 
5142 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5143 
5144 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5145 		return;
5146 
5147 	tx = dmu_tx_create(os);
5148 
5149 	cb_data[0] = ztest_create_cb_data(os, 0);
5150 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5151 
5152 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
5153 
5154 	/* Every once in a while, abort the transaction on purpose */
5155 	if (ztest_random(100) == 0)
5156 		error = -1;
5157 
5158 	if (!error)
5159 		error = dmu_tx_assign(tx, TXG_NOWAIT);
5160 
5161 	txg = error ? 0 : dmu_tx_get_txg(tx);
5162 
5163 	cb_data[0]->zcd_txg = txg;
5164 	cb_data[1] = ztest_create_cb_data(os, txg);
5165 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5166 
5167 	if (error) {
5168 		/*
5169 		 * It's not a strict requirement to call the registered
5170 		 * callbacks from inside dmu_tx_abort(), but that's what
5171 		 * it's supposed to happen in the current implementation
5172 		 * so we will check for that.
5173 		 */
5174 		for (i = 0; i < 2; i++) {
5175 			cb_data[i]->zcd_expected_err = ECANCELED;
5176 			VERIFY(!cb_data[i]->zcd_called);
5177 		}
5178 
5179 		dmu_tx_abort(tx);
5180 
5181 		for (i = 0; i < 2; i++) {
5182 			VERIFY(cb_data[i]->zcd_called);
5183 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5184 		}
5185 
5186 		return;
5187 	}
5188 
5189 	cb_data[2] = ztest_create_cb_data(os, txg);
5190 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5191 
5192 	/*
5193 	 * Read existing data to make sure there isn't a future leak.
5194 	 */
5195 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
5196 	    &old_txg, DMU_READ_PREFETCH));
5197 
5198 	if (old_txg > txg)
5199 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
5200 		    old_txg, txg);
5201 
5202 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
5203 
5204 	mutex_enter(&zcl.zcl_callbacks_lock);
5205 
5206 	/*
5207 	 * Since commit callbacks don't have any ordering requirement and since
5208 	 * it is theoretically possible for a commit callback to be called
5209 	 * after an arbitrary amount of time has elapsed since its txg has been
5210 	 * synced, it is difficult to reliably determine whether a commit
5211 	 * callback hasn't been called due to high load or due to a flawed
5212 	 * implementation.
5213 	 *
5214 	 * In practice, we will assume that if after a certain number of txgs a
5215 	 * commit callback hasn't been called, then most likely there's an
5216 	 * implementation bug..
5217 	 */
5218 	tmp_cb = list_head(&zcl.zcl_callbacks);
5219 	if (tmp_cb != NULL &&
5220 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
5221 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
5222 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
5223 	}
5224 
5225 	/*
5226 	 * Let's find the place to insert our callbacks.
5227 	 *
5228 	 * Even though the list is ordered by txg, it is possible for the
5229 	 * insertion point to not be the end because our txg may already be
5230 	 * quiescing at this point and other callbacks in the open txg
5231 	 * (from other objsets) may have sneaked in.
5232 	 */
5233 	tmp_cb = list_tail(&zcl.zcl_callbacks);
5234 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5235 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5236 
5237 	/* Add the 3 callbacks to the list */
5238 	for (i = 0; i < 3; i++) {
5239 		if (tmp_cb == NULL)
5240 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5241 		else
5242 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5243 			    cb_data[i]);
5244 
5245 		cb_data[i]->zcd_added = B_TRUE;
5246 		VERIFY(!cb_data[i]->zcd_called);
5247 
5248 		tmp_cb = cb_data[i];
5249 	}
5250 
5251 	mutex_exit(&zcl.zcl_callbacks_lock);
5252 
5253 	dmu_tx_commit(tx);
5254 }
5255 
5256 /*
5257  * Visit each object in the dataset. Verify that its properties
5258  * are consistent what was stored in the block tag when it was created,
5259  * and that its unused bonus buffer space has not been overwritten.
5260  */
5261 void
5262 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5263 {
5264 	objset_t *os = zd->zd_os;
5265 	uint64_t obj;
5266 	int err = 0;
5267 
5268 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5269 		ztest_block_tag_t *bt = NULL;
5270 		dmu_object_info_t doi;
5271 		dmu_buf_t *db;
5272 
5273 		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0)
5274 			continue;
5275 
5276 		dmu_object_info_from_db(db, &doi);
5277 		if (doi.doi_bonus_size >= sizeof (*bt))
5278 			bt = ztest_bt_bonus(db);
5279 
5280 		if (bt && bt->bt_magic == BT_MAGIC) {
5281 			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5282 			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
5283 			    bt->bt_crtxg);
5284 			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5285 		}
5286 
5287 		dmu_buf_rele(db, FTAG);
5288 	}
5289 }
5290 
5291 /* ARGSUSED */
5292 void
5293 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5294 {
5295 	zfs_prop_t proplist[] = {
5296 		ZFS_PROP_CHECKSUM,
5297 		ZFS_PROP_COMPRESSION,
5298 		ZFS_PROP_COPIES,
5299 		ZFS_PROP_DEDUP
5300 	};
5301 
5302 	rw_enter(&ztest_name_lock, RW_READER);
5303 
5304 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5305 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5306 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5307 
5308 	rw_exit(&ztest_name_lock);
5309 }
5310 
5311 /* ARGSUSED */
5312 void
5313 ztest_remap_blocks(ztest_ds_t *zd, uint64_t id)
5314 {
5315 	rw_enter(&ztest_name_lock, RW_READER);
5316 
5317 	int error = dmu_objset_remap_indirects(zd->zd_name);
5318 	if (error == ENOSPC)
5319 		error = 0;
5320 	ASSERT0(error);
5321 
5322 	rw_exit(&ztest_name_lock);
5323 }
5324 
5325 /* ARGSUSED */
5326 void
5327 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5328 {
5329 	nvlist_t *props = NULL;
5330 
5331 	rw_enter(&ztest_name_lock, RW_READER);
5332 
5333 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
5334 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
5335 
5336 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_AUTOTRIM, ztest_random(2));
5337 
5338 	VERIFY0(spa_prop_get(ztest_spa, &props));
5339 
5340 	if (ztest_opts.zo_verbose >= 6)
5341 		dump_nvlist(props, 4);
5342 
5343 	nvlist_free(props);
5344 
5345 	rw_exit(&ztest_name_lock);
5346 }
5347 
5348 void
5349 ztest_fletcher(ztest_ds_t *zd, uint64_t id)
5350 {
5351 	hrtime_t end = gethrtime() + NANOSEC;
5352 
5353 	while (gethrtime() <= end) {
5354 		int run_count = 100;
5355 		void *buf;
5356 		struct abd *abd_data, *abd_meta;
5357 		uint32_t size;
5358 		uint_t *ptr;
5359 		int i;
5360 		zio_cksum_t zc_ref;
5361 		zio_cksum_t zc_ref_byteswap;
5362 
5363 		size = ztest_random_blocksize();
5364 
5365 		buf = umem_alloc(size, UMEM_NOFAIL);
5366 		abd_data = abd_alloc(size, B_FALSE);
5367 		abd_meta = abd_alloc(size, B_TRUE);
5368 
5369 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5370 			*ptr = ztest_random(UINT_MAX);
5371 
5372 		abd_copy_from_buf_off(abd_data, buf, 0, size);
5373 		abd_copy_from_buf_off(abd_meta, buf, 0, size);
5374 
5375 		VERIFY0(fletcher_4_impl_set("scalar"));
5376 		fletcher_4_native(buf, size, NULL, &zc_ref);
5377 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_byteswap);
5378 
5379 		VERIFY0(fletcher_4_impl_set("cycle"));
5380 		while (run_count-- > 0) {
5381 			zio_cksum_t zc;
5382 			zio_cksum_t zc_byteswap;
5383 
5384 			fletcher_4_byteswap(buf, size, NULL, &zc_byteswap);
5385 			fletcher_4_native(buf, size, NULL, &zc);
5386 
5387 			VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5388 			VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5389 			    sizeof (zc_byteswap)));
5390 
5391 			/* Test ABD - data */
5392 			abd_fletcher_4_byteswap(abd_data, size, NULL,
5393 			    &zc_byteswap);
5394 			abd_fletcher_4_native(abd_data, size, NULL, &zc);
5395 
5396 			VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5397 			VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5398 			    sizeof (zc_byteswap)));
5399 
5400 			/* Test ABD - metadata */
5401 			abd_fletcher_4_byteswap(abd_meta, size, NULL,
5402 			    &zc_byteswap);
5403 			abd_fletcher_4_native(abd_meta, size, NULL, &zc);
5404 
5405 			VERIFY0(bcmp(&zc, &zc_ref, sizeof (zc)));
5406 			VERIFY0(bcmp(&zc_byteswap, &zc_ref_byteswap,
5407 			    sizeof (zc_byteswap)));
5408 
5409 		}
5410 
5411 		umem_free(buf, size);
5412 		abd_free(abd_data);
5413 		abd_free(abd_meta);
5414 	}
5415 }
5416 
5417 void
5418 ztest_fletcher_incr(ztest_ds_t *zd, uint64_t id)
5419 {
5420 	void *buf;
5421 	size_t size;
5422 	uint_t *ptr;
5423 	int i;
5424 	zio_cksum_t zc_ref;
5425 	zio_cksum_t zc_ref_bswap;
5426 
5427 	hrtime_t end = gethrtime() + NANOSEC;
5428 
5429 	while (gethrtime() <= end) {
5430 		int run_count = 100;
5431 
5432 		size = ztest_random_blocksize();
5433 		buf = umem_alloc(size, UMEM_NOFAIL);
5434 
5435 		for (i = 0, ptr = buf; i < size / sizeof (*ptr); i++, ptr++)
5436 			*ptr = ztest_random(UINT_MAX);
5437 
5438 		VERIFY0(fletcher_4_impl_set("scalar"));
5439 		fletcher_4_native(buf, size, NULL,  &zc_ref);
5440 		fletcher_4_byteswap(buf, size, NULL, &zc_ref_bswap);
5441 
5442 		VERIFY0(fletcher_4_impl_set("cycle"));
5443 
5444 		while (run_count-- > 0) {
5445 			zio_cksum_t zc;
5446 			zio_cksum_t zc_bswap;
5447 			size_t pos = 0;
5448 
5449 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
5450 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
5451 
5452 			while (pos < size) {
5453 				size_t inc = 64 * ztest_random(size / 67);
5454 				/* sometimes add few bytes to test non-simd */
5455 				if (ztest_random(100) < 10)
5456 					inc += P2ALIGN(ztest_random(64),
5457 					    sizeof (uint32_t));
5458 
5459 				if (inc > (size - pos))
5460 					inc = size - pos;
5461 
5462 				fletcher_4_incremental_native(buf + pos, inc,
5463 				    &zc);
5464 				fletcher_4_incremental_byteswap(buf + pos, inc,
5465 				    &zc_bswap);
5466 
5467 				pos += inc;
5468 			}
5469 
5470 			VERIFY3U(pos, ==, size);
5471 
5472 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
5473 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
5474 
5475 			/*
5476 			 * verify if incremental on the whole buffer is
5477 			 * equivalent to non-incremental version
5478 			 */
5479 			ZIO_SET_CHECKSUM(&zc, 0, 0, 0, 0);
5480 			ZIO_SET_CHECKSUM(&zc_bswap, 0, 0, 0, 0);
5481 
5482 			fletcher_4_incremental_native(buf, size, &zc);
5483 			fletcher_4_incremental_byteswap(buf, size, &zc_bswap);
5484 
5485 			VERIFY(ZIO_CHECKSUM_EQUAL(zc, zc_ref));
5486 			VERIFY(ZIO_CHECKSUM_EQUAL(zc_bswap, zc_ref_bswap));
5487 		}
5488 
5489 		umem_free(buf, size);
5490 	}
5491 }
5492 
5493 static int
5494 user_release_one(const char *snapname, const char *holdname)
5495 {
5496 	nvlist_t *snaps, *holds;
5497 	int error;
5498 
5499 	snaps = fnvlist_alloc();
5500 	holds = fnvlist_alloc();
5501 	fnvlist_add_boolean(holds, holdname);
5502 	fnvlist_add_nvlist(snaps, snapname, holds);
5503 	fnvlist_free(holds);
5504 	error = dsl_dataset_user_release(snaps, NULL);
5505 	fnvlist_free(snaps);
5506 	return (error);
5507 }
5508 
5509 /*
5510  * Test snapshot hold/release and deferred destroy.
5511  */
5512 void
5513 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5514 {
5515 	int error;
5516 	objset_t *os = zd->zd_os;
5517 	objset_t *origin;
5518 	char snapname[100];
5519 	char fullname[100];
5520 	char clonename[100];
5521 	char tag[100];
5522 	char osname[ZFS_MAX_DATASET_NAME_LEN];
5523 	nvlist_t *holds;
5524 
5525 	rw_enter(&ztest_name_lock, RW_READER);
5526 
5527 	dmu_objset_name(os, osname);
5528 
5529 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
5530 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5531 	(void) snprintf(clonename, sizeof (clonename),
5532 	    "%s/ch1_%llu", osname, id);
5533 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
5534 
5535 	/*
5536 	 * Clean up from any previous run.
5537 	 */
5538 	error = dsl_destroy_head(clonename);
5539 	if (error != ENOENT)
5540 		ASSERT0(error);
5541 	error = user_release_one(fullname, tag);
5542 	if (error != ESRCH && error != ENOENT)
5543 		ASSERT0(error);
5544 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5545 	if (error != ENOENT)
5546 		ASSERT0(error);
5547 
5548 	/*
5549 	 * Create snapshot, clone it, mark snap for deferred destroy,
5550 	 * destroy clone, verify snap was also destroyed.
5551 	 */
5552 	error = dmu_objset_snapshot_one(osname, snapname);
5553 	if (error) {
5554 		if (error == ENOSPC) {
5555 			ztest_record_enospc("dmu_objset_snapshot");
5556 			goto out;
5557 		}
5558 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5559 	}
5560 
5561 	error = dmu_objset_clone(clonename, fullname);
5562 	if (error) {
5563 		if (error == ENOSPC) {
5564 			ztest_record_enospc("dmu_objset_clone");
5565 			goto out;
5566 		}
5567 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
5568 	}
5569 
5570 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5571 	if (error) {
5572 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5573 		    fullname, error);
5574 	}
5575 
5576 	error = dsl_destroy_head(clonename);
5577 	if (error)
5578 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
5579 
5580 	error = dmu_objset_hold(fullname, FTAG, &origin);
5581 	if (error != ENOENT)
5582 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
5583 
5584 	/*
5585 	 * Create snapshot, add temporary hold, verify that we can't
5586 	 * destroy a held snapshot, mark for deferred destroy,
5587 	 * release hold, verify snapshot was destroyed.
5588 	 */
5589 	error = dmu_objset_snapshot_one(osname, snapname);
5590 	if (error) {
5591 		if (error == ENOSPC) {
5592 			ztest_record_enospc("dmu_objset_snapshot");
5593 			goto out;
5594 		}
5595 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5596 	}
5597 
5598 	holds = fnvlist_alloc();
5599 	fnvlist_add_string(holds, fullname, tag);
5600 	error = dsl_dataset_user_hold(holds, 0, NULL);
5601 	fnvlist_free(holds);
5602 
5603 	if (error == ENOSPC) {
5604 		ztest_record_enospc("dsl_dataset_user_hold");
5605 		goto out;
5606 	} else if (error) {
5607 		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5608 		    fullname, tag, error);
5609 	}
5610 
5611 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5612 	if (error != EBUSY) {
5613 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5614 		    fullname, error);
5615 	}
5616 
5617 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5618 	if (error) {
5619 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5620 		    fullname, error);
5621 	}
5622 
5623 	error = user_release_one(fullname, tag);
5624 	if (error)
5625 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
5626 
5627 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5628 
5629 out:
5630 	rw_exit(&ztest_name_lock);
5631 }
5632 
5633 /*
5634  * Inject random faults into the on-disk data.
5635  */
5636 /* ARGSUSED */
5637 void
5638 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5639 {
5640 	ztest_shared_t *zs = ztest_shared;
5641 	spa_t *spa = ztest_spa;
5642 	int fd;
5643 	uint64_t offset;
5644 	uint64_t leaves;
5645 	uint64_t bad = 0x1990c0ffeedecade;
5646 	uint64_t top, leaf;
5647 	char path0[MAXPATHLEN];
5648 	char pathrand[MAXPATHLEN];
5649 	size_t fsize;
5650 	int bshift = SPA_MAXBLOCKSHIFT + 2;
5651 	int iters = 1000;
5652 	int maxfaults;
5653 	int mirror_save;
5654 	vdev_t *vd0 = NULL;
5655 	uint64_t guid0 = 0;
5656 	boolean_t islog = B_FALSE;
5657 
5658 	mutex_enter(&ztest_vdev_lock);
5659 
5660 	/*
5661 	 * Device removal is in progress, fault injection must be disabled
5662 	 * until it completes and the pool is scrubbed.  The fault injection
5663 	 * strategy for damaging blocks does not take in to account evacuated
5664 	 * blocks which may have already been damaged.
5665 	 */
5666 	if (ztest_device_removal_active) {
5667 		mutex_exit(&ztest_vdev_lock);
5668 		return;
5669 	}
5670 
5671 	maxfaults = MAXFAULTS();
5672 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5673 	mirror_save = zs->zs_mirrors;
5674 	mutex_exit(&ztest_vdev_lock);
5675 
5676 	ASSERT(leaves >= 1);
5677 
5678 	/*
5679 	 * Grab the name lock as reader. There are some operations
5680 	 * which don't like to have their vdevs changed while
5681 	 * they are in progress (i.e. spa_change_guid). Those
5682 	 * operations will have grabbed the name lock as writer.
5683 	 */
5684 	rw_enter(&ztest_name_lock, RW_READER);
5685 
5686 	/*
5687 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5688 	 */
5689 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5690 
5691 	if (ztest_random(2) == 0) {
5692 		/*
5693 		 * Inject errors on a normal data device or slog device.
5694 		 */
5695 		top = ztest_random_vdev_top(spa, B_TRUE);
5696 		leaf = ztest_random(leaves) + zs->zs_splits;
5697 
5698 		/*
5699 		 * Generate paths to the first leaf in this top-level vdev,
5700 		 * and to the random leaf we selected.  We'll induce transient
5701 		 * write failures and random online/offline activity on leaf 0,
5702 		 * and we'll write random garbage to the randomly chosen leaf.
5703 		 */
5704 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
5705 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
5706 		    top * leaves + zs->zs_splits);
5707 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
5708 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
5709 		    top * leaves + leaf);
5710 
5711 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5712 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5713 			islog = B_TRUE;
5714 
5715 		/*
5716 		 * If the top-level vdev needs to be resilvered
5717 		 * then we only allow faults on the device that is
5718 		 * resilvering.
5719 		 */
5720 		if (vd0 != NULL && maxfaults != 1 &&
5721 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5722 		    vd0->vdev_resilver_txg != 0)) {
5723 			/*
5724 			 * Make vd0 explicitly claim to be unreadable,
5725 			 * or unwriteable, or reach behind its back
5726 			 * and close the underlying fd.  We can do this if
5727 			 * maxfaults == 0 because we'll fail and reexecute,
5728 			 * and we can do it if maxfaults >= 2 because we'll
5729 			 * have enough redundancy.  If maxfaults == 1, the
5730 			 * combination of this with injection of random data
5731 			 * corruption below exceeds the pool's fault tolerance.
5732 			 */
5733 			vdev_file_t *vf = vd0->vdev_tsd;
5734 
5735 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
5736 			    (long long)vd0->vdev_id, (int)maxfaults);
5737 
5738 			if (vf != NULL && ztest_random(3) == 0) {
5739 				(void) close(vf->vf_vnode->v_fd);
5740 				vf->vf_vnode->v_fd = -1;
5741 			} else if (ztest_random(2) == 0) {
5742 				vd0->vdev_cant_read = B_TRUE;
5743 			} else {
5744 				vd0->vdev_cant_write = B_TRUE;
5745 			}
5746 			guid0 = vd0->vdev_guid;
5747 		}
5748 	} else {
5749 		/*
5750 		 * Inject errors on an l2cache device.
5751 		 */
5752 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
5753 
5754 		if (sav->sav_count == 0) {
5755 			spa_config_exit(spa, SCL_STATE, FTAG);
5756 			rw_exit(&ztest_name_lock);
5757 			return;
5758 		}
5759 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5760 		guid0 = vd0->vdev_guid;
5761 		(void) strcpy(path0, vd0->vdev_path);
5762 		(void) strcpy(pathrand, vd0->vdev_path);
5763 
5764 		leaf = 0;
5765 		leaves = 1;
5766 		maxfaults = INT_MAX;	/* no limit on cache devices */
5767 	}
5768 
5769 	spa_config_exit(spa, SCL_STATE, FTAG);
5770 	rw_exit(&ztest_name_lock);
5771 
5772 	/*
5773 	 * If we can tolerate two or more faults, or we're dealing
5774 	 * with a slog, randomly online/offline vd0.
5775 	 */
5776 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
5777 		if (ztest_random(10) < 6) {
5778 			int flags = (ztest_random(2) == 0 ?
5779 			    ZFS_OFFLINE_TEMPORARY : 0);
5780 
5781 			/*
5782 			 * We have to grab the zs_name_lock as writer to
5783 			 * prevent a race between offlining a slog and
5784 			 * destroying a dataset. Offlining the slog will
5785 			 * grab a reference on the dataset which may cause
5786 			 * dmu_objset_destroy() to fail with EBUSY thus
5787 			 * leaving the dataset in an inconsistent state.
5788 			 */
5789 			if (islog)
5790 				rw_enter(&ztest_name_lock, RW_WRITER);
5791 
5792 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5793 
5794 			if (islog)
5795 				rw_exit(&ztest_name_lock);
5796 		} else {
5797 			/*
5798 			 * Ideally we would like to be able to randomly
5799 			 * call vdev_[on|off]line without holding locks
5800 			 * to force unpredictable failures but the side
5801 			 * effects of vdev_[on|off]line prevent us from
5802 			 * doing so. We grab the ztest_vdev_lock here to
5803 			 * prevent a race between injection testing and
5804 			 * aux_vdev removal.
5805 			 */
5806 			mutex_enter(&ztest_vdev_lock);
5807 			(void) vdev_online(spa, guid0, 0, NULL);
5808 			mutex_exit(&ztest_vdev_lock);
5809 		}
5810 	}
5811 
5812 	if (maxfaults == 0)
5813 		return;
5814 
5815 	/*
5816 	 * We have at least single-fault tolerance, so inject data corruption.
5817 	 */
5818 	fd = open(pathrand, O_RDWR);
5819 
5820 	if (fd == -1) /* we hit a gap in the device namespace */
5821 		return;
5822 
5823 	fsize = lseek(fd, 0, SEEK_END);
5824 
5825 	while (--iters != 0) {
5826 		/*
5827 		 * The offset must be chosen carefully to ensure that
5828 		 * we do not inject a given logical block with errors
5829 		 * on two different leaf devices, because ZFS can not
5830 		 * tolerate that (if maxfaults==1).
5831 		 *
5832 		 * We divide each leaf into chunks of size
5833 		 * (# leaves * SPA_MAXBLOCKSIZE * 4).  Within each chunk
5834 		 * there is a series of ranges to which we can inject errors.
5835 		 * Each range can accept errors on only a single leaf vdev.
5836 		 * The error injection ranges are separated by ranges
5837 		 * which we will not inject errors on any device (DMZs).
5838 		 * Each DMZ must be large enough such that a single block
5839 		 * can not straddle it, so that a single block can not be
5840 		 * a target in two different injection ranges (on different
5841 		 * leaf vdevs).
5842 		 *
5843 		 * For example, with 3 leaves, each chunk looks like:
5844 		 *    0 to  32M: injection range for leaf 0
5845 		 *  32M to  64M: DMZ - no injection allowed
5846 		 *  64M to  96M: injection range for leaf 1
5847 		 *  96M to 128M: DMZ - no injection allowed
5848 		 * 128M to 160M: injection range for leaf 2
5849 		 * 160M to 192M: DMZ - no injection allowed
5850 		 */
5851 		offset = ztest_random(fsize / (leaves << bshift)) *
5852 		    (leaves << bshift) + (leaf << bshift) +
5853 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5854 
5855 		/*
5856 		 * Only allow damage to the labels at one end of the vdev.
5857 		 *
5858 		 * If all labels are damaged, the device will be totally
5859 		 * inaccessible, which will result in loss of data,
5860 		 * because we also damage (parts of) the other side of
5861 		 * the mirror/raidz.
5862 		 *
5863 		 * Additionally, we will always have both an even and an
5864 		 * odd label, so that we can handle crashes in the
5865 		 * middle of vdev_config_sync().
5866 		 */
5867 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5868 			continue;
5869 
5870 		/*
5871 		 * The two end labels are stored at the "end" of the disk, but
5872 		 * the end of the disk (vdev_psize) is aligned to
5873 		 * sizeof (vdev_label_t).
5874 		 */
5875 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5876 		if ((leaf & 1) == 1 &&
5877 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5878 			continue;
5879 
5880 		mutex_enter(&ztest_vdev_lock);
5881 		if (mirror_save != zs->zs_mirrors) {
5882 			mutex_exit(&ztest_vdev_lock);
5883 			(void) close(fd);
5884 			return;
5885 		}
5886 
5887 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5888 			fatal(1, "can't inject bad word at 0x%llx in %s",
5889 			    offset, pathrand);
5890 
5891 		mutex_exit(&ztest_vdev_lock);
5892 
5893 		if (ztest_opts.zo_verbose >= 7)
5894 			(void) printf("injected bad word into %s,"
5895 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5896 	}
5897 
5898 	(void) close(fd);
5899 }
5900 
5901 /*
5902  * Verify that DDT repair works as expected.
5903  */
5904 void
5905 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5906 {
5907 	ztest_shared_t *zs = ztest_shared;
5908 	spa_t *spa = ztest_spa;
5909 	objset_t *os = zd->zd_os;
5910 	ztest_od_t od[1];
5911 	uint64_t object, blocksize, txg, pattern, psize;
5912 	enum zio_checksum checksum = spa_dedup_checksum(spa);
5913 	dmu_buf_t *db;
5914 	dmu_tx_t *tx;
5915 	abd_t *abd;
5916 	blkptr_t blk;
5917 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
5918 
5919 	blocksize = ztest_random_blocksize();
5920 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
5921 
5922 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
5923 	    0, 0);
5924 
5925 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5926 		return;
5927 
5928 	/*
5929 	 * Take the name lock as writer to prevent anyone else from changing
5930 	 * the pool and dataset properies we need to maintain during this test.
5931 	 */
5932 	rw_enter(&ztest_name_lock, RW_WRITER);
5933 
5934 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5935 	    B_FALSE) != 0 ||
5936 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5937 	    B_FALSE) != 0) {
5938 		rw_exit(&ztest_name_lock);
5939 		return;
5940 	}
5941 
5942 	dmu_objset_stats_t dds;
5943 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5944 	dmu_objset_fast_stat(os, &dds);
5945 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5946 
5947 	object = od[0].od_object;
5948 	blocksize = od[0].od_blocksize;
5949 	pattern = zs->zs_guid ^ dds.dds_guid;
5950 
5951 	ASSERT(object != 0);
5952 
5953 	tx = dmu_tx_create(os);
5954 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5955 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5956 	if (txg == 0) {
5957 		rw_exit(&ztest_name_lock);
5958 		return;
5959 	}
5960 
5961 	/*
5962 	 * Write all the copies of our block.
5963 	 */
5964 	for (int i = 0; i < copies; i++) {
5965 		uint64_t offset = i * blocksize;
5966 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5967 		    DMU_READ_NO_PREFETCH);
5968 		if (error != 0) {
5969 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5970 			    os, (long long)object, (long long) offset, error);
5971 		}
5972 		ASSERT(db->db_offset == offset);
5973 		ASSERT(db->db_size == blocksize);
5974 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5975 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5976 		dmu_buf_will_fill(db, tx);
5977 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5978 		dmu_buf_rele(db, FTAG);
5979 	}
5980 
5981 	dmu_tx_commit(tx);
5982 	txg_wait_synced(spa_get_dsl(spa), txg);
5983 
5984 	/*
5985 	 * Find out what block we got.
5986 	 */
5987 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5988 	    DMU_READ_NO_PREFETCH));
5989 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5990 	dmu_buf_rele(db, FTAG);
5991 
5992 	/*
5993 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5994 	 */
5995 	psize = BP_GET_PSIZE(&blk);
5996 	abd = abd_alloc_linear(psize, B_TRUE);
5997 	ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5998 
5999 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
6000 	    abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
6001 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
6002 
6003 	abd_free(abd);
6004 
6005 	rw_exit(&ztest_name_lock);
6006 }
6007 
6008 /*
6009  * Scrub the pool.
6010  */
6011 /* ARGSUSED */
6012 void
6013 ztest_scrub(ztest_ds_t *zd, uint64_t id)
6014 {
6015 	spa_t *spa = ztest_spa;
6016 
6017 	/*
6018 	 * Scrub in progress by device removal.
6019 	 */
6020 	if (ztest_device_removal_active)
6021 		return;
6022 
6023 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
6024 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
6025 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
6026 }
6027 
6028 /*
6029  * Change the guid for the pool.
6030  */
6031 /* ARGSUSED */
6032 void
6033 ztest_reguid(ztest_ds_t *zd, uint64_t id)
6034 {
6035 	spa_t *spa = ztest_spa;
6036 	uint64_t orig, load;
6037 	int error;
6038 
6039 	if (ztest_opts.zo_mmp_test)
6040 		return;
6041 
6042 	orig = spa_guid(spa);
6043 	load = spa_load_guid(spa);
6044 
6045 	rw_enter(&ztest_name_lock, RW_WRITER);
6046 	error = spa_change_guid(spa);
6047 	rw_exit(&ztest_name_lock);
6048 
6049 	if (error != 0)
6050 		return;
6051 
6052 	if (ztest_opts.zo_verbose >= 4) {
6053 		(void) printf("Changed guid old %llu -> %llu\n",
6054 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
6055 	}
6056 
6057 	VERIFY3U(orig, !=, spa_guid(spa));
6058 	VERIFY3U(load, ==, spa_load_guid(spa));
6059 }
6060 
6061 static vdev_t *
6062 ztest_random_concrete_vdev_leaf(vdev_t *vd)
6063 {
6064 	if (vd == NULL)
6065 		return (NULL);
6066 
6067 	if (vd->vdev_children == 0)
6068 		return (vd);
6069 
6070 	vdev_t *eligible[vd->vdev_children];
6071 	int eligible_idx = 0, i;
6072 	for (i = 0; i < vd->vdev_children; i++) {
6073 		vdev_t *cvd = vd->vdev_child[i];
6074 		if (cvd->vdev_top->vdev_removing)
6075 			continue;
6076 		if (cvd->vdev_children > 0 ||
6077 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
6078 			eligible[eligible_idx++] = cvd;
6079 		}
6080 	}
6081 	VERIFY(eligible_idx > 0);
6082 
6083 	uint64_t child_no = ztest_random(eligible_idx);
6084 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
6085 }
6086 
6087 /* ARGSUSED */
6088 void
6089 ztest_initialize(ztest_ds_t *zd, uint64_t id)
6090 {
6091 	spa_t *spa = ztest_spa;
6092 	int error = 0;
6093 
6094 	mutex_enter(&ztest_vdev_lock);
6095 
6096 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6097 
6098 	/* Random leaf vdev */
6099 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6100 	if (rand_vd == NULL) {
6101 		spa_config_exit(spa, SCL_VDEV, FTAG);
6102 		mutex_exit(&ztest_vdev_lock);
6103 		return;
6104 	}
6105 
6106 	/*
6107 	 * The random vdev we've selected may change as soon as we
6108 	 * drop the spa_config_lock. We create local copies of things
6109 	 * we're interested in.
6110 	 */
6111 	uint64_t guid = rand_vd->vdev_guid;
6112 	char *path = strdup(rand_vd->vdev_path);
6113 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
6114 
6115 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, guid);
6116 	spa_config_exit(spa, SCL_VDEV, FTAG);
6117 
6118 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
6119 
6120 	nvlist_t *vdev_guids = fnvlist_alloc();
6121 	nvlist_t *vdev_errlist = fnvlist_alloc();
6122 	fnvlist_add_uint64(vdev_guids, path, guid);
6123 	error = spa_vdev_initialize(spa, vdev_guids, cmd, vdev_errlist);
6124 	fnvlist_free(vdev_guids);
6125 	fnvlist_free(vdev_errlist);
6126 
6127 	switch (cmd) {
6128 	case POOL_INITIALIZE_CANCEL:
6129 		if (ztest_opts.zo_verbose >= 4) {
6130 			(void) printf("Cancel initialize %s", path);
6131 			if (!active)
6132 				(void) printf(" failed (no initialize active)");
6133 			(void) printf("\n");
6134 		}
6135 		break;
6136 	case POOL_INITIALIZE_START:
6137 		if (ztest_opts.zo_verbose >= 4) {
6138 			(void) printf("Start initialize %s", path);
6139 			if (active && error == 0)
6140 				(void) printf(" failed (already active)");
6141 			else if (error != 0)
6142 				(void) printf(" failed (error %d)", error);
6143 			(void) printf("\n");
6144 		}
6145 		break;
6146 	case POOL_INITIALIZE_SUSPEND:
6147 		if (ztest_opts.zo_verbose >= 4) {
6148 			(void) printf("Suspend initialize %s", path);
6149 			if (!active)
6150 				(void) printf(" failed (no initialize active)");
6151 			(void) printf("\n");
6152 		}
6153 		break;
6154 	}
6155 	free(path);
6156 	mutex_exit(&ztest_vdev_lock);
6157 }
6158 
6159 /* ARGSUSED */
6160 void
6161 ztest_trim(ztest_ds_t *zd, uint64_t id)
6162 {
6163 	spa_t *spa = ztest_spa;
6164 	int error = 0;
6165 
6166 	mutex_enter(&ztest_vdev_lock);
6167 
6168 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
6169 
6170 	/* Random leaf vdev */
6171 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
6172 	if (rand_vd == NULL) {
6173 		spa_config_exit(spa, SCL_VDEV, FTAG);
6174 		mutex_exit(&ztest_vdev_lock);
6175 		return;
6176 	}
6177 
6178 	/*
6179 	 * The random vdev we've selected may change as soon as we
6180 	 * drop the spa_config_lock. We create local copies of things
6181 	 * we're interested in.
6182 	 */
6183 	uint64_t guid = rand_vd->vdev_guid;
6184 	char *path = strdup(rand_vd->vdev_path);
6185 	boolean_t active = rand_vd->vdev_trim_thread != NULL;
6186 
6187 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, guid);
6188 	spa_config_exit(spa, SCL_VDEV, FTAG);
6189 
6190 	uint64_t cmd = ztest_random(POOL_TRIM_FUNCS);
6191 	uint64_t rate = 1 << ztest_random(30);
6192 	boolean_t partial = (ztest_random(5) > 0);
6193 	boolean_t secure = (ztest_random(5) > 0);
6194 
6195 	nvlist_t *vdev_guids = fnvlist_alloc();
6196 	nvlist_t *vdev_errlist = fnvlist_alloc();
6197 	fnvlist_add_uint64(vdev_guids, path, guid);
6198 	error = spa_vdev_trim(spa, vdev_guids, cmd, rate, partial,
6199 	    secure, vdev_errlist);
6200 	fnvlist_free(vdev_guids);
6201 	fnvlist_free(vdev_errlist);
6202 
6203 	switch (cmd) {
6204 	case POOL_TRIM_CANCEL:
6205 		if (ztest_opts.zo_verbose >= 4) {
6206 			(void) printf("Cancel TRIM %s", path);
6207 			if (!active)
6208 				(void) printf(" failed (no TRIM active)");
6209 			(void) printf("\n");
6210 		}
6211 		break;
6212 	case POOL_TRIM_START:
6213 		if (ztest_opts.zo_verbose >= 4) {
6214 			(void) printf("Start TRIM %s", path);
6215 			if (active && error == 0)
6216 				(void) printf(" failed (already active)");
6217 			else if (error != 0)
6218 				(void) printf(" failed (error %d)", error);
6219 			(void) printf("\n");
6220 		}
6221 		break;
6222 	case POOL_TRIM_SUSPEND:
6223 		if (ztest_opts.zo_verbose >= 4) {
6224 			(void) printf("Suspend TRIM %s", path);
6225 			if (!active)
6226 				(void) printf(" failed (no TRIM active)");
6227 			(void) printf("\n");
6228 		}
6229 		break;
6230 	}
6231 	free(path);
6232 	mutex_exit(&ztest_vdev_lock);
6233 }
6234 
6235 /*
6236  * Verify pool integrity by running zdb.
6237  */
6238 static void
6239 ztest_run_zdb(char *pool)
6240 {
6241 	int status;
6242 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
6243 	char zbuf[1024];
6244 	FILE *fp;
6245 
6246 	(void) snprintf(zdb, sizeof (zdb),
6247 	    "/usr/sbin/zdb -bcc%s%s -G -d -U %s "
6248 	    "-o zfs_reconstruct_indirect_combinations_max=65536 %s",
6249 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
6250 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
6251 	    spa_config_path,
6252 	    pool);
6253 
6254 	if (ztest_opts.zo_verbose >= 5)
6255 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
6256 
6257 	fp = popen(zdb, "r");
6258 
6259 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
6260 		if (ztest_opts.zo_verbose >= 3)
6261 			(void) printf("%s", zbuf);
6262 
6263 	status = pclose(fp);
6264 
6265 	if (status == 0)
6266 		return;
6267 
6268 	ztest_dump_core = 0;
6269 	if (WIFEXITED(status))
6270 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
6271 	else
6272 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
6273 }
6274 
6275 static void
6276 ztest_walk_pool_directory(char *header)
6277 {
6278 	spa_t *spa = NULL;
6279 
6280 	if (ztest_opts.zo_verbose >= 6)
6281 		(void) printf("%s\n", header);
6282 
6283 	mutex_enter(&spa_namespace_lock);
6284 	while ((spa = spa_next(spa)) != NULL)
6285 		if (ztest_opts.zo_verbose >= 6)
6286 			(void) printf("\t%s\n", spa_name(spa));
6287 	mutex_exit(&spa_namespace_lock);
6288 }
6289 
6290 static void
6291 ztest_spa_import_export(char *oldname, char *newname)
6292 {
6293 	nvlist_t *config, *newconfig;
6294 	uint64_t pool_guid;
6295 	spa_t *spa;
6296 	int error;
6297 
6298 	if (ztest_opts.zo_verbose >= 4) {
6299 		(void) printf("import/export: old = %s, new = %s\n",
6300 		    oldname, newname);
6301 	}
6302 
6303 	/*
6304 	 * Clean up from previous runs.
6305 	 */
6306 	(void) spa_destroy(newname);
6307 
6308 	/*
6309 	 * Get the pool's configuration and guid.
6310 	 */
6311 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
6312 
6313 	/*
6314 	 * Kick off a scrub to tickle scrub/export races.
6315 	 */
6316 	if (ztest_random(2) == 0)
6317 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
6318 
6319 	pool_guid = spa_guid(spa);
6320 	spa_close(spa, FTAG);
6321 
6322 	ztest_walk_pool_directory("pools before export");
6323 
6324 	/*
6325 	 * Export it.
6326 	 */
6327 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
6328 
6329 	ztest_walk_pool_directory("pools after export");
6330 
6331 	/*
6332 	 * Try to import it.
6333 	 */
6334 	newconfig = spa_tryimport(config);
6335 	ASSERT(newconfig != NULL);
6336 	nvlist_free(newconfig);
6337 
6338 	/*
6339 	 * Import it under the new name.
6340 	 */
6341 	error = spa_import(newname, config, NULL, 0);
6342 	if (error != 0) {
6343 		dump_nvlist(config, 0);
6344 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
6345 		    oldname, newname, error);
6346 	}
6347 
6348 	ztest_walk_pool_directory("pools after import");
6349 
6350 	/*
6351 	 * Try to import it again -- should fail with EEXIST.
6352 	 */
6353 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
6354 
6355 	/*
6356 	 * Try to import it under a different name -- should fail with EEXIST.
6357 	 */
6358 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
6359 
6360 	/*
6361 	 * Verify that the pool is no longer visible under the old name.
6362 	 */
6363 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
6364 
6365 	/*
6366 	 * Verify that we can open and close the pool using the new name.
6367 	 */
6368 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
6369 	ASSERT(pool_guid == spa_guid(spa));
6370 	spa_close(spa, FTAG);
6371 
6372 	nvlist_free(config);
6373 }
6374 
6375 static void
6376 ztest_resume(spa_t *spa)
6377 {
6378 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
6379 		(void) printf("resuming from suspended state\n");
6380 	spa_vdev_state_enter(spa, SCL_NONE);
6381 	vdev_clear(spa, NULL);
6382 	(void) spa_vdev_state_exit(spa, NULL, 0);
6383 	(void) zio_resume(spa);
6384 }
6385 
6386 static void *
6387 ztest_resume_thread(void *arg)
6388 {
6389 	spa_t *spa = arg;
6390 
6391 	while (!ztest_exiting) {
6392 		if (spa_suspended(spa))
6393 			ztest_resume(spa);
6394 		(void) poll(NULL, 0, 100);
6395 
6396 		/*
6397 		 * Periodically change the zfs_compressed_arc_enabled setting.
6398 		 */
6399 		if (ztest_random(10) == 0)
6400 			zfs_compressed_arc_enabled = ztest_random(2);
6401 
6402 		/*
6403 		 * Periodically change the zfs_abd_scatter_enabled setting.
6404 		 */
6405 		if (ztest_random(10) == 0)
6406 			zfs_abd_scatter_enabled = ztest_random(2);
6407 	}
6408 	return (NULL);
6409 }
6410 
6411 static void *
6412 ztest_deadman_thread(void *arg)
6413 {
6414 	ztest_shared_t *zs = arg;
6415 	spa_t *spa = ztest_spa;
6416 	hrtime_t delta, total = 0;
6417 
6418 	for (;;) {
6419 		delta = zs->zs_thread_stop - zs->zs_thread_start +
6420 		    MSEC2NSEC(zfs_deadman_synctime_ms);
6421 
6422 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
6423 
6424 		/*
6425 		 * If the pool is suspended then fail immediately. Otherwise,
6426 		 * check to see if the pool is making any progress. If
6427 		 * vdev_deadman() discovers that there hasn't been any recent
6428 		 * I/Os then it will end up aborting the tests.
6429 		 */
6430 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
6431 			fatal(0, "aborting test after %llu seconds because "
6432 			    "pool has transitioned to a suspended state.",
6433 			    zfs_deadman_synctime_ms / 1000);
6434 			return (NULL);
6435 		}
6436 		vdev_deadman(spa->spa_root_vdev);
6437 
6438 		total += zfs_deadman_synctime_ms/1000;
6439 		(void) printf("ztest has been running for %lld seconds\n",
6440 		    total);
6441 	}
6442 }
6443 
6444 static void
6445 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
6446 {
6447 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
6448 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
6449 	hrtime_t functime = gethrtime();
6450 
6451 	for (int i = 0; i < zi->zi_iters; i++)
6452 		zi->zi_func(zd, id);
6453 
6454 	functime = gethrtime() - functime;
6455 
6456 	atomic_add_64(&zc->zc_count, 1);
6457 	atomic_add_64(&zc->zc_time, functime);
6458 
6459 	if (ztest_opts.zo_verbose >= 4) {
6460 		Dl_info dli;
6461 		(void) dladdr((void *)zi->zi_func, &dli);
6462 		(void) printf("%6.2f sec in %s\n",
6463 		    (double)functime / NANOSEC, dli.dli_sname);
6464 	}
6465 }
6466 
6467 static void *
6468 ztest_thread(void *arg)
6469 {
6470 	int rand;
6471 	uint64_t id = (uintptr_t)arg;
6472 	ztest_shared_t *zs = ztest_shared;
6473 	uint64_t call_next;
6474 	hrtime_t now;
6475 	ztest_info_t *zi;
6476 	ztest_shared_callstate_t *zc;
6477 
6478 	while ((now = gethrtime()) < zs->zs_thread_stop) {
6479 		/*
6480 		 * See if it's time to force a crash.
6481 		 */
6482 		if (now > zs->zs_thread_kill)
6483 			ztest_kill(zs);
6484 
6485 		/*
6486 		 * If we're getting ENOSPC with some regularity, stop.
6487 		 */
6488 		if (zs->zs_enospc_count > 10)
6489 			break;
6490 
6491 		/*
6492 		 * Pick a random function to execute.
6493 		 */
6494 		rand = ztest_random(ZTEST_FUNCS);
6495 		zi = &ztest_info[rand];
6496 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
6497 		call_next = zc->zc_next;
6498 
6499 		if (now >= call_next &&
6500 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
6501 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
6502 			ztest_execute(rand, zi, id);
6503 		}
6504 	}
6505 
6506 	return (NULL);
6507 }
6508 
6509 static void
6510 ztest_dataset_name(char *dsname, char *pool, int d)
6511 {
6512 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
6513 }
6514 
6515 static void
6516 ztest_dataset_destroy(int d)
6517 {
6518 	char name[ZFS_MAX_DATASET_NAME_LEN];
6519 
6520 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
6521 
6522 	if (ztest_opts.zo_verbose >= 3)
6523 		(void) printf("Destroying %s to free up space\n", name);
6524 
6525 	/*
6526 	 * Cleanup any non-standard clones and snapshots.  In general,
6527 	 * ztest thread t operates on dataset (t % zopt_datasets),
6528 	 * so there may be more than one thing to clean up.
6529 	 */
6530 	for (int t = d; t < ztest_opts.zo_threads;
6531 	    t += ztest_opts.zo_datasets) {
6532 		ztest_dsl_dataset_cleanup(name, t);
6533 	}
6534 
6535 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
6536 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
6537 }
6538 
6539 static void
6540 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
6541 {
6542 	uint64_t usedobjs, dirobjs, scratch;
6543 
6544 	/*
6545 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
6546 	 * Therefore, the number of objects in use should equal the
6547 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6548 	 * If not, we have an object leak.
6549 	 *
6550 	 * Note that we can only check this in ztest_dataset_open(),
6551 	 * when the open-context and syncing-context values agree.
6552 	 * That's because zap_count() returns the open-context value,
6553 	 * while dmu_objset_space() returns the rootbp fill count.
6554 	 */
6555 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
6556 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
6557 	ASSERT3U(dirobjs + 1, ==, usedobjs);
6558 }
6559 
6560 static int
6561 ztest_dataset_open(int d)
6562 {
6563 	ztest_ds_t *zd = &ztest_ds[d];
6564 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
6565 	objset_t *os;
6566 	zilog_t *zilog;
6567 	char name[ZFS_MAX_DATASET_NAME_LEN];
6568 	int error;
6569 
6570 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
6571 
6572 	rw_enter(&ztest_name_lock, RW_READER);
6573 
6574 	error = ztest_dataset_create(name);
6575 	if (error == ENOSPC) {
6576 		rw_exit(&ztest_name_lock);
6577 		ztest_record_enospc(FTAG);
6578 		return (error);
6579 	}
6580 	ASSERT(error == 0 || error == EEXIST);
6581 
6582 	VERIFY0(ztest_dmu_objset_own(name, DMU_OST_OTHER, B_FALSE,
6583 	    B_TRUE, zd, &os));
6584 	rw_exit(&ztest_name_lock);
6585 
6586 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
6587 
6588 	zilog = zd->zd_zilog;
6589 
6590 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
6591 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
6592 		fatal(0, "missing log records: claimed %llu < committed %llu",
6593 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
6594 
6595 	ztest_dataset_dirobj_verify(zd);
6596 
6597 	zil_replay(os, zd, ztest_replay_vector);
6598 
6599 	ztest_dataset_dirobj_verify(zd);
6600 
6601 	if (ztest_opts.zo_verbose >= 6)
6602 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6603 		    zd->zd_name,
6604 		    (u_longlong_t)zilog->zl_parse_blk_count,
6605 		    (u_longlong_t)zilog->zl_parse_lr_count,
6606 		    (u_longlong_t)zilog->zl_replaying_seq);
6607 
6608 	zilog = zil_open(os, ztest_get_data);
6609 
6610 	if (zilog->zl_replaying_seq != 0 &&
6611 	    zilog->zl_replaying_seq < committed_seq)
6612 		fatal(0, "missing log records: replayed %llu < committed %llu",
6613 		    zilog->zl_replaying_seq, committed_seq);
6614 
6615 	return (0);
6616 }
6617 
6618 static void
6619 ztest_dataset_close(int d)
6620 {
6621 	ztest_ds_t *zd = &ztest_ds[d];
6622 
6623 	zil_close(zd->zd_zilog);
6624 	dmu_objset_disown(zd->zd_os, B_TRUE, zd);
6625 
6626 	ztest_zd_fini(zd);
6627 }
6628 
6629 /*
6630  * Kick off threads to run tests on all datasets in parallel.
6631  */
6632 static void
6633 ztest_run(ztest_shared_t *zs)
6634 {
6635 	thread_t *tid;
6636 	spa_t *spa;
6637 	objset_t *os;
6638 	thread_t resume_tid;
6639 	int error;
6640 
6641 	ztest_exiting = B_FALSE;
6642 
6643 	/*
6644 	 * Initialize parent/child shared state.
6645 	 */
6646 	mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);
6647 	mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
6648 	rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
6649 
6650 	zs->zs_thread_start = gethrtime();
6651 	zs->zs_thread_stop =
6652 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
6653 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
6654 	zs->zs_thread_kill = zs->zs_thread_stop;
6655 	if (ztest_random(100) < ztest_opts.zo_killrate) {
6656 		zs->zs_thread_kill -=
6657 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
6658 	}
6659 
6660 	mutex_init(&zcl.zcl_callbacks_lock, NULL, USYNC_THREAD, NULL);
6661 
6662 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
6663 	    offsetof(ztest_cb_data_t, zcd_node));
6664 
6665 	/*
6666 	 * Open our pool.
6667 	 */
6668 	kernel_init(FREAD | FWRITE);
6669 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
6670 	metaslab_preload_limit = ztest_random(20) + 1;
6671 	ztest_spa = spa;
6672 
6673 	dmu_objset_stats_t dds;
6674 	VERIFY0(ztest_dmu_objset_own(ztest_opts.zo_pool,
6675 	    DMU_OST_ANY, B_TRUE, B_TRUE, FTAG, &os));
6676 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
6677 	dmu_objset_fast_stat(os, &dds);
6678 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
6679 	zs->zs_guid = dds.dds_guid;
6680 	dmu_objset_disown(os, B_TRUE, FTAG);
6681 
6682 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
6683 
6684 	/*
6685 	 * We don't expect the pool to suspend unless maxfaults == 0,
6686 	 * in which case ztest_fault_inject() temporarily takes away
6687 	 * the only valid replica.
6688 	 */
6689 	if (MAXFAULTS() == 0)
6690 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
6691 	else
6692 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
6693 
6694 	/*
6695 	 * Create a thread to periodically resume suspended I/O.
6696 	 */
6697 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
6698 	    &resume_tid) == 0);
6699 
6700 	/*
6701 	 * Create a deadman thread to abort() if we hang.
6702 	 */
6703 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
6704 	    NULL) == 0);
6705 
6706 	/*
6707 	 * Verify that we can safely inquire about any object,
6708 	 * whether it's allocated or not.  To make it interesting,
6709 	 * we probe a 5-wide window around each power of two.
6710 	 * This hits all edge cases, including zero and the max.
6711 	 */
6712 	for (int t = 0; t < 64; t++) {
6713 		for (int d = -5; d <= 5; d++) {
6714 			error = dmu_object_info(spa->spa_meta_objset,
6715 			    (1ULL << t) + d, NULL);
6716 			ASSERT(error == 0 || error == ENOENT ||
6717 			    error == EINVAL);
6718 		}
6719 	}
6720 
6721 	/*
6722 	 * If we got any ENOSPC errors on the previous run, destroy something.
6723 	 */
6724 	if (zs->zs_enospc_count != 0) {
6725 		int d = ztest_random(ztest_opts.zo_datasets);
6726 		ztest_dataset_destroy(d);
6727 	}
6728 	zs->zs_enospc_count = 0;
6729 
6730 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
6731 	    UMEM_NOFAIL);
6732 
6733 	if (ztest_opts.zo_verbose >= 4)
6734 		(void) printf("starting main threads...\n");
6735 
6736 	/*
6737 	 * Kick off all the tests that run in parallel.
6738 	 */
6739 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
6740 		if (t < ztest_opts.zo_datasets &&
6741 		    ztest_dataset_open(t) != 0)
6742 			return;
6743 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
6744 		    THR_BOUND, &tid[t]) == 0);
6745 	}
6746 
6747 	/*
6748 	 * Wait for all of the tests to complete.  We go in reverse order
6749 	 * so we don't close datasets while threads are still using them.
6750 	 */
6751 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
6752 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
6753 		if (t < ztest_opts.zo_datasets)
6754 			ztest_dataset_close(t);
6755 	}
6756 
6757 	txg_wait_synced(spa_get_dsl(spa), 0);
6758 
6759 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
6760 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
6761 	zfs_dbgmsg_print(FTAG);
6762 
6763 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
6764 
6765 	/* Kill the resume thread */
6766 	ztest_exiting = B_TRUE;
6767 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
6768 	ztest_resume(spa);
6769 
6770 	/*
6771 	 * Right before closing the pool, kick off a bunch of async I/O;
6772 	 * spa_close() should wait for it to complete.
6773 	 */
6774 	for (uint64_t object = 1; object < 50; object++) {
6775 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
6776 		    ZIO_PRIORITY_SYNC_READ);
6777 	}
6778 
6779 	spa_close(spa, FTAG);
6780 
6781 	/*
6782 	 * Verify that we can loop over all pools.
6783 	 */
6784 	mutex_enter(&spa_namespace_lock);
6785 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
6786 		if (ztest_opts.zo_verbose > 3)
6787 			(void) printf("spa_next: found %s\n", spa_name(spa));
6788 	mutex_exit(&spa_namespace_lock);
6789 
6790 	/*
6791 	 * Verify that we can export the pool and reimport it under a
6792 	 * different name.
6793 	 */
6794 	if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
6795 		char name[ZFS_MAX_DATASET_NAME_LEN];
6796 		(void) snprintf(name, sizeof (name), "%s_import",
6797 		    ztest_opts.zo_pool);
6798 		ztest_spa_import_export(ztest_opts.zo_pool, name);
6799 		ztest_spa_import_export(name, ztest_opts.zo_pool);
6800 	}
6801 
6802 	kernel_fini();
6803 
6804 	list_destroy(&zcl.zcl_callbacks);
6805 
6806 	mutex_destroy(&zcl.zcl_callbacks_lock);
6807 
6808 	rw_destroy(&ztest_name_lock);
6809 	mutex_destroy(&ztest_vdev_lock);
6810 	mutex_destroy(&ztest_checkpoint_lock);
6811 }
6812 
6813 static void
6814 ztest_freeze(void)
6815 {
6816 	ztest_ds_t *zd = &ztest_ds[0];
6817 	spa_t *spa;
6818 	int numloops = 0;
6819 
6820 	if (ztest_opts.zo_verbose >= 3)
6821 		(void) printf("testing spa_freeze()...\n");
6822 
6823 	kernel_init(FREAD | FWRITE);
6824 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6825 	VERIFY3U(0, ==, ztest_dataset_open(0));
6826 	ztest_spa = spa;
6827 
6828 	/*
6829 	 * Force the first log block to be transactionally allocated.
6830 	 * We have to do this before we freeze the pool -- otherwise
6831 	 * the log chain won't be anchored.
6832 	 */
6833 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6834 		ztest_dmu_object_alloc_free(zd, 0);
6835 		zil_commit(zd->zd_zilog, 0);
6836 	}
6837 
6838 	txg_wait_synced(spa_get_dsl(spa), 0);
6839 
6840 	/*
6841 	 * Freeze the pool.  This stops spa_sync() from doing anything,
6842 	 * so that the only way to record changes from now on is the ZIL.
6843 	 */
6844 	spa_freeze(spa);
6845 
6846 	/*
6847 	 * Because it is hard to predict how much space a write will actually
6848 	 * require beforehand, we leave ourselves some fudge space to write over
6849 	 * capacity.
6850 	 */
6851 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6852 
6853 	/*
6854 	 * Run tests that generate log records but don't alter the pool config
6855 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6856 	 * We do a txg_wait_synced() after each iteration to force the txg
6857 	 * to increase well beyond the last synced value in the uberblock.
6858 	 * The ZIL should be OK with that.
6859 	 *
6860 	 * Run a random number of times less than zo_maxloops and ensure we do
6861 	 * not run out of space on the pool.
6862 	 */
6863 	while (ztest_random(10) != 0 &&
6864 	    numloops++ < ztest_opts.zo_maxloops &&
6865 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6866 		ztest_od_t od;
6867 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6868 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6869 		ztest_io(zd, od.od_object,
6870 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6871 		txg_wait_synced(spa_get_dsl(spa), 0);
6872 	}
6873 
6874 	/*
6875 	 * Commit all of the changes we just generated.
6876 	 */
6877 	zil_commit(zd->zd_zilog, 0);
6878 	txg_wait_synced(spa_get_dsl(spa), 0);
6879 
6880 	/*
6881 	 * Close our dataset and close the pool.
6882 	 */
6883 	ztest_dataset_close(0);
6884 	spa_close(spa, FTAG);
6885 	kernel_fini();
6886 
6887 	/*
6888 	 * Open and close the pool and dataset to induce log replay.
6889 	 */
6890 	kernel_init(FREAD | FWRITE);
6891 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6892 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6893 	VERIFY3U(0, ==, ztest_dataset_open(0));
6894 	ztest_spa = spa;
6895 	txg_wait_synced(spa_get_dsl(spa), 0);
6896 	ztest_dataset_close(0);
6897 	ztest_reguid(NULL, 0);
6898 
6899 	spa_close(spa, FTAG);
6900 	kernel_fini();
6901 }
6902 
6903 void
6904 print_time(hrtime_t t, char *timebuf)
6905 {
6906 	hrtime_t s = t / NANOSEC;
6907 	hrtime_t m = s / 60;
6908 	hrtime_t h = m / 60;
6909 	hrtime_t d = h / 24;
6910 
6911 	s -= m * 60;
6912 	m -= h * 60;
6913 	h -= d * 24;
6914 
6915 	timebuf[0] = '\0';
6916 
6917 	if (d)
6918 		(void) sprintf(timebuf,
6919 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
6920 	else if (h)
6921 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6922 	else if (m)
6923 		(void) sprintf(timebuf, "%llum%02llus", m, s);
6924 	else
6925 		(void) sprintf(timebuf, "%llus", s);
6926 }
6927 
6928 static nvlist_t *
6929 make_random_props()
6930 {
6931 	nvlist_t *props;
6932 
6933 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6934 
6935 	if (ztest_random(2) == 0)
6936 		return (props);
6937 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6938 
6939 	return (props);
6940 }
6941 
6942 /*
6943  * Import a storage pool with the given name.
6944  */
6945 static void
6946 ztest_import(ztest_shared_t *zs)
6947 {
6948 	importargs_t args = { 0 };
6949 	spa_t *spa;
6950 	nvlist_t *cfg = NULL;
6951 	int nsearch = 1;
6952 	char *searchdirs[nsearch];
6953 	char *name = ztest_opts.zo_pool;
6954 	int flags = ZFS_IMPORT_MISSING_LOG;
6955 	int error;
6956 
6957 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6958 	rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
6959 
6960 	kernel_init(FREAD | FWRITE);
6961 
6962 	searchdirs[0] = ztest_opts.zo_dir;
6963 	args.paths = nsearch;
6964 	args.path = searchdirs;
6965 	args.can_be_active = B_FALSE;
6966 
6967 	error = zpool_find_config(NULL, name, &cfg, &args,
6968 	    &libzpool_config_ops);
6969 	if (error)
6970 		(void) fatal(0, "No pools found\n");
6971 
6972 	VERIFY0(spa_import(name, cfg, NULL, flags));
6973 	VERIFY0(spa_open(name, &spa, FTAG));
6974 	zs->zs_metaslab_sz =
6975 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6976 	spa_close(spa, FTAG);
6977 
6978 	kernel_fini();
6979 
6980 	if (!ztest_opts.zo_mmp_test) {
6981 		ztest_run_zdb(ztest_opts.zo_pool);
6982 		ztest_freeze();
6983 		ztest_run_zdb(ztest_opts.zo_pool);
6984 	}
6985 
6986 	rw_destroy(&ztest_name_lock);
6987 	mutex_destroy(&ztest_vdev_lock);
6988 }
6989 
6990 /*
6991  * Create a storage pool with the given name and initial vdev size.
6992  * Then test spa_freeze() functionality.
6993  */
6994 static void
6995 ztest_init(ztest_shared_t *zs)
6996 {
6997 	spa_t *spa;
6998 	nvlist_t *nvroot, *props;
6999 
7000 	mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
7001 	mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);
7002 	rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
7003 
7004 	kernel_init(FREAD | FWRITE);
7005 
7006 	/*
7007 	 * Create the storage pool.
7008 	 */
7009 	(void) spa_destroy(ztest_opts.zo_pool);
7010 	ztest_shared->zs_vdev_next_leaf = 0;
7011 	zs->zs_splits = 0;
7012 	zs->zs_mirrors = ztest_opts.zo_mirrors;
7013 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
7014 	    NULL, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
7015 	props = make_random_props();
7016 	for (int i = 0; i < SPA_FEATURES; i++) {
7017 		char buf[1024];
7018 
7019 		/*
7020 		 * 75% chance of using the log space map feature. We want ztest
7021 		 * to exercise both the code paths that use the log space map
7022 		 * feature and the ones that don't.
7023 		 */
7024 		if (i == SPA_FEATURE_LOG_SPACEMAP && ztest_random(4) == 0)
7025 			continue;
7026 
7027 		(void) snprintf(buf, sizeof (buf), "feature@%s",
7028 		    spa_feature_table[i].fi_uname);
7029 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
7030 	}
7031 	VERIFY3U(0, ==,
7032 	    spa_create(ztest_opts.zo_pool, nvroot, props, NULL, NULL));
7033 	nvlist_free(nvroot);
7034 	nvlist_free(props);
7035 
7036 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
7037 	zs->zs_metaslab_sz =
7038 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
7039 
7040 	spa_close(spa, FTAG);
7041 
7042 	kernel_fini();
7043 
7044 	if (!ztest_opts.zo_mmp_test) {
7045 		ztest_run_zdb(ztest_opts.zo_pool);
7046 		ztest_freeze();
7047 		ztest_run_zdb(ztest_opts.zo_pool);
7048 	}
7049 
7050 	rw_destroy(&ztest_name_lock);
7051 	mutex_destroy(&ztest_vdev_lock);
7052 	mutex_destroy(&ztest_checkpoint_lock);
7053 }
7054 
7055 static void
7056 setup_data_fd(void)
7057 {
7058 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
7059 
7060 	ztest_fd_data = mkstemp(ztest_name_data);
7061 	ASSERT3S(ztest_fd_data, >=, 0);
7062 	(void) unlink(ztest_name_data);
7063 }
7064 
7065 static int
7066 shared_data_size(ztest_shared_hdr_t *hdr)
7067 {
7068 	int size;
7069 
7070 	size = hdr->zh_hdr_size;
7071 	size += hdr->zh_opts_size;
7072 	size += hdr->zh_size;
7073 	size += hdr->zh_stats_size * hdr->zh_stats_count;
7074 	size += hdr->zh_ds_size * hdr->zh_ds_count;
7075 
7076 	return (size);
7077 }
7078 
7079 static void
7080 setup_hdr(void)
7081 {
7082 	int size;
7083 	ztest_shared_hdr_t *hdr;
7084 
7085 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7086 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7087 	ASSERT(hdr != MAP_FAILED);
7088 
7089 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
7090 
7091 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
7092 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
7093 	hdr->zh_size = sizeof (ztest_shared_t);
7094 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
7095 	hdr->zh_stats_count = ZTEST_FUNCS;
7096 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
7097 	hdr->zh_ds_count = ztest_opts.zo_datasets;
7098 
7099 	size = shared_data_size(hdr);
7100 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
7101 
7102 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7103 }
7104 
7105 static void
7106 setup_data(void)
7107 {
7108 	int size, offset;
7109 	ztest_shared_hdr_t *hdr;
7110 	uint8_t *buf;
7111 
7112 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
7113 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
7114 	ASSERT(hdr != MAP_FAILED);
7115 
7116 	size = shared_data_size(hdr);
7117 
7118 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
7119 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
7120 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
7121 	ASSERT(hdr != MAP_FAILED);
7122 	buf = (uint8_t *)hdr;
7123 
7124 	offset = hdr->zh_hdr_size;
7125 	ztest_shared_opts = (void *)&buf[offset];
7126 	offset += hdr->zh_opts_size;
7127 	ztest_shared = (void *)&buf[offset];
7128 	offset += hdr->zh_size;
7129 	ztest_shared_callstate = (void *)&buf[offset];
7130 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
7131 	ztest_shared_ds = (void *)&buf[offset];
7132 }
7133 
7134 static boolean_t
7135 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
7136 {
7137 	pid_t pid;
7138 	int status;
7139 	char *cmdbuf = NULL;
7140 
7141 	pid = fork();
7142 
7143 	if (cmd == NULL) {
7144 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
7145 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
7146 		cmd = cmdbuf;
7147 	}
7148 
7149 	if (pid == -1)
7150 		fatal(1, "fork failed");
7151 
7152 	if (pid == 0) {	/* child */
7153 		char *emptyargv[2] = { cmd, NULL };
7154 		char fd_data_str[12];
7155 
7156 		struct rlimit rl = { 1024, 1024 };
7157 		(void) setrlimit(RLIMIT_NOFILE, &rl);
7158 
7159 		(void) close(ztest_fd_rand);
7160 		VERIFY3U(11, >=,
7161 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
7162 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
7163 
7164 		(void) enable_extended_FILE_stdio(-1, -1);
7165 		if (libpath != NULL)
7166 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
7167 		(void) execv(cmd, emptyargv);
7168 		ztest_dump_core = B_FALSE;
7169 		fatal(B_TRUE, "exec failed: %s", cmd);
7170 	}
7171 
7172 	if (cmdbuf != NULL) {
7173 		umem_free(cmdbuf, MAXPATHLEN);
7174 		cmd = NULL;
7175 	}
7176 
7177 	while (waitpid(pid, &status, 0) != pid)
7178 		continue;
7179 	if (statusp != NULL)
7180 		*statusp = status;
7181 
7182 	if (WIFEXITED(status)) {
7183 		if (WEXITSTATUS(status) != 0) {
7184 			(void) fprintf(stderr, "child exited with code %d\n",
7185 			    WEXITSTATUS(status));
7186 			exit(2);
7187 		}
7188 		return (B_FALSE);
7189 	} else if (WIFSIGNALED(status)) {
7190 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
7191 			(void) fprintf(stderr, "child died with signal %d\n",
7192 			    WTERMSIG(status));
7193 			exit(3);
7194 		}
7195 		return (B_TRUE);
7196 	} else {
7197 		(void) fprintf(stderr, "something strange happened to child\n");
7198 		exit(4);
7199 		/* NOTREACHED */
7200 	}
7201 }
7202 
7203 static void
7204 ztest_run_init(void)
7205 {
7206 	ztest_shared_t *zs = ztest_shared;
7207 
7208 	/*
7209 	 * Blow away any existing copy of zpool.cache
7210 	 */
7211 	(void) remove(spa_config_path);
7212 
7213 	if (ztest_opts.zo_init == 0) {
7214 		if (ztest_opts.zo_verbose >= 1)
7215 			(void) printf("Importing pool %s\n",
7216 			    ztest_opts.zo_pool);
7217 		ztest_import(zs);
7218 		return;
7219 	}
7220 
7221 	/*
7222 	 * Create and initialize our storage pool.
7223 	 */
7224 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
7225 		bzero(zs, sizeof (ztest_shared_t));
7226 		if (ztest_opts.zo_verbose >= 3 &&
7227 		    ztest_opts.zo_init != 1) {
7228 			(void) printf("ztest_init(), pass %d\n", i);
7229 		}
7230 		ztest_init(zs);
7231 	}
7232 }
7233 
7234 int
7235 main(int argc, char **argv)
7236 {
7237 	int kills = 0;
7238 	int iters = 0;
7239 	int older = 0;
7240 	int newer = 0;
7241 	ztest_shared_t *zs;
7242 	ztest_info_t *zi;
7243 	ztest_shared_callstate_t *zc;
7244 	char timebuf[100];
7245 	char numbuf[NN_NUMBUF_SZ];
7246 	char *cmd;
7247 	boolean_t hasalt;
7248 	char *fd_data_str = getenv("ZTEST_FD_DATA");
7249 
7250 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
7251 
7252 	dprintf_setup(&argc, argv);
7253 	zfs_deadman_synctime_ms = 300000;
7254 	/*
7255 	 * As two-word space map entries may not come up often (especially
7256 	 * if pool and vdev sizes are small) we want to force at least some
7257 	 * of them so the feature get tested.
7258 	 */
7259 	zfs_force_some_double_word_sm_entries = B_TRUE;
7260 
7261 	/*
7262 	 * Verify that even extensively damaged split blocks with many
7263 	 * segments can be reconstructed in a reasonable amount of time
7264 	 * when reconstruction is known to be possible.
7265 	 */
7266 	zfs_reconstruct_indirect_damage_fraction = 4;
7267 
7268 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
7269 	ASSERT3S(ztest_fd_rand, >=, 0);
7270 
7271 	if (!fd_data_str) {
7272 		process_options(argc, argv);
7273 
7274 		setup_data_fd();
7275 		setup_hdr();
7276 		setup_data();
7277 		bcopy(&ztest_opts, ztest_shared_opts,
7278 		    sizeof (*ztest_shared_opts));
7279 	} else {
7280 		ztest_fd_data = atoi(fd_data_str);
7281 		setup_data();
7282 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
7283 	}
7284 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
7285 
7286 	/* Override location of zpool.cache */
7287 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
7288 	    ztest_opts.zo_dir), !=, -1);
7289 
7290 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
7291 	    UMEM_NOFAIL);
7292 	zs = ztest_shared;
7293 
7294 	if (fd_data_str) {
7295 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
7296 		metaslab_df_alloc_threshold =
7297 		    zs->zs_metaslab_df_alloc_threshold;
7298 
7299 		if (zs->zs_do_init)
7300 			ztest_run_init();
7301 		else
7302 			ztest_run(zs);
7303 		exit(0);
7304 	}
7305 
7306 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
7307 
7308 	if (ztest_opts.zo_verbose >= 1) {
7309 		(void) printf("%llu vdevs, %d datasets, %d threads,"
7310 		    " %llu seconds...\n",
7311 		    (u_longlong_t)ztest_opts.zo_vdevs,
7312 		    ztest_opts.zo_datasets,
7313 		    ztest_opts.zo_threads,
7314 		    (u_longlong_t)ztest_opts.zo_time);
7315 	}
7316 
7317 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
7318 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
7319 
7320 	zs->zs_do_init = B_TRUE;
7321 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
7322 		if (ztest_opts.zo_verbose >= 1) {
7323 			(void) printf("Executing older ztest for "
7324 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
7325 		}
7326 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
7327 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
7328 	} else {
7329 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
7330 	}
7331 	zs->zs_do_init = B_FALSE;
7332 
7333 	zs->zs_proc_start = gethrtime();
7334 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
7335 
7336 	for (int f = 0; f < ZTEST_FUNCS; f++) {
7337 		zi = &ztest_info[f];
7338 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
7339 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
7340 			zc->zc_next = UINT64_MAX;
7341 		else
7342 			zc->zc_next = zs->zs_proc_start +
7343 			    ztest_random(2 * zi->zi_interval[0] + 1);
7344 	}
7345 
7346 	/*
7347 	 * Run the tests in a loop.  These tests include fault injection
7348 	 * to verify that self-healing data works, and forced crashes
7349 	 * to verify that we never lose on-disk consistency.
7350 	 */
7351 	while (gethrtime() < zs->zs_proc_stop) {
7352 		int status;
7353 		boolean_t killed;
7354 
7355 		/*
7356 		 * Initialize the workload counters for each function.
7357 		 */
7358 		for (int f = 0; f < ZTEST_FUNCS; f++) {
7359 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
7360 			zc->zc_count = 0;
7361 			zc->zc_time = 0;
7362 		}
7363 
7364 		/* Set the allocation switch size */
7365 		zs->zs_metaslab_df_alloc_threshold =
7366 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
7367 
7368 		if (!hasalt || ztest_random(2) == 0) {
7369 			if (hasalt && ztest_opts.zo_verbose >= 1) {
7370 				(void) printf("Executing newer ztest: %s\n",
7371 				    cmd);
7372 			}
7373 			newer++;
7374 			killed = exec_child(cmd, NULL, B_TRUE, &status);
7375 		} else {
7376 			if (hasalt && ztest_opts.zo_verbose >= 1) {
7377 				(void) printf("Executing older ztest: %s\n",
7378 				    ztest_opts.zo_alt_ztest);
7379 			}
7380 			older++;
7381 			killed = exec_child(ztest_opts.zo_alt_ztest,
7382 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
7383 		}
7384 
7385 		if (killed)
7386 			kills++;
7387 		iters++;
7388 
7389 		if (ztest_opts.zo_verbose >= 1) {
7390 			hrtime_t now = gethrtime();
7391 
7392 			now = MIN(now, zs->zs_proc_stop);
7393 			print_time(zs->zs_proc_stop - now, timebuf);
7394 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
7395 
7396 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
7397 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
7398 			    iters,
7399 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
7400 			    (u_longlong_t)zs->zs_enospc_count,
7401 			    100.0 * zs->zs_alloc / zs->zs_space,
7402 			    numbuf,
7403 			    100.0 * (now - zs->zs_proc_start) /
7404 			    (ztest_opts.zo_time * NANOSEC), timebuf);
7405 		}
7406 
7407 		if (ztest_opts.zo_verbose >= 2) {
7408 			(void) printf("\nWorkload summary:\n\n");
7409 			(void) printf("%7s %9s   %s\n",
7410 			    "Calls", "Time", "Function");
7411 			(void) printf("%7s %9s   %s\n",
7412 			    "-----", "----", "--------");
7413 			for (int f = 0; f < ZTEST_FUNCS; f++) {
7414 				Dl_info dli;
7415 
7416 				zi = &ztest_info[f];
7417 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
7418 				print_time(zc->zc_time, timebuf);
7419 				(void) dladdr((void *)zi->zi_func, &dli);
7420 				(void) printf("%7llu %9s   %s\n",
7421 				    (u_longlong_t)zc->zc_count, timebuf,
7422 				    dli.dli_sname);
7423 			}
7424 			(void) printf("\n");
7425 		}
7426 
7427 		if (!ztest_opts.zo_mmp_test)
7428 			ztest_run_zdb(ztest_opts.zo_pool);
7429 	}
7430 
7431 	if (ztest_opts.zo_verbose >= 1) {
7432 		if (hasalt) {
7433 			(void) printf("%d runs of older ztest: %s\n", older,
7434 			    ztest_opts.zo_alt_ztest);
7435 			(void) printf("%d runs of newer ztest: %s\n", newer,
7436 			    cmd);
7437 		}
7438 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
7439 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
7440 	}
7441 
7442 	umem_free(cmd, MAXNAMELEN);
7443 
7444 	return (0);
7445 }
7446