xref: /illumos-gate/usr/src/cmd/ztest/ztest.c (revision 89c86e32293a30cdd7af530c38b2073fee01411c)
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, 2014 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  */
27 
28 /*
29  * The objective of this program is to provide a DMU/ZAP/SPA stress test
30  * that runs entirely in userland, is easy to use, and easy to extend.
31  *
32  * The overall design of the ztest program is as follows:
33  *
34  * (1) For each major functional area (e.g. adding vdevs to a pool,
35  *     creating and destroying datasets, reading and writing objects, etc)
36  *     we have a simple routine to test that functionality.  These
37  *     individual routines do not have to do anything "stressful".
38  *
39  * (2) We turn these simple functionality tests into a stress test by
40  *     running them all in parallel, with as many threads as desired,
41  *     and spread across as many datasets, objects, and vdevs as desired.
42  *
43  * (3) While all this is happening, we inject faults into the pool to
44  *     verify that self-healing data really works.
45  *
46  * (4) Every time we open a dataset, we change its checksum and compression
47  *     functions.  Thus even individual objects vary from block to block
48  *     in which checksum they use and whether they're compressed.
49  *
50  * (5) To verify that we never lose on-disk consistency after a crash,
51  *     we run the entire test in a child of the main process.
52  *     At random times, the child self-immolates with a SIGKILL.
53  *     This is the software equivalent of pulling the power cord.
54  *     The parent then runs the test again, using the existing
55  *     storage pool, as many times as desired. If backwards compatibility
56  *     testing is enabled ztest will sometimes run the "older" version
57  *     of ztest after a SIGKILL.
58  *
59  * (6) To verify that we don't have future leaks or temporal incursions,
60  *     many of the functional tests record the transaction group number
61  *     as part of their data.  When reading old data, they verify that
62  *     the transaction group number is less than the current, open txg.
63  *     If you add a new test, please do this if applicable.
64  *
65  * When run with no arguments, ztest runs for about five minutes and
66  * produces no output if successful.  To get a little bit of information,
67  * specify -V.  To get more information, specify -VV, and so on.
68  *
69  * To turn this into an overnight stress test, use -T to specify run time.
70  *
71  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
72  * to increase the pool capacity, fanout, and overall stress level.
73  *
74  * Use the -k option to set the desired frequency of kills.
75  *
76  * When ztest invokes itself it passes all relevant information through a
77  * temporary file which is mmap-ed in the child process. This allows shared
78  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
79  * stored at offset 0 of this file and contains information on the size and
80  * number of shared structures in the file. The information stored in this file
81  * must remain backwards compatible with older versions of ztest so that
82  * ztest can invoke them during backwards compatibility testing (-B).
83  */
84 
85 #include <sys/zfs_context.h>
86 #include <sys/spa.h>
87 #include <sys/dmu.h>
88 #include <sys/txg.h>
89 #include <sys/dbuf.h>
90 #include <sys/zap.h>
91 #include <sys/dmu_objset.h>
92 #include <sys/poll.h>
93 #include <sys/stat.h>
94 #include <sys/time.h>
95 #include <sys/wait.h>
96 #include <sys/mman.h>
97 #include <sys/resource.h>
98 #include <sys/zio.h>
99 #include <sys/zil.h>
100 #include <sys/zil_impl.h>
101 #include <sys/vdev_impl.h>
102 #include <sys/vdev_file.h>
103 #include <sys/spa_impl.h>
104 #include <sys/metaslab_impl.h>
105 #include <sys/dsl_prop.h>
106 #include <sys/dsl_dataset.h>
107 #include <sys/dsl_destroy.h>
108 #include <sys/dsl_scan.h>
109 #include <sys/zio_checksum.h>
110 #include <sys/refcount.h>
111 #include <sys/zfeature.h>
112 #include <sys/dsl_userhold.h>
113 #include <stdio.h>
114 #include <stdio_ext.h>
115 #include <stdlib.h>
116 #include <unistd.h>
117 #include <signal.h>
118 #include <umem.h>
119 #include <dlfcn.h>
120 #include <ctype.h>
121 #include <math.h>
122 #include <sys/fs/zfs.h>
123 #include <libnvpair.h>
124 
125 static int ztest_fd_data = -1;
126 static int ztest_fd_rand = -1;
127 
128 typedef struct ztest_shared_hdr {
129 	uint64_t	zh_hdr_size;
130 	uint64_t	zh_opts_size;
131 	uint64_t	zh_size;
132 	uint64_t	zh_stats_size;
133 	uint64_t	zh_stats_count;
134 	uint64_t	zh_ds_size;
135 	uint64_t	zh_ds_count;
136 } ztest_shared_hdr_t;
137 
138 static ztest_shared_hdr_t *ztest_shared_hdr;
139 
140 typedef struct ztest_shared_opts {
141 	char zo_pool[MAXNAMELEN];
142 	char zo_dir[MAXNAMELEN];
143 	char zo_alt_ztest[MAXNAMELEN];
144 	char zo_alt_libpath[MAXNAMELEN];
145 	uint64_t zo_vdevs;
146 	uint64_t zo_vdevtime;
147 	size_t zo_vdev_size;
148 	int zo_ashift;
149 	int zo_mirrors;
150 	int zo_raidz;
151 	int zo_raidz_parity;
152 	int zo_datasets;
153 	int zo_threads;
154 	uint64_t zo_passtime;
155 	uint64_t zo_killrate;
156 	int zo_verbose;
157 	int zo_init;
158 	uint64_t zo_time;
159 	uint64_t zo_maxloops;
160 	uint64_t zo_metaslab_gang_bang;
161 } ztest_shared_opts_t;
162 
163 static const ztest_shared_opts_t ztest_opts_defaults = {
164 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
165 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
166 	.zo_alt_ztest = { '\0' },
167 	.zo_alt_libpath = { '\0' },
168 	.zo_vdevs = 5,
169 	.zo_ashift = SPA_MINBLOCKSHIFT,
170 	.zo_mirrors = 2,
171 	.zo_raidz = 4,
172 	.zo_raidz_parity = 1,
173 	.zo_vdev_size = SPA_MINDEVSIZE * 2,
174 	.zo_datasets = 7,
175 	.zo_threads = 23,
176 	.zo_passtime = 60,		/* 60 seconds */
177 	.zo_killrate = 70,		/* 70% kill rate */
178 	.zo_verbose = 0,
179 	.zo_init = 1,
180 	.zo_time = 300,			/* 5 minutes */
181 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
182 	.zo_metaslab_gang_bang = 32 << 10
183 };
184 
185 extern uint64_t metaslab_gang_bang;
186 extern uint64_t metaslab_df_alloc_threshold;
187 extern uint64_t zfs_deadman_synctime_ms;
188 extern int metaslab_preload_limit;
189 
190 static ztest_shared_opts_t *ztest_shared_opts;
191 static ztest_shared_opts_t ztest_opts;
192 
193 typedef struct ztest_shared_ds {
194 	uint64_t	zd_seq;
195 } ztest_shared_ds_t;
196 
197 static ztest_shared_ds_t *ztest_shared_ds;
198 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
199 
200 #define	BT_MAGIC	0x123456789abcdefULL
201 #define	MAXFAULTS() \
202 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
203 
204 enum ztest_io_type {
205 	ZTEST_IO_WRITE_TAG,
206 	ZTEST_IO_WRITE_PATTERN,
207 	ZTEST_IO_WRITE_ZEROES,
208 	ZTEST_IO_TRUNCATE,
209 	ZTEST_IO_SETATTR,
210 	ZTEST_IO_REWRITE,
211 	ZTEST_IO_TYPES
212 };
213 
214 typedef struct ztest_block_tag {
215 	uint64_t	bt_magic;
216 	uint64_t	bt_objset;
217 	uint64_t	bt_object;
218 	uint64_t	bt_offset;
219 	uint64_t	bt_gen;
220 	uint64_t	bt_txg;
221 	uint64_t	bt_crtxg;
222 } ztest_block_tag_t;
223 
224 typedef struct bufwad {
225 	uint64_t	bw_index;
226 	uint64_t	bw_txg;
227 	uint64_t	bw_data;
228 } bufwad_t;
229 
230 /*
231  * XXX -- fix zfs range locks to be generic so we can use them here.
232  */
233 typedef enum {
234 	RL_READER,
235 	RL_WRITER,
236 	RL_APPEND
237 } rl_type_t;
238 
239 typedef struct rll {
240 	void		*rll_writer;
241 	int		rll_readers;
242 	mutex_t		rll_lock;
243 	cond_t		rll_cv;
244 } rll_t;
245 
246 typedef struct rl {
247 	uint64_t	rl_object;
248 	uint64_t	rl_offset;
249 	uint64_t	rl_size;
250 	rll_t		*rl_lock;
251 } rl_t;
252 
253 #define	ZTEST_RANGE_LOCKS	64
254 #define	ZTEST_OBJECT_LOCKS	64
255 
256 /*
257  * Object descriptor.  Used as a template for object lookup/create/remove.
258  */
259 typedef struct ztest_od {
260 	uint64_t	od_dir;
261 	uint64_t	od_object;
262 	dmu_object_type_t od_type;
263 	dmu_object_type_t od_crtype;
264 	uint64_t	od_blocksize;
265 	uint64_t	od_crblocksize;
266 	uint64_t	od_gen;
267 	uint64_t	od_crgen;
268 	char		od_name[MAXNAMELEN];
269 } ztest_od_t;
270 
271 /*
272  * Per-dataset state.
273  */
274 typedef struct ztest_ds {
275 	ztest_shared_ds_t *zd_shared;
276 	objset_t	*zd_os;
277 	rwlock_t	zd_zilog_lock;
278 	zilog_t		*zd_zilog;
279 	ztest_od_t	*zd_od;		/* debugging aid */
280 	char		zd_name[MAXNAMELEN];
281 	mutex_t		zd_dirobj_lock;
282 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
283 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
284 } ztest_ds_t;
285 
286 /*
287  * Per-iteration state.
288  */
289 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
290 
291 typedef struct ztest_info {
292 	ztest_func_t	*zi_func;	/* test function */
293 	uint64_t	zi_iters;	/* iterations per execution */
294 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
295 } ztest_info_t;
296 
297 typedef struct ztest_shared_callstate {
298 	uint64_t	zc_count;	/* per-pass count */
299 	uint64_t	zc_time;	/* per-pass time */
300 	uint64_t	zc_next;	/* next time to call this function */
301 } ztest_shared_callstate_t;
302 
303 static ztest_shared_callstate_t *ztest_shared_callstate;
304 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
305 
306 /*
307  * Note: these aren't static because we want dladdr() to work.
308  */
309 ztest_func_t ztest_dmu_read_write;
310 ztest_func_t ztest_dmu_write_parallel;
311 ztest_func_t ztest_dmu_object_alloc_free;
312 ztest_func_t ztest_dmu_commit_callbacks;
313 ztest_func_t ztest_zap;
314 ztest_func_t ztest_zap_parallel;
315 ztest_func_t ztest_zil_commit;
316 ztest_func_t ztest_zil_remount;
317 ztest_func_t ztest_dmu_read_write_zcopy;
318 ztest_func_t ztest_dmu_objset_create_destroy;
319 ztest_func_t ztest_dmu_prealloc;
320 ztest_func_t ztest_fzap;
321 ztest_func_t ztest_dmu_snapshot_create_destroy;
322 ztest_func_t ztest_dsl_prop_get_set;
323 ztest_func_t ztest_spa_prop_get_set;
324 ztest_func_t ztest_spa_create_destroy;
325 ztest_func_t ztest_fault_inject;
326 ztest_func_t ztest_ddt_repair;
327 ztest_func_t ztest_dmu_snapshot_hold;
328 ztest_func_t ztest_spa_rename;
329 ztest_func_t ztest_scrub;
330 ztest_func_t ztest_dsl_dataset_promote_busy;
331 ztest_func_t ztest_vdev_attach_detach;
332 ztest_func_t ztest_vdev_LUN_growth;
333 ztest_func_t ztest_vdev_add_remove;
334 ztest_func_t ztest_vdev_aux_add_remove;
335 ztest_func_t ztest_split_pool;
336 ztest_func_t ztest_reguid;
337 ztest_func_t ztest_spa_upgrade;
338 
339 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
340 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
341 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
342 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
343 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
344 
345 ztest_info_t ztest_info[] = {
346 	{ ztest_dmu_read_write,			1,	&zopt_always	},
347 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
348 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
349 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
350 	{ ztest_zap,				30,	&zopt_always	},
351 	{ ztest_zap_parallel,			100,	&zopt_always	},
352 	{ ztest_split_pool,			1,	&zopt_always	},
353 	{ ztest_zil_commit,			1,	&zopt_incessant	},
354 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
355 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
356 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
357 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
358 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
359 #if 0
360 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
361 #endif
362 	{ ztest_fzap,				1,	&zopt_sometimes	},
363 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
364 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
365 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
366 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
367 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
368 	{ ztest_reguid,				1,	&zopt_rarely	},
369 	{ ztest_spa_rename,			1,	&zopt_rarely	},
370 	{ ztest_scrub,				1,	&zopt_rarely	},
371 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
372 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
373 	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
374 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
375 	{ ztest_vdev_add_remove,		1,
376 	    &ztest_opts.zo_vdevtime				},
377 	{ ztest_vdev_aux_add_remove,		1,
378 	    &ztest_opts.zo_vdevtime				},
379 };
380 
381 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
382 
383 /*
384  * The following struct is used to hold a list of uncalled commit callbacks.
385  * The callbacks are ordered by txg number.
386  */
387 typedef struct ztest_cb_list {
388 	mutex_t	zcl_callbacks_lock;
389 	list_t	zcl_callbacks;
390 } ztest_cb_list_t;
391 
392 /*
393  * Stuff we need to share writably between parent and child.
394  */
395 typedef struct ztest_shared {
396 	boolean_t	zs_do_init;
397 	hrtime_t	zs_proc_start;
398 	hrtime_t	zs_proc_stop;
399 	hrtime_t	zs_thread_start;
400 	hrtime_t	zs_thread_stop;
401 	hrtime_t	zs_thread_kill;
402 	uint64_t	zs_enospc_count;
403 	uint64_t	zs_vdev_next_leaf;
404 	uint64_t	zs_vdev_aux;
405 	uint64_t	zs_alloc;
406 	uint64_t	zs_space;
407 	uint64_t	zs_splits;
408 	uint64_t	zs_mirrors;
409 	uint64_t	zs_metaslab_sz;
410 	uint64_t	zs_metaslab_df_alloc_threshold;
411 	uint64_t	zs_guid;
412 } ztest_shared_t;
413 
414 #define	ID_PARALLEL	-1ULL
415 
416 static char ztest_dev_template[] = "%s/%s.%llua";
417 static char ztest_aux_template[] = "%s/%s.%s.%llu";
418 ztest_shared_t *ztest_shared;
419 
420 static spa_t *ztest_spa = NULL;
421 static ztest_ds_t *ztest_ds;
422 
423 static mutex_t ztest_vdev_lock;
424 
425 /*
426  * The ztest_name_lock protects the pool and dataset namespace used by
427  * the individual tests. To modify the namespace, consumers must grab
428  * this lock as writer. Grabbing the lock as reader will ensure that the
429  * namespace does not change while the lock is held.
430  */
431 static rwlock_t ztest_name_lock;
432 
433 static boolean_t ztest_dump_core = B_TRUE;
434 static boolean_t ztest_exiting;
435 
436 /* Global commit callback list */
437 static ztest_cb_list_t zcl;
438 
439 enum ztest_object {
440 	ZTEST_META_DNODE = 0,
441 	ZTEST_DIROBJ,
442 	ZTEST_OBJECTS
443 };
444 
445 static void usage(boolean_t) __NORETURN;
446 
447 /*
448  * These libumem hooks provide a reasonable set of defaults for the allocator's
449  * debugging facilities.
450  */
451 const char *
452 _umem_debug_init()
453 {
454 	return ("default,verbose"); /* $UMEM_DEBUG setting */
455 }
456 
457 const char *
458 _umem_logging_init(void)
459 {
460 	return ("fail,contents"); /* $UMEM_LOGGING setting */
461 }
462 
463 #define	FATAL_MSG_SZ	1024
464 
465 char *fatal_msg;
466 
467 static void
468 fatal(int do_perror, char *message, ...)
469 {
470 	va_list args;
471 	int save_errno = errno;
472 	char buf[FATAL_MSG_SZ];
473 
474 	(void) fflush(stdout);
475 
476 	va_start(args, message);
477 	(void) sprintf(buf, "ztest: ");
478 	/* LINTED */
479 	(void) vsprintf(buf + strlen(buf), message, args);
480 	va_end(args);
481 	if (do_perror) {
482 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
483 		    ": %s", strerror(save_errno));
484 	}
485 	(void) fprintf(stderr, "%s\n", buf);
486 	fatal_msg = buf;			/* to ease debugging */
487 	if (ztest_dump_core)
488 		abort();
489 	exit(3);
490 }
491 
492 static int
493 str2shift(const char *buf)
494 {
495 	const char *ends = "BKMGTPEZ";
496 	int i;
497 
498 	if (buf[0] == '\0')
499 		return (0);
500 	for (i = 0; i < strlen(ends); i++) {
501 		if (toupper(buf[0]) == ends[i])
502 			break;
503 	}
504 	if (i == strlen(ends)) {
505 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
506 		    buf);
507 		usage(B_FALSE);
508 	}
509 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
510 		return (10*i);
511 	}
512 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
513 	usage(B_FALSE);
514 	/* NOTREACHED */
515 }
516 
517 static uint64_t
518 nicenumtoull(const char *buf)
519 {
520 	char *end;
521 	uint64_t val;
522 
523 	val = strtoull(buf, &end, 0);
524 	if (end == buf) {
525 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
526 		usage(B_FALSE);
527 	} else if (end[0] == '.') {
528 		double fval = strtod(buf, &end);
529 		fval *= pow(2, str2shift(end));
530 		if (fval > UINT64_MAX) {
531 			(void) fprintf(stderr, "ztest: value too large: %s\n",
532 			    buf);
533 			usage(B_FALSE);
534 		}
535 		val = (uint64_t)fval;
536 	} else {
537 		int shift = str2shift(end);
538 		if (shift >= 64 || (val << shift) >> shift != val) {
539 			(void) fprintf(stderr, "ztest: value too large: %s\n",
540 			    buf);
541 			usage(B_FALSE);
542 		}
543 		val <<= shift;
544 	}
545 	return (val);
546 }
547 
548 static void
549 usage(boolean_t requested)
550 {
551 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
552 
553 	char nice_vdev_size[10];
554 	char nice_gang_bang[10];
555 	FILE *fp = requested ? stdout : stderr;
556 
557 	nicenum(zo->zo_vdev_size, nice_vdev_size);
558 	nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
559 
560 	(void) fprintf(fp, "Usage: %s\n"
561 	    "\t[-v vdevs (default: %llu)]\n"
562 	    "\t[-s size_of_each_vdev (default: %s)]\n"
563 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
564 	    "\t[-m mirror_copies (default: %d)]\n"
565 	    "\t[-r raidz_disks (default: %d)]\n"
566 	    "\t[-R raidz_parity (default: %d)]\n"
567 	    "\t[-d datasets (default: %d)]\n"
568 	    "\t[-t threads (default: %d)]\n"
569 	    "\t[-g gang_block_threshold (default: %s)]\n"
570 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
571 	    "\t[-k kill_percentage (default: %llu%%)]\n"
572 	    "\t[-p pool_name (default: %s)]\n"
573 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
574 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
575 	    "\t[-E] use existing pool instead of creating new one\n"
576 	    "\t[-T time (default: %llu sec)] total run time\n"
577 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
578 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
579 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
580 	    "\t[-h] (print help)\n"
581 	    "",
582 	    zo->zo_pool,
583 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
584 	    nice_vdev_size,				/* -s */
585 	    zo->zo_ashift,				/* -a */
586 	    zo->zo_mirrors,				/* -m */
587 	    zo->zo_raidz,				/* -r */
588 	    zo->zo_raidz_parity,			/* -R */
589 	    zo->zo_datasets,				/* -d */
590 	    zo->zo_threads,				/* -t */
591 	    nice_gang_bang,				/* -g */
592 	    zo->zo_init,				/* -i */
593 	    (u_longlong_t)zo->zo_killrate,		/* -k */
594 	    zo->zo_pool,				/* -p */
595 	    zo->zo_dir,					/* -f */
596 	    (u_longlong_t)zo->zo_time,			/* -T */
597 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
598 	    (u_longlong_t)zo->zo_passtime);
599 	exit(requested ? 0 : 1);
600 }
601 
602 static void
603 process_options(int argc, char **argv)
604 {
605 	char *path;
606 	ztest_shared_opts_t *zo = &ztest_opts;
607 
608 	int opt;
609 	uint64_t value;
610 	char altdir[MAXNAMELEN] = { 0 };
611 
612 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
613 
614 	while ((opt = getopt(argc, argv,
615 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
616 		value = 0;
617 		switch (opt) {
618 		case 'v':
619 		case 's':
620 		case 'a':
621 		case 'm':
622 		case 'r':
623 		case 'R':
624 		case 'd':
625 		case 't':
626 		case 'g':
627 		case 'i':
628 		case 'k':
629 		case 'T':
630 		case 'P':
631 		case 'F':
632 			value = nicenumtoull(optarg);
633 		}
634 		switch (opt) {
635 		case 'v':
636 			zo->zo_vdevs = value;
637 			break;
638 		case 's':
639 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
640 			break;
641 		case 'a':
642 			zo->zo_ashift = value;
643 			break;
644 		case 'm':
645 			zo->zo_mirrors = value;
646 			break;
647 		case 'r':
648 			zo->zo_raidz = MAX(1, value);
649 			break;
650 		case 'R':
651 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
652 			break;
653 		case 'd':
654 			zo->zo_datasets = MAX(1, value);
655 			break;
656 		case 't':
657 			zo->zo_threads = MAX(1, value);
658 			break;
659 		case 'g':
660 			zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
661 			    value);
662 			break;
663 		case 'i':
664 			zo->zo_init = value;
665 			break;
666 		case 'k':
667 			zo->zo_killrate = value;
668 			break;
669 		case 'p':
670 			(void) strlcpy(zo->zo_pool, optarg,
671 			    sizeof (zo->zo_pool));
672 			break;
673 		case 'f':
674 			path = realpath(optarg, NULL);
675 			if (path == NULL) {
676 				(void) fprintf(stderr, "error: %s: %s\n",
677 				    optarg, strerror(errno));
678 				usage(B_FALSE);
679 			} else {
680 				(void) strlcpy(zo->zo_dir, path,
681 				    sizeof (zo->zo_dir));
682 			}
683 			break;
684 		case 'V':
685 			zo->zo_verbose++;
686 			break;
687 		case 'E':
688 			zo->zo_init = 0;
689 			break;
690 		case 'T':
691 			zo->zo_time = value;
692 			break;
693 		case 'P':
694 			zo->zo_passtime = MAX(1, value);
695 			break;
696 		case 'F':
697 			zo->zo_maxloops = MAX(1, value);
698 			break;
699 		case 'B':
700 			(void) strlcpy(altdir, optarg, sizeof (altdir));
701 			break;
702 		case 'h':
703 			usage(B_TRUE);
704 			break;
705 		case '?':
706 		default:
707 			usage(B_FALSE);
708 			break;
709 		}
710 	}
711 
712 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
713 
714 	zo->zo_vdevtime =
715 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
716 	    UINT64_MAX >> 2);
717 
718 	if (strlen(altdir) > 0) {
719 		char *cmd;
720 		char *realaltdir;
721 		char *bin;
722 		char *ztest;
723 		char *isa;
724 		int isalen;
725 
726 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
727 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
728 
729 		VERIFY(NULL != realpath(getexecname(), cmd));
730 		if (0 != access(altdir, F_OK)) {
731 			ztest_dump_core = B_FALSE;
732 			fatal(B_TRUE, "invalid alternate ztest path: %s",
733 			    altdir);
734 		}
735 		VERIFY(NULL != realpath(altdir, realaltdir));
736 
737 		/*
738 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
739 		 * We want to extract <isa> to determine if we should use
740 		 * 32 or 64 bit binaries.
741 		 */
742 		bin = strstr(cmd, "/usr/bin/");
743 		ztest = strstr(bin, "/ztest");
744 		isa = bin + 9;
745 		isalen = ztest - isa;
746 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
747 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
748 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
749 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
750 
751 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
752 			ztest_dump_core = B_FALSE;
753 			fatal(B_TRUE, "invalid alternate ztest: %s",
754 			    zo->zo_alt_ztest);
755 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
756 			ztest_dump_core = B_FALSE;
757 			fatal(B_TRUE, "invalid alternate lib directory %s",
758 			    zo->zo_alt_libpath);
759 		}
760 
761 		umem_free(cmd, MAXPATHLEN);
762 		umem_free(realaltdir, MAXPATHLEN);
763 	}
764 }
765 
766 static void
767 ztest_kill(ztest_shared_t *zs)
768 {
769 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
770 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
771 
772 	/*
773 	 * Before we kill off ztest, make sure that the config is updated.
774 	 * See comment above spa_config_sync().
775 	 */
776 	mutex_enter(&spa_namespace_lock);
777 	spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
778 	mutex_exit(&spa_namespace_lock);
779 
780 	zfs_dbgmsg_print(FTAG);
781 	(void) kill(getpid(), SIGKILL);
782 }
783 
784 static uint64_t
785 ztest_random(uint64_t range)
786 {
787 	uint64_t r;
788 
789 	ASSERT3S(ztest_fd_rand, >=, 0);
790 
791 	if (range == 0)
792 		return (0);
793 
794 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
795 		fatal(1, "short read from /dev/urandom");
796 
797 	return (r % range);
798 }
799 
800 /* ARGSUSED */
801 static void
802 ztest_record_enospc(const char *s)
803 {
804 	ztest_shared->zs_enospc_count++;
805 }
806 
807 static uint64_t
808 ztest_get_ashift(void)
809 {
810 	if (ztest_opts.zo_ashift == 0)
811 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
812 	return (ztest_opts.zo_ashift);
813 }
814 
815 static nvlist_t *
816 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
817 {
818 	char pathbuf[MAXPATHLEN];
819 	uint64_t vdev;
820 	nvlist_t *file;
821 
822 	if (ashift == 0)
823 		ashift = ztest_get_ashift();
824 
825 	if (path == NULL) {
826 		path = pathbuf;
827 
828 		if (aux != NULL) {
829 			vdev = ztest_shared->zs_vdev_aux;
830 			(void) snprintf(path, sizeof (pathbuf),
831 			    ztest_aux_template, ztest_opts.zo_dir,
832 			    pool == NULL ? ztest_opts.zo_pool : pool,
833 			    aux, vdev);
834 		} else {
835 			vdev = ztest_shared->zs_vdev_next_leaf++;
836 			(void) snprintf(path, sizeof (pathbuf),
837 			    ztest_dev_template, ztest_opts.zo_dir,
838 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
839 		}
840 	}
841 
842 	if (size != 0) {
843 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
844 		if (fd == -1)
845 			fatal(1, "can't open %s", path);
846 		if (ftruncate(fd, size) != 0)
847 			fatal(1, "can't ftruncate %s", path);
848 		(void) close(fd);
849 	}
850 
851 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
852 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
853 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
854 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
855 
856 	return (file);
857 }
858 
859 static nvlist_t *
860 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
861     uint64_t ashift, int r)
862 {
863 	nvlist_t *raidz, **child;
864 	int c;
865 
866 	if (r < 2)
867 		return (make_vdev_file(path, aux, pool, size, ashift));
868 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
869 
870 	for (c = 0; c < r; c++)
871 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
872 
873 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
874 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
875 	    VDEV_TYPE_RAIDZ) == 0);
876 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
877 	    ztest_opts.zo_raidz_parity) == 0);
878 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
879 	    child, r) == 0);
880 
881 	for (c = 0; c < r; c++)
882 		nvlist_free(child[c]);
883 
884 	umem_free(child, r * sizeof (nvlist_t *));
885 
886 	return (raidz);
887 }
888 
889 static nvlist_t *
890 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
891     uint64_t ashift, int r, int m)
892 {
893 	nvlist_t *mirror, **child;
894 	int c;
895 
896 	if (m < 1)
897 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
898 
899 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
900 
901 	for (c = 0; c < m; c++)
902 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
903 
904 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
905 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
906 	    VDEV_TYPE_MIRROR) == 0);
907 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
908 	    child, m) == 0);
909 
910 	for (c = 0; c < m; c++)
911 		nvlist_free(child[c]);
912 
913 	umem_free(child, m * sizeof (nvlist_t *));
914 
915 	return (mirror);
916 }
917 
918 static nvlist_t *
919 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
920     int log, int r, int m, int t)
921 {
922 	nvlist_t *root, **child;
923 	int c;
924 
925 	ASSERT(t > 0);
926 
927 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
928 
929 	for (c = 0; c < t; c++) {
930 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
931 		    r, m);
932 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
933 		    log) == 0);
934 	}
935 
936 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
937 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
938 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
939 	    child, t) == 0);
940 
941 	for (c = 0; c < t; c++)
942 		nvlist_free(child[c]);
943 
944 	umem_free(child, t * sizeof (nvlist_t *));
945 
946 	return (root);
947 }
948 
949 /*
950  * Find a random spa version. Returns back a random spa version in the
951  * range [initial_version, SPA_VERSION_FEATURES].
952  */
953 static uint64_t
954 ztest_random_spa_version(uint64_t initial_version)
955 {
956 	uint64_t version = initial_version;
957 
958 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
959 		version = version +
960 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
961 	}
962 
963 	if (version > SPA_VERSION_BEFORE_FEATURES)
964 		version = SPA_VERSION_FEATURES;
965 
966 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
967 	return (version);
968 }
969 
970 /*
971  * Find the largest ashift used
972  */
973 static uint64_t
974 ztest_spa_get_ashift() {
975 	uint64_t i;
976 	uint64_t ashift = SPA_MINBLOCKSHIFT;
977 	vdev_t *rvd = ztest_spa->spa_root_vdev;
978 
979 	for (i = 0; i < rvd->vdev_children; i++) {
980 		ashift = MAX(ashift, rvd->vdev_child[i]->vdev_ashift);
981 	}
982 	return (ashift);
983 }
984 
985 static int
986 ztest_random_blocksize(void)
987 {
988 	uint64_t block_shift;
989 	/*
990 	 * Choose a block size >= the ashift.
991 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
992 	 */
993 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
994 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
995 		maxbs = 20;
996 	block_shift = ztest_random(maxbs - ztest_spa_get_ashift() + 1);
997 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
998 }
999 
1000 static int
1001 ztest_random_ibshift(void)
1002 {
1003 	return (DN_MIN_INDBLKSHIFT +
1004 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1005 }
1006 
1007 static uint64_t
1008 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1009 {
1010 	uint64_t top;
1011 	vdev_t *rvd = spa->spa_root_vdev;
1012 	vdev_t *tvd;
1013 
1014 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1015 
1016 	do {
1017 		top = ztest_random(rvd->vdev_children);
1018 		tvd = rvd->vdev_child[top];
1019 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1020 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1021 
1022 	return (top);
1023 }
1024 
1025 static uint64_t
1026 ztest_random_dsl_prop(zfs_prop_t prop)
1027 {
1028 	uint64_t value;
1029 
1030 	do {
1031 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1032 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1033 
1034 	return (value);
1035 }
1036 
1037 static int
1038 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1039     boolean_t inherit)
1040 {
1041 	const char *propname = zfs_prop_to_name(prop);
1042 	const char *valname;
1043 	char setpoint[MAXPATHLEN];
1044 	uint64_t curval;
1045 	int error;
1046 
1047 	error = dsl_prop_set_int(osname, propname,
1048 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1049 
1050 	if (error == ENOSPC) {
1051 		ztest_record_enospc(FTAG);
1052 		return (error);
1053 	}
1054 	ASSERT0(error);
1055 
1056 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1057 
1058 	if (ztest_opts.zo_verbose >= 6) {
1059 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1060 		(void) printf("%s %s = %s at '%s'\n",
1061 		    osname, propname, valname, setpoint);
1062 	}
1063 
1064 	return (error);
1065 }
1066 
1067 static int
1068 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1069 {
1070 	spa_t *spa = ztest_spa;
1071 	nvlist_t *props = NULL;
1072 	int error;
1073 
1074 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1075 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1076 
1077 	error = spa_prop_set(spa, props);
1078 
1079 	nvlist_free(props);
1080 
1081 	if (error == ENOSPC) {
1082 		ztest_record_enospc(FTAG);
1083 		return (error);
1084 	}
1085 	ASSERT0(error);
1086 
1087 	return (error);
1088 }
1089 
1090 static void
1091 ztest_rll_init(rll_t *rll)
1092 {
1093 	rll->rll_writer = NULL;
1094 	rll->rll_readers = 0;
1095 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1096 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1097 }
1098 
1099 static void
1100 ztest_rll_destroy(rll_t *rll)
1101 {
1102 	ASSERT(rll->rll_writer == NULL);
1103 	ASSERT(rll->rll_readers == 0);
1104 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1105 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
1106 }
1107 
1108 static void
1109 ztest_rll_lock(rll_t *rll, rl_type_t type)
1110 {
1111 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1112 
1113 	if (type == RL_READER) {
1114 		while (rll->rll_writer != NULL)
1115 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1116 		rll->rll_readers++;
1117 	} else {
1118 		while (rll->rll_writer != NULL || rll->rll_readers)
1119 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1120 		rll->rll_writer = curthread;
1121 	}
1122 
1123 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1124 }
1125 
1126 static void
1127 ztest_rll_unlock(rll_t *rll)
1128 {
1129 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1130 
1131 	if (rll->rll_writer) {
1132 		ASSERT(rll->rll_readers == 0);
1133 		rll->rll_writer = NULL;
1134 	} else {
1135 		ASSERT(rll->rll_readers != 0);
1136 		ASSERT(rll->rll_writer == NULL);
1137 		rll->rll_readers--;
1138 	}
1139 
1140 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1141 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1142 
1143 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1144 }
1145 
1146 static void
1147 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1148 {
1149 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1150 
1151 	ztest_rll_lock(rll, type);
1152 }
1153 
1154 static void
1155 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1156 {
1157 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1158 
1159 	ztest_rll_unlock(rll);
1160 }
1161 
1162 static rl_t *
1163 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1164     uint64_t size, rl_type_t type)
1165 {
1166 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1167 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1168 	rl_t *rl;
1169 
1170 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1171 	rl->rl_object = object;
1172 	rl->rl_offset = offset;
1173 	rl->rl_size = size;
1174 	rl->rl_lock = rll;
1175 
1176 	ztest_rll_lock(rll, type);
1177 
1178 	return (rl);
1179 }
1180 
1181 static void
1182 ztest_range_unlock(rl_t *rl)
1183 {
1184 	rll_t *rll = rl->rl_lock;
1185 
1186 	ztest_rll_unlock(rll);
1187 
1188 	umem_free(rl, sizeof (*rl));
1189 }
1190 
1191 static void
1192 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1193 {
1194 	zd->zd_os = os;
1195 	zd->zd_zilog = dmu_objset_zil(os);
1196 	zd->zd_shared = szd;
1197 	dmu_objset_name(os, zd->zd_name);
1198 
1199 	if (zd->zd_shared != NULL)
1200 		zd->zd_shared->zd_seq = 0;
1201 
1202 	VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1203 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1204 
1205 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1206 		ztest_rll_init(&zd->zd_object_lock[l]);
1207 
1208 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1209 		ztest_rll_init(&zd->zd_range_lock[l]);
1210 }
1211 
1212 static void
1213 ztest_zd_fini(ztest_ds_t *zd)
1214 {
1215 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1216 
1217 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1218 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1219 
1220 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1221 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1222 }
1223 
1224 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1225 
1226 static uint64_t
1227 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1228 {
1229 	uint64_t txg;
1230 	int error;
1231 
1232 	/*
1233 	 * Attempt to assign tx to some transaction group.
1234 	 */
1235 	error = dmu_tx_assign(tx, txg_how);
1236 	if (error) {
1237 		if (error == ERESTART) {
1238 			ASSERT(txg_how == TXG_NOWAIT);
1239 			dmu_tx_wait(tx);
1240 		} else {
1241 			ASSERT3U(error, ==, ENOSPC);
1242 			ztest_record_enospc(tag);
1243 		}
1244 		dmu_tx_abort(tx);
1245 		return (0);
1246 	}
1247 	txg = dmu_tx_get_txg(tx);
1248 	ASSERT(txg != 0);
1249 	return (txg);
1250 }
1251 
1252 static void
1253 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1254 {
1255 	uint64_t *ip = buf;
1256 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1257 
1258 	while (ip < ip_end)
1259 		*ip++ = value;
1260 }
1261 
1262 static boolean_t
1263 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1264 {
1265 	uint64_t *ip = buf;
1266 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1267 	uint64_t diff = 0;
1268 
1269 	while (ip < ip_end)
1270 		diff |= (value - *ip++);
1271 
1272 	return (diff == 0);
1273 }
1274 
1275 static void
1276 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1277     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1278 {
1279 	bt->bt_magic = BT_MAGIC;
1280 	bt->bt_objset = dmu_objset_id(os);
1281 	bt->bt_object = object;
1282 	bt->bt_offset = offset;
1283 	bt->bt_gen = gen;
1284 	bt->bt_txg = txg;
1285 	bt->bt_crtxg = crtxg;
1286 }
1287 
1288 static void
1289 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1290     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1291 {
1292 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1293 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1294 	ASSERT3U(bt->bt_object, ==, object);
1295 	ASSERT3U(bt->bt_offset, ==, offset);
1296 	ASSERT3U(bt->bt_gen, <=, gen);
1297 	ASSERT3U(bt->bt_txg, <=, txg);
1298 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1299 }
1300 
1301 static ztest_block_tag_t *
1302 ztest_bt_bonus(dmu_buf_t *db)
1303 {
1304 	dmu_object_info_t doi;
1305 	ztest_block_tag_t *bt;
1306 
1307 	dmu_object_info_from_db(db, &doi);
1308 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1309 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1310 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1311 
1312 	return (bt);
1313 }
1314 
1315 /*
1316  * ZIL logging ops
1317  */
1318 
1319 #define	lrz_type	lr_mode
1320 #define	lrz_blocksize	lr_uid
1321 #define	lrz_ibshift	lr_gid
1322 #define	lrz_bonustype	lr_rdev
1323 #define	lrz_bonuslen	lr_crtime[1]
1324 
1325 static void
1326 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1327 {
1328 	char *name = (void *)(lr + 1);		/* name follows lr */
1329 	size_t namesize = strlen(name) + 1;
1330 	itx_t *itx;
1331 
1332 	if (zil_replaying(zd->zd_zilog, tx))
1333 		return;
1334 
1335 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1336 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1337 	    sizeof (*lr) + namesize - sizeof (lr_t));
1338 
1339 	zil_itx_assign(zd->zd_zilog, itx, tx);
1340 }
1341 
1342 static void
1343 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1344 {
1345 	char *name = (void *)(lr + 1);		/* name follows lr */
1346 	size_t namesize = strlen(name) + 1;
1347 	itx_t *itx;
1348 
1349 	if (zil_replaying(zd->zd_zilog, tx))
1350 		return;
1351 
1352 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1353 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1354 	    sizeof (*lr) + namesize - sizeof (lr_t));
1355 
1356 	itx->itx_oid = object;
1357 	zil_itx_assign(zd->zd_zilog, itx, tx);
1358 }
1359 
1360 static void
1361 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1362 {
1363 	itx_t *itx;
1364 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1365 
1366 	if (zil_replaying(zd->zd_zilog, tx))
1367 		return;
1368 
1369 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1370 		write_state = WR_INDIRECT;
1371 
1372 	itx = zil_itx_create(TX_WRITE,
1373 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1374 
1375 	if (write_state == WR_COPIED &&
1376 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1377 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1378 		zil_itx_destroy(itx);
1379 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1380 		write_state = WR_NEED_COPY;
1381 	}
1382 	itx->itx_private = zd;
1383 	itx->itx_wr_state = write_state;
1384 	itx->itx_sync = (ztest_random(8) == 0);
1385 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1386 
1387 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1388 	    sizeof (*lr) - sizeof (lr_t));
1389 
1390 	zil_itx_assign(zd->zd_zilog, itx, tx);
1391 }
1392 
1393 static void
1394 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1395 {
1396 	itx_t *itx;
1397 
1398 	if (zil_replaying(zd->zd_zilog, tx))
1399 		return;
1400 
1401 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1402 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1403 	    sizeof (*lr) - sizeof (lr_t));
1404 
1405 	itx->itx_sync = B_FALSE;
1406 	zil_itx_assign(zd->zd_zilog, itx, tx);
1407 }
1408 
1409 static void
1410 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1411 {
1412 	itx_t *itx;
1413 
1414 	if (zil_replaying(zd->zd_zilog, tx))
1415 		return;
1416 
1417 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1418 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1419 	    sizeof (*lr) - sizeof (lr_t));
1420 
1421 	itx->itx_sync = B_FALSE;
1422 	zil_itx_assign(zd->zd_zilog, itx, tx);
1423 }
1424 
1425 /*
1426  * ZIL replay ops
1427  */
1428 static int
1429 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1430 {
1431 	char *name = (void *)(lr + 1);		/* name follows lr */
1432 	objset_t *os = zd->zd_os;
1433 	ztest_block_tag_t *bbt;
1434 	dmu_buf_t *db;
1435 	dmu_tx_t *tx;
1436 	uint64_t txg;
1437 	int error = 0;
1438 
1439 	if (byteswap)
1440 		byteswap_uint64_array(lr, sizeof (*lr));
1441 
1442 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1443 	ASSERT(name[0] != '\0');
1444 
1445 	tx = dmu_tx_create(os);
1446 
1447 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1448 
1449 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1450 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1451 	} else {
1452 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1453 	}
1454 
1455 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1456 	if (txg == 0)
1457 		return (ENOSPC);
1458 
1459 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1460 
1461 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1462 		if (lr->lr_foid == 0) {
1463 			lr->lr_foid = zap_create(os,
1464 			    lr->lrz_type, lr->lrz_bonustype,
1465 			    lr->lrz_bonuslen, tx);
1466 		} else {
1467 			error = zap_create_claim(os, lr->lr_foid,
1468 			    lr->lrz_type, lr->lrz_bonustype,
1469 			    lr->lrz_bonuslen, tx);
1470 		}
1471 	} else {
1472 		if (lr->lr_foid == 0) {
1473 			lr->lr_foid = dmu_object_alloc(os,
1474 			    lr->lrz_type, 0, lr->lrz_bonustype,
1475 			    lr->lrz_bonuslen, tx);
1476 		} else {
1477 			error = dmu_object_claim(os, lr->lr_foid,
1478 			    lr->lrz_type, 0, lr->lrz_bonustype,
1479 			    lr->lrz_bonuslen, tx);
1480 		}
1481 	}
1482 
1483 	if (error) {
1484 		ASSERT3U(error, ==, EEXIST);
1485 		ASSERT(zd->zd_zilog->zl_replay);
1486 		dmu_tx_commit(tx);
1487 		return (error);
1488 	}
1489 
1490 	ASSERT(lr->lr_foid != 0);
1491 
1492 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1493 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1494 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1495 
1496 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1497 	bbt = ztest_bt_bonus(db);
1498 	dmu_buf_will_dirty(db, tx);
1499 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1500 	dmu_buf_rele(db, FTAG);
1501 
1502 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1503 	    &lr->lr_foid, tx));
1504 
1505 	(void) ztest_log_create(zd, tx, lr);
1506 
1507 	dmu_tx_commit(tx);
1508 
1509 	return (0);
1510 }
1511 
1512 static int
1513 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1514 {
1515 	char *name = (void *)(lr + 1);		/* name follows lr */
1516 	objset_t *os = zd->zd_os;
1517 	dmu_object_info_t doi;
1518 	dmu_tx_t *tx;
1519 	uint64_t object, txg;
1520 
1521 	if (byteswap)
1522 		byteswap_uint64_array(lr, sizeof (*lr));
1523 
1524 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1525 	ASSERT(name[0] != '\0');
1526 
1527 	VERIFY3U(0, ==,
1528 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1529 	ASSERT(object != 0);
1530 
1531 	ztest_object_lock(zd, object, RL_WRITER);
1532 
1533 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1534 
1535 	tx = dmu_tx_create(os);
1536 
1537 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1538 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1539 
1540 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1541 	if (txg == 0) {
1542 		ztest_object_unlock(zd, object);
1543 		return (ENOSPC);
1544 	}
1545 
1546 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1547 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1548 	} else {
1549 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1550 	}
1551 
1552 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1553 
1554 	(void) ztest_log_remove(zd, tx, lr, object);
1555 
1556 	dmu_tx_commit(tx);
1557 
1558 	ztest_object_unlock(zd, object);
1559 
1560 	return (0);
1561 }
1562 
1563 static int
1564 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1565 {
1566 	objset_t *os = zd->zd_os;
1567 	void *data = lr + 1;			/* data follows lr */
1568 	uint64_t offset, length;
1569 	ztest_block_tag_t *bt = data;
1570 	ztest_block_tag_t *bbt;
1571 	uint64_t gen, txg, lrtxg, crtxg;
1572 	dmu_object_info_t doi;
1573 	dmu_tx_t *tx;
1574 	dmu_buf_t *db;
1575 	arc_buf_t *abuf = NULL;
1576 	rl_t *rl;
1577 
1578 	if (byteswap)
1579 		byteswap_uint64_array(lr, sizeof (*lr));
1580 
1581 	offset = lr->lr_offset;
1582 	length = lr->lr_length;
1583 
1584 	/* If it's a dmu_sync() block, write the whole block */
1585 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1586 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1587 		if (length < blocksize) {
1588 			offset -= offset % blocksize;
1589 			length = blocksize;
1590 		}
1591 	}
1592 
1593 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1594 		byteswap_uint64_array(bt, sizeof (*bt));
1595 
1596 	if (bt->bt_magic != BT_MAGIC)
1597 		bt = NULL;
1598 
1599 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1600 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1601 
1602 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1603 
1604 	dmu_object_info_from_db(db, &doi);
1605 
1606 	bbt = ztest_bt_bonus(db);
1607 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1608 	gen = bbt->bt_gen;
1609 	crtxg = bbt->bt_crtxg;
1610 	lrtxg = lr->lr_common.lrc_txg;
1611 
1612 	tx = dmu_tx_create(os);
1613 
1614 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1615 
1616 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1617 	    P2PHASE(offset, length) == 0)
1618 		abuf = dmu_request_arcbuf(db, length);
1619 
1620 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1621 	if (txg == 0) {
1622 		if (abuf != NULL)
1623 			dmu_return_arcbuf(abuf);
1624 		dmu_buf_rele(db, FTAG);
1625 		ztest_range_unlock(rl);
1626 		ztest_object_unlock(zd, lr->lr_foid);
1627 		return (ENOSPC);
1628 	}
1629 
1630 	if (bt != NULL) {
1631 		/*
1632 		 * Usually, verify the old data before writing new data --
1633 		 * but not always, because we also want to verify correct
1634 		 * behavior when the data was not recently read into cache.
1635 		 */
1636 		ASSERT(offset % doi.doi_data_block_size == 0);
1637 		if (ztest_random(4) != 0) {
1638 			int prefetch = ztest_random(2) ?
1639 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1640 			ztest_block_tag_t rbt;
1641 
1642 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1643 			    sizeof (rbt), &rbt, prefetch) == 0);
1644 			if (rbt.bt_magic == BT_MAGIC) {
1645 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1646 				    offset, gen, txg, crtxg);
1647 			}
1648 		}
1649 
1650 		/*
1651 		 * Writes can appear to be newer than the bonus buffer because
1652 		 * the ztest_get_data() callback does a dmu_read() of the
1653 		 * open-context data, which may be different than the data
1654 		 * as it was when the write was generated.
1655 		 */
1656 		if (zd->zd_zilog->zl_replay) {
1657 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1658 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1659 			    bt->bt_crtxg);
1660 		}
1661 
1662 		/*
1663 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1664 		 * so that all of the usual ASSERTs will work.
1665 		 */
1666 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1667 	}
1668 
1669 	if (abuf == NULL) {
1670 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1671 	} else {
1672 		bcopy(data, abuf->b_data, length);
1673 		dmu_assign_arcbuf(db, offset, abuf, tx);
1674 	}
1675 
1676 	(void) ztest_log_write(zd, tx, lr);
1677 
1678 	dmu_buf_rele(db, FTAG);
1679 
1680 	dmu_tx_commit(tx);
1681 
1682 	ztest_range_unlock(rl);
1683 	ztest_object_unlock(zd, lr->lr_foid);
1684 
1685 	return (0);
1686 }
1687 
1688 static int
1689 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1690 {
1691 	objset_t *os = zd->zd_os;
1692 	dmu_tx_t *tx;
1693 	uint64_t txg;
1694 	rl_t *rl;
1695 
1696 	if (byteswap)
1697 		byteswap_uint64_array(lr, sizeof (*lr));
1698 
1699 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1700 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1701 	    RL_WRITER);
1702 
1703 	tx = dmu_tx_create(os);
1704 
1705 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1706 
1707 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1708 	if (txg == 0) {
1709 		ztest_range_unlock(rl);
1710 		ztest_object_unlock(zd, lr->lr_foid);
1711 		return (ENOSPC);
1712 	}
1713 
1714 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1715 	    lr->lr_length, tx) == 0);
1716 
1717 	(void) ztest_log_truncate(zd, tx, lr);
1718 
1719 	dmu_tx_commit(tx);
1720 
1721 	ztest_range_unlock(rl);
1722 	ztest_object_unlock(zd, lr->lr_foid);
1723 
1724 	return (0);
1725 }
1726 
1727 static int
1728 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1729 {
1730 	objset_t *os = zd->zd_os;
1731 	dmu_tx_t *tx;
1732 	dmu_buf_t *db;
1733 	ztest_block_tag_t *bbt;
1734 	uint64_t txg, lrtxg, crtxg;
1735 
1736 	if (byteswap)
1737 		byteswap_uint64_array(lr, sizeof (*lr));
1738 
1739 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1740 
1741 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1742 
1743 	tx = dmu_tx_create(os);
1744 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1745 
1746 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1747 	if (txg == 0) {
1748 		dmu_buf_rele(db, FTAG);
1749 		ztest_object_unlock(zd, lr->lr_foid);
1750 		return (ENOSPC);
1751 	}
1752 
1753 	bbt = ztest_bt_bonus(db);
1754 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1755 	crtxg = bbt->bt_crtxg;
1756 	lrtxg = lr->lr_common.lrc_txg;
1757 
1758 	if (zd->zd_zilog->zl_replay) {
1759 		ASSERT(lr->lr_size != 0);
1760 		ASSERT(lr->lr_mode != 0);
1761 		ASSERT(lrtxg != 0);
1762 	} else {
1763 		/*
1764 		 * Randomly change the size and increment the generation.
1765 		 */
1766 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1767 		    sizeof (*bbt);
1768 		lr->lr_mode = bbt->bt_gen + 1;
1769 		ASSERT(lrtxg == 0);
1770 	}
1771 
1772 	/*
1773 	 * Verify that the current bonus buffer is not newer than our txg.
1774 	 */
1775 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1776 	    MAX(txg, lrtxg), crtxg);
1777 
1778 	dmu_buf_will_dirty(db, tx);
1779 
1780 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1781 	ASSERT3U(lr->lr_size, <=, db->db_size);
1782 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1783 	bbt = ztest_bt_bonus(db);
1784 
1785 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1786 
1787 	dmu_buf_rele(db, FTAG);
1788 
1789 	(void) ztest_log_setattr(zd, tx, lr);
1790 
1791 	dmu_tx_commit(tx);
1792 
1793 	ztest_object_unlock(zd, lr->lr_foid);
1794 
1795 	return (0);
1796 }
1797 
1798 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1799 	NULL,			/* 0 no such transaction type */
1800 	ztest_replay_create,	/* TX_CREATE */
1801 	NULL,			/* TX_MKDIR */
1802 	NULL,			/* TX_MKXATTR */
1803 	NULL,			/* TX_SYMLINK */
1804 	ztest_replay_remove,	/* TX_REMOVE */
1805 	NULL,			/* TX_RMDIR */
1806 	NULL,			/* TX_LINK */
1807 	NULL,			/* TX_RENAME */
1808 	ztest_replay_write,	/* TX_WRITE */
1809 	ztest_replay_truncate,	/* TX_TRUNCATE */
1810 	ztest_replay_setattr,	/* TX_SETATTR */
1811 	NULL,			/* TX_ACL */
1812 	NULL,			/* TX_CREATE_ACL */
1813 	NULL,			/* TX_CREATE_ATTR */
1814 	NULL,			/* TX_CREATE_ACL_ATTR */
1815 	NULL,			/* TX_MKDIR_ACL */
1816 	NULL,			/* TX_MKDIR_ATTR */
1817 	NULL,			/* TX_MKDIR_ACL_ATTR */
1818 	NULL,			/* TX_WRITE2 */
1819 };
1820 
1821 /*
1822  * ZIL get_data callbacks
1823  */
1824 
1825 static void
1826 ztest_get_done(zgd_t *zgd, int error)
1827 {
1828 	ztest_ds_t *zd = zgd->zgd_private;
1829 	uint64_t object = zgd->zgd_rl->rl_object;
1830 
1831 	if (zgd->zgd_db)
1832 		dmu_buf_rele(zgd->zgd_db, zgd);
1833 
1834 	ztest_range_unlock(zgd->zgd_rl);
1835 	ztest_object_unlock(zd, object);
1836 
1837 	if (error == 0 && zgd->zgd_bp)
1838 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1839 
1840 	umem_free(zgd, sizeof (*zgd));
1841 }
1842 
1843 static int
1844 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1845 {
1846 	ztest_ds_t *zd = arg;
1847 	objset_t *os = zd->zd_os;
1848 	uint64_t object = lr->lr_foid;
1849 	uint64_t offset = lr->lr_offset;
1850 	uint64_t size = lr->lr_length;
1851 	blkptr_t *bp = &lr->lr_blkptr;
1852 	uint64_t txg = lr->lr_common.lrc_txg;
1853 	uint64_t crtxg;
1854 	dmu_object_info_t doi;
1855 	dmu_buf_t *db;
1856 	zgd_t *zgd;
1857 	int error;
1858 
1859 	ztest_object_lock(zd, object, RL_READER);
1860 	error = dmu_bonus_hold(os, object, FTAG, &db);
1861 	if (error) {
1862 		ztest_object_unlock(zd, object);
1863 		return (error);
1864 	}
1865 
1866 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1867 
1868 	if (crtxg == 0 || crtxg > txg) {
1869 		dmu_buf_rele(db, FTAG);
1870 		ztest_object_unlock(zd, object);
1871 		return (ENOENT);
1872 	}
1873 
1874 	dmu_object_info_from_db(db, &doi);
1875 	dmu_buf_rele(db, FTAG);
1876 	db = NULL;
1877 
1878 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1879 	zgd->zgd_zilog = zd->zd_zilog;
1880 	zgd->zgd_private = zd;
1881 
1882 	if (buf != NULL) {	/* immediate write */
1883 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1884 		    RL_READER);
1885 
1886 		error = dmu_read(os, object, offset, size, buf,
1887 		    DMU_READ_NO_PREFETCH);
1888 		ASSERT(error == 0);
1889 	} else {
1890 		size = doi.doi_data_block_size;
1891 		if (ISP2(size)) {
1892 			offset = P2ALIGN(offset, size);
1893 		} else {
1894 			ASSERT(offset < size);
1895 			offset = 0;
1896 		}
1897 
1898 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1899 		    RL_READER);
1900 
1901 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1902 		    DMU_READ_NO_PREFETCH);
1903 
1904 		if (error == 0) {
1905 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1906 			if (obp) {
1907 				ASSERT(BP_IS_HOLE(bp));
1908 				*bp = *obp;
1909 			}
1910 
1911 			zgd->zgd_db = db;
1912 			zgd->zgd_bp = bp;
1913 
1914 			ASSERT(db->db_offset == offset);
1915 			ASSERT(db->db_size == size);
1916 
1917 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1918 			    ztest_get_done, zgd);
1919 
1920 			if (error == 0)
1921 				return (0);
1922 		}
1923 	}
1924 
1925 	ztest_get_done(zgd, error);
1926 
1927 	return (error);
1928 }
1929 
1930 static void *
1931 ztest_lr_alloc(size_t lrsize, char *name)
1932 {
1933 	char *lr;
1934 	size_t namesize = name ? strlen(name) + 1 : 0;
1935 
1936 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1937 
1938 	if (name)
1939 		bcopy(name, lr + lrsize, namesize);
1940 
1941 	return (lr);
1942 }
1943 
1944 void
1945 ztest_lr_free(void *lr, size_t lrsize, char *name)
1946 {
1947 	size_t namesize = name ? strlen(name) + 1 : 0;
1948 
1949 	umem_free(lr, lrsize + namesize);
1950 }
1951 
1952 /*
1953  * Lookup a bunch of objects.  Returns the number of objects not found.
1954  */
1955 static int
1956 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1957 {
1958 	int missing = 0;
1959 	int error;
1960 
1961 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1962 
1963 	for (int i = 0; i < count; i++, od++) {
1964 		od->od_object = 0;
1965 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1966 		    sizeof (uint64_t), 1, &od->od_object);
1967 		if (error) {
1968 			ASSERT(error == ENOENT);
1969 			ASSERT(od->od_object == 0);
1970 			missing++;
1971 		} else {
1972 			dmu_buf_t *db;
1973 			ztest_block_tag_t *bbt;
1974 			dmu_object_info_t doi;
1975 
1976 			ASSERT(od->od_object != 0);
1977 			ASSERT(missing == 0);	/* there should be no gaps */
1978 
1979 			ztest_object_lock(zd, od->od_object, RL_READER);
1980 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1981 			    od->od_object, FTAG, &db));
1982 			dmu_object_info_from_db(db, &doi);
1983 			bbt = ztest_bt_bonus(db);
1984 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1985 			od->od_type = doi.doi_type;
1986 			od->od_blocksize = doi.doi_data_block_size;
1987 			od->od_gen = bbt->bt_gen;
1988 			dmu_buf_rele(db, FTAG);
1989 			ztest_object_unlock(zd, od->od_object);
1990 		}
1991 	}
1992 
1993 	return (missing);
1994 }
1995 
1996 static int
1997 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1998 {
1999 	int missing = 0;
2000 
2001 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2002 
2003 	for (int i = 0; i < count; i++, od++) {
2004 		if (missing) {
2005 			od->od_object = 0;
2006 			missing++;
2007 			continue;
2008 		}
2009 
2010 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2011 
2012 		lr->lr_doid = od->od_dir;
2013 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2014 		lr->lrz_type = od->od_crtype;
2015 		lr->lrz_blocksize = od->od_crblocksize;
2016 		lr->lrz_ibshift = ztest_random_ibshift();
2017 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2018 		lr->lrz_bonuslen = dmu_bonus_max();
2019 		lr->lr_gen = od->od_crgen;
2020 		lr->lr_crtime[0] = time(NULL);
2021 
2022 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2023 			ASSERT(missing == 0);
2024 			od->od_object = 0;
2025 			missing++;
2026 		} else {
2027 			od->od_object = lr->lr_foid;
2028 			od->od_type = od->od_crtype;
2029 			od->od_blocksize = od->od_crblocksize;
2030 			od->od_gen = od->od_crgen;
2031 			ASSERT(od->od_object != 0);
2032 		}
2033 
2034 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2035 	}
2036 
2037 	return (missing);
2038 }
2039 
2040 static int
2041 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2042 {
2043 	int missing = 0;
2044 	int error;
2045 
2046 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2047 
2048 	od += count - 1;
2049 
2050 	for (int i = count - 1; i >= 0; i--, od--) {
2051 		if (missing) {
2052 			missing++;
2053 			continue;
2054 		}
2055 
2056 		/*
2057 		 * No object was found.
2058 		 */
2059 		if (od->od_object == 0)
2060 			continue;
2061 
2062 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2063 
2064 		lr->lr_doid = od->od_dir;
2065 
2066 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2067 			ASSERT3U(error, ==, ENOSPC);
2068 			missing++;
2069 		} else {
2070 			od->od_object = 0;
2071 		}
2072 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2073 	}
2074 
2075 	return (missing);
2076 }
2077 
2078 static int
2079 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2080     void *data)
2081 {
2082 	lr_write_t *lr;
2083 	int error;
2084 
2085 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2086 
2087 	lr->lr_foid = object;
2088 	lr->lr_offset = offset;
2089 	lr->lr_length = size;
2090 	lr->lr_blkoff = 0;
2091 	BP_ZERO(&lr->lr_blkptr);
2092 
2093 	bcopy(data, lr + 1, size);
2094 
2095 	error = ztest_replay_write(zd, lr, B_FALSE);
2096 
2097 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2098 
2099 	return (error);
2100 }
2101 
2102 static int
2103 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2104 {
2105 	lr_truncate_t *lr;
2106 	int error;
2107 
2108 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2109 
2110 	lr->lr_foid = object;
2111 	lr->lr_offset = offset;
2112 	lr->lr_length = size;
2113 
2114 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2115 
2116 	ztest_lr_free(lr, sizeof (*lr), NULL);
2117 
2118 	return (error);
2119 }
2120 
2121 static int
2122 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2123 {
2124 	lr_setattr_t *lr;
2125 	int error;
2126 
2127 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2128 
2129 	lr->lr_foid = object;
2130 	lr->lr_size = 0;
2131 	lr->lr_mode = 0;
2132 
2133 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2134 
2135 	ztest_lr_free(lr, sizeof (*lr), NULL);
2136 
2137 	return (error);
2138 }
2139 
2140 static void
2141 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2142 {
2143 	objset_t *os = zd->zd_os;
2144 	dmu_tx_t *tx;
2145 	uint64_t txg;
2146 	rl_t *rl;
2147 
2148 	txg_wait_synced(dmu_objset_pool(os), 0);
2149 
2150 	ztest_object_lock(zd, object, RL_READER);
2151 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2152 
2153 	tx = dmu_tx_create(os);
2154 
2155 	dmu_tx_hold_write(tx, object, offset, size);
2156 
2157 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2158 
2159 	if (txg != 0) {
2160 		dmu_prealloc(os, object, offset, size, tx);
2161 		dmu_tx_commit(tx);
2162 		txg_wait_synced(dmu_objset_pool(os), txg);
2163 	} else {
2164 		(void) dmu_free_long_range(os, object, offset, size);
2165 	}
2166 
2167 	ztest_range_unlock(rl);
2168 	ztest_object_unlock(zd, object);
2169 }
2170 
2171 static void
2172 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2173 {
2174 	int err;
2175 	ztest_block_tag_t wbt;
2176 	dmu_object_info_t doi;
2177 	enum ztest_io_type io_type;
2178 	uint64_t blocksize;
2179 	void *data;
2180 
2181 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2182 	blocksize = doi.doi_data_block_size;
2183 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2184 
2185 	/*
2186 	 * Pick an i/o type at random, biased toward writing block tags.
2187 	 */
2188 	io_type = ztest_random(ZTEST_IO_TYPES);
2189 	if (ztest_random(2) == 0)
2190 		io_type = ZTEST_IO_WRITE_TAG;
2191 
2192 	(void) rw_rdlock(&zd->zd_zilog_lock);
2193 
2194 	switch (io_type) {
2195 
2196 	case ZTEST_IO_WRITE_TAG:
2197 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2198 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2199 		break;
2200 
2201 	case ZTEST_IO_WRITE_PATTERN:
2202 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2203 		if (ztest_random(2) == 0) {
2204 			/*
2205 			 * Induce fletcher2 collisions to ensure that
2206 			 * zio_ddt_collision() detects and resolves them
2207 			 * when using fletcher2-verify for deduplication.
2208 			 */
2209 			((uint64_t *)data)[0] ^= 1ULL << 63;
2210 			((uint64_t *)data)[4] ^= 1ULL << 63;
2211 		}
2212 		(void) ztest_write(zd, object, offset, blocksize, data);
2213 		break;
2214 
2215 	case ZTEST_IO_WRITE_ZEROES:
2216 		bzero(data, blocksize);
2217 		(void) ztest_write(zd, object, offset, blocksize, data);
2218 		break;
2219 
2220 	case ZTEST_IO_TRUNCATE:
2221 		(void) ztest_truncate(zd, object, offset, blocksize);
2222 		break;
2223 
2224 	case ZTEST_IO_SETATTR:
2225 		(void) ztest_setattr(zd, object);
2226 		break;
2227 
2228 	case ZTEST_IO_REWRITE:
2229 		(void) rw_rdlock(&ztest_name_lock);
2230 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2231 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2232 		    B_FALSE);
2233 		VERIFY(err == 0 || err == ENOSPC);
2234 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2235 		    ZFS_PROP_COMPRESSION,
2236 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2237 		    B_FALSE);
2238 		VERIFY(err == 0 || err == ENOSPC);
2239 		(void) rw_unlock(&ztest_name_lock);
2240 
2241 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2242 		    DMU_READ_NO_PREFETCH));
2243 
2244 		(void) ztest_write(zd, object, offset, blocksize, data);
2245 		break;
2246 	}
2247 
2248 	(void) rw_unlock(&zd->zd_zilog_lock);
2249 
2250 	umem_free(data, blocksize);
2251 }
2252 
2253 /*
2254  * Initialize an object description template.
2255  */
2256 static void
2257 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2258     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2259 {
2260 	od->od_dir = ZTEST_DIROBJ;
2261 	od->od_object = 0;
2262 
2263 	od->od_crtype = type;
2264 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2265 	od->od_crgen = gen;
2266 
2267 	od->od_type = DMU_OT_NONE;
2268 	od->od_blocksize = 0;
2269 	od->od_gen = 0;
2270 
2271 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2272 	    tag, (int64_t)id, index);
2273 }
2274 
2275 /*
2276  * Lookup or create the objects for a test using the od template.
2277  * If the objects do not all exist, or if 'remove' is specified,
2278  * remove any existing objects and create new ones.  Otherwise,
2279  * use the existing objects.
2280  */
2281 static int
2282 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2283 {
2284 	int count = size / sizeof (*od);
2285 	int rv = 0;
2286 
2287 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2288 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2289 	    (ztest_remove(zd, od, count) != 0 ||
2290 	    ztest_create(zd, od, count) != 0))
2291 		rv = -1;
2292 	zd->zd_od = od;
2293 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2294 
2295 	return (rv);
2296 }
2297 
2298 /* ARGSUSED */
2299 void
2300 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2301 {
2302 	zilog_t *zilog = zd->zd_zilog;
2303 
2304 	(void) rw_rdlock(&zd->zd_zilog_lock);
2305 
2306 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2307 
2308 	/*
2309 	 * Remember the committed values in zd, which is in parent/child
2310 	 * shared memory.  If we die, the next iteration of ztest_run()
2311 	 * will verify that the log really does contain this record.
2312 	 */
2313 	mutex_enter(&zilog->zl_lock);
2314 	ASSERT(zd->zd_shared != NULL);
2315 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2316 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2317 	mutex_exit(&zilog->zl_lock);
2318 
2319 	(void) rw_unlock(&zd->zd_zilog_lock);
2320 }
2321 
2322 /*
2323  * This function is designed to simulate the operations that occur during a
2324  * mount/unmount operation.  We hold the dataset across these operations in an
2325  * attempt to expose any implicit assumptions about ZIL management.
2326  */
2327 /* ARGSUSED */
2328 void
2329 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2330 {
2331 	objset_t *os = zd->zd_os;
2332 
2333 	/*
2334 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2335 	 * updating the zil (i.e. adding in-memory log records) and the
2336 	 * zd_zilog_lock to block any I/O.
2337 	 */
2338 	VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2339 	(void) rw_wrlock(&zd->zd_zilog_lock);
2340 
2341 	/* zfsvfs_teardown() */
2342 	zil_close(zd->zd_zilog);
2343 
2344 	/* zfsvfs_setup() */
2345 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2346 	zil_replay(os, zd, ztest_replay_vector);
2347 
2348 	(void) rw_unlock(&zd->zd_zilog_lock);
2349 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2350 }
2351 
2352 /*
2353  * Verify that we can't destroy an active pool, create an existing pool,
2354  * or create a pool with a bad vdev spec.
2355  */
2356 /* ARGSUSED */
2357 void
2358 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2359 {
2360 	ztest_shared_opts_t *zo = &ztest_opts;
2361 	spa_t *spa;
2362 	nvlist_t *nvroot;
2363 
2364 	/*
2365 	 * Attempt to create using a bad file.
2366 	 */
2367 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2368 	VERIFY3U(ENOENT, ==,
2369 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2370 	nvlist_free(nvroot);
2371 
2372 	/*
2373 	 * Attempt to create using a bad mirror.
2374 	 */
2375 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2376 	VERIFY3U(ENOENT, ==,
2377 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2378 	nvlist_free(nvroot);
2379 
2380 	/*
2381 	 * Attempt to create an existing pool.  It shouldn't matter
2382 	 * what's in the nvroot; we should fail with EEXIST.
2383 	 */
2384 	(void) rw_rdlock(&ztest_name_lock);
2385 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2386 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2387 	nvlist_free(nvroot);
2388 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2389 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2390 	spa_close(spa, FTAG);
2391 
2392 	(void) rw_unlock(&ztest_name_lock);
2393 }
2394 
2395 /* ARGSUSED */
2396 void
2397 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2398 {
2399 	spa_t *spa;
2400 	uint64_t initial_version = SPA_VERSION_INITIAL;
2401 	uint64_t version, newversion;
2402 	nvlist_t *nvroot, *props;
2403 	char *name;
2404 
2405 	VERIFY0(mutex_lock(&ztest_vdev_lock));
2406 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2407 
2408 	/*
2409 	 * Clean up from previous runs.
2410 	 */
2411 	(void) spa_destroy(name);
2412 
2413 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2414 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2415 
2416 	/*
2417 	 * If we're configuring a RAIDZ device then make sure that the
2418 	 * the initial version is capable of supporting that feature.
2419 	 */
2420 	switch (ztest_opts.zo_raidz_parity) {
2421 	case 0:
2422 	case 1:
2423 		initial_version = SPA_VERSION_INITIAL;
2424 		break;
2425 	case 2:
2426 		initial_version = SPA_VERSION_RAIDZ2;
2427 		break;
2428 	case 3:
2429 		initial_version = SPA_VERSION_RAIDZ3;
2430 		break;
2431 	}
2432 
2433 	/*
2434 	 * Create a pool with a spa version that can be upgraded. Pick
2435 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2436 	 */
2437 	do {
2438 		version = ztest_random_spa_version(initial_version);
2439 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2440 
2441 	props = fnvlist_alloc();
2442 	fnvlist_add_uint64(props,
2443 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2444 	VERIFY0(spa_create(name, nvroot, props, NULL));
2445 	fnvlist_free(nvroot);
2446 	fnvlist_free(props);
2447 
2448 	VERIFY0(spa_open(name, &spa, FTAG));
2449 	VERIFY3U(spa_version(spa), ==, version);
2450 	newversion = ztest_random_spa_version(version + 1);
2451 
2452 	if (ztest_opts.zo_verbose >= 4) {
2453 		(void) printf("upgrading spa version from %llu to %llu\n",
2454 		    (u_longlong_t)version, (u_longlong_t)newversion);
2455 	}
2456 
2457 	spa_upgrade(spa, newversion);
2458 	VERIFY3U(spa_version(spa), >, version);
2459 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2460 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2461 	spa_close(spa, FTAG);
2462 
2463 	strfree(name);
2464 	VERIFY0(mutex_unlock(&ztest_vdev_lock));
2465 }
2466 
2467 static vdev_t *
2468 vdev_lookup_by_path(vdev_t *vd, const char *path)
2469 {
2470 	vdev_t *mvd;
2471 
2472 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2473 		return (vd);
2474 
2475 	for (int c = 0; c < vd->vdev_children; c++)
2476 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2477 		    NULL)
2478 			return (mvd);
2479 
2480 	return (NULL);
2481 }
2482 
2483 /*
2484  * Find the first available hole which can be used as a top-level.
2485  */
2486 int
2487 find_vdev_hole(spa_t *spa)
2488 {
2489 	vdev_t *rvd = spa->spa_root_vdev;
2490 	int c;
2491 
2492 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2493 
2494 	for (c = 0; c < rvd->vdev_children; c++) {
2495 		vdev_t *cvd = rvd->vdev_child[c];
2496 
2497 		if (cvd->vdev_ishole)
2498 			break;
2499 	}
2500 	return (c);
2501 }
2502 
2503 /*
2504  * Verify that vdev_add() works as expected.
2505  */
2506 /* ARGSUSED */
2507 void
2508 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2509 {
2510 	ztest_shared_t *zs = ztest_shared;
2511 	spa_t *spa = ztest_spa;
2512 	uint64_t leaves;
2513 	uint64_t guid;
2514 	nvlist_t *nvroot;
2515 	int error;
2516 
2517 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2518 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2519 
2520 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2521 
2522 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2523 
2524 	/*
2525 	 * If we have slogs then remove them 1/4 of the time.
2526 	 */
2527 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2528 		/*
2529 		 * Grab the guid from the head of the log class rotor.
2530 		 */
2531 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2532 
2533 		spa_config_exit(spa, SCL_VDEV, FTAG);
2534 
2535 		/*
2536 		 * We have to grab the zs_name_lock as writer to
2537 		 * prevent a race between removing a slog (dmu_objset_find)
2538 		 * and destroying a dataset. Removing the slog will
2539 		 * grab a reference on the dataset which may cause
2540 		 * dmu_objset_destroy() to fail with EBUSY thus
2541 		 * leaving the dataset in an inconsistent state.
2542 		 */
2543 		VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2544 		error = spa_vdev_remove(spa, guid, B_FALSE);
2545 		VERIFY(rw_unlock(&ztest_name_lock) == 0);
2546 
2547 		if (error && error != EEXIST)
2548 			fatal(0, "spa_vdev_remove() = %d", error);
2549 	} else {
2550 		spa_config_exit(spa, SCL_VDEV, FTAG);
2551 
2552 		/*
2553 		 * Make 1/4 of the devices be log devices.
2554 		 */
2555 		nvroot = make_vdev_root(NULL, NULL, NULL,
2556 		    ztest_opts.zo_vdev_size, 0,
2557 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2558 		    zs->zs_mirrors, 1);
2559 
2560 		error = spa_vdev_add(spa, nvroot);
2561 		nvlist_free(nvroot);
2562 
2563 		if (error == ENOSPC)
2564 			ztest_record_enospc("spa_vdev_add");
2565 		else if (error != 0)
2566 			fatal(0, "spa_vdev_add() = %d", error);
2567 	}
2568 
2569 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2570 }
2571 
2572 /*
2573  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2574  */
2575 /* ARGSUSED */
2576 void
2577 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2578 {
2579 	ztest_shared_t *zs = ztest_shared;
2580 	spa_t *spa = ztest_spa;
2581 	vdev_t *rvd = spa->spa_root_vdev;
2582 	spa_aux_vdev_t *sav;
2583 	char *aux;
2584 	uint64_t guid = 0;
2585 	int error;
2586 
2587 	if (ztest_random(2) == 0) {
2588 		sav = &spa->spa_spares;
2589 		aux = ZPOOL_CONFIG_SPARES;
2590 	} else {
2591 		sav = &spa->spa_l2cache;
2592 		aux = ZPOOL_CONFIG_L2CACHE;
2593 	}
2594 
2595 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2596 
2597 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2598 
2599 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2600 		/*
2601 		 * Pick a random device to remove.
2602 		 */
2603 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2604 	} else {
2605 		/*
2606 		 * Find an unused device we can add.
2607 		 */
2608 		zs->zs_vdev_aux = 0;
2609 		for (;;) {
2610 			char path[MAXPATHLEN];
2611 			int c;
2612 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2613 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2614 			    zs->zs_vdev_aux);
2615 			for (c = 0; c < sav->sav_count; c++)
2616 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2617 				    path) == 0)
2618 					break;
2619 			if (c == sav->sav_count &&
2620 			    vdev_lookup_by_path(rvd, path) == NULL)
2621 				break;
2622 			zs->zs_vdev_aux++;
2623 		}
2624 	}
2625 
2626 	spa_config_exit(spa, SCL_VDEV, FTAG);
2627 
2628 	if (guid == 0) {
2629 		/*
2630 		 * Add a new device.
2631 		 */
2632 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2633 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2634 		error = spa_vdev_add(spa, nvroot);
2635 		if (error != 0)
2636 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2637 		nvlist_free(nvroot);
2638 	} else {
2639 		/*
2640 		 * Remove an existing device.  Sometimes, dirty its
2641 		 * vdev state first to make sure we handle removal
2642 		 * of devices that have pending state changes.
2643 		 */
2644 		if (ztest_random(2) == 0)
2645 			(void) vdev_online(spa, guid, 0, NULL);
2646 
2647 		error = spa_vdev_remove(spa, guid, B_FALSE);
2648 		if (error != 0 && error != EBUSY)
2649 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2650 	}
2651 
2652 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2653 }
2654 
2655 /*
2656  * split a pool if it has mirror tlvdevs
2657  */
2658 /* ARGSUSED */
2659 void
2660 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2661 {
2662 	ztest_shared_t *zs = ztest_shared;
2663 	spa_t *spa = ztest_spa;
2664 	vdev_t *rvd = spa->spa_root_vdev;
2665 	nvlist_t *tree, **child, *config, *split, **schild;
2666 	uint_t c, children, schildren = 0, lastlogid = 0;
2667 	int error = 0;
2668 
2669 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2670 
2671 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2672 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2673 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2674 		return;
2675 	}
2676 
2677 	/* clean up the old pool, if any */
2678 	(void) spa_destroy("splitp");
2679 
2680 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2681 
2682 	/* generate a config from the existing config */
2683 	mutex_enter(&spa->spa_props_lock);
2684 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2685 	    &tree) == 0);
2686 	mutex_exit(&spa->spa_props_lock);
2687 
2688 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2689 	    &children) == 0);
2690 
2691 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2692 	for (c = 0; c < children; c++) {
2693 		vdev_t *tvd = rvd->vdev_child[c];
2694 		nvlist_t **mchild;
2695 		uint_t mchildren;
2696 
2697 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2698 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2699 			    0) == 0);
2700 			VERIFY(nvlist_add_string(schild[schildren],
2701 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2702 			VERIFY(nvlist_add_uint64(schild[schildren],
2703 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2704 			if (lastlogid == 0)
2705 				lastlogid = schildren;
2706 			++schildren;
2707 			continue;
2708 		}
2709 		lastlogid = 0;
2710 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2711 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2712 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2713 	}
2714 
2715 	/* OK, create a config that can be used to split */
2716 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2717 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2718 	    VDEV_TYPE_ROOT) == 0);
2719 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2720 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2721 
2722 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2723 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2724 
2725 	for (c = 0; c < schildren; c++)
2726 		nvlist_free(schild[c]);
2727 	free(schild);
2728 	nvlist_free(split);
2729 
2730 	spa_config_exit(spa, SCL_VDEV, FTAG);
2731 
2732 	(void) rw_wrlock(&ztest_name_lock);
2733 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2734 	(void) rw_unlock(&ztest_name_lock);
2735 
2736 	nvlist_free(config);
2737 
2738 	if (error == 0) {
2739 		(void) printf("successful split - results:\n");
2740 		mutex_enter(&spa_namespace_lock);
2741 		show_pool_stats(spa);
2742 		show_pool_stats(spa_lookup("splitp"));
2743 		mutex_exit(&spa_namespace_lock);
2744 		++zs->zs_splits;
2745 		--zs->zs_mirrors;
2746 	}
2747 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2748 
2749 }
2750 
2751 /*
2752  * Verify that we can attach and detach devices.
2753  */
2754 /* ARGSUSED */
2755 void
2756 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2757 {
2758 	ztest_shared_t *zs = ztest_shared;
2759 	spa_t *spa = ztest_spa;
2760 	spa_aux_vdev_t *sav = &spa->spa_spares;
2761 	vdev_t *rvd = spa->spa_root_vdev;
2762 	vdev_t *oldvd, *newvd, *pvd;
2763 	nvlist_t *root;
2764 	uint64_t leaves;
2765 	uint64_t leaf, top;
2766 	uint64_t ashift = ztest_get_ashift();
2767 	uint64_t oldguid, pguid;
2768 	uint64_t oldsize, newsize;
2769 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2770 	int replacing;
2771 	int oldvd_has_siblings = B_FALSE;
2772 	int newvd_is_spare = B_FALSE;
2773 	int oldvd_is_log;
2774 	int error, expected_error;
2775 
2776 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2777 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2778 
2779 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2780 
2781 	/*
2782 	 * Decide whether to do an attach or a replace.
2783 	 */
2784 	replacing = ztest_random(2);
2785 
2786 	/*
2787 	 * Pick a random top-level vdev.
2788 	 */
2789 	top = ztest_random_vdev_top(spa, B_TRUE);
2790 
2791 	/*
2792 	 * Pick a random leaf within it.
2793 	 */
2794 	leaf = ztest_random(leaves);
2795 
2796 	/*
2797 	 * Locate this vdev.
2798 	 */
2799 	oldvd = rvd->vdev_child[top];
2800 	if (zs->zs_mirrors >= 1) {
2801 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2802 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2803 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2804 	}
2805 	if (ztest_opts.zo_raidz > 1) {
2806 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2807 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2808 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2809 	}
2810 
2811 	/*
2812 	 * If we're already doing an attach or replace, oldvd may be a
2813 	 * mirror vdev -- in which case, pick a random child.
2814 	 */
2815 	while (oldvd->vdev_children != 0) {
2816 		oldvd_has_siblings = B_TRUE;
2817 		ASSERT(oldvd->vdev_children >= 2);
2818 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2819 	}
2820 
2821 	oldguid = oldvd->vdev_guid;
2822 	oldsize = vdev_get_min_asize(oldvd);
2823 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2824 	(void) strcpy(oldpath, oldvd->vdev_path);
2825 	pvd = oldvd->vdev_parent;
2826 	pguid = pvd->vdev_guid;
2827 
2828 	/*
2829 	 * If oldvd has siblings, then half of the time, detach it.
2830 	 */
2831 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2832 		spa_config_exit(spa, SCL_VDEV, FTAG);
2833 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2834 		if (error != 0 && error != ENODEV && error != EBUSY &&
2835 		    error != ENOTSUP)
2836 			fatal(0, "detach (%s) returned %d", oldpath, error);
2837 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2838 		return;
2839 	}
2840 
2841 	/*
2842 	 * For the new vdev, choose with equal probability between the two
2843 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2844 	 */
2845 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2846 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2847 		newvd_is_spare = B_TRUE;
2848 		(void) strcpy(newpath, newvd->vdev_path);
2849 	} else {
2850 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2851 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2852 		    top * leaves + leaf);
2853 		if (ztest_random(2) == 0)
2854 			newpath[strlen(newpath) - 1] = 'b';
2855 		newvd = vdev_lookup_by_path(rvd, newpath);
2856 	}
2857 
2858 	if (newvd) {
2859 		newsize = vdev_get_min_asize(newvd);
2860 	} else {
2861 		/*
2862 		 * Make newsize a little bigger or smaller than oldsize.
2863 		 * If it's smaller, the attach should fail.
2864 		 * If it's larger, and we're doing a replace,
2865 		 * we should get dynamic LUN growth when we're done.
2866 		 */
2867 		newsize = 10 * oldsize / (9 + ztest_random(3));
2868 	}
2869 
2870 	/*
2871 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2872 	 * unless it's a replace; in that case any non-replacing parent is OK.
2873 	 *
2874 	 * If newvd is already part of the pool, it should fail with EBUSY.
2875 	 *
2876 	 * If newvd is too small, it should fail with EOVERFLOW.
2877 	 */
2878 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2879 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2880 	    pvd->vdev_ops == &vdev_replacing_ops ||
2881 	    pvd->vdev_ops == &vdev_spare_ops))
2882 		expected_error = ENOTSUP;
2883 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2884 		expected_error = ENOTSUP;
2885 	else if (newvd == oldvd)
2886 		expected_error = replacing ? 0 : EBUSY;
2887 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2888 		expected_error = EBUSY;
2889 	else if (newsize < oldsize)
2890 		expected_error = EOVERFLOW;
2891 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2892 		expected_error = EDOM;
2893 	else
2894 		expected_error = 0;
2895 
2896 	spa_config_exit(spa, SCL_VDEV, FTAG);
2897 
2898 	/*
2899 	 * Build the nvlist describing newpath.
2900 	 */
2901 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2902 	    ashift, 0, 0, 0, 1);
2903 
2904 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2905 
2906 	nvlist_free(root);
2907 
2908 	/*
2909 	 * If our parent was the replacing vdev, but the replace completed,
2910 	 * then instead of failing with ENOTSUP we may either succeed,
2911 	 * fail with ENODEV, or fail with EOVERFLOW.
2912 	 */
2913 	if (expected_error == ENOTSUP &&
2914 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2915 		expected_error = error;
2916 
2917 	/*
2918 	 * If someone grew the LUN, the replacement may be too small.
2919 	 */
2920 	if (error == EOVERFLOW || error == EBUSY)
2921 		expected_error = error;
2922 
2923 	/* XXX workaround 6690467 */
2924 	if (error != expected_error && expected_error != EBUSY) {
2925 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2926 		    "returned %d, expected %d",
2927 		    oldpath, oldsize, newpath,
2928 		    newsize, replacing, error, expected_error);
2929 	}
2930 
2931 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2932 }
2933 
2934 /*
2935  * Callback function which expands the physical size of the vdev.
2936  */
2937 vdev_t *
2938 grow_vdev(vdev_t *vd, void *arg)
2939 {
2940 	spa_t *spa = vd->vdev_spa;
2941 	size_t *newsize = arg;
2942 	size_t fsize;
2943 	int fd;
2944 
2945 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2946 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2947 
2948 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2949 		return (vd);
2950 
2951 	fsize = lseek(fd, 0, SEEK_END);
2952 	(void) ftruncate(fd, *newsize);
2953 
2954 	if (ztest_opts.zo_verbose >= 6) {
2955 		(void) printf("%s grew from %lu to %lu bytes\n",
2956 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2957 	}
2958 	(void) close(fd);
2959 	return (NULL);
2960 }
2961 
2962 /*
2963  * Callback function which expands a given vdev by calling vdev_online().
2964  */
2965 /* ARGSUSED */
2966 vdev_t *
2967 online_vdev(vdev_t *vd, void *arg)
2968 {
2969 	spa_t *spa = vd->vdev_spa;
2970 	vdev_t *tvd = vd->vdev_top;
2971 	uint64_t guid = vd->vdev_guid;
2972 	uint64_t generation = spa->spa_config_generation + 1;
2973 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2974 	int error;
2975 
2976 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2977 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2978 
2979 	/* Calling vdev_online will initialize the new metaslabs */
2980 	spa_config_exit(spa, SCL_STATE, spa);
2981 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2982 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2983 
2984 	/*
2985 	 * If vdev_online returned an error or the underlying vdev_open
2986 	 * failed then we abort the expand. The only way to know that
2987 	 * vdev_open fails is by checking the returned newstate.
2988 	 */
2989 	if (error || newstate != VDEV_STATE_HEALTHY) {
2990 		if (ztest_opts.zo_verbose >= 5) {
2991 			(void) printf("Unable to expand vdev, state %llu, "
2992 			    "error %d\n", (u_longlong_t)newstate, error);
2993 		}
2994 		return (vd);
2995 	}
2996 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2997 
2998 	/*
2999 	 * Since we dropped the lock we need to ensure that we're
3000 	 * still talking to the original vdev. It's possible this
3001 	 * vdev may have been detached/replaced while we were
3002 	 * trying to online it.
3003 	 */
3004 	if (generation != spa->spa_config_generation) {
3005 		if (ztest_opts.zo_verbose >= 5) {
3006 			(void) printf("vdev configuration has changed, "
3007 			    "guid %llu, state %llu, expected gen %llu, "
3008 			    "got gen %llu\n",
3009 			    (u_longlong_t)guid,
3010 			    (u_longlong_t)tvd->vdev_state,
3011 			    (u_longlong_t)generation,
3012 			    (u_longlong_t)spa->spa_config_generation);
3013 		}
3014 		return (vd);
3015 	}
3016 	return (NULL);
3017 }
3018 
3019 /*
3020  * Traverse the vdev tree calling the supplied function.
3021  * We continue to walk the tree until we either have walked all
3022  * children or we receive a non-NULL return from the callback.
3023  * If a NULL callback is passed, then we just return back the first
3024  * leaf vdev we encounter.
3025  */
3026 vdev_t *
3027 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3028 {
3029 	if (vd->vdev_ops->vdev_op_leaf) {
3030 		if (func == NULL)
3031 			return (vd);
3032 		else
3033 			return (func(vd, arg));
3034 	}
3035 
3036 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3037 		vdev_t *cvd = vd->vdev_child[c];
3038 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3039 			return (cvd);
3040 	}
3041 	return (NULL);
3042 }
3043 
3044 /*
3045  * Verify that dynamic LUN growth works as expected.
3046  */
3047 /* ARGSUSED */
3048 void
3049 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3050 {
3051 	spa_t *spa = ztest_spa;
3052 	vdev_t *vd, *tvd;
3053 	metaslab_class_t *mc;
3054 	metaslab_group_t *mg;
3055 	size_t psize, newsize;
3056 	uint64_t top;
3057 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3058 
3059 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3060 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3061 
3062 	top = ztest_random_vdev_top(spa, B_TRUE);
3063 
3064 	tvd = spa->spa_root_vdev->vdev_child[top];
3065 	mg = tvd->vdev_mg;
3066 	mc = mg->mg_class;
3067 	old_ms_count = tvd->vdev_ms_count;
3068 	old_class_space = metaslab_class_get_space(mc);
3069 
3070 	/*
3071 	 * Determine the size of the first leaf vdev associated with
3072 	 * our top-level device.
3073 	 */
3074 	vd = vdev_walk_tree(tvd, NULL, NULL);
3075 	ASSERT3P(vd, !=, NULL);
3076 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3077 
3078 	psize = vd->vdev_psize;
3079 
3080 	/*
3081 	 * We only try to expand the vdev if it's healthy, less than 4x its
3082 	 * original size, and it has a valid psize.
3083 	 */
3084 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3085 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3086 		spa_config_exit(spa, SCL_STATE, spa);
3087 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3088 		return;
3089 	}
3090 	ASSERT(psize > 0);
3091 	newsize = psize + psize / 8;
3092 	ASSERT3U(newsize, >, psize);
3093 
3094 	if (ztest_opts.zo_verbose >= 6) {
3095 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3096 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3097 	}
3098 
3099 	/*
3100 	 * Growing the vdev is a two step process:
3101 	 *	1). expand the physical size (i.e. relabel)
3102 	 *	2). online the vdev to create the new metaslabs
3103 	 */
3104 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3105 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3106 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3107 		if (ztest_opts.zo_verbose >= 5) {
3108 			(void) printf("Could not expand LUN because "
3109 			    "the vdev configuration changed.\n");
3110 		}
3111 		spa_config_exit(spa, SCL_STATE, spa);
3112 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3113 		return;
3114 	}
3115 
3116 	spa_config_exit(spa, SCL_STATE, spa);
3117 
3118 	/*
3119 	 * Expanding the LUN will update the config asynchronously,
3120 	 * thus we must wait for the async thread to complete any
3121 	 * pending tasks before proceeding.
3122 	 */
3123 	for (;;) {
3124 		boolean_t done;
3125 		mutex_enter(&spa->spa_async_lock);
3126 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3127 		mutex_exit(&spa->spa_async_lock);
3128 		if (done)
3129 			break;
3130 		txg_wait_synced(spa_get_dsl(spa), 0);
3131 		(void) poll(NULL, 0, 100);
3132 	}
3133 
3134 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3135 
3136 	tvd = spa->spa_root_vdev->vdev_child[top];
3137 	new_ms_count = tvd->vdev_ms_count;
3138 	new_class_space = metaslab_class_get_space(mc);
3139 
3140 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3141 		if (ztest_opts.zo_verbose >= 5) {
3142 			(void) printf("Could not verify LUN expansion due to "
3143 			    "intervening vdev offline or remove.\n");
3144 		}
3145 		spa_config_exit(spa, SCL_STATE, spa);
3146 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3147 		return;
3148 	}
3149 
3150 	/*
3151 	 * Make sure we were able to grow the vdev.
3152 	 */
3153 	if (new_ms_count <= old_ms_count)
3154 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3155 		    old_ms_count, new_ms_count);
3156 
3157 	/*
3158 	 * Make sure we were able to grow the pool.
3159 	 */
3160 	if (new_class_space <= old_class_space)
3161 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3162 		    old_class_space, new_class_space);
3163 
3164 	if (ztest_opts.zo_verbose >= 5) {
3165 		char oldnumbuf[6], newnumbuf[6];
3166 
3167 		nicenum(old_class_space, oldnumbuf);
3168 		nicenum(new_class_space, newnumbuf);
3169 		(void) printf("%s grew from %s to %s\n",
3170 		    spa->spa_name, oldnumbuf, newnumbuf);
3171 	}
3172 
3173 	spa_config_exit(spa, SCL_STATE, spa);
3174 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3175 }
3176 
3177 /*
3178  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3179  */
3180 /* ARGSUSED */
3181 static void
3182 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3183 {
3184 	/*
3185 	 * Create the objects common to all ztest datasets.
3186 	 */
3187 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3188 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3189 }
3190 
3191 static int
3192 ztest_dataset_create(char *dsname)
3193 {
3194 	uint64_t zilset = ztest_random(100);
3195 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3196 	    ztest_objset_create_cb, NULL);
3197 
3198 	if (err || zilset < 80)
3199 		return (err);
3200 
3201 	if (ztest_opts.zo_verbose >= 6)
3202 		(void) printf("Setting dataset %s to sync always\n", dsname);
3203 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3204 	    ZFS_SYNC_ALWAYS, B_FALSE));
3205 }
3206 
3207 /* ARGSUSED */
3208 static int
3209 ztest_objset_destroy_cb(const char *name, void *arg)
3210 {
3211 	objset_t *os;
3212 	dmu_object_info_t doi;
3213 	int error;
3214 
3215 	/*
3216 	 * Verify that the dataset contains a directory object.
3217 	 */
3218 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3219 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3220 	if (error != ENOENT) {
3221 		/* We could have crashed in the middle of destroying it */
3222 		ASSERT0(error);
3223 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3224 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3225 	}
3226 	dmu_objset_disown(os, FTAG);
3227 
3228 	/*
3229 	 * Destroy the dataset.
3230 	 */
3231 	if (strchr(name, '@') != NULL) {
3232 		VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3233 	} else {
3234 		VERIFY0(dsl_destroy_head(name));
3235 	}
3236 	return (0);
3237 }
3238 
3239 static boolean_t
3240 ztest_snapshot_create(char *osname, uint64_t id)
3241 {
3242 	char snapname[MAXNAMELEN];
3243 	int error;
3244 
3245 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3246 
3247 	error = dmu_objset_snapshot_one(osname, snapname);
3248 	if (error == ENOSPC) {
3249 		ztest_record_enospc(FTAG);
3250 		return (B_FALSE);
3251 	}
3252 	if (error != 0 && error != EEXIST) {
3253 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3254 		    snapname, error);
3255 	}
3256 	return (B_TRUE);
3257 }
3258 
3259 static boolean_t
3260 ztest_snapshot_destroy(char *osname, uint64_t id)
3261 {
3262 	char snapname[MAXNAMELEN];
3263 	int error;
3264 
3265 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3266 	    (u_longlong_t)id);
3267 
3268 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3269 	if (error != 0 && error != ENOENT)
3270 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3271 	return (B_TRUE);
3272 }
3273 
3274 /* ARGSUSED */
3275 void
3276 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3277 {
3278 	ztest_ds_t zdtmp;
3279 	int iters;
3280 	int error;
3281 	objset_t *os, *os2;
3282 	char name[MAXNAMELEN];
3283 	zilog_t *zilog;
3284 
3285 	(void) rw_rdlock(&ztest_name_lock);
3286 
3287 	(void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3288 	    ztest_opts.zo_pool, (u_longlong_t)id);
3289 
3290 	/*
3291 	 * If this dataset exists from a previous run, process its replay log
3292 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3293 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3294 	 */
3295 	if (ztest_random(2) == 0 &&
3296 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3297 		ztest_zd_init(&zdtmp, NULL, os);
3298 		zil_replay(os, &zdtmp, ztest_replay_vector);
3299 		ztest_zd_fini(&zdtmp);
3300 		dmu_objset_disown(os, FTAG);
3301 	}
3302 
3303 	/*
3304 	 * There may be an old instance of the dataset we're about to
3305 	 * create lying around from a previous run.  If so, destroy it
3306 	 * and all of its snapshots.
3307 	 */
3308 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3309 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3310 
3311 	/*
3312 	 * Verify that the destroyed dataset is no longer in the namespace.
3313 	 */
3314 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3315 	    FTAG, &os));
3316 
3317 	/*
3318 	 * Verify that we can create a new dataset.
3319 	 */
3320 	error = ztest_dataset_create(name);
3321 	if (error) {
3322 		if (error == ENOSPC) {
3323 			ztest_record_enospc(FTAG);
3324 			(void) rw_unlock(&ztest_name_lock);
3325 			return;
3326 		}
3327 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3328 	}
3329 
3330 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3331 
3332 	ztest_zd_init(&zdtmp, NULL, os);
3333 
3334 	/*
3335 	 * Open the intent log for it.
3336 	 */
3337 	zilog = zil_open(os, ztest_get_data);
3338 
3339 	/*
3340 	 * Put some objects in there, do a little I/O to them,
3341 	 * and randomly take a couple of snapshots along the way.
3342 	 */
3343 	iters = ztest_random(5);
3344 	for (int i = 0; i < iters; i++) {
3345 		ztest_dmu_object_alloc_free(&zdtmp, id);
3346 		if (ztest_random(iters) == 0)
3347 			(void) ztest_snapshot_create(name, i);
3348 	}
3349 
3350 	/*
3351 	 * Verify that we cannot create an existing dataset.
3352 	 */
3353 	VERIFY3U(EEXIST, ==,
3354 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3355 
3356 	/*
3357 	 * Verify that we can hold an objset that is also owned.
3358 	 */
3359 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3360 	dmu_objset_rele(os2, FTAG);
3361 
3362 	/*
3363 	 * Verify that we cannot own an objset that is already owned.
3364 	 */
3365 	VERIFY3U(EBUSY, ==,
3366 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3367 
3368 	zil_close(zilog);
3369 	dmu_objset_disown(os, FTAG);
3370 	ztest_zd_fini(&zdtmp);
3371 
3372 	(void) rw_unlock(&ztest_name_lock);
3373 }
3374 
3375 /*
3376  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3377  */
3378 void
3379 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3380 {
3381 	(void) rw_rdlock(&ztest_name_lock);
3382 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3383 	(void) ztest_snapshot_create(zd->zd_name, id);
3384 	(void) rw_unlock(&ztest_name_lock);
3385 }
3386 
3387 /*
3388  * Cleanup non-standard snapshots and clones.
3389  */
3390 void
3391 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3392 {
3393 	char snap1name[MAXNAMELEN];
3394 	char clone1name[MAXNAMELEN];
3395 	char snap2name[MAXNAMELEN];
3396 	char clone2name[MAXNAMELEN];
3397 	char snap3name[MAXNAMELEN];
3398 	int error;
3399 
3400 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3401 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3402 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3403 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3404 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3405 
3406 	error = dsl_destroy_head(clone2name);
3407 	if (error && error != ENOENT)
3408 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3409 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3410 	if (error && error != ENOENT)
3411 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3412 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3413 	if (error && error != ENOENT)
3414 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3415 	error = dsl_destroy_head(clone1name);
3416 	if (error && error != ENOENT)
3417 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3418 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3419 	if (error && error != ENOENT)
3420 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3421 }
3422 
3423 /*
3424  * Verify dsl_dataset_promote handles EBUSY
3425  */
3426 void
3427 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3428 {
3429 	objset_t *os;
3430 	char snap1name[MAXNAMELEN];
3431 	char clone1name[MAXNAMELEN];
3432 	char snap2name[MAXNAMELEN];
3433 	char clone2name[MAXNAMELEN];
3434 	char snap3name[MAXNAMELEN];
3435 	char *osname = zd->zd_name;
3436 	int error;
3437 
3438 	(void) rw_rdlock(&ztest_name_lock);
3439 
3440 	ztest_dsl_dataset_cleanup(osname, id);
3441 
3442 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3443 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3444 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3445 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3446 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3447 
3448 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3449 	if (error && error != EEXIST) {
3450 		if (error == ENOSPC) {
3451 			ztest_record_enospc(FTAG);
3452 			goto out;
3453 		}
3454 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3455 	}
3456 
3457 	error = dmu_objset_clone(clone1name, snap1name);
3458 	if (error) {
3459 		if (error == ENOSPC) {
3460 			ztest_record_enospc(FTAG);
3461 			goto out;
3462 		}
3463 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3464 	}
3465 
3466 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3467 	if (error && error != EEXIST) {
3468 		if (error == ENOSPC) {
3469 			ztest_record_enospc(FTAG);
3470 			goto out;
3471 		}
3472 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3473 	}
3474 
3475 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3476 	if (error && error != EEXIST) {
3477 		if (error == ENOSPC) {
3478 			ztest_record_enospc(FTAG);
3479 			goto out;
3480 		}
3481 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3482 	}
3483 
3484 	error = dmu_objset_clone(clone2name, snap3name);
3485 	if (error) {
3486 		if (error == ENOSPC) {
3487 			ztest_record_enospc(FTAG);
3488 			goto out;
3489 		}
3490 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3491 	}
3492 
3493 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3494 	if (error)
3495 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3496 	error = dsl_dataset_promote(clone2name, NULL);
3497 	if (error == ENOSPC) {
3498 		dmu_objset_disown(os, FTAG);
3499 		ztest_record_enospc(FTAG);
3500 		goto out;
3501 	}
3502 	if (error != EBUSY)
3503 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3504 		    error);
3505 	dmu_objset_disown(os, FTAG);
3506 
3507 out:
3508 	ztest_dsl_dataset_cleanup(osname, id);
3509 
3510 	(void) rw_unlock(&ztest_name_lock);
3511 }
3512 
3513 /*
3514  * Verify that dmu_object_{alloc,free} work as expected.
3515  */
3516 void
3517 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3518 {
3519 	ztest_od_t od[4];
3520 	int batchsize = sizeof (od) / sizeof (od[0]);
3521 
3522 	for (int b = 0; b < batchsize; b++)
3523 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3524 
3525 	/*
3526 	 * Destroy the previous batch of objects, create a new batch,
3527 	 * and do some I/O on the new objects.
3528 	 */
3529 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3530 		return;
3531 
3532 	while (ztest_random(4 * batchsize) != 0)
3533 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3534 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3535 }
3536 
3537 /*
3538  * Verify that dmu_{read,write} work as expected.
3539  */
3540 void
3541 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3542 {
3543 	objset_t *os = zd->zd_os;
3544 	ztest_od_t od[2];
3545 	dmu_tx_t *tx;
3546 	int i, freeit, error;
3547 	uint64_t n, s, txg;
3548 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3549 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3550 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3551 	uint64_t regions = 997;
3552 	uint64_t stride = 123456789ULL;
3553 	uint64_t width = 40;
3554 	int free_percent = 5;
3555 
3556 	/*
3557 	 * This test uses two objects, packobj and bigobj, that are always
3558 	 * updated together (i.e. in the same tx) so that their contents are
3559 	 * in sync and can be compared.  Their contents relate to each other
3560 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3561 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3562 	 * for any index n, there are three bufwads that should be identical:
3563 	 *
3564 	 *	packobj, at offset n * sizeof (bufwad_t)
3565 	 *	bigobj, at the head of the nth chunk
3566 	 *	bigobj, at the tail of the nth chunk
3567 	 *
3568 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3569 	 * and it doesn't have any relation to the object blocksize.
3570 	 * The only requirement is that it can hold at least two bufwads.
3571 	 *
3572 	 * Normally, we write the bufwad to each of these locations.
3573 	 * However, free_percent of the time we instead write zeroes to
3574 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3575 	 * bigobj to packobj, we can verify that the DMU is correctly
3576 	 * tracking which parts of an object are allocated and free,
3577 	 * and that the contents of the allocated blocks are correct.
3578 	 */
3579 
3580 	/*
3581 	 * Read the directory info.  If it's the first time, set things up.
3582 	 */
3583 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3584 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3585 
3586 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3587 		return;
3588 
3589 	bigobj = od[0].od_object;
3590 	packobj = od[1].od_object;
3591 	chunksize = od[0].od_gen;
3592 	ASSERT(chunksize == od[1].od_gen);
3593 
3594 	/*
3595 	 * Prefetch a random chunk of the big object.
3596 	 * Our aim here is to get some async reads in flight
3597 	 * for blocks that we may free below; the DMU should
3598 	 * handle this race correctly.
3599 	 */
3600 	n = ztest_random(regions) * stride + ztest_random(width);
3601 	s = 1 + ztest_random(2 * width - 1);
3602 	dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3603 
3604 	/*
3605 	 * Pick a random index and compute the offsets into packobj and bigobj.
3606 	 */
3607 	n = ztest_random(regions) * stride + ztest_random(width);
3608 	s = 1 + ztest_random(width - 1);
3609 
3610 	packoff = n * sizeof (bufwad_t);
3611 	packsize = s * sizeof (bufwad_t);
3612 
3613 	bigoff = n * chunksize;
3614 	bigsize = s * chunksize;
3615 
3616 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3617 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3618 
3619 	/*
3620 	 * free_percent of the time, free a range of bigobj rather than
3621 	 * overwriting it.
3622 	 */
3623 	freeit = (ztest_random(100) < free_percent);
3624 
3625 	/*
3626 	 * Read the current contents of our objects.
3627 	 */
3628 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3629 	    DMU_READ_PREFETCH);
3630 	ASSERT0(error);
3631 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3632 	    DMU_READ_PREFETCH);
3633 	ASSERT0(error);
3634 
3635 	/*
3636 	 * Get a tx for the mods to both packobj and bigobj.
3637 	 */
3638 	tx = dmu_tx_create(os);
3639 
3640 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3641 
3642 	if (freeit)
3643 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3644 	else
3645 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3646 
3647 	/* This accounts for setting the checksum/compression. */
3648 	dmu_tx_hold_bonus(tx, bigobj);
3649 
3650 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3651 	if (txg == 0) {
3652 		umem_free(packbuf, packsize);
3653 		umem_free(bigbuf, bigsize);
3654 		return;
3655 	}
3656 
3657 	enum zio_checksum cksum;
3658 	do {
3659 		cksum = (enum zio_checksum)
3660 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3661 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3662 	dmu_object_set_checksum(os, bigobj, cksum, tx);
3663 
3664 	enum zio_compress comp;
3665 	do {
3666 		comp = (enum zio_compress)
3667 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3668 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3669 	dmu_object_set_compress(os, bigobj, comp, tx);
3670 
3671 	/*
3672 	 * For each index from n to n + s, verify that the existing bufwad
3673 	 * in packobj matches the bufwads at the head and tail of the
3674 	 * corresponding chunk in bigobj.  Then update all three bufwads
3675 	 * with the new values we want to write out.
3676 	 */
3677 	for (i = 0; i < s; i++) {
3678 		/* LINTED */
3679 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3680 		/* LINTED */
3681 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3682 		/* LINTED */
3683 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3684 
3685 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3686 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3687 
3688 		if (pack->bw_txg > txg)
3689 			fatal(0, "future leak: got %llx, open txg is %llx",
3690 			    pack->bw_txg, txg);
3691 
3692 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3693 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3694 			    pack->bw_index, n, i);
3695 
3696 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3697 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3698 
3699 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3700 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3701 
3702 		if (freeit) {
3703 			bzero(pack, sizeof (bufwad_t));
3704 		} else {
3705 			pack->bw_index = n + i;
3706 			pack->bw_txg = txg;
3707 			pack->bw_data = 1 + ztest_random(-2ULL);
3708 		}
3709 		*bigH = *pack;
3710 		*bigT = *pack;
3711 	}
3712 
3713 	/*
3714 	 * We've verified all the old bufwads, and made new ones.
3715 	 * Now write them out.
3716 	 */
3717 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3718 
3719 	if (freeit) {
3720 		if (ztest_opts.zo_verbose >= 7) {
3721 			(void) printf("freeing offset %llx size %llx"
3722 			    " txg %llx\n",
3723 			    (u_longlong_t)bigoff,
3724 			    (u_longlong_t)bigsize,
3725 			    (u_longlong_t)txg);
3726 		}
3727 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3728 	} else {
3729 		if (ztest_opts.zo_verbose >= 7) {
3730 			(void) printf("writing offset %llx size %llx"
3731 			    " txg %llx\n",
3732 			    (u_longlong_t)bigoff,
3733 			    (u_longlong_t)bigsize,
3734 			    (u_longlong_t)txg);
3735 		}
3736 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3737 	}
3738 
3739 	dmu_tx_commit(tx);
3740 
3741 	/*
3742 	 * Sanity check the stuff we just wrote.
3743 	 */
3744 	{
3745 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3746 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3747 
3748 		VERIFY(0 == dmu_read(os, packobj, packoff,
3749 		    packsize, packcheck, DMU_READ_PREFETCH));
3750 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3751 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3752 
3753 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3754 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3755 
3756 		umem_free(packcheck, packsize);
3757 		umem_free(bigcheck, bigsize);
3758 	}
3759 
3760 	umem_free(packbuf, packsize);
3761 	umem_free(bigbuf, bigsize);
3762 }
3763 
3764 void
3765 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3766     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3767 {
3768 	uint64_t i;
3769 	bufwad_t *pack;
3770 	bufwad_t *bigH;
3771 	bufwad_t *bigT;
3772 
3773 	/*
3774 	 * For each index from n to n + s, verify that the existing bufwad
3775 	 * in packobj matches the bufwads at the head and tail of the
3776 	 * corresponding chunk in bigobj.  Then update all three bufwads
3777 	 * with the new values we want to write out.
3778 	 */
3779 	for (i = 0; i < s; i++) {
3780 		/* LINTED */
3781 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3782 		/* LINTED */
3783 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3784 		/* LINTED */
3785 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3786 
3787 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3788 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3789 
3790 		if (pack->bw_txg > txg)
3791 			fatal(0, "future leak: got %llx, open txg is %llx",
3792 			    pack->bw_txg, txg);
3793 
3794 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3795 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3796 			    pack->bw_index, n, i);
3797 
3798 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3799 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3800 
3801 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3802 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3803 
3804 		pack->bw_index = n + i;
3805 		pack->bw_txg = txg;
3806 		pack->bw_data = 1 + ztest_random(-2ULL);
3807 
3808 		*bigH = *pack;
3809 		*bigT = *pack;
3810 	}
3811 }
3812 
3813 void
3814 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3815 {
3816 	objset_t *os = zd->zd_os;
3817 	ztest_od_t od[2];
3818 	dmu_tx_t *tx;
3819 	uint64_t i;
3820 	int error;
3821 	uint64_t n, s, txg;
3822 	bufwad_t *packbuf, *bigbuf;
3823 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3824 	uint64_t blocksize = ztest_random_blocksize();
3825 	uint64_t chunksize = blocksize;
3826 	uint64_t regions = 997;
3827 	uint64_t stride = 123456789ULL;
3828 	uint64_t width = 9;
3829 	dmu_buf_t *bonus_db;
3830 	arc_buf_t **bigbuf_arcbufs;
3831 	dmu_object_info_t doi;
3832 
3833 	/*
3834 	 * This test uses two objects, packobj and bigobj, that are always
3835 	 * updated together (i.e. in the same tx) so that their contents are
3836 	 * in sync and can be compared.  Their contents relate to each other
3837 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3838 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3839 	 * for any index n, there are three bufwads that should be identical:
3840 	 *
3841 	 *	packobj, at offset n * sizeof (bufwad_t)
3842 	 *	bigobj, at the head of the nth chunk
3843 	 *	bigobj, at the tail of the nth chunk
3844 	 *
3845 	 * The chunk size is set equal to bigobj block size so that
3846 	 * dmu_assign_arcbuf() can be tested for object updates.
3847 	 */
3848 
3849 	/*
3850 	 * Read the directory info.  If it's the first time, set things up.
3851 	 */
3852 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3853 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3854 
3855 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3856 		return;
3857 
3858 	bigobj = od[0].od_object;
3859 	packobj = od[1].od_object;
3860 	blocksize = od[0].od_blocksize;
3861 	chunksize = blocksize;
3862 	ASSERT(chunksize == od[1].od_gen);
3863 
3864 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3865 	VERIFY(ISP2(doi.doi_data_block_size));
3866 	VERIFY(chunksize == doi.doi_data_block_size);
3867 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3868 
3869 	/*
3870 	 * Pick a random index and compute the offsets into packobj and bigobj.
3871 	 */
3872 	n = ztest_random(regions) * stride + ztest_random(width);
3873 	s = 1 + ztest_random(width - 1);
3874 
3875 	packoff = n * sizeof (bufwad_t);
3876 	packsize = s * sizeof (bufwad_t);
3877 
3878 	bigoff = n * chunksize;
3879 	bigsize = s * chunksize;
3880 
3881 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3882 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3883 
3884 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3885 
3886 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3887 
3888 	/*
3889 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3890 	 * Iteration 1 test zcopy to already referenced dbufs.
3891 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3892 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3893 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3894 	 * Iteration 5 test zcopy when it can't be done.
3895 	 * Iteration 6 one more zcopy write.
3896 	 */
3897 	for (i = 0; i < 7; i++) {
3898 		uint64_t j;
3899 		uint64_t off;
3900 
3901 		/*
3902 		 * In iteration 5 (i == 5) use arcbufs
3903 		 * that don't match bigobj blksz to test
3904 		 * dmu_assign_arcbuf() when it can't directly
3905 		 * assign an arcbuf to a dbuf.
3906 		 */
3907 		for (j = 0; j < s; j++) {
3908 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3909 				bigbuf_arcbufs[j] =
3910 				    dmu_request_arcbuf(bonus_db, chunksize);
3911 			} else {
3912 				bigbuf_arcbufs[2 * j] =
3913 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3914 				bigbuf_arcbufs[2 * j + 1] =
3915 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3916 			}
3917 		}
3918 
3919 		/*
3920 		 * Get a tx for the mods to both packobj and bigobj.
3921 		 */
3922 		tx = dmu_tx_create(os);
3923 
3924 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3925 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3926 
3927 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3928 		if (txg == 0) {
3929 			umem_free(packbuf, packsize);
3930 			umem_free(bigbuf, bigsize);
3931 			for (j = 0; j < s; j++) {
3932 				if (i != 5 ||
3933 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
3934 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3935 				} else {
3936 					dmu_return_arcbuf(
3937 					    bigbuf_arcbufs[2 * j]);
3938 					dmu_return_arcbuf(
3939 					    bigbuf_arcbufs[2 * j + 1]);
3940 				}
3941 			}
3942 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3943 			dmu_buf_rele(bonus_db, FTAG);
3944 			return;
3945 		}
3946 
3947 		/*
3948 		 * 50% of the time don't read objects in the 1st iteration to
3949 		 * test dmu_assign_arcbuf() for the case when there're no
3950 		 * existing dbufs for the specified offsets.
3951 		 */
3952 		if (i != 0 || ztest_random(2) != 0) {
3953 			error = dmu_read(os, packobj, packoff,
3954 			    packsize, packbuf, DMU_READ_PREFETCH);
3955 			ASSERT0(error);
3956 			error = dmu_read(os, bigobj, bigoff, bigsize,
3957 			    bigbuf, DMU_READ_PREFETCH);
3958 			ASSERT0(error);
3959 		}
3960 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3961 		    n, chunksize, txg);
3962 
3963 		/*
3964 		 * We've verified all the old bufwads, and made new ones.
3965 		 * Now write them out.
3966 		 */
3967 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3968 		if (ztest_opts.zo_verbose >= 7) {
3969 			(void) printf("writing offset %llx size %llx"
3970 			    " txg %llx\n",
3971 			    (u_longlong_t)bigoff,
3972 			    (u_longlong_t)bigsize,
3973 			    (u_longlong_t)txg);
3974 		}
3975 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3976 			dmu_buf_t *dbt;
3977 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3978 				bcopy((caddr_t)bigbuf + (off - bigoff),
3979 				    bigbuf_arcbufs[j]->b_data, chunksize);
3980 			} else {
3981 				bcopy((caddr_t)bigbuf + (off - bigoff),
3982 				    bigbuf_arcbufs[2 * j]->b_data,
3983 				    chunksize / 2);
3984 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3985 				    chunksize / 2,
3986 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3987 				    chunksize / 2);
3988 			}
3989 
3990 			if (i == 1) {
3991 				VERIFY(dmu_buf_hold(os, bigobj, off,
3992 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3993 			}
3994 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3995 				dmu_assign_arcbuf(bonus_db, off,
3996 				    bigbuf_arcbufs[j], tx);
3997 			} else {
3998 				dmu_assign_arcbuf(bonus_db, off,
3999 				    bigbuf_arcbufs[2 * j], tx);
4000 				dmu_assign_arcbuf(bonus_db,
4001 				    off + chunksize / 2,
4002 				    bigbuf_arcbufs[2 * j + 1], tx);
4003 			}
4004 			if (i == 1) {
4005 				dmu_buf_rele(dbt, FTAG);
4006 			}
4007 		}
4008 		dmu_tx_commit(tx);
4009 
4010 		/*
4011 		 * Sanity check the stuff we just wrote.
4012 		 */
4013 		{
4014 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4015 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4016 
4017 			VERIFY(0 == dmu_read(os, packobj, packoff,
4018 			    packsize, packcheck, DMU_READ_PREFETCH));
4019 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4020 			    bigsize, bigcheck, DMU_READ_PREFETCH));
4021 
4022 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4023 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4024 
4025 			umem_free(packcheck, packsize);
4026 			umem_free(bigcheck, bigsize);
4027 		}
4028 		if (i == 2) {
4029 			txg_wait_open(dmu_objset_pool(os), 0);
4030 		} else if (i == 3) {
4031 			txg_wait_synced(dmu_objset_pool(os), 0);
4032 		}
4033 	}
4034 
4035 	dmu_buf_rele(bonus_db, FTAG);
4036 	umem_free(packbuf, packsize);
4037 	umem_free(bigbuf, bigsize);
4038 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4039 }
4040 
4041 /* ARGSUSED */
4042 void
4043 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4044 {
4045 	ztest_od_t od[1];
4046 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4047 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4048 
4049 	/*
4050 	 * Have multiple threads write to large offsets in an object
4051 	 * to verify that parallel writes to an object -- even to the
4052 	 * same blocks within the object -- doesn't cause any trouble.
4053 	 */
4054 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4055 
4056 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4057 		return;
4058 
4059 	while (ztest_random(10) != 0)
4060 		ztest_io(zd, od[0].od_object, offset);
4061 }
4062 
4063 void
4064 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4065 {
4066 	ztest_od_t od[1];
4067 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4068 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4069 	uint64_t count = ztest_random(20) + 1;
4070 	uint64_t blocksize = ztest_random_blocksize();
4071 	void *data;
4072 
4073 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4074 
4075 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4076 		return;
4077 
4078 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4079 		return;
4080 
4081 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4082 
4083 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4084 
4085 	while (ztest_random(count) != 0) {
4086 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4087 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4088 		    data) != 0)
4089 			break;
4090 		while (ztest_random(4) != 0)
4091 			ztest_io(zd, od[0].od_object, randoff);
4092 	}
4093 
4094 	umem_free(data, blocksize);
4095 }
4096 
4097 /*
4098  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4099  */
4100 #define	ZTEST_ZAP_MIN_INTS	1
4101 #define	ZTEST_ZAP_MAX_INTS	4
4102 #define	ZTEST_ZAP_MAX_PROPS	1000
4103 
4104 void
4105 ztest_zap(ztest_ds_t *zd, uint64_t id)
4106 {
4107 	objset_t *os = zd->zd_os;
4108 	ztest_od_t od[1];
4109 	uint64_t object;
4110 	uint64_t txg, last_txg;
4111 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4112 	uint64_t zl_ints, zl_intsize, prop;
4113 	int i, ints;
4114 	dmu_tx_t *tx;
4115 	char propname[100], txgname[100];
4116 	int error;
4117 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4118 
4119 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4120 
4121 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4122 		return;
4123 
4124 	object = od[0].od_object;
4125 
4126 	/*
4127 	 * Generate a known hash collision, and verify that
4128 	 * we can lookup and remove both entries.
4129 	 */
4130 	tx = dmu_tx_create(os);
4131 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4132 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4133 	if (txg == 0)
4134 		return;
4135 	for (i = 0; i < 2; i++) {
4136 		value[i] = i;
4137 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4138 		    1, &value[i], tx));
4139 	}
4140 	for (i = 0; i < 2; i++) {
4141 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4142 		    sizeof (uint64_t), 1, &value[i], tx));
4143 		VERIFY3U(0, ==,
4144 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4145 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4146 		ASSERT3U(zl_ints, ==, 1);
4147 	}
4148 	for (i = 0; i < 2; i++) {
4149 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4150 	}
4151 	dmu_tx_commit(tx);
4152 
4153 	/*
4154 	 * Generate a buch of random entries.
4155 	 */
4156 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4157 
4158 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4159 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4160 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4161 	bzero(value, sizeof (value));
4162 	last_txg = 0;
4163 
4164 	/*
4165 	 * If these zap entries already exist, validate their contents.
4166 	 */
4167 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4168 	if (error == 0) {
4169 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4170 		ASSERT3U(zl_ints, ==, 1);
4171 
4172 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4173 		    zl_ints, &last_txg) == 0);
4174 
4175 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4176 		    &zl_ints) == 0);
4177 
4178 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4179 		ASSERT3U(zl_ints, ==, ints);
4180 
4181 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4182 		    zl_ints, value) == 0);
4183 
4184 		for (i = 0; i < ints; i++) {
4185 			ASSERT3U(value[i], ==, last_txg + object + i);
4186 		}
4187 	} else {
4188 		ASSERT3U(error, ==, ENOENT);
4189 	}
4190 
4191 	/*
4192 	 * Atomically update two entries in our zap object.
4193 	 * The first is named txg_%llu, and contains the txg
4194 	 * in which the property was last updated.  The second
4195 	 * is named prop_%llu, and the nth element of its value
4196 	 * should be txg + object + n.
4197 	 */
4198 	tx = dmu_tx_create(os);
4199 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4200 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4201 	if (txg == 0)
4202 		return;
4203 
4204 	if (last_txg > txg)
4205 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4206 
4207 	for (i = 0; i < ints; i++)
4208 		value[i] = txg + object + i;
4209 
4210 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4211 	    1, &txg, tx));
4212 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4213 	    ints, value, tx));
4214 
4215 	dmu_tx_commit(tx);
4216 
4217 	/*
4218 	 * Remove a random pair of entries.
4219 	 */
4220 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4221 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4222 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4223 
4224 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4225 
4226 	if (error == ENOENT)
4227 		return;
4228 
4229 	ASSERT0(error);
4230 
4231 	tx = dmu_tx_create(os);
4232 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4233 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4234 	if (txg == 0)
4235 		return;
4236 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4237 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4238 	dmu_tx_commit(tx);
4239 }
4240 
4241 /*
4242  * Testcase to test the upgrading of a microzap to fatzap.
4243  */
4244 void
4245 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4246 {
4247 	objset_t *os = zd->zd_os;
4248 	ztest_od_t od[1];
4249 	uint64_t object, txg;
4250 
4251 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4252 
4253 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4254 		return;
4255 
4256 	object = od[0].od_object;
4257 
4258 	/*
4259 	 * Add entries to this ZAP and make sure it spills over
4260 	 * and gets upgraded to a fatzap. Also, since we are adding
4261 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4262 	 */
4263 	for (int i = 0; i < 2050; i++) {
4264 		char name[MAXNAMELEN];
4265 		uint64_t value = i;
4266 		dmu_tx_t *tx;
4267 		int error;
4268 
4269 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4270 		    id, value);
4271 
4272 		tx = dmu_tx_create(os);
4273 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4274 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4275 		if (txg == 0)
4276 			return;
4277 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4278 		    &value, tx);
4279 		ASSERT(error == 0 || error == EEXIST);
4280 		dmu_tx_commit(tx);
4281 	}
4282 }
4283 
4284 /* ARGSUSED */
4285 void
4286 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4287 {
4288 	objset_t *os = zd->zd_os;
4289 	ztest_od_t od[1];
4290 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4291 	dmu_tx_t *tx;
4292 	int i, namelen, error;
4293 	int micro = ztest_random(2);
4294 	char name[20], string_value[20];
4295 	void *data;
4296 
4297 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4298 
4299 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4300 		return;
4301 
4302 	object = od[0].od_object;
4303 
4304 	/*
4305 	 * Generate a random name of the form 'xxx.....' where each
4306 	 * x is a random printable character and the dots are dots.
4307 	 * There are 94 such characters, and the name length goes from
4308 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4309 	 */
4310 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4311 
4312 	for (i = 0; i < 3; i++)
4313 		name[i] = '!' + ztest_random('~' - '!' + 1);
4314 	for (; i < namelen - 1; i++)
4315 		name[i] = '.';
4316 	name[i] = '\0';
4317 
4318 	if ((namelen & 1) || micro) {
4319 		wsize = sizeof (txg);
4320 		wc = 1;
4321 		data = &txg;
4322 	} else {
4323 		wsize = 1;
4324 		wc = namelen;
4325 		data = string_value;
4326 	}
4327 
4328 	count = -1ULL;
4329 	VERIFY0(zap_count(os, object, &count));
4330 	ASSERT(count != -1ULL);
4331 
4332 	/*
4333 	 * Select an operation: length, lookup, add, update, remove.
4334 	 */
4335 	i = ztest_random(5);
4336 
4337 	if (i >= 2) {
4338 		tx = dmu_tx_create(os);
4339 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4340 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4341 		if (txg == 0)
4342 			return;
4343 		bcopy(name, string_value, namelen);
4344 	} else {
4345 		tx = NULL;
4346 		txg = 0;
4347 		bzero(string_value, namelen);
4348 	}
4349 
4350 	switch (i) {
4351 
4352 	case 0:
4353 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4354 		if (error == 0) {
4355 			ASSERT3U(wsize, ==, zl_wsize);
4356 			ASSERT3U(wc, ==, zl_wc);
4357 		} else {
4358 			ASSERT3U(error, ==, ENOENT);
4359 		}
4360 		break;
4361 
4362 	case 1:
4363 		error = zap_lookup(os, object, name, wsize, wc, data);
4364 		if (error == 0) {
4365 			if (data == string_value &&
4366 			    bcmp(name, data, namelen) != 0)
4367 				fatal(0, "name '%s' != val '%s' len %d",
4368 				    name, data, namelen);
4369 		} else {
4370 			ASSERT3U(error, ==, ENOENT);
4371 		}
4372 		break;
4373 
4374 	case 2:
4375 		error = zap_add(os, object, name, wsize, wc, data, tx);
4376 		ASSERT(error == 0 || error == EEXIST);
4377 		break;
4378 
4379 	case 3:
4380 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4381 		break;
4382 
4383 	case 4:
4384 		error = zap_remove(os, object, name, tx);
4385 		ASSERT(error == 0 || error == ENOENT);
4386 		break;
4387 	}
4388 
4389 	if (tx != NULL)
4390 		dmu_tx_commit(tx);
4391 }
4392 
4393 /*
4394  * Commit callback data.
4395  */
4396 typedef struct ztest_cb_data {
4397 	list_node_t		zcd_node;
4398 	uint64_t		zcd_txg;
4399 	int			zcd_expected_err;
4400 	boolean_t		zcd_added;
4401 	boolean_t		zcd_called;
4402 	spa_t			*zcd_spa;
4403 } ztest_cb_data_t;
4404 
4405 /* This is the actual commit callback function */
4406 static void
4407 ztest_commit_callback(void *arg, int error)
4408 {
4409 	ztest_cb_data_t *data = arg;
4410 	uint64_t synced_txg;
4411 
4412 	VERIFY(data != NULL);
4413 	VERIFY3S(data->zcd_expected_err, ==, error);
4414 	VERIFY(!data->zcd_called);
4415 
4416 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4417 	if (data->zcd_txg > synced_txg)
4418 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4419 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4420 		    synced_txg);
4421 
4422 	data->zcd_called = B_TRUE;
4423 
4424 	if (error == ECANCELED) {
4425 		ASSERT0(data->zcd_txg);
4426 		ASSERT(!data->zcd_added);
4427 
4428 		/*
4429 		 * The private callback data should be destroyed here, but
4430 		 * since we are going to check the zcd_called field after
4431 		 * dmu_tx_abort(), we will destroy it there.
4432 		 */
4433 		return;
4434 	}
4435 
4436 	/* Was this callback added to the global callback list? */
4437 	if (!data->zcd_added)
4438 		goto out;
4439 
4440 	ASSERT3U(data->zcd_txg, !=, 0);
4441 
4442 	/* Remove our callback from the list */
4443 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4444 	list_remove(&zcl.zcl_callbacks, data);
4445 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4446 
4447 out:
4448 	umem_free(data, sizeof (ztest_cb_data_t));
4449 }
4450 
4451 /* Allocate and initialize callback data structure */
4452 static ztest_cb_data_t *
4453 ztest_create_cb_data(objset_t *os, uint64_t txg)
4454 {
4455 	ztest_cb_data_t *cb_data;
4456 
4457 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4458 
4459 	cb_data->zcd_txg = txg;
4460 	cb_data->zcd_spa = dmu_objset_spa(os);
4461 
4462 	return (cb_data);
4463 }
4464 
4465 /*
4466  * If a number of txgs equal to this threshold have been created after a commit
4467  * callback has been registered but not called, then we assume there is an
4468  * implementation bug.
4469  */
4470 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4471 
4472 /*
4473  * Commit callback test.
4474  */
4475 void
4476 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4477 {
4478 	objset_t *os = zd->zd_os;
4479 	ztest_od_t od[1];
4480 	dmu_tx_t *tx;
4481 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4482 	uint64_t old_txg, txg;
4483 	int i, error;
4484 
4485 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4486 
4487 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4488 		return;
4489 
4490 	tx = dmu_tx_create(os);
4491 
4492 	cb_data[0] = ztest_create_cb_data(os, 0);
4493 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4494 
4495 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4496 
4497 	/* Every once in a while, abort the transaction on purpose */
4498 	if (ztest_random(100) == 0)
4499 		error = -1;
4500 
4501 	if (!error)
4502 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4503 
4504 	txg = error ? 0 : dmu_tx_get_txg(tx);
4505 
4506 	cb_data[0]->zcd_txg = txg;
4507 	cb_data[1] = ztest_create_cb_data(os, txg);
4508 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4509 
4510 	if (error) {
4511 		/*
4512 		 * It's not a strict requirement to call the registered
4513 		 * callbacks from inside dmu_tx_abort(), but that's what
4514 		 * it's supposed to happen in the current implementation
4515 		 * so we will check for that.
4516 		 */
4517 		for (i = 0; i < 2; i++) {
4518 			cb_data[i]->zcd_expected_err = ECANCELED;
4519 			VERIFY(!cb_data[i]->zcd_called);
4520 		}
4521 
4522 		dmu_tx_abort(tx);
4523 
4524 		for (i = 0; i < 2; i++) {
4525 			VERIFY(cb_data[i]->zcd_called);
4526 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4527 		}
4528 
4529 		return;
4530 	}
4531 
4532 	cb_data[2] = ztest_create_cb_data(os, txg);
4533 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4534 
4535 	/*
4536 	 * Read existing data to make sure there isn't a future leak.
4537 	 */
4538 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4539 	    &old_txg, DMU_READ_PREFETCH));
4540 
4541 	if (old_txg > txg)
4542 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4543 		    old_txg, txg);
4544 
4545 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4546 
4547 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4548 
4549 	/*
4550 	 * Since commit callbacks don't have any ordering requirement and since
4551 	 * it is theoretically possible for a commit callback to be called
4552 	 * after an arbitrary amount of time has elapsed since its txg has been
4553 	 * synced, it is difficult to reliably determine whether a commit
4554 	 * callback hasn't been called due to high load or due to a flawed
4555 	 * implementation.
4556 	 *
4557 	 * In practice, we will assume that if after a certain number of txgs a
4558 	 * commit callback hasn't been called, then most likely there's an
4559 	 * implementation bug..
4560 	 */
4561 	tmp_cb = list_head(&zcl.zcl_callbacks);
4562 	if (tmp_cb != NULL &&
4563 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4564 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4565 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4566 	}
4567 
4568 	/*
4569 	 * Let's find the place to insert our callbacks.
4570 	 *
4571 	 * Even though the list is ordered by txg, it is possible for the
4572 	 * insertion point to not be the end because our txg may already be
4573 	 * quiescing at this point and other callbacks in the open txg
4574 	 * (from other objsets) may have sneaked in.
4575 	 */
4576 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4577 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4578 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4579 
4580 	/* Add the 3 callbacks to the list */
4581 	for (i = 0; i < 3; i++) {
4582 		if (tmp_cb == NULL)
4583 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4584 		else
4585 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4586 			    cb_data[i]);
4587 
4588 		cb_data[i]->zcd_added = B_TRUE;
4589 		VERIFY(!cb_data[i]->zcd_called);
4590 
4591 		tmp_cb = cb_data[i];
4592 	}
4593 
4594 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4595 
4596 	dmu_tx_commit(tx);
4597 }
4598 
4599 /* ARGSUSED */
4600 void
4601 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4602 {
4603 	zfs_prop_t proplist[] = {
4604 		ZFS_PROP_CHECKSUM,
4605 		ZFS_PROP_COMPRESSION,
4606 		ZFS_PROP_COPIES,
4607 		ZFS_PROP_DEDUP
4608 	};
4609 
4610 	(void) rw_rdlock(&ztest_name_lock);
4611 
4612 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4613 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4614 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4615 
4616 	(void) rw_unlock(&ztest_name_lock);
4617 }
4618 
4619 /* ARGSUSED */
4620 void
4621 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4622 {
4623 	nvlist_t *props = NULL;
4624 
4625 	(void) rw_rdlock(&ztest_name_lock);
4626 
4627 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4628 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4629 
4630 	VERIFY0(spa_prop_get(ztest_spa, &props));
4631 
4632 	if (ztest_opts.zo_verbose >= 6)
4633 		dump_nvlist(props, 4);
4634 
4635 	nvlist_free(props);
4636 
4637 	(void) rw_unlock(&ztest_name_lock);
4638 }
4639 
4640 static int
4641 user_release_one(const char *snapname, const char *holdname)
4642 {
4643 	nvlist_t *snaps, *holds;
4644 	int error;
4645 
4646 	snaps = fnvlist_alloc();
4647 	holds = fnvlist_alloc();
4648 	fnvlist_add_boolean(holds, holdname);
4649 	fnvlist_add_nvlist(snaps, snapname, holds);
4650 	fnvlist_free(holds);
4651 	error = dsl_dataset_user_release(snaps, NULL);
4652 	fnvlist_free(snaps);
4653 	return (error);
4654 }
4655 
4656 /*
4657  * Test snapshot hold/release and deferred destroy.
4658  */
4659 void
4660 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4661 {
4662 	int error;
4663 	objset_t *os = zd->zd_os;
4664 	objset_t *origin;
4665 	char snapname[100];
4666 	char fullname[100];
4667 	char clonename[100];
4668 	char tag[100];
4669 	char osname[MAXNAMELEN];
4670 	nvlist_t *holds;
4671 
4672 	(void) rw_rdlock(&ztest_name_lock);
4673 
4674 	dmu_objset_name(os, osname);
4675 
4676 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4677 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4678 	(void) snprintf(clonename, sizeof (clonename),
4679 	    "%s/ch1_%llu", osname, id);
4680 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4681 
4682 	/*
4683 	 * Clean up from any previous run.
4684 	 */
4685 	error = dsl_destroy_head(clonename);
4686 	if (error != ENOENT)
4687 		ASSERT0(error);
4688 	error = user_release_one(fullname, tag);
4689 	if (error != ESRCH && error != ENOENT)
4690 		ASSERT0(error);
4691 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4692 	if (error != ENOENT)
4693 		ASSERT0(error);
4694 
4695 	/*
4696 	 * Create snapshot, clone it, mark snap for deferred destroy,
4697 	 * destroy clone, verify snap was also destroyed.
4698 	 */
4699 	error = dmu_objset_snapshot_one(osname, snapname);
4700 	if (error) {
4701 		if (error == ENOSPC) {
4702 			ztest_record_enospc("dmu_objset_snapshot");
4703 			goto out;
4704 		}
4705 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4706 	}
4707 
4708 	error = dmu_objset_clone(clonename, fullname);
4709 	if (error) {
4710 		if (error == ENOSPC) {
4711 			ztest_record_enospc("dmu_objset_clone");
4712 			goto out;
4713 		}
4714 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4715 	}
4716 
4717 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4718 	if (error) {
4719 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4720 		    fullname, error);
4721 	}
4722 
4723 	error = dsl_destroy_head(clonename);
4724 	if (error)
4725 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4726 
4727 	error = dmu_objset_hold(fullname, FTAG, &origin);
4728 	if (error != ENOENT)
4729 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4730 
4731 	/*
4732 	 * Create snapshot, add temporary hold, verify that we can't
4733 	 * destroy a held snapshot, mark for deferred destroy,
4734 	 * release hold, verify snapshot was destroyed.
4735 	 */
4736 	error = dmu_objset_snapshot_one(osname, snapname);
4737 	if (error) {
4738 		if (error == ENOSPC) {
4739 			ztest_record_enospc("dmu_objset_snapshot");
4740 			goto out;
4741 		}
4742 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4743 	}
4744 
4745 	holds = fnvlist_alloc();
4746 	fnvlist_add_string(holds, fullname, tag);
4747 	error = dsl_dataset_user_hold(holds, 0, NULL);
4748 	fnvlist_free(holds);
4749 
4750 	if (error == ENOSPC) {
4751 		ztest_record_enospc("dsl_dataset_user_hold");
4752 		goto out;
4753 	} else if (error) {
4754 		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4755 		    fullname, tag, error);
4756 	}
4757 
4758 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4759 	if (error != EBUSY) {
4760 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4761 		    fullname, error);
4762 	}
4763 
4764 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4765 	if (error) {
4766 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4767 		    fullname, error);
4768 	}
4769 
4770 	error = user_release_one(fullname, tag);
4771 	if (error)
4772 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4773 
4774 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4775 
4776 out:
4777 	(void) rw_unlock(&ztest_name_lock);
4778 }
4779 
4780 /*
4781  * Inject random faults into the on-disk data.
4782  */
4783 /* ARGSUSED */
4784 void
4785 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4786 {
4787 	ztest_shared_t *zs = ztest_shared;
4788 	spa_t *spa = ztest_spa;
4789 	int fd;
4790 	uint64_t offset;
4791 	uint64_t leaves;
4792 	uint64_t bad = 0x1990c0ffeedecade;
4793 	uint64_t top, leaf;
4794 	char path0[MAXPATHLEN];
4795 	char pathrand[MAXPATHLEN];
4796 	size_t fsize;
4797 	int bshift = SPA_OLD_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
4798 	int iters = 1000;
4799 	int maxfaults;
4800 	int mirror_save;
4801 	vdev_t *vd0 = NULL;
4802 	uint64_t guid0 = 0;
4803 	boolean_t islog = B_FALSE;
4804 
4805 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4806 	maxfaults = MAXFAULTS();
4807 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4808 	mirror_save = zs->zs_mirrors;
4809 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4810 
4811 	ASSERT(leaves >= 1);
4812 
4813 	/*
4814 	 * Grab the name lock as reader. There are some operations
4815 	 * which don't like to have their vdevs changed while
4816 	 * they are in progress (i.e. spa_change_guid). Those
4817 	 * operations will have grabbed the name lock as writer.
4818 	 */
4819 	(void) rw_rdlock(&ztest_name_lock);
4820 
4821 	/*
4822 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4823 	 */
4824 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4825 
4826 	if (ztest_random(2) == 0) {
4827 		/*
4828 		 * Inject errors on a normal data device or slog device.
4829 		 */
4830 		top = ztest_random_vdev_top(spa, B_TRUE);
4831 		leaf = ztest_random(leaves) + zs->zs_splits;
4832 
4833 		/*
4834 		 * Generate paths to the first leaf in this top-level vdev,
4835 		 * and to the random leaf we selected.  We'll induce transient
4836 		 * write failures and random online/offline activity on leaf 0,
4837 		 * and we'll write random garbage to the randomly chosen leaf.
4838 		 */
4839 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4840 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4841 		    top * leaves + zs->zs_splits);
4842 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4843 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4844 		    top * leaves + leaf);
4845 
4846 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4847 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4848 			islog = B_TRUE;
4849 
4850 		/*
4851 		 * If the top-level vdev needs to be resilvered
4852 		 * then we only allow faults on the device that is
4853 		 * resilvering.
4854 		 */
4855 		if (vd0 != NULL && maxfaults != 1 &&
4856 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4857 		    vd0->vdev_resilver_txg != 0)) {
4858 			/*
4859 			 * Make vd0 explicitly claim to be unreadable,
4860 			 * or unwriteable, or reach behind its back
4861 			 * and close the underlying fd.  We can do this if
4862 			 * maxfaults == 0 because we'll fail and reexecute,
4863 			 * and we can do it if maxfaults >= 2 because we'll
4864 			 * have enough redundancy.  If maxfaults == 1, the
4865 			 * combination of this with injection of random data
4866 			 * corruption below exceeds the pool's fault tolerance.
4867 			 */
4868 			vdev_file_t *vf = vd0->vdev_tsd;
4869 
4870 			if (vf != NULL && ztest_random(3) == 0) {
4871 				(void) close(vf->vf_vnode->v_fd);
4872 				vf->vf_vnode->v_fd = -1;
4873 			} else if (ztest_random(2) == 0) {
4874 				vd0->vdev_cant_read = B_TRUE;
4875 			} else {
4876 				vd0->vdev_cant_write = B_TRUE;
4877 			}
4878 			guid0 = vd0->vdev_guid;
4879 		}
4880 	} else {
4881 		/*
4882 		 * Inject errors on an l2cache device.
4883 		 */
4884 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4885 
4886 		if (sav->sav_count == 0) {
4887 			spa_config_exit(spa, SCL_STATE, FTAG);
4888 			(void) rw_unlock(&ztest_name_lock);
4889 			return;
4890 		}
4891 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4892 		guid0 = vd0->vdev_guid;
4893 		(void) strcpy(path0, vd0->vdev_path);
4894 		(void) strcpy(pathrand, vd0->vdev_path);
4895 
4896 		leaf = 0;
4897 		leaves = 1;
4898 		maxfaults = INT_MAX;	/* no limit on cache devices */
4899 	}
4900 
4901 	spa_config_exit(spa, SCL_STATE, FTAG);
4902 	(void) rw_unlock(&ztest_name_lock);
4903 
4904 	/*
4905 	 * If we can tolerate two or more faults, or we're dealing
4906 	 * with a slog, randomly online/offline vd0.
4907 	 */
4908 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4909 		if (ztest_random(10) < 6) {
4910 			int flags = (ztest_random(2) == 0 ?
4911 			    ZFS_OFFLINE_TEMPORARY : 0);
4912 
4913 			/*
4914 			 * We have to grab the zs_name_lock as writer to
4915 			 * prevent a race between offlining a slog and
4916 			 * destroying a dataset. Offlining the slog will
4917 			 * grab a reference on the dataset which may cause
4918 			 * dmu_objset_destroy() to fail with EBUSY thus
4919 			 * leaving the dataset in an inconsistent state.
4920 			 */
4921 			if (islog)
4922 				(void) rw_wrlock(&ztest_name_lock);
4923 
4924 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4925 
4926 			if (islog)
4927 				(void) rw_unlock(&ztest_name_lock);
4928 		} else {
4929 			/*
4930 			 * Ideally we would like to be able to randomly
4931 			 * call vdev_[on|off]line without holding locks
4932 			 * to force unpredictable failures but the side
4933 			 * effects of vdev_[on|off]line prevent us from
4934 			 * doing so. We grab the ztest_vdev_lock here to
4935 			 * prevent a race between injection testing and
4936 			 * aux_vdev removal.
4937 			 */
4938 			VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4939 			(void) vdev_online(spa, guid0, 0, NULL);
4940 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4941 		}
4942 	}
4943 
4944 	if (maxfaults == 0)
4945 		return;
4946 
4947 	/*
4948 	 * We have at least single-fault tolerance, so inject data corruption.
4949 	 */
4950 	fd = open(pathrand, O_RDWR);
4951 
4952 	if (fd == -1)	/* we hit a gap in the device namespace */
4953 		return;
4954 
4955 	fsize = lseek(fd, 0, SEEK_END);
4956 
4957 	while (--iters != 0) {
4958 		offset = ztest_random(fsize / (leaves << bshift)) *
4959 		    (leaves << bshift) + (leaf << bshift) +
4960 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4961 
4962 		if (offset >= fsize)
4963 			continue;
4964 
4965 		VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4966 		if (mirror_save != zs->zs_mirrors) {
4967 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4968 			(void) close(fd);
4969 			return;
4970 		}
4971 
4972 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4973 			fatal(1, "can't inject bad word at 0x%llx in %s",
4974 			    offset, pathrand);
4975 
4976 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4977 
4978 		if (ztest_opts.zo_verbose >= 7)
4979 			(void) printf("injected bad word into %s,"
4980 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4981 	}
4982 
4983 	(void) close(fd);
4984 }
4985 
4986 /*
4987  * Verify that DDT repair works as expected.
4988  */
4989 void
4990 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4991 {
4992 	ztest_shared_t *zs = ztest_shared;
4993 	spa_t *spa = ztest_spa;
4994 	objset_t *os = zd->zd_os;
4995 	ztest_od_t od[1];
4996 	uint64_t object, blocksize, txg, pattern, psize;
4997 	enum zio_checksum checksum = spa_dedup_checksum(spa);
4998 	dmu_buf_t *db;
4999 	dmu_tx_t *tx;
5000 	void *buf;
5001 	blkptr_t blk;
5002 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
5003 
5004 	blocksize = ztest_random_blocksize();
5005 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
5006 
5007 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5008 
5009 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5010 		return;
5011 
5012 	/*
5013 	 * Take the name lock as writer to prevent anyone else from changing
5014 	 * the pool and dataset properies we need to maintain during this test.
5015 	 */
5016 	(void) rw_wrlock(&ztest_name_lock);
5017 
5018 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5019 	    B_FALSE) != 0 ||
5020 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5021 	    B_FALSE) != 0) {
5022 		(void) rw_unlock(&ztest_name_lock);
5023 		return;
5024 	}
5025 
5026 	object = od[0].od_object;
5027 	blocksize = od[0].od_blocksize;
5028 	pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5029 
5030 	ASSERT(object != 0);
5031 
5032 	tx = dmu_tx_create(os);
5033 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5034 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5035 	if (txg == 0) {
5036 		(void) rw_unlock(&ztest_name_lock);
5037 		return;
5038 	}
5039 
5040 	/*
5041 	 * Write all the copies of our block.
5042 	 */
5043 	for (int i = 0; i < copies; i++) {
5044 		uint64_t offset = i * blocksize;
5045 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5046 		    DMU_READ_NO_PREFETCH);
5047 		if (error != 0) {
5048 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5049 			    os, (long long)object, (long long) offset, error);
5050 		}
5051 		ASSERT(db->db_offset == offset);
5052 		ASSERT(db->db_size == blocksize);
5053 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5054 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5055 		dmu_buf_will_fill(db, tx);
5056 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5057 		dmu_buf_rele(db, FTAG);
5058 	}
5059 
5060 	dmu_tx_commit(tx);
5061 	txg_wait_synced(spa_get_dsl(spa), txg);
5062 
5063 	/*
5064 	 * Find out what block we got.
5065 	 */
5066 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5067 	    DMU_READ_NO_PREFETCH));
5068 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5069 	dmu_buf_rele(db, FTAG);
5070 
5071 	/*
5072 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5073 	 */
5074 	psize = BP_GET_PSIZE(&blk);
5075 	buf = zio_buf_alloc(psize);
5076 	ztest_pattern_set(buf, psize, ~pattern);
5077 
5078 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5079 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5080 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5081 
5082 	zio_buf_free(buf, psize);
5083 
5084 	(void) rw_unlock(&ztest_name_lock);
5085 }
5086 
5087 /*
5088  * Scrub the pool.
5089  */
5090 /* ARGSUSED */
5091 void
5092 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5093 {
5094 	spa_t *spa = ztest_spa;
5095 
5096 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5097 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5098 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5099 }
5100 
5101 /*
5102  * Change the guid for the pool.
5103  */
5104 /* ARGSUSED */
5105 void
5106 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5107 {
5108 	spa_t *spa = ztest_spa;
5109 	uint64_t orig, load;
5110 	int error;
5111 
5112 	orig = spa_guid(spa);
5113 	load = spa_load_guid(spa);
5114 
5115 	(void) rw_wrlock(&ztest_name_lock);
5116 	error = spa_change_guid(spa);
5117 	(void) rw_unlock(&ztest_name_lock);
5118 
5119 	if (error != 0)
5120 		return;
5121 
5122 	if (ztest_opts.zo_verbose >= 4) {
5123 		(void) printf("Changed guid old %llu -> %llu\n",
5124 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5125 	}
5126 
5127 	VERIFY3U(orig, !=, spa_guid(spa));
5128 	VERIFY3U(load, ==, spa_load_guid(spa));
5129 }
5130 
5131 /*
5132  * Rename the pool to a different name and then rename it back.
5133  */
5134 /* ARGSUSED */
5135 void
5136 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5137 {
5138 	char *oldname, *newname;
5139 	spa_t *spa;
5140 
5141 	(void) rw_wrlock(&ztest_name_lock);
5142 
5143 	oldname = ztest_opts.zo_pool;
5144 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5145 	(void) strcpy(newname, oldname);
5146 	(void) strcat(newname, "_tmp");
5147 
5148 	/*
5149 	 * Do the rename
5150 	 */
5151 	VERIFY3U(0, ==, spa_rename(oldname, newname));
5152 
5153 	/*
5154 	 * Try to open it under the old name, which shouldn't exist
5155 	 */
5156 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5157 
5158 	/*
5159 	 * Open it under the new name and make sure it's still the same spa_t.
5160 	 */
5161 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5162 
5163 	ASSERT(spa == ztest_spa);
5164 	spa_close(spa, FTAG);
5165 
5166 	/*
5167 	 * Rename it back to the original
5168 	 */
5169 	VERIFY3U(0, ==, spa_rename(newname, oldname));
5170 
5171 	/*
5172 	 * Make sure it can still be opened
5173 	 */
5174 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5175 
5176 	ASSERT(spa == ztest_spa);
5177 	spa_close(spa, FTAG);
5178 
5179 	umem_free(newname, strlen(newname) + 1);
5180 
5181 	(void) rw_unlock(&ztest_name_lock);
5182 }
5183 
5184 /*
5185  * Verify pool integrity by running zdb.
5186  */
5187 static void
5188 ztest_run_zdb(char *pool)
5189 {
5190 	int status;
5191 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5192 	char zbuf[1024];
5193 	char *bin;
5194 	char *ztest;
5195 	char *isa;
5196 	int isalen;
5197 	FILE *fp;
5198 
5199 	(void) realpath(getexecname(), zdb);
5200 
5201 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5202 	bin = strstr(zdb, "/usr/bin/");
5203 	ztest = strstr(bin, "/ztest");
5204 	isa = bin + 8;
5205 	isalen = ztest - isa;
5206 	isa = strdup(isa);
5207 	/* LINTED */
5208 	(void) sprintf(bin,
5209 	    "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5210 	    isalen,
5211 	    isa,
5212 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5213 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5214 	    spa_config_path,
5215 	    pool);
5216 	free(isa);
5217 
5218 	if (ztest_opts.zo_verbose >= 5)
5219 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5220 
5221 	fp = popen(zdb, "r");
5222 
5223 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5224 		if (ztest_opts.zo_verbose >= 3)
5225 			(void) printf("%s", zbuf);
5226 
5227 	status = pclose(fp);
5228 
5229 	if (status == 0)
5230 		return;
5231 
5232 	ztest_dump_core = 0;
5233 	if (WIFEXITED(status))
5234 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5235 	else
5236 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5237 }
5238 
5239 static void
5240 ztest_walk_pool_directory(char *header)
5241 {
5242 	spa_t *spa = NULL;
5243 
5244 	if (ztest_opts.zo_verbose >= 6)
5245 		(void) printf("%s\n", header);
5246 
5247 	mutex_enter(&spa_namespace_lock);
5248 	while ((spa = spa_next(spa)) != NULL)
5249 		if (ztest_opts.zo_verbose >= 6)
5250 			(void) printf("\t%s\n", spa_name(spa));
5251 	mutex_exit(&spa_namespace_lock);
5252 }
5253 
5254 static void
5255 ztest_spa_import_export(char *oldname, char *newname)
5256 {
5257 	nvlist_t *config, *newconfig;
5258 	uint64_t pool_guid;
5259 	spa_t *spa;
5260 	int error;
5261 
5262 	if (ztest_opts.zo_verbose >= 4) {
5263 		(void) printf("import/export: old = %s, new = %s\n",
5264 		    oldname, newname);
5265 	}
5266 
5267 	/*
5268 	 * Clean up from previous runs.
5269 	 */
5270 	(void) spa_destroy(newname);
5271 
5272 	/*
5273 	 * Get the pool's configuration and guid.
5274 	 */
5275 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5276 
5277 	/*
5278 	 * Kick off a scrub to tickle scrub/export races.
5279 	 */
5280 	if (ztest_random(2) == 0)
5281 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5282 
5283 	pool_guid = spa_guid(spa);
5284 	spa_close(spa, FTAG);
5285 
5286 	ztest_walk_pool_directory("pools before export");
5287 
5288 	/*
5289 	 * Export it.
5290 	 */
5291 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5292 
5293 	ztest_walk_pool_directory("pools after export");
5294 
5295 	/*
5296 	 * Try to import it.
5297 	 */
5298 	newconfig = spa_tryimport(config);
5299 	ASSERT(newconfig != NULL);
5300 	nvlist_free(newconfig);
5301 
5302 	/*
5303 	 * Import it under the new name.
5304 	 */
5305 	error = spa_import(newname, config, NULL, 0);
5306 	if (error != 0) {
5307 		dump_nvlist(config, 0);
5308 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5309 		    oldname, newname, error);
5310 	}
5311 
5312 	ztest_walk_pool_directory("pools after import");
5313 
5314 	/*
5315 	 * Try to import it again -- should fail with EEXIST.
5316 	 */
5317 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5318 
5319 	/*
5320 	 * Try to import it under a different name -- should fail with EEXIST.
5321 	 */
5322 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5323 
5324 	/*
5325 	 * Verify that the pool is no longer visible under the old name.
5326 	 */
5327 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5328 
5329 	/*
5330 	 * Verify that we can open and close the pool using the new name.
5331 	 */
5332 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5333 	ASSERT(pool_guid == spa_guid(spa));
5334 	spa_close(spa, FTAG);
5335 
5336 	nvlist_free(config);
5337 }
5338 
5339 static void
5340 ztest_resume(spa_t *spa)
5341 {
5342 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5343 		(void) printf("resuming from suspended state\n");
5344 	spa_vdev_state_enter(spa, SCL_NONE);
5345 	vdev_clear(spa, NULL);
5346 	(void) spa_vdev_state_exit(spa, NULL, 0);
5347 	(void) zio_resume(spa);
5348 }
5349 
5350 static void *
5351 ztest_resume_thread(void *arg)
5352 {
5353 	spa_t *spa = arg;
5354 
5355 	while (!ztest_exiting) {
5356 		if (spa_suspended(spa))
5357 			ztest_resume(spa);
5358 		(void) poll(NULL, 0, 100);
5359 	}
5360 	return (NULL);
5361 }
5362 
5363 static void *
5364 ztest_deadman_thread(void *arg)
5365 {
5366 	ztest_shared_t *zs = arg;
5367 	spa_t *spa = ztest_spa;
5368 	hrtime_t delta, total = 0;
5369 
5370 	for (;;) {
5371 		delta = zs->zs_thread_stop - zs->zs_thread_start +
5372 		    MSEC2NSEC(zfs_deadman_synctime_ms);
5373 
5374 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5375 
5376 		/*
5377 		 * If the pool is suspended then fail immediately. Otherwise,
5378 		 * check to see if the pool is making any progress. If
5379 		 * vdev_deadman() discovers that there hasn't been any recent
5380 		 * I/Os then it will end up aborting the tests.
5381 		 */
5382 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5383 			fatal(0, "aborting test after %llu seconds because "
5384 			    "pool has transitioned to a suspended state.",
5385 			    zfs_deadman_synctime_ms / 1000);
5386 			return (NULL);
5387 		}
5388 		vdev_deadman(spa->spa_root_vdev);
5389 
5390 		total += zfs_deadman_synctime_ms/1000;
5391 		(void) printf("ztest has been running for %lld seconds\n",
5392 		    total);
5393 	}
5394 }
5395 
5396 static void
5397 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5398 {
5399 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5400 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5401 	hrtime_t functime = gethrtime();
5402 
5403 	for (int i = 0; i < zi->zi_iters; i++)
5404 		zi->zi_func(zd, id);
5405 
5406 	functime = gethrtime() - functime;
5407 
5408 	atomic_add_64(&zc->zc_count, 1);
5409 	atomic_add_64(&zc->zc_time, functime);
5410 
5411 	if (ztest_opts.zo_verbose >= 4) {
5412 		Dl_info dli;
5413 		(void) dladdr((void *)zi->zi_func, &dli);
5414 		(void) printf("%6.2f sec in %s\n",
5415 		    (double)functime / NANOSEC, dli.dli_sname);
5416 	}
5417 }
5418 
5419 static void *
5420 ztest_thread(void *arg)
5421 {
5422 	int rand;
5423 	uint64_t id = (uintptr_t)arg;
5424 	ztest_shared_t *zs = ztest_shared;
5425 	uint64_t call_next;
5426 	hrtime_t now;
5427 	ztest_info_t *zi;
5428 	ztest_shared_callstate_t *zc;
5429 
5430 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5431 		/*
5432 		 * See if it's time to force a crash.
5433 		 */
5434 		if (now > zs->zs_thread_kill)
5435 			ztest_kill(zs);
5436 
5437 		/*
5438 		 * If we're getting ENOSPC with some regularity, stop.
5439 		 */
5440 		if (zs->zs_enospc_count > 10)
5441 			break;
5442 
5443 		/*
5444 		 * Pick a random function to execute.
5445 		 */
5446 		rand = ztest_random(ZTEST_FUNCS);
5447 		zi = &ztest_info[rand];
5448 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5449 		call_next = zc->zc_next;
5450 
5451 		if (now >= call_next &&
5452 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5453 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5454 			ztest_execute(rand, zi, id);
5455 		}
5456 	}
5457 
5458 	return (NULL);
5459 }
5460 
5461 static void
5462 ztest_dataset_name(char *dsname, char *pool, int d)
5463 {
5464 	(void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5465 }
5466 
5467 static void
5468 ztest_dataset_destroy(int d)
5469 {
5470 	char name[MAXNAMELEN];
5471 
5472 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5473 
5474 	if (ztest_opts.zo_verbose >= 3)
5475 		(void) printf("Destroying %s to free up space\n", name);
5476 
5477 	/*
5478 	 * Cleanup any non-standard clones and snapshots.  In general,
5479 	 * ztest thread t operates on dataset (t % zopt_datasets),
5480 	 * so there may be more than one thing to clean up.
5481 	 */
5482 	for (int t = d; t < ztest_opts.zo_threads;
5483 	    t += ztest_opts.zo_datasets) {
5484 		ztest_dsl_dataset_cleanup(name, t);
5485 	}
5486 
5487 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5488 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5489 }
5490 
5491 static void
5492 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5493 {
5494 	uint64_t usedobjs, dirobjs, scratch;
5495 
5496 	/*
5497 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5498 	 * Therefore, the number of objects in use should equal the
5499 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5500 	 * If not, we have an object leak.
5501 	 *
5502 	 * Note that we can only check this in ztest_dataset_open(),
5503 	 * when the open-context and syncing-context values agree.
5504 	 * That's because zap_count() returns the open-context value,
5505 	 * while dmu_objset_space() returns the rootbp fill count.
5506 	 */
5507 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5508 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5509 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5510 }
5511 
5512 static int
5513 ztest_dataset_open(int d)
5514 {
5515 	ztest_ds_t *zd = &ztest_ds[d];
5516 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5517 	objset_t *os;
5518 	zilog_t *zilog;
5519 	char name[MAXNAMELEN];
5520 	int error;
5521 
5522 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5523 
5524 	(void) rw_rdlock(&ztest_name_lock);
5525 
5526 	error = ztest_dataset_create(name);
5527 	if (error == ENOSPC) {
5528 		(void) rw_unlock(&ztest_name_lock);
5529 		ztest_record_enospc(FTAG);
5530 		return (error);
5531 	}
5532 	ASSERT(error == 0 || error == EEXIST);
5533 
5534 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5535 	(void) rw_unlock(&ztest_name_lock);
5536 
5537 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5538 
5539 	zilog = zd->zd_zilog;
5540 
5541 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5542 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5543 		fatal(0, "missing log records: claimed %llu < committed %llu",
5544 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5545 
5546 	ztest_dataset_dirobj_verify(zd);
5547 
5548 	zil_replay(os, zd, ztest_replay_vector);
5549 
5550 	ztest_dataset_dirobj_verify(zd);
5551 
5552 	if (ztest_opts.zo_verbose >= 6)
5553 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5554 		    zd->zd_name,
5555 		    (u_longlong_t)zilog->zl_parse_blk_count,
5556 		    (u_longlong_t)zilog->zl_parse_lr_count,
5557 		    (u_longlong_t)zilog->zl_replaying_seq);
5558 
5559 	zilog = zil_open(os, ztest_get_data);
5560 
5561 	if (zilog->zl_replaying_seq != 0 &&
5562 	    zilog->zl_replaying_seq < committed_seq)
5563 		fatal(0, "missing log records: replayed %llu < committed %llu",
5564 		    zilog->zl_replaying_seq, committed_seq);
5565 
5566 	return (0);
5567 }
5568 
5569 static void
5570 ztest_dataset_close(int d)
5571 {
5572 	ztest_ds_t *zd = &ztest_ds[d];
5573 
5574 	zil_close(zd->zd_zilog);
5575 	dmu_objset_disown(zd->zd_os, zd);
5576 
5577 	ztest_zd_fini(zd);
5578 }
5579 
5580 /*
5581  * Kick off threads to run tests on all datasets in parallel.
5582  */
5583 static void
5584 ztest_run(ztest_shared_t *zs)
5585 {
5586 	thread_t *tid;
5587 	spa_t *spa;
5588 	objset_t *os;
5589 	thread_t resume_tid;
5590 	int error;
5591 
5592 	ztest_exiting = B_FALSE;
5593 
5594 	/*
5595 	 * Initialize parent/child shared state.
5596 	 */
5597 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5598 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5599 
5600 	zs->zs_thread_start = gethrtime();
5601 	zs->zs_thread_stop =
5602 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5603 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5604 	zs->zs_thread_kill = zs->zs_thread_stop;
5605 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5606 		zs->zs_thread_kill -=
5607 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5608 	}
5609 
5610 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5611 
5612 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5613 	    offsetof(ztest_cb_data_t, zcd_node));
5614 
5615 	/*
5616 	 * Open our pool.
5617 	 */
5618 	kernel_init(FREAD | FWRITE);
5619 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5620 	spa->spa_debug = B_TRUE;
5621 	metaslab_preload_limit = ztest_random(20) + 1;
5622 	ztest_spa = spa;
5623 
5624 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5625 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
5626 	zs->zs_guid = dmu_objset_fsid_guid(os);
5627 	dmu_objset_disown(os, FTAG);
5628 
5629 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5630 
5631 	/*
5632 	 * We don't expect the pool to suspend unless maxfaults == 0,
5633 	 * in which case ztest_fault_inject() temporarily takes away
5634 	 * the only valid replica.
5635 	 */
5636 	if (MAXFAULTS() == 0)
5637 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5638 	else
5639 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5640 
5641 	/*
5642 	 * Create a thread to periodically resume suspended I/O.
5643 	 */
5644 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5645 	    &resume_tid) == 0);
5646 
5647 	/*
5648 	 * Create a deadman thread to abort() if we hang.
5649 	 */
5650 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5651 	    NULL) == 0);
5652 
5653 	/*
5654 	 * Verify that we can safely inquire about about any object,
5655 	 * whether it's allocated or not.  To make it interesting,
5656 	 * we probe a 5-wide window around each power of two.
5657 	 * This hits all edge cases, including zero and the max.
5658 	 */
5659 	for (int t = 0; t < 64; t++) {
5660 		for (int d = -5; d <= 5; d++) {
5661 			error = dmu_object_info(spa->spa_meta_objset,
5662 			    (1ULL << t) + d, NULL);
5663 			ASSERT(error == 0 || error == ENOENT ||
5664 			    error == EINVAL);
5665 		}
5666 	}
5667 
5668 	/*
5669 	 * If we got any ENOSPC errors on the previous run, destroy something.
5670 	 */
5671 	if (zs->zs_enospc_count != 0) {
5672 		int d = ztest_random(ztest_opts.zo_datasets);
5673 		ztest_dataset_destroy(d);
5674 	}
5675 	zs->zs_enospc_count = 0;
5676 
5677 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5678 	    UMEM_NOFAIL);
5679 
5680 	if (ztest_opts.zo_verbose >= 4)
5681 		(void) printf("starting main threads...\n");
5682 
5683 	/*
5684 	 * Kick off all the tests that run in parallel.
5685 	 */
5686 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
5687 		if (t < ztest_opts.zo_datasets &&
5688 		    ztest_dataset_open(t) != 0)
5689 			return;
5690 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5691 		    THR_BOUND, &tid[t]) == 0);
5692 	}
5693 
5694 	/*
5695 	 * Wait for all of the tests to complete.  We go in reverse order
5696 	 * so we don't close datasets while threads are still using them.
5697 	 */
5698 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5699 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5700 		if (t < ztest_opts.zo_datasets)
5701 			ztest_dataset_close(t);
5702 	}
5703 
5704 	txg_wait_synced(spa_get_dsl(spa), 0);
5705 
5706 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5707 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5708 	zfs_dbgmsg_print(FTAG);
5709 
5710 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5711 
5712 	/* Kill the resume thread */
5713 	ztest_exiting = B_TRUE;
5714 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5715 	ztest_resume(spa);
5716 
5717 	/*
5718 	 * Right before closing the pool, kick off a bunch of async I/O;
5719 	 * spa_close() should wait for it to complete.
5720 	 */
5721 	for (uint64_t object = 1; object < 50; object++)
5722 		dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5723 
5724 	spa_close(spa, FTAG);
5725 
5726 	/*
5727 	 * Verify that we can loop over all pools.
5728 	 */
5729 	mutex_enter(&spa_namespace_lock);
5730 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5731 		if (ztest_opts.zo_verbose > 3)
5732 			(void) printf("spa_next: found %s\n", spa_name(spa));
5733 	mutex_exit(&spa_namespace_lock);
5734 
5735 	/*
5736 	 * Verify that we can export the pool and reimport it under a
5737 	 * different name.
5738 	 */
5739 	if (ztest_random(2) == 0) {
5740 		char name[MAXNAMELEN];
5741 		(void) snprintf(name, MAXNAMELEN, "%s_import",
5742 		    ztest_opts.zo_pool);
5743 		ztest_spa_import_export(ztest_opts.zo_pool, name);
5744 		ztest_spa_import_export(name, ztest_opts.zo_pool);
5745 	}
5746 
5747 	kernel_fini();
5748 
5749 	list_destroy(&zcl.zcl_callbacks);
5750 
5751 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5752 
5753 	(void) rwlock_destroy(&ztest_name_lock);
5754 	(void) _mutex_destroy(&ztest_vdev_lock);
5755 }
5756 
5757 static void
5758 ztest_freeze(void)
5759 {
5760 	ztest_ds_t *zd = &ztest_ds[0];
5761 	spa_t *spa;
5762 	int numloops = 0;
5763 
5764 	if (ztest_opts.zo_verbose >= 3)
5765 		(void) printf("testing spa_freeze()...\n");
5766 
5767 	kernel_init(FREAD | FWRITE);
5768 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5769 	VERIFY3U(0, ==, ztest_dataset_open(0));
5770 	spa->spa_debug = B_TRUE;
5771 	ztest_spa = spa;
5772 
5773 	/*
5774 	 * Force the first log block to be transactionally allocated.
5775 	 * We have to do this before we freeze the pool -- otherwise
5776 	 * the log chain won't be anchored.
5777 	 */
5778 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5779 		ztest_dmu_object_alloc_free(zd, 0);
5780 		zil_commit(zd->zd_zilog, 0);
5781 	}
5782 
5783 	txg_wait_synced(spa_get_dsl(spa), 0);
5784 
5785 	/*
5786 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5787 	 * so that the only way to record changes from now on is the ZIL.
5788 	 */
5789 	spa_freeze(spa);
5790 
5791 	/*
5792 	 * Because it is hard to predict how much space a write will actually
5793 	 * require beforehand, we leave ourselves some fudge space to write over
5794 	 * capacity.
5795 	 */
5796 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5797 
5798 	/*
5799 	 * Run tests that generate log records but don't alter the pool config
5800 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5801 	 * We do a txg_wait_synced() after each iteration to force the txg
5802 	 * to increase well beyond the last synced value in the uberblock.
5803 	 * The ZIL should be OK with that.
5804 	 *
5805 	 * Run a random number of times less than zo_maxloops and ensure we do
5806 	 * not run out of space on the pool.
5807 	 */
5808 	while (ztest_random(10) != 0 &&
5809 	    numloops++ < ztest_opts.zo_maxloops &&
5810 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5811 		ztest_od_t od;
5812 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5813 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5814 		ztest_io(zd, od.od_object,
5815 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5816 		txg_wait_synced(spa_get_dsl(spa), 0);
5817 	}
5818 
5819 	/*
5820 	 * Commit all of the changes we just generated.
5821 	 */
5822 	zil_commit(zd->zd_zilog, 0);
5823 	txg_wait_synced(spa_get_dsl(spa), 0);
5824 
5825 	/*
5826 	 * Close our dataset and close the pool.
5827 	 */
5828 	ztest_dataset_close(0);
5829 	spa_close(spa, FTAG);
5830 	kernel_fini();
5831 
5832 	/*
5833 	 * Open and close the pool and dataset to induce log replay.
5834 	 */
5835 	kernel_init(FREAD | FWRITE);
5836 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5837 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5838 	VERIFY3U(0, ==, ztest_dataset_open(0));
5839 	ztest_dataset_close(0);
5840 
5841 	spa->spa_debug = B_TRUE;
5842 	ztest_spa = spa;
5843 	txg_wait_synced(spa_get_dsl(spa), 0);
5844 	ztest_reguid(NULL, 0);
5845 
5846 	spa_close(spa, FTAG);
5847 	kernel_fini();
5848 }
5849 
5850 void
5851 print_time(hrtime_t t, char *timebuf)
5852 {
5853 	hrtime_t s = t / NANOSEC;
5854 	hrtime_t m = s / 60;
5855 	hrtime_t h = m / 60;
5856 	hrtime_t d = h / 24;
5857 
5858 	s -= m * 60;
5859 	m -= h * 60;
5860 	h -= d * 24;
5861 
5862 	timebuf[0] = '\0';
5863 
5864 	if (d)
5865 		(void) sprintf(timebuf,
5866 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5867 	else if (h)
5868 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5869 	else if (m)
5870 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5871 	else
5872 		(void) sprintf(timebuf, "%llus", s);
5873 }
5874 
5875 static nvlist_t *
5876 make_random_props()
5877 {
5878 	nvlist_t *props;
5879 
5880 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5881 	if (ztest_random(2) == 0)
5882 		return (props);
5883 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5884 
5885 	return (props);
5886 }
5887 
5888 /*
5889  * Create a storage pool with the given name and initial vdev size.
5890  * Then test spa_freeze() functionality.
5891  */
5892 static void
5893 ztest_init(ztest_shared_t *zs)
5894 {
5895 	spa_t *spa;
5896 	nvlist_t *nvroot, *props;
5897 
5898 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5899 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5900 
5901 	kernel_init(FREAD | FWRITE);
5902 
5903 	/*
5904 	 * Create the storage pool.
5905 	 */
5906 	(void) spa_destroy(ztest_opts.zo_pool);
5907 	ztest_shared->zs_vdev_next_leaf = 0;
5908 	zs->zs_splits = 0;
5909 	zs->zs_mirrors = ztest_opts.zo_mirrors;
5910 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5911 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5912 	props = make_random_props();
5913 	for (int i = 0; i < SPA_FEATURES; i++) {
5914 		char buf[1024];
5915 		(void) snprintf(buf, sizeof (buf), "feature@%s",
5916 		    spa_feature_table[i].fi_uname);
5917 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5918 	}
5919 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5920 	nvlist_free(nvroot);
5921 
5922 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5923 	zs->zs_metaslab_sz =
5924 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5925 
5926 	spa_close(spa, FTAG);
5927 
5928 	kernel_fini();
5929 
5930 	ztest_run_zdb(ztest_opts.zo_pool);
5931 
5932 	ztest_freeze();
5933 
5934 	ztest_run_zdb(ztest_opts.zo_pool);
5935 
5936 	(void) rwlock_destroy(&ztest_name_lock);
5937 	(void) _mutex_destroy(&ztest_vdev_lock);
5938 }
5939 
5940 static void
5941 setup_data_fd(void)
5942 {
5943 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5944 
5945 	ztest_fd_data = mkstemp(ztest_name_data);
5946 	ASSERT3S(ztest_fd_data, >=, 0);
5947 	(void) unlink(ztest_name_data);
5948 }
5949 
5950 
5951 static int
5952 shared_data_size(ztest_shared_hdr_t *hdr)
5953 {
5954 	int size;
5955 
5956 	size = hdr->zh_hdr_size;
5957 	size += hdr->zh_opts_size;
5958 	size += hdr->zh_size;
5959 	size += hdr->zh_stats_size * hdr->zh_stats_count;
5960 	size += hdr->zh_ds_size * hdr->zh_ds_count;
5961 
5962 	return (size);
5963 }
5964 
5965 static void
5966 setup_hdr(void)
5967 {
5968 	int size;
5969 	ztest_shared_hdr_t *hdr;
5970 
5971 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5972 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5973 	ASSERT(hdr != MAP_FAILED);
5974 
5975 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5976 
5977 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5978 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5979 	hdr->zh_size = sizeof (ztest_shared_t);
5980 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5981 	hdr->zh_stats_count = ZTEST_FUNCS;
5982 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5983 	hdr->zh_ds_count = ztest_opts.zo_datasets;
5984 
5985 	size = shared_data_size(hdr);
5986 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5987 
5988 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5989 }
5990 
5991 static void
5992 setup_data(void)
5993 {
5994 	int size, offset;
5995 	ztest_shared_hdr_t *hdr;
5996 	uint8_t *buf;
5997 
5998 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5999 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6000 	ASSERT(hdr != MAP_FAILED);
6001 
6002 	size = shared_data_size(hdr);
6003 
6004 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6005 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6006 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6007 	ASSERT(hdr != MAP_FAILED);
6008 	buf = (uint8_t *)hdr;
6009 
6010 	offset = hdr->zh_hdr_size;
6011 	ztest_shared_opts = (void *)&buf[offset];
6012 	offset += hdr->zh_opts_size;
6013 	ztest_shared = (void *)&buf[offset];
6014 	offset += hdr->zh_size;
6015 	ztest_shared_callstate = (void *)&buf[offset];
6016 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
6017 	ztest_shared_ds = (void *)&buf[offset];
6018 }
6019 
6020 static boolean_t
6021 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6022 {
6023 	pid_t pid;
6024 	int status;
6025 	char *cmdbuf = NULL;
6026 
6027 	pid = fork();
6028 
6029 	if (cmd == NULL) {
6030 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6031 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6032 		cmd = cmdbuf;
6033 	}
6034 
6035 	if (pid == -1)
6036 		fatal(1, "fork failed");
6037 
6038 	if (pid == 0) {	/* child */
6039 		char *emptyargv[2] = { cmd, NULL };
6040 		char fd_data_str[12];
6041 
6042 		struct rlimit rl = { 1024, 1024 };
6043 		(void) setrlimit(RLIMIT_NOFILE, &rl);
6044 
6045 		(void) close(ztest_fd_rand);
6046 		VERIFY3U(11, >=,
6047 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6048 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6049 
6050 		(void) enable_extended_FILE_stdio(-1, -1);
6051 		if (libpath != NULL)
6052 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6053 		(void) execv(cmd, emptyargv);
6054 		ztest_dump_core = B_FALSE;
6055 		fatal(B_TRUE, "exec failed: %s", cmd);
6056 	}
6057 
6058 	if (cmdbuf != NULL) {
6059 		umem_free(cmdbuf, MAXPATHLEN);
6060 		cmd = NULL;
6061 	}
6062 
6063 	while (waitpid(pid, &status, 0) != pid)
6064 		continue;
6065 	if (statusp != NULL)
6066 		*statusp = status;
6067 
6068 	if (WIFEXITED(status)) {
6069 		if (WEXITSTATUS(status) != 0) {
6070 			(void) fprintf(stderr, "child exited with code %d\n",
6071 			    WEXITSTATUS(status));
6072 			exit(2);
6073 		}
6074 		return (B_FALSE);
6075 	} else if (WIFSIGNALED(status)) {
6076 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6077 			(void) fprintf(stderr, "child died with signal %d\n",
6078 			    WTERMSIG(status));
6079 			exit(3);
6080 		}
6081 		return (B_TRUE);
6082 	} else {
6083 		(void) fprintf(stderr, "something strange happened to child\n");
6084 		exit(4);
6085 		/* NOTREACHED */
6086 	}
6087 }
6088 
6089 static void
6090 ztest_run_init(void)
6091 {
6092 	ztest_shared_t *zs = ztest_shared;
6093 
6094 	ASSERT(ztest_opts.zo_init != 0);
6095 
6096 	/*
6097 	 * Blow away any existing copy of zpool.cache
6098 	 */
6099 	(void) remove(spa_config_path);
6100 
6101 	/*
6102 	 * Create and initialize our storage pool.
6103 	 */
6104 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
6105 		bzero(zs, sizeof (ztest_shared_t));
6106 		if (ztest_opts.zo_verbose >= 3 &&
6107 		    ztest_opts.zo_init != 1) {
6108 			(void) printf("ztest_init(), pass %d\n", i);
6109 		}
6110 		ztest_init(zs);
6111 	}
6112 }
6113 
6114 int
6115 main(int argc, char **argv)
6116 {
6117 	int kills = 0;
6118 	int iters = 0;
6119 	int older = 0;
6120 	int newer = 0;
6121 	ztest_shared_t *zs;
6122 	ztest_info_t *zi;
6123 	ztest_shared_callstate_t *zc;
6124 	char timebuf[100];
6125 	char numbuf[6];
6126 	spa_t *spa;
6127 	char *cmd;
6128 	boolean_t hasalt;
6129 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6130 
6131 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6132 
6133 	dprintf_setup(&argc, argv);
6134 	zfs_deadman_synctime_ms = 300000;
6135 
6136 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6137 	ASSERT3S(ztest_fd_rand, >=, 0);
6138 
6139 	if (!fd_data_str) {
6140 		process_options(argc, argv);
6141 
6142 		setup_data_fd();
6143 		setup_hdr();
6144 		setup_data();
6145 		bcopy(&ztest_opts, ztest_shared_opts,
6146 		    sizeof (*ztest_shared_opts));
6147 	} else {
6148 		ztest_fd_data = atoi(fd_data_str);
6149 		setup_data();
6150 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6151 	}
6152 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6153 
6154 	/* Override location of zpool.cache */
6155 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6156 	    ztest_opts.zo_dir), !=, -1);
6157 
6158 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6159 	    UMEM_NOFAIL);
6160 	zs = ztest_shared;
6161 
6162 	if (fd_data_str) {
6163 		metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6164 		metaslab_df_alloc_threshold =
6165 		    zs->zs_metaslab_df_alloc_threshold;
6166 
6167 		if (zs->zs_do_init)
6168 			ztest_run_init();
6169 		else
6170 			ztest_run(zs);
6171 		exit(0);
6172 	}
6173 
6174 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6175 
6176 	if (ztest_opts.zo_verbose >= 1) {
6177 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6178 		    " %llu seconds...\n",
6179 		    (u_longlong_t)ztest_opts.zo_vdevs,
6180 		    ztest_opts.zo_datasets,
6181 		    ztest_opts.zo_threads,
6182 		    (u_longlong_t)ztest_opts.zo_time);
6183 	}
6184 
6185 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6186 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6187 
6188 	zs->zs_do_init = B_TRUE;
6189 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6190 		if (ztest_opts.zo_verbose >= 1) {
6191 			(void) printf("Executing older ztest for "
6192 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6193 		}
6194 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6195 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6196 	} else {
6197 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6198 	}
6199 	zs->zs_do_init = B_FALSE;
6200 
6201 	zs->zs_proc_start = gethrtime();
6202 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6203 
6204 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6205 		zi = &ztest_info[f];
6206 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6207 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6208 			zc->zc_next = UINT64_MAX;
6209 		else
6210 			zc->zc_next = zs->zs_proc_start +
6211 			    ztest_random(2 * zi->zi_interval[0] + 1);
6212 	}
6213 
6214 	/*
6215 	 * Run the tests in a loop.  These tests include fault injection
6216 	 * to verify that self-healing data works, and forced crashes
6217 	 * to verify that we never lose on-disk consistency.
6218 	 */
6219 	while (gethrtime() < zs->zs_proc_stop) {
6220 		int status;
6221 		boolean_t killed;
6222 
6223 		/*
6224 		 * Initialize the workload counters for each function.
6225 		 */
6226 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6227 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6228 			zc->zc_count = 0;
6229 			zc->zc_time = 0;
6230 		}
6231 
6232 		/* Set the allocation switch size */
6233 		zs->zs_metaslab_df_alloc_threshold =
6234 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6235 
6236 		if (!hasalt || ztest_random(2) == 0) {
6237 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6238 				(void) printf("Executing newer ztest: %s\n",
6239 				    cmd);
6240 			}
6241 			newer++;
6242 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6243 		} else {
6244 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6245 				(void) printf("Executing older ztest: %s\n",
6246 				    ztest_opts.zo_alt_ztest);
6247 			}
6248 			older++;
6249 			killed = exec_child(ztest_opts.zo_alt_ztest,
6250 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6251 		}
6252 
6253 		if (killed)
6254 			kills++;
6255 		iters++;
6256 
6257 		if (ztest_opts.zo_verbose >= 1) {
6258 			hrtime_t now = gethrtime();
6259 
6260 			now = MIN(now, zs->zs_proc_stop);
6261 			print_time(zs->zs_proc_stop - now, timebuf);
6262 			nicenum(zs->zs_space, numbuf);
6263 
6264 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6265 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6266 			    iters,
6267 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6268 			    (u_longlong_t)zs->zs_enospc_count,
6269 			    100.0 * zs->zs_alloc / zs->zs_space,
6270 			    numbuf,
6271 			    100.0 * (now - zs->zs_proc_start) /
6272 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6273 		}
6274 
6275 		if (ztest_opts.zo_verbose >= 2) {
6276 			(void) printf("\nWorkload summary:\n\n");
6277 			(void) printf("%7s %9s   %s\n",
6278 			    "Calls", "Time", "Function");
6279 			(void) printf("%7s %9s   %s\n",
6280 			    "-----", "----", "--------");
6281 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6282 				Dl_info dli;
6283 
6284 				zi = &ztest_info[f];
6285 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6286 				print_time(zc->zc_time, timebuf);
6287 				(void) dladdr((void *)zi->zi_func, &dli);
6288 				(void) printf("%7llu %9s   %s\n",
6289 				    (u_longlong_t)zc->zc_count, timebuf,
6290 				    dli.dli_sname);
6291 			}
6292 			(void) printf("\n");
6293 		}
6294 
6295 		/*
6296 		 * It's possible that we killed a child during a rename test,
6297 		 * in which case we'll have a 'ztest_tmp' pool lying around
6298 		 * instead of 'ztest'.  Do a blind rename in case this happened.
6299 		 */
6300 		kernel_init(FREAD);
6301 		if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6302 			spa_close(spa, FTAG);
6303 		} else {
6304 			char tmpname[MAXNAMELEN];
6305 			kernel_fini();
6306 			kernel_init(FREAD | FWRITE);
6307 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6308 			    ztest_opts.zo_pool);
6309 			(void) spa_rename(tmpname, ztest_opts.zo_pool);
6310 		}
6311 		kernel_fini();
6312 
6313 		ztest_run_zdb(ztest_opts.zo_pool);
6314 	}
6315 
6316 	if (ztest_opts.zo_verbose >= 1) {
6317 		if (hasalt) {
6318 			(void) printf("%d runs of older ztest: %s\n", older,
6319 			    ztest_opts.zo_alt_ztest);
6320 			(void) printf("%d runs of newer ztest: %s\n", newer,
6321 			    cmd);
6322 		}
6323 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6324 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6325 	}
6326 
6327 	umem_free(cmd, MAXNAMELEN);
6328 
6329 	return (0);
6330 }
6331