xref: /titanic_52/usr/src/cmd/ztest/ztest.c (revision 25c28e83beb90e7c80452a7c818c5e6f73a07dc8)
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 	// Choose a block size >= the ashift.
989 	uint64_t block_shift =
990 	    ztest_random(SPA_MAXBLOCKSHIFT - ztest_spa_get_ashift() + 1);
991 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
992 }
993 
994 static int
995 ztest_random_ibshift(void)
996 {
997 	return (DN_MIN_INDBLKSHIFT +
998 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
999 }
1000 
1001 static uint64_t
1002 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1003 {
1004 	uint64_t top;
1005 	vdev_t *rvd = spa->spa_root_vdev;
1006 	vdev_t *tvd;
1007 
1008 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1009 
1010 	do {
1011 		top = ztest_random(rvd->vdev_children);
1012 		tvd = rvd->vdev_child[top];
1013 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1014 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1015 
1016 	return (top);
1017 }
1018 
1019 static uint64_t
1020 ztest_random_dsl_prop(zfs_prop_t prop)
1021 {
1022 	uint64_t value;
1023 
1024 	do {
1025 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1026 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1027 
1028 	return (value);
1029 }
1030 
1031 static int
1032 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1033     boolean_t inherit)
1034 {
1035 	const char *propname = zfs_prop_to_name(prop);
1036 	const char *valname;
1037 	char setpoint[MAXPATHLEN];
1038 	uint64_t curval;
1039 	int error;
1040 
1041 	error = dsl_prop_set_int(osname, propname,
1042 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1043 
1044 	if (error == ENOSPC) {
1045 		ztest_record_enospc(FTAG);
1046 		return (error);
1047 	}
1048 	ASSERT0(error);
1049 
1050 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1051 
1052 	if (ztest_opts.zo_verbose >= 6) {
1053 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1054 		(void) printf("%s %s = %s at '%s'\n",
1055 		    osname, propname, valname, setpoint);
1056 	}
1057 
1058 	return (error);
1059 }
1060 
1061 static int
1062 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1063 {
1064 	spa_t *spa = ztest_spa;
1065 	nvlist_t *props = NULL;
1066 	int error;
1067 
1068 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1069 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1070 
1071 	error = spa_prop_set(spa, props);
1072 
1073 	nvlist_free(props);
1074 
1075 	if (error == ENOSPC) {
1076 		ztest_record_enospc(FTAG);
1077 		return (error);
1078 	}
1079 	ASSERT0(error);
1080 
1081 	return (error);
1082 }
1083 
1084 static void
1085 ztest_rll_init(rll_t *rll)
1086 {
1087 	rll->rll_writer = NULL;
1088 	rll->rll_readers = 0;
1089 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1090 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1091 }
1092 
1093 static void
1094 ztest_rll_destroy(rll_t *rll)
1095 {
1096 	ASSERT(rll->rll_writer == NULL);
1097 	ASSERT(rll->rll_readers == 0);
1098 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1099 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
1100 }
1101 
1102 static void
1103 ztest_rll_lock(rll_t *rll, rl_type_t type)
1104 {
1105 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1106 
1107 	if (type == RL_READER) {
1108 		while (rll->rll_writer != NULL)
1109 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1110 		rll->rll_readers++;
1111 	} else {
1112 		while (rll->rll_writer != NULL || rll->rll_readers)
1113 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1114 		rll->rll_writer = curthread;
1115 	}
1116 
1117 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1118 }
1119 
1120 static void
1121 ztest_rll_unlock(rll_t *rll)
1122 {
1123 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1124 
1125 	if (rll->rll_writer) {
1126 		ASSERT(rll->rll_readers == 0);
1127 		rll->rll_writer = NULL;
1128 	} else {
1129 		ASSERT(rll->rll_readers != 0);
1130 		ASSERT(rll->rll_writer == NULL);
1131 		rll->rll_readers--;
1132 	}
1133 
1134 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1135 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1136 
1137 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1138 }
1139 
1140 static void
1141 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1142 {
1143 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1144 
1145 	ztest_rll_lock(rll, type);
1146 }
1147 
1148 static void
1149 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1150 {
1151 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1152 
1153 	ztest_rll_unlock(rll);
1154 }
1155 
1156 static rl_t *
1157 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1158     uint64_t size, rl_type_t type)
1159 {
1160 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1161 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1162 	rl_t *rl;
1163 
1164 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1165 	rl->rl_object = object;
1166 	rl->rl_offset = offset;
1167 	rl->rl_size = size;
1168 	rl->rl_lock = rll;
1169 
1170 	ztest_rll_lock(rll, type);
1171 
1172 	return (rl);
1173 }
1174 
1175 static void
1176 ztest_range_unlock(rl_t *rl)
1177 {
1178 	rll_t *rll = rl->rl_lock;
1179 
1180 	ztest_rll_unlock(rll);
1181 
1182 	umem_free(rl, sizeof (*rl));
1183 }
1184 
1185 static void
1186 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1187 {
1188 	zd->zd_os = os;
1189 	zd->zd_zilog = dmu_objset_zil(os);
1190 	zd->zd_shared = szd;
1191 	dmu_objset_name(os, zd->zd_name);
1192 
1193 	if (zd->zd_shared != NULL)
1194 		zd->zd_shared->zd_seq = 0;
1195 
1196 	VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1197 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1198 
1199 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1200 		ztest_rll_init(&zd->zd_object_lock[l]);
1201 
1202 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1203 		ztest_rll_init(&zd->zd_range_lock[l]);
1204 }
1205 
1206 static void
1207 ztest_zd_fini(ztest_ds_t *zd)
1208 {
1209 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1210 
1211 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1212 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1213 
1214 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1215 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1216 }
1217 
1218 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1219 
1220 static uint64_t
1221 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1222 {
1223 	uint64_t txg;
1224 	int error;
1225 
1226 	/*
1227 	 * Attempt to assign tx to some transaction group.
1228 	 */
1229 	error = dmu_tx_assign(tx, txg_how);
1230 	if (error) {
1231 		if (error == ERESTART) {
1232 			ASSERT(txg_how == TXG_NOWAIT);
1233 			dmu_tx_wait(tx);
1234 		} else {
1235 			ASSERT3U(error, ==, ENOSPC);
1236 			ztest_record_enospc(tag);
1237 		}
1238 		dmu_tx_abort(tx);
1239 		return (0);
1240 	}
1241 	txg = dmu_tx_get_txg(tx);
1242 	ASSERT(txg != 0);
1243 	return (txg);
1244 }
1245 
1246 static void
1247 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1248 {
1249 	uint64_t *ip = buf;
1250 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1251 
1252 	while (ip < ip_end)
1253 		*ip++ = value;
1254 }
1255 
1256 static boolean_t
1257 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1258 {
1259 	uint64_t *ip = buf;
1260 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1261 	uint64_t diff = 0;
1262 
1263 	while (ip < ip_end)
1264 		diff |= (value - *ip++);
1265 
1266 	return (diff == 0);
1267 }
1268 
1269 static void
1270 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1271     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1272 {
1273 	bt->bt_magic = BT_MAGIC;
1274 	bt->bt_objset = dmu_objset_id(os);
1275 	bt->bt_object = object;
1276 	bt->bt_offset = offset;
1277 	bt->bt_gen = gen;
1278 	bt->bt_txg = txg;
1279 	bt->bt_crtxg = crtxg;
1280 }
1281 
1282 static void
1283 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1284     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1285 {
1286 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1287 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1288 	ASSERT3U(bt->bt_object, ==, object);
1289 	ASSERT3U(bt->bt_offset, ==, offset);
1290 	ASSERT3U(bt->bt_gen, <=, gen);
1291 	ASSERT3U(bt->bt_txg, <=, txg);
1292 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1293 }
1294 
1295 static ztest_block_tag_t *
1296 ztest_bt_bonus(dmu_buf_t *db)
1297 {
1298 	dmu_object_info_t doi;
1299 	ztest_block_tag_t *bt;
1300 
1301 	dmu_object_info_from_db(db, &doi);
1302 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1303 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1304 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1305 
1306 	return (bt);
1307 }
1308 
1309 /*
1310  * ZIL logging ops
1311  */
1312 
1313 #define	lrz_type	lr_mode
1314 #define	lrz_blocksize	lr_uid
1315 #define	lrz_ibshift	lr_gid
1316 #define	lrz_bonustype	lr_rdev
1317 #define	lrz_bonuslen	lr_crtime[1]
1318 
1319 static void
1320 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1321 {
1322 	char *name = (void *)(lr + 1);		/* name follows lr */
1323 	size_t namesize = strlen(name) + 1;
1324 	itx_t *itx;
1325 
1326 	if (zil_replaying(zd->zd_zilog, tx))
1327 		return;
1328 
1329 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1330 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1331 	    sizeof (*lr) + namesize - sizeof (lr_t));
1332 
1333 	zil_itx_assign(zd->zd_zilog, itx, tx);
1334 }
1335 
1336 static void
1337 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1338 {
1339 	char *name = (void *)(lr + 1);		/* name follows lr */
1340 	size_t namesize = strlen(name) + 1;
1341 	itx_t *itx;
1342 
1343 	if (zil_replaying(zd->zd_zilog, tx))
1344 		return;
1345 
1346 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1347 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1348 	    sizeof (*lr) + namesize - sizeof (lr_t));
1349 
1350 	itx->itx_oid = object;
1351 	zil_itx_assign(zd->zd_zilog, itx, tx);
1352 }
1353 
1354 static void
1355 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1356 {
1357 	itx_t *itx;
1358 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1359 
1360 	if (zil_replaying(zd->zd_zilog, tx))
1361 		return;
1362 
1363 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1364 		write_state = WR_INDIRECT;
1365 
1366 	itx = zil_itx_create(TX_WRITE,
1367 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1368 
1369 	if (write_state == WR_COPIED &&
1370 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1371 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1372 		zil_itx_destroy(itx);
1373 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1374 		write_state = WR_NEED_COPY;
1375 	}
1376 	itx->itx_private = zd;
1377 	itx->itx_wr_state = write_state;
1378 	itx->itx_sync = (ztest_random(8) == 0);
1379 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1380 
1381 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1382 	    sizeof (*lr) - sizeof (lr_t));
1383 
1384 	zil_itx_assign(zd->zd_zilog, itx, tx);
1385 }
1386 
1387 static void
1388 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1389 {
1390 	itx_t *itx;
1391 
1392 	if (zil_replaying(zd->zd_zilog, tx))
1393 		return;
1394 
1395 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1396 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1397 	    sizeof (*lr) - sizeof (lr_t));
1398 
1399 	itx->itx_sync = B_FALSE;
1400 	zil_itx_assign(zd->zd_zilog, itx, tx);
1401 }
1402 
1403 static void
1404 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1405 {
1406 	itx_t *itx;
1407 
1408 	if (zil_replaying(zd->zd_zilog, tx))
1409 		return;
1410 
1411 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1412 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1413 	    sizeof (*lr) - sizeof (lr_t));
1414 
1415 	itx->itx_sync = B_FALSE;
1416 	zil_itx_assign(zd->zd_zilog, itx, tx);
1417 }
1418 
1419 /*
1420  * ZIL replay ops
1421  */
1422 static int
1423 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1424 {
1425 	char *name = (void *)(lr + 1);		/* name follows lr */
1426 	objset_t *os = zd->zd_os;
1427 	ztest_block_tag_t *bbt;
1428 	dmu_buf_t *db;
1429 	dmu_tx_t *tx;
1430 	uint64_t txg;
1431 	int error = 0;
1432 
1433 	if (byteswap)
1434 		byteswap_uint64_array(lr, sizeof (*lr));
1435 
1436 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1437 	ASSERT(name[0] != '\0');
1438 
1439 	tx = dmu_tx_create(os);
1440 
1441 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1442 
1443 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1444 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1445 	} else {
1446 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1447 	}
1448 
1449 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1450 	if (txg == 0)
1451 		return (ENOSPC);
1452 
1453 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1454 
1455 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1456 		if (lr->lr_foid == 0) {
1457 			lr->lr_foid = zap_create(os,
1458 			    lr->lrz_type, lr->lrz_bonustype,
1459 			    lr->lrz_bonuslen, tx);
1460 		} else {
1461 			error = zap_create_claim(os, lr->lr_foid,
1462 			    lr->lrz_type, lr->lrz_bonustype,
1463 			    lr->lrz_bonuslen, tx);
1464 		}
1465 	} else {
1466 		if (lr->lr_foid == 0) {
1467 			lr->lr_foid = dmu_object_alloc(os,
1468 			    lr->lrz_type, 0, lr->lrz_bonustype,
1469 			    lr->lrz_bonuslen, tx);
1470 		} else {
1471 			error = dmu_object_claim(os, lr->lr_foid,
1472 			    lr->lrz_type, 0, lr->lrz_bonustype,
1473 			    lr->lrz_bonuslen, tx);
1474 		}
1475 	}
1476 
1477 	if (error) {
1478 		ASSERT3U(error, ==, EEXIST);
1479 		ASSERT(zd->zd_zilog->zl_replay);
1480 		dmu_tx_commit(tx);
1481 		return (error);
1482 	}
1483 
1484 	ASSERT(lr->lr_foid != 0);
1485 
1486 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1487 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1488 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1489 
1490 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1491 	bbt = ztest_bt_bonus(db);
1492 	dmu_buf_will_dirty(db, tx);
1493 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1494 	dmu_buf_rele(db, FTAG);
1495 
1496 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1497 	    &lr->lr_foid, tx));
1498 
1499 	(void) ztest_log_create(zd, tx, lr);
1500 
1501 	dmu_tx_commit(tx);
1502 
1503 	return (0);
1504 }
1505 
1506 static int
1507 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1508 {
1509 	char *name = (void *)(lr + 1);		/* name follows lr */
1510 	objset_t *os = zd->zd_os;
1511 	dmu_object_info_t doi;
1512 	dmu_tx_t *tx;
1513 	uint64_t object, txg;
1514 
1515 	if (byteswap)
1516 		byteswap_uint64_array(lr, sizeof (*lr));
1517 
1518 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1519 	ASSERT(name[0] != '\0');
1520 
1521 	VERIFY3U(0, ==,
1522 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1523 	ASSERT(object != 0);
1524 
1525 	ztest_object_lock(zd, object, RL_WRITER);
1526 
1527 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1528 
1529 	tx = dmu_tx_create(os);
1530 
1531 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1532 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1533 
1534 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1535 	if (txg == 0) {
1536 		ztest_object_unlock(zd, object);
1537 		return (ENOSPC);
1538 	}
1539 
1540 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1541 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1542 	} else {
1543 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1544 	}
1545 
1546 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1547 
1548 	(void) ztest_log_remove(zd, tx, lr, object);
1549 
1550 	dmu_tx_commit(tx);
1551 
1552 	ztest_object_unlock(zd, object);
1553 
1554 	return (0);
1555 }
1556 
1557 static int
1558 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1559 {
1560 	objset_t *os = zd->zd_os;
1561 	void *data = lr + 1;			/* data follows lr */
1562 	uint64_t offset, length;
1563 	ztest_block_tag_t *bt = data;
1564 	ztest_block_tag_t *bbt;
1565 	uint64_t gen, txg, lrtxg, crtxg;
1566 	dmu_object_info_t doi;
1567 	dmu_tx_t *tx;
1568 	dmu_buf_t *db;
1569 	arc_buf_t *abuf = NULL;
1570 	rl_t *rl;
1571 
1572 	if (byteswap)
1573 		byteswap_uint64_array(lr, sizeof (*lr));
1574 
1575 	offset = lr->lr_offset;
1576 	length = lr->lr_length;
1577 
1578 	/* If it's a dmu_sync() block, write the whole block */
1579 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1580 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1581 		if (length < blocksize) {
1582 			offset -= offset % blocksize;
1583 			length = blocksize;
1584 		}
1585 	}
1586 
1587 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1588 		byteswap_uint64_array(bt, sizeof (*bt));
1589 
1590 	if (bt->bt_magic != BT_MAGIC)
1591 		bt = NULL;
1592 
1593 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1594 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1595 
1596 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1597 
1598 	dmu_object_info_from_db(db, &doi);
1599 
1600 	bbt = ztest_bt_bonus(db);
1601 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1602 	gen = bbt->bt_gen;
1603 	crtxg = bbt->bt_crtxg;
1604 	lrtxg = lr->lr_common.lrc_txg;
1605 
1606 	tx = dmu_tx_create(os);
1607 
1608 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1609 
1610 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1611 	    P2PHASE(offset, length) == 0)
1612 		abuf = dmu_request_arcbuf(db, length);
1613 
1614 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1615 	if (txg == 0) {
1616 		if (abuf != NULL)
1617 			dmu_return_arcbuf(abuf);
1618 		dmu_buf_rele(db, FTAG);
1619 		ztest_range_unlock(rl);
1620 		ztest_object_unlock(zd, lr->lr_foid);
1621 		return (ENOSPC);
1622 	}
1623 
1624 	if (bt != NULL) {
1625 		/*
1626 		 * Usually, verify the old data before writing new data --
1627 		 * but not always, because we also want to verify correct
1628 		 * behavior when the data was not recently read into cache.
1629 		 */
1630 		ASSERT(offset % doi.doi_data_block_size == 0);
1631 		if (ztest_random(4) != 0) {
1632 			int prefetch = ztest_random(2) ?
1633 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1634 			ztest_block_tag_t rbt;
1635 
1636 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1637 			    sizeof (rbt), &rbt, prefetch) == 0);
1638 			if (rbt.bt_magic == BT_MAGIC) {
1639 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1640 				    offset, gen, txg, crtxg);
1641 			}
1642 		}
1643 
1644 		/*
1645 		 * Writes can appear to be newer than the bonus buffer because
1646 		 * the ztest_get_data() callback does a dmu_read() of the
1647 		 * open-context data, which may be different than the data
1648 		 * as it was when the write was generated.
1649 		 */
1650 		if (zd->zd_zilog->zl_replay) {
1651 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1652 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1653 			    bt->bt_crtxg);
1654 		}
1655 
1656 		/*
1657 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1658 		 * so that all of the usual ASSERTs will work.
1659 		 */
1660 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1661 	}
1662 
1663 	if (abuf == NULL) {
1664 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1665 	} else {
1666 		bcopy(data, abuf->b_data, length);
1667 		dmu_assign_arcbuf(db, offset, abuf, tx);
1668 	}
1669 
1670 	(void) ztest_log_write(zd, tx, lr);
1671 
1672 	dmu_buf_rele(db, FTAG);
1673 
1674 	dmu_tx_commit(tx);
1675 
1676 	ztest_range_unlock(rl);
1677 	ztest_object_unlock(zd, lr->lr_foid);
1678 
1679 	return (0);
1680 }
1681 
1682 static int
1683 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1684 {
1685 	objset_t *os = zd->zd_os;
1686 	dmu_tx_t *tx;
1687 	uint64_t txg;
1688 	rl_t *rl;
1689 
1690 	if (byteswap)
1691 		byteswap_uint64_array(lr, sizeof (*lr));
1692 
1693 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1694 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1695 	    RL_WRITER);
1696 
1697 	tx = dmu_tx_create(os);
1698 
1699 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1700 
1701 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1702 	if (txg == 0) {
1703 		ztest_range_unlock(rl);
1704 		ztest_object_unlock(zd, lr->lr_foid);
1705 		return (ENOSPC);
1706 	}
1707 
1708 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1709 	    lr->lr_length, tx) == 0);
1710 
1711 	(void) ztest_log_truncate(zd, tx, lr);
1712 
1713 	dmu_tx_commit(tx);
1714 
1715 	ztest_range_unlock(rl);
1716 	ztest_object_unlock(zd, lr->lr_foid);
1717 
1718 	return (0);
1719 }
1720 
1721 static int
1722 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1723 {
1724 	objset_t *os = zd->zd_os;
1725 	dmu_tx_t *tx;
1726 	dmu_buf_t *db;
1727 	ztest_block_tag_t *bbt;
1728 	uint64_t txg, lrtxg, crtxg;
1729 
1730 	if (byteswap)
1731 		byteswap_uint64_array(lr, sizeof (*lr));
1732 
1733 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1734 
1735 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1736 
1737 	tx = dmu_tx_create(os);
1738 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1739 
1740 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1741 	if (txg == 0) {
1742 		dmu_buf_rele(db, FTAG);
1743 		ztest_object_unlock(zd, lr->lr_foid);
1744 		return (ENOSPC);
1745 	}
1746 
1747 	bbt = ztest_bt_bonus(db);
1748 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1749 	crtxg = bbt->bt_crtxg;
1750 	lrtxg = lr->lr_common.lrc_txg;
1751 
1752 	if (zd->zd_zilog->zl_replay) {
1753 		ASSERT(lr->lr_size != 0);
1754 		ASSERT(lr->lr_mode != 0);
1755 		ASSERT(lrtxg != 0);
1756 	} else {
1757 		/*
1758 		 * Randomly change the size and increment the generation.
1759 		 */
1760 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1761 		    sizeof (*bbt);
1762 		lr->lr_mode = bbt->bt_gen + 1;
1763 		ASSERT(lrtxg == 0);
1764 	}
1765 
1766 	/*
1767 	 * Verify that the current bonus buffer is not newer than our txg.
1768 	 */
1769 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1770 	    MAX(txg, lrtxg), crtxg);
1771 
1772 	dmu_buf_will_dirty(db, tx);
1773 
1774 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1775 	ASSERT3U(lr->lr_size, <=, db->db_size);
1776 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1777 	bbt = ztest_bt_bonus(db);
1778 
1779 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1780 
1781 	dmu_buf_rele(db, FTAG);
1782 
1783 	(void) ztest_log_setattr(zd, tx, lr);
1784 
1785 	dmu_tx_commit(tx);
1786 
1787 	ztest_object_unlock(zd, lr->lr_foid);
1788 
1789 	return (0);
1790 }
1791 
1792 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1793 	NULL,			/* 0 no such transaction type */
1794 	ztest_replay_create,	/* TX_CREATE */
1795 	NULL,			/* TX_MKDIR */
1796 	NULL,			/* TX_MKXATTR */
1797 	NULL,			/* TX_SYMLINK */
1798 	ztest_replay_remove,	/* TX_REMOVE */
1799 	NULL,			/* TX_RMDIR */
1800 	NULL,			/* TX_LINK */
1801 	NULL,			/* TX_RENAME */
1802 	ztest_replay_write,	/* TX_WRITE */
1803 	ztest_replay_truncate,	/* TX_TRUNCATE */
1804 	ztest_replay_setattr,	/* TX_SETATTR */
1805 	NULL,			/* TX_ACL */
1806 	NULL,			/* TX_CREATE_ACL */
1807 	NULL,			/* TX_CREATE_ATTR */
1808 	NULL,			/* TX_CREATE_ACL_ATTR */
1809 	NULL,			/* TX_MKDIR_ACL */
1810 	NULL,			/* TX_MKDIR_ATTR */
1811 	NULL,			/* TX_MKDIR_ACL_ATTR */
1812 	NULL,			/* TX_WRITE2 */
1813 };
1814 
1815 /*
1816  * ZIL get_data callbacks
1817  */
1818 
1819 static void
1820 ztest_get_done(zgd_t *zgd, int error)
1821 {
1822 	ztest_ds_t *zd = zgd->zgd_private;
1823 	uint64_t object = zgd->zgd_rl->rl_object;
1824 
1825 	if (zgd->zgd_db)
1826 		dmu_buf_rele(zgd->zgd_db, zgd);
1827 
1828 	ztest_range_unlock(zgd->zgd_rl);
1829 	ztest_object_unlock(zd, object);
1830 
1831 	if (error == 0 && zgd->zgd_bp)
1832 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1833 
1834 	umem_free(zgd, sizeof (*zgd));
1835 }
1836 
1837 static int
1838 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1839 {
1840 	ztest_ds_t *zd = arg;
1841 	objset_t *os = zd->zd_os;
1842 	uint64_t object = lr->lr_foid;
1843 	uint64_t offset = lr->lr_offset;
1844 	uint64_t size = lr->lr_length;
1845 	blkptr_t *bp = &lr->lr_blkptr;
1846 	uint64_t txg = lr->lr_common.lrc_txg;
1847 	uint64_t crtxg;
1848 	dmu_object_info_t doi;
1849 	dmu_buf_t *db;
1850 	zgd_t *zgd;
1851 	int error;
1852 
1853 	ztest_object_lock(zd, object, RL_READER);
1854 	error = dmu_bonus_hold(os, object, FTAG, &db);
1855 	if (error) {
1856 		ztest_object_unlock(zd, object);
1857 		return (error);
1858 	}
1859 
1860 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1861 
1862 	if (crtxg == 0 || crtxg > txg) {
1863 		dmu_buf_rele(db, FTAG);
1864 		ztest_object_unlock(zd, object);
1865 		return (ENOENT);
1866 	}
1867 
1868 	dmu_object_info_from_db(db, &doi);
1869 	dmu_buf_rele(db, FTAG);
1870 	db = NULL;
1871 
1872 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1873 	zgd->zgd_zilog = zd->zd_zilog;
1874 	zgd->zgd_private = zd;
1875 
1876 	if (buf != NULL) {	/* immediate write */
1877 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1878 		    RL_READER);
1879 
1880 		error = dmu_read(os, object, offset, size, buf,
1881 		    DMU_READ_NO_PREFETCH);
1882 		ASSERT(error == 0);
1883 	} else {
1884 		size = doi.doi_data_block_size;
1885 		if (ISP2(size)) {
1886 			offset = P2ALIGN(offset, size);
1887 		} else {
1888 			ASSERT(offset < size);
1889 			offset = 0;
1890 		}
1891 
1892 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1893 		    RL_READER);
1894 
1895 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1896 		    DMU_READ_NO_PREFETCH);
1897 
1898 		if (error == 0) {
1899 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1900 			if (obp) {
1901 				ASSERT(BP_IS_HOLE(bp));
1902 				*bp = *obp;
1903 			}
1904 
1905 			zgd->zgd_db = db;
1906 			zgd->zgd_bp = bp;
1907 
1908 			ASSERT(db->db_offset == offset);
1909 			ASSERT(db->db_size == size);
1910 
1911 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1912 			    ztest_get_done, zgd);
1913 
1914 			if (error == 0)
1915 				return (0);
1916 		}
1917 	}
1918 
1919 	ztest_get_done(zgd, error);
1920 
1921 	return (error);
1922 }
1923 
1924 static void *
1925 ztest_lr_alloc(size_t lrsize, char *name)
1926 {
1927 	char *lr;
1928 	size_t namesize = name ? strlen(name) + 1 : 0;
1929 
1930 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1931 
1932 	if (name)
1933 		bcopy(name, lr + lrsize, namesize);
1934 
1935 	return (lr);
1936 }
1937 
1938 void
1939 ztest_lr_free(void *lr, size_t lrsize, char *name)
1940 {
1941 	size_t namesize = name ? strlen(name) + 1 : 0;
1942 
1943 	umem_free(lr, lrsize + namesize);
1944 }
1945 
1946 /*
1947  * Lookup a bunch of objects.  Returns the number of objects not found.
1948  */
1949 static int
1950 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1951 {
1952 	int missing = 0;
1953 	int error;
1954 
1955 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1956 
1957 	for (int i = 0; i < count; i++, od++) {
1958 		od->od_object = 0;
1959 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1960 		    sizeof (uint64_t), 1, &od->od_object);
1961 		if (error) {
1962 			ASSERT(error == ENOENT);
1963 			ASSERT(od->od_object == 0);
1964 			missing++;
1965 		} else {
1966 			dmu_buf_t *db;
1967 			ztest_block_tag_t *bbt;
1968 			dmu_object_info_t doi;
1969 
1970 			ASSERT(od->od_object != 0);
1971 			ASSERT(missing == 0);	/* there should be no gaps */
1972 
1973 			ztest_object_lock(zd, od->od_object, RL_READER);
1974 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1975 			    od->od_object, FTAG, &db));
1976 			dmu_object_info_from_db(db, &doi);
1977 			bbt = ztest_bt_bonus(db);
1978 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1979 			od->od_type = doi.doi_type;
1980 			od->od_blocksize = doi.doi_data_block_size;
1981 			od->od_gen = bbt->bt_gen;
1982 			dmu_buf_rele(db, FTAG);
1983 			ztest_object_unlock(zd, od->od_object);
1984 		}
1985 	}
1986 
1987 	return (missing);
1988 }
1989 
1990 static int
1991 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1992 {
1993 	int missing = 0;
1994 
1995 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1996 
1997 	for (int i = 0; i < count; i++, od++) {
1998 		if (missing) {
1999 			od->od_object = 0;
2000 			missing++;
2001 			continue;
2002 		}
2003 
2004 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2005 
2006 		lr->lr_doid = od->od_dir;
2007 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2008 		lr->lrz_type = od->od_crtype;
2009 		lr->lrz_blocksize = od->od_crblocksize;
2010 		lr->lrz_ibshift = ztest_random_ibshift();
2011 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2012 		lr->lrz_bonuslen = dmu_bonus_max();
2013 		lr->lr_gen = od->od_crgen;
2014 		lr->lr_crtime[0] = time(NULL);
2015 
2016 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2017 			ASSERT(missing == 0);
2018 			od->od_object = 0;
2019 			missing++;
2020 		} else {
2021 			od->od_object = lr->lr_foid;
2022 			od->od_type = od->od_crtype;
2023 			od->od_blocksize = od->od_crblocksize;
2024 			od->od_gen = od->od_crgen;
2025 			ASSERT(od->od_object != 0);
2026 		}
2027 
2028 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2029 	}
2030 
2031 	return (missing);
2032 }
2033 
2034 static int
2035 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2036 {
2037 	int missing = 0;
2038 	int error;
2039 
2040 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2041 
2042 	od += count - 1;
2043 
2044 	for (int i = count - 1; i >= 0; i--, od--) {
2045 		if (missing) {
2046 			missing++;
2047 			continue;
2048 		}
2049 
2050 		/*
2051 		 * No object was found.
2052 		 */
2053 		if (od->od_object == 0)
2054 			continue;
2055 
2056 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2057 
2058 		lr->lr_doid = od->od_dir;
2059 
2060 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2061 			ASSERT3U(error, ==, ENOSPC);
2062 			missing++;
2063 		} else {
2064 			od->od_object = 0;
2065 		}
2066 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2067 	}
2068 
2069 	return (missing);
2070 }
2071 
2072 static int
2073 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2074     void *data)
2075 {
2076 	lr_write_t *lr;
2077 	int error;
2078 
2079 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2080 
2081 	lr->lr_foid = object;
2082 	lr->lr_offset = offset;
2083 	lr->lr_length = size;
2084 	lr->lr_blkoff = 0;
2085 	BP_ZERO(&lr->lr_blkptr);
2086 
2087 	bcopy(data, lr + 1, size);
2088 
2089 	error = ztest_replay_write(zd, lr, B_FALSE);
2090 
2091 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2092 
2093 	return (error);
2094 }
2095 
2096 static int
2097 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2098 {
2099 	lr_truncate_t *lr;
2100 	int error;
2101 
2102 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2103 
2104 	lr->lr_foid = object;
2105 	lr->lr_offset = offset;
2106 	lr->lr_length = size;
2107 
2108 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2109 
2110 	ztest_lr_free(lr, sizeof (*lr), NULL);
2111 
2112 	return (error);
2113 }
2114 
2115 static int
2116 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2117 {
2118 	lr_setattr_t *lr;
2119 	int error;
2120 
2121 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2122 
2123 	lr->lr_foid = object;
2124 	lr->lr_size = 0;
2125 	lr->lr_mode = 0;
2126 
2127 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2128 
2129 	ztest_lr_free(lr, sizeof (*lr), NULL);
2130 
2131 	return (error);
2132 }
2133 
2134 static void
2135 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2136 {
2137 	objset_t *os = zd->zd_os;
2138 	dmu_tx_t *tx;
2139 	uint64_t txg;
2140 	rl_t *rl;
2141 
2142 	txg_wait_synced(dmu_objset_pool(os), 0);
2143 
2144 	ztest_object_lock(zd, object, RL_READER);
2145 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2146 
2147 	tx = dmu_tx_create(os);
2148 
2149 	dmu_tx_hold_write(tx, object, offset, size);
2150 
2151 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2152 
2153 	if (txg != 0) {
2154 		dmu_prealloc(os, object, offset, size, tx);
2155 		dmu_tx_commit(tx);
2156 		txg_wait_synced(dmu_objset_pool(os), txg);
2157 	} else {
2158 		(void) dmu_free_long_range(os, object, offset, size);
2159 	}
2160 
2161 	ztest_range_unlock(rl);
2162 	ztest_object_unlock(zd, object);
2163 }
2164 
2165 static void
2166 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2167 {
2168 	int err;
2169 	ztest_block_tag_t wbt;
2170 	dmu_object_info_t doi;
2171 	enum ztest_io_type io_type;
2172 	uint64_t blocksize;
2173 	void *data;
2174 
2175 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2176 	blocksize = doi.doi_data_block_size;
2177 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2178 
2179 	/*
2180 	 * Pick an i/o type at random, biased toward writing block tags.
2181 	 */
2182 	io_type = ztest_random(ZTEST_IO_TYPES);
2183 	if (ztest_random(2) == 0)
2184 		io_type = ZTEST_IO_WRITE_TAG;
2185 
2186 	(void) rw_rdlock(&zd->zd_zilog_lock);
2187 
2188 	switch (io_type) {
2189 
2190 	case ZTEST_IO_WRITE_TAG:
2191 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2192 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2193 		break;
2194 
2195 	case ZTEST_IO_WRITE_PATTERN:
2196 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2197 		if (ztest_random(2) == 0) {
2198 			/*
2199 			 * Induce fletcher2 collisions to ensure that
2200 			 * zio_ddt_collision() detects and resolves them
2201 			 * when using fletcher2-verify for deduplication.
2202 			 */
2203 			((uint64_t *)data)[0] ^= 1ULL << 63;
2204 			((uint64_t *)data)[4] ^= 1ULL << 63;
2205 		}
2206 		(void) ztest_write(zd, object, offset, blocksize, data);
2207 		break;
2208 
2209 	case ZTEST_IO_WRITE_ZEROES:
2210 		bzero(data, blocksize);
2211 		(void) ztest_write(zd, object, offset, blocksize, data);
2212 		break;
2213 
2214 	case ZTEST_IO_TRUNCATE:
2215 		(void) ztest_truncate(zd, object, offset, blocksize);
2216 		break;
2217 
2218 	case ZTEST_IO_SETATTR:
2219 		(void) ztest_setattr(zd, object);
2220 		break;
2221 
2222 	case ZTEST_IO_REWRITE:
2223 		(void) rw_rdlock(&ztest_name_lock);
2224 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2225 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2226 		    B_FALSE);
2227 		VERIFY(err == 0 || err == ENOSPC);
2228 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2229 		    ZFS_PROP_COMPRESSION,
2230 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2231 		    B_FALSE);
2232 		VERIFY(err == 0 || err == ENOSPC);
2233 		(void) rw_unlock(&ztest_name_lock);
2234 
2235 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2236 		    DMU_READ_NO_PREFETCH));
2237 
2238 		(void) ztest_write(zd, object, offset, blocksize, data);
2239 		break;
2240 	}
2241 
2242 	(void) rw_unlock(&zd->zd_zilog_lock);
2243 
2244 	umem_free(data, blocksize);
2245 }
2246 
2247 /*
2248  * Initialize an object description template.
2249  */
2250 static void
2251 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2252     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2253 {
2254 	od->od_dir = ZTEST_DIROBJ;
2255 	od->od_object = 0;
2256 
2257 	od->od_crtype = type;
2258 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2259 	od->od_crgen = gen;
2260 
2261 	od->od_type = DMU_OT_NONE;
2262 	od->od_blocksize = 0;
2263 	od->od_gen = 0;
2264 
2265 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2266 	    tag, (int64_t)id, index);
2267 }
2268 
2269 /*
2270  * Lookup or create the objects for a test using the od template.
2271  * If the objects do not all exist, or if 'remove' is specified,
2272  * remove any existing objects and create new ones.  Otherwise,
2273  * use the existing objects.
2274  */
2275 static int
2276 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2277 {
2278 	int count = size / sizeof (*od);
2279 	int rv = 0;
2280 
2281 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2282 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2283 	    (ztest_remove(zd, od, count) != 0 ||
2284 	    ztest_create(zd, od, count) != 0))
2285 		rv = -1;
2286 	zd->zd_od = od;
2287 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2288 
2289 	return (rv);
2290 }
2291 
2292 /* ARGSUSED */
2293 void
2294 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2295 {
2296 	zilog_t *zilog = zd->zd_zilog;
2297 
2298 	(void) rw_rdlock(&zd->zd_zilog_lock);
2299 
2300 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2301 
2302 	/*
2303 	 * Remember the committed values in zd, which is in parent/child
2304 	 * shared memory.  If we die, the next iteration of ztest_run()
2305 	 * will verify that the log really does contain this record.
2306 	 */
2307 	mutex_enter(&zilog->zl_lock);
2308 	ASSERT(zd->zd_shared != NULL);
2309 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2310 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2311 	mutex_exit(&zilog->zl_lock);
2312 
2313 	(void) rw_unlock(&zd->zd_zilog_lock);
2314 }
2315 
2316 /*
2317  * This function is designed to simulate the operations that occur during a
2318  * mount/unmount operation.  We hold the dataset across these operations in an
2319  * attempt to expose any implicit assumptions about ZIL management.
2320  */
2321 /* ARGSUSED */
2322 void
2323 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2324 {
2325 	objset_t *os = zd->zd_os;
2326 
2327 	/*
2328 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2329 	 * updating the zil (i.e. adding in-memory log records) and the
2330 	 * zd_zilog_lock to block any I/O.
2331 	 */
2332 	VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2333 	(void) rw_wrlock(&zd->zd_zilog_lock);
2334 
2335 	/* zfsvfs_teardown() */
2336 	zil_close(zd->zd_zilog);
2337 
2338 	/* zfsvfs_setup() */
2339 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2340 	zil_replay(os, zd, ztest_replay_vector);
2341 
2342 	(void) rw_unlock(&zd->zd_zilog_lock);
2343 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2344 }
2345 
2346 /*
2347  * Verify that we can't destroy an active pool, create an existing pool,
2348  * or create a pool with a bad vdev spec.
2349  */
2350 /* ARGSUSED */
2351 void
2352 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2353 {
2354 	ztest_shared_opts_t *zo = &ztest_opts;
2355 	spa_t *spa;
2356 	nvlist_t *nvroot;
2357 
2358 	/*
2359 	 * Attempt to create using a bad file.
2360 	 */
2361 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2362 	VERIFY3U(ENOENT, ==,
2363 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2364 	nvlist_free(nvroot);
2365 
2366 	/*
2367 	 * Attempt to create using a bad mirror.
2368 	 */
2369 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2370 	VERIFY3U(ENOENT, ==,
2371 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2372 	nvlist_free(nvroot);
2373 
2374 	/*
2375 	 * Attempt to create an existing pool.  It shouldn't matter
2376 	 * what's in the nvroot; we should fail with EEXIST.
2377 	 */
2378 	(void) rw_rdlock(&ztest_name_lock);
2379 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2380 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2381 	nvlist_free(nvroot);
2382 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2383 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2384 	spa_close(spa, FTAG);
2385 
2386 	(void) rw_unlock(&ztest_name_lock);
2387 }
2388 
2389 /* ARGSUSED */
2390 void
2391 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2392 {
2393 	spa_t *spa;
2394 	uint64_t initial_version = SPA_VERSION_INITIAL;
2395 	uint64_t version, newversion;
2396 	nvlist_t *nvroot, *props;
2397 	char *name;
2398 
2399 	VERIFY0(mutex_lock(&ztest_vdev_lock));
2400 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2401 
2402 	/*
2403 	 * Clean up from previous runs.
2404 	 */
2405 	(void) spa_destroy(name);
2406 
2407 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2408 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2409 
2410 	/*
2411 	 * If we're configuring a RAIDZ device then make sure that the
2412 	 * the initial version is capable of supporting that feature.
2413 	 */
2414 	switch (ztest_opts.zo_raidz_parity) {
2415 	case 0:
2416 	case 1:
2417 		initial_version = SPA_VERSION_INITIAL;
2418 		break;
2419 	case 2:
2420 		initial_version = SPA_VERSION_RAIDZ2;
2421 		break;
2422 	case 3:
2423 		initial_version = SPA_VERSION_RAIDZ3;
2424 		break;
2425 	}
2426 
2427 	/*
2428 	 * Create a pool with a spa version that can be upgraded. Pick
2429 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2430 	 */
2431 	do {
2432 		version = ztest_random_spa_version(initial_version);
2433 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2434 
2435 	props = fnvlist_alloc();
2436 	fnvlist_add_uint64(props,
2437 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2438 	VERIFY0(spa_create(name, nvroot, props, NULL));
2439 	fnvlist_free(nvroot);
2440 	fnvlist_free(props);
2441 
2442 	VERIFY0(spa_open(name, &spa, FTAG));
2443 	VERIFY3U(spa_version(spa), ==, version);
2444 	newversion = ztest_random_spa_version(version + 1);
2445 
2446 	if (ztest_opts.zo_verbose >= 4) {
2447 		(void) printf("upgrading spa version from %llu to %llu\n",
2448 		    (u_longlong_t)version, (u_longlong_t)newversion);
2449 	}
2450 
2451 	spa_upgrade(spa, newversion);
2452 	VERIFY3U(spa_version(spa), >, version);
2453 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2454 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2455 	spa_close(spa, FTAG);
2456 
2457 	strfree(name);
2458 	VERIFY0(mutex_unlock(&ztest_vdev_lock));
2459 }
2460 
2461 static vdev_t *
2462 vdev_lookup_by_path(vdev_t *vd, const char *path)
2463 {
2464 	vdev_t *mvd;
2465 
2466 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2467 		return (vd);
2468 
2469 	for (int c = 0; c < vd->vdev_children; c++)
2470 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2471 		    NULL)
2472 			return (mvd);
2473 
2474 	return (NULL);
2475 }
2476 
2477 /*
2478  * Find the first available hole which can be used as a top-level.
2479  */
2480 int
2481 find_vdev_hole(spa_t *spa)
2482 {
2483 	vdev_t *rvd = spa->spa_root_vdev;
2484 	int c;
2485 
2486 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2487 
2488 	for (c = 0; c < rvd->vdev_children; c++) {
2489 		vdev_t *cvd = rvd->vdev_child[c];
2490 
2491 		if (cvd->vdev_ishole)
2492 			break;
2493 	}
2494 	return (c);
2495 }
2496 
2497 /*
2498  * Verify that vdev_add() works as expected.
2499  */
2500 /* ARGSUSED */
2501 void
2502 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2503 {
2504 	ztest_shared_t *zs = ztest_shared;
2505 	spa_t *spa = ztest_spa;
2506 	uint64_t leaves;
2507 	uint64_t guid;
2508 	nvlist_t *nvroot;
2509 	int error;
2510 
2511 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2512 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2513 
2514 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2515 
2516 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2517 
2518 	/*
2519 	 * If we have slogs then remove them 1/4 of the time.
2520 	 */
2521 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2522 		/*
2523 		 * Grab the guid from the head of the log class rotor.
2524 		 */
2525 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2526 
2527 		spa_config_exit(spa, SCL_VDEV, FTAG);
2528 
2529 		/*
2530 		 * We have to grab the zs_name_lock as writer to
2531 		 * prevent a race between removing a slog (dmu_objset_find)
2532 		 * and destroying a dataset. Removing the slog will
2533 		 * grab a reference on the dataset which may cause
2534 		 * dmu_objset_destroy() to fail with EBUSY thus
2535 		 * leaving the dataset in an inconsistent state.
2536 		 */
2537 		VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2538 		error = spa_vdev_remove(spa, guid, B_FALSE);
2539 		VERIFY(rw_unlock(&ztest_name_lock) == 0);
2540 
2541 		if (error && error != EEXIST)
2542 			fatal(0, "spa_vdev_remove() = %d", error);
2543 	} else {
2544 		spa_config_exit(spa, SCL_VDEV, FTAG);
2545 
2546 		/*
2547 		 * Make 1/4 of the devices be log devices.
2548 		 */
2549 		nvroot = make_vdev_root(NULL, NULL, NULL,
2550 		    ztest_opts.zo_vdev_size, 0,
2551 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2552 		    zs->zs_mirrors, 1);
2553 
2554 		error = spa_vdev_add(spa, nvroot);
2555 		nvlist_free(nvroot);
2556 
2557 		if (error == ENOSPC)
2558 			ztest_record_enospc("spa_vdev_add");
2559 		else if (error != 0)
2560 			fatal(0, "spa_vdev_add() = %d", error);
2561 	}
2562 
2563 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2564 }
2565 
2566 /*
2567  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2568  */
2569 /* ARGSUSED */
2570 void
2571 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2572 {
2573 	ztest_shared_t *zs = ztest_shared;
2574 	spa_t *spa = ztest_spa;
2575 	vdev_t *rvd = spa->spa_root_vdev;
2576 	spa_aux_vdev_t *sav;
2577 	char *aux;
2578 	uint64_t guid = 0;
2579 	int error;
2580 
2581 	if (ztest_random(2) == 0) {
2582 		sav = &spa->spa_spares;
2583 		aux = ZPOOL_CONFIG_SPARES;
2584 	} else {
2585 		sav = &spa->spa_l2cache;
2586 		aux = ZPOOL_CONFIG_L2CACHE;
2587 	}
2588 
2589 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2590 
2591 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2592 
2593 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2594 		/*
2595 		 * Pick a random device to remove.
2596 		 */
2597 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2598 	} else {
2599 		/*
2600 		 * Find an unused device we can add.
2601 		 */
2602 		zs->zs_vdev_aux = 0;
2603 		for (;;) {
2604 			char path[MAXPATHLEN];
2605 			int c;
2606 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2607 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2608 			    zs->zs_vdev_aux);
2609 			for (c = 0; c < sav->sav_count; c++)
2610 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2611 				    path) == 0)
2612 					break;
2613 			if (c == sav->sav_count &&
2614 			    vdev_lookup_by_path(rvd, path) == NULL)
2615 				break;
2616 			zs->zs_vdev_aux++;
2617 		}
2618 	}
2619 
2620 	spa_config_exit(spa, SCL_VDEV, FTAG);
2621 
2622 	if (guid == 0) {
2623 		/*
2624 		 * Add a new device.
2625 		 */
2626 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2627 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2628 		error = spa_vdev_add(spa, nvroot);
2629 		if (error != 0)
2630 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2631 		nvlist_free(nvroot);
2632 	} else {
2633 		/*
2634 		 * Remove an existing device.  Sometimes, dirty its
2635 		 * vdev state first to make sure we handle removal
2636 		 * of devices that have pending state changes.
2637 		 */
2638 		if (ztest_random(2) == 0)
2639 			(void) vdev_online(spa, guid, 0, NULL);
2640 
2641 		error = spa_vdev_remove(spa, guid, B_FALSE);
2642 		if (error != 0 && error != EBUSY)
2643 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2644 	}
2645 
2646 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2647 }
2648 
2649 /*
2650  * split a pool if it has mirror tlvdevs
2651  */
2652 /* ARGSUSED */
2653 void
2654 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2655 {
2656 	ztest_shared_t *zs = ztest_shared;
2657 	spa_t *spa = ztest_spa;
2658 	vdev_t *rvd = spa->spa_root_vdev;
2659 	nvlist_t *tree, **child, *config, *split, **schild;
2660 	uint_t c, children, schildren = 0, lastlogid = 0;
2661 	int error = 0;
2662 
2663 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2664 
2665 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2666 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2667 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2668 		return;
2669 	}
2670 
2671 	/* clean up the old pool, if any */
2672 	(void) spa_destroy("splitp");
2673 
2674 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2675 
2676 	/* generate a config from the existing config */
2677 	mutex_enter(&spa->spa_props_lock);
2678 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2679 	    &tree) == 0);
2680 	mutex_exit(&spa->spa_props_lock);
2681 
2682 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2683 	    &children) == 0);
2684 
2685 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2686 	for (c = 0; c < children; c++) {
2687 		vdev_t *tvd = rvd->vdev_child[c];
2688 		nvlist_t **mchild;
2689 		uint_t mchildren;
2690 
2691 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2692 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2693 			    0) == 0);
2694 			VERIFY(nvlist_add_string(schild[schildren],
2695 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2696 			VERIFY(nvlist_add_uint64(schild[schildren],
2697 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2698 			if (lastlogid == 0)
2699 				lastlogid = schildren;
2700 			++schildren;
2701 			continue;
2702 		}
2703 		lastlogid = 0;
2704 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2705 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2706 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2707 	}
2708 
2709 	/* OK, create a config that can be used to split */
2710 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2711 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2712 	    VDEV_TYPE_ROOT) == 0);
2713 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2714 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2715 
2716 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2717 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2718 
2719 	for (c = 0; c < schildren; c++)
2720 		nvlist_free(schild[c]);
2721 	free(schild);
2722 	nvlist_free(split);
2723 
2724 	spa_config_exit(spa, SCL_VDEV, FTAG);
2725 
2726 	(void) rw_wrlock(&ztest_name_lock);
2727 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2728 	(void) rw_unlock(&ztest_name_lock);
2729 
2730 	nvlist_free(config);
2731 
2732 	if (error == 0) {
2733 		(void) printf("successful split - results:\n");
2734 		mutex_enter(&spa_namespace_lock);
2735 		show_pool_stats(spa);
2736 		show_pool_stats(spa_lookup("splitp"));
2737 		mutex_exit(&spa_namespace_lock);
2738 		++zs->zs_splits;
2739 		--zs->zs_mirrors;
2740 	}
2741 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2742 
2743 }
2744 
2745 /*
2746  * Verify that we can attach and detach devices.
2747  */
2748 /* ARGSUSED */
2749 void
2750 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2751 {
2752 	ztest_shared_t *zs = ztest_shared;
2753 	spa_t *spa = ztest_spa;
2754 	spa_aux_vdev_t *sav = &spa->spa_spares;
2755 	vdev_t *rvd = spa->spa_root_vdev;
2756 	vdev_t *oldvd, *newvd, *pvd;
2757 	nvlist_t *root;
2758 	uint64_t leaves;
2759 	uint64_t leaf, top;
2760 	uint64_t ashift = ztest_get_ashift();
2761 	uint64_t oldguid, pguid;
2762 	uint64_t oldsize, newsize;
2763 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2764 	int replacing;
2765 	int oldvd_has_siblings = B_FALSE;
2766 	int newvd_is_spare = B_FALSE;
2767 	int oldvd_is_log;
2768 	int error, expected_error;
2769 
2770 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2771 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2772 
2773 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2774 
2775 	/*
2776 	 * Decide whether to do an attach or a replace.
2777 	 */
2778 	replacing = ztest_random(2);
2779 
2780 	/*
2781 	 * Pick a random top-level vdev.
2782 	 */
2783 	top = ztest_random_vdev_top(spa, B_TRUE);
2784 
2785 	/*
2786 	 * Pick a random leaf within it.
2787 	 */
2788 	leaf = ztest_random(leaves);
2789 
2790 	/*
2791 	 * Locate this vdev.
2792 	 */
2793 	oldvd = rvd->vdev_child[top];
2794 	if (zs->zs_mirrors >= 1) {
2795 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2796 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2797 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2798 	}
2799 	if (ztest_opts.zo_raidz > 1) {
2800 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2801 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2802 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2803 	}
2804 
2805 	/*
2806 	 * If we're already doing an attach or replace, oldvd may be a
2807 	 * mirror vdev -- in which case, pick a random child.
2808 	 */
2809 	while (oldvd->vdev_children != 0) {
2810 		oldvd_has_siblings = B_TRUE;
2811 		ASSERT(oldvd->vdev_children >= 2);
2812 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2813 	}
2814 
2815 	oldguid = oldvd->vdev_guid;
2816 	oldsize = vdev_get_min_asize(oldvd);
2817 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2818 	(void) strcpy(oldpath, oldvd->vdev_path);
2819 	pvd = oldvd->vdev_parent;
2820 	pguid = pvd->vdev_guid;
2821 
2822 	/*
2823 	 * If oldvd has siblings, then half of the time, detach it.
2824 	 */
2825 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2826 		spa_config_exit(spa, SCL_VDEV, FTAG);
2827 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2828 		if (error != 0 && error != ENODEV && error != EBUSY &&
2829 		    error != ENOTSUP)
2830 			fatal(0, "detach (%s) returned %d", oldpath, error);
2831 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2832 		return;
2833 	}
2834 
2835 	/*
2836 	 * For the new vdev, choose with equal probability between the two
2837 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2838 	 */
2839 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2840 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2841 		newvd_is_spare = B_TRUE;
2842 		(void) strcpy(newpath, newvd->vdev_path);
2843 	} else {
2844 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2845 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2846 		    top * leaves + leaf);
2847 		if (ztest_random(2) == 0)
2848 			newpath[strlen(newpath) - 1] = 'b';
2849 		newvd = vdev_lookup_by_path(rvd, newpath);
2850 	}
2851 
2852 	if (newvd) {
2853 		newsize = vdev_get_min_asize(newvd);
2854 	} else {
2855 		/*
2856 		 * Make newsize a little bigger or smaller than oldsize.
2857 		 * If it's smaller, the attach should fail.
2858 		 * If it's larger, and we're doing a replace,
2859 		 * we should get dynamic LUN growth when we're done.
2860 		 */
2861 		newsize = 10 * oldsize / (9 + ztest_random(3));
2862 	}
2863 
2864 	/*
2865 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2866 	 * unless it's a replace; in that case any non-replacing parent is OK.
2867 	 *
2868 	 * If newvd is already part of the pool, it should fail with EBUSY.
2869 	 *
2870 	 * If newvd is too small, it should fail with EOVERFLOW.
2871 	 */
2872 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2873 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2874 	    pvd->vdev_ops == &vdev_replacing_ops ||
2875 	    pvd->vdev_ops == &vdev_spare_ops))
2876 		expected_error = ENOTSUP;
2877 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2878 		expected_error = ENOTSUP;
2879 	else if (newvd == oldvd)
2880 		expected_error = replacing ? 0 : EBUSY;
2881 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2882 		expected_error = EBUSY;
2883 	else if (newsize < oldsize)
2884 		expected_error = EOVERFLOW;
2885 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2886 		expected_error = EDOM;
2887 	else
2888 		expected_error = 0;
2889 
2890 	spa_config_exit(spa, SCL_VDEV, FTAG);
2891 
2892 	/*
2893 	 * Build the nvlist describing newpath.
2894 	 */
2895 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2896 	    ashift, 0, 0, 0, 1);
2897 
2898 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2899 
2900 	nvlist_free(root);
2901 
2902 	/*
2903 	 * If our parent was the replacing vdev, but the replace completed,
2904 	 * then instead of failing with ENOTSUP we may either succeed,
2905 	 * fail with ENODEV, or fail with EOVERFLOW.
2906 	 */
2907 	if (expected_error == ENOTSUP &&
2908 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2909 		expected_error = error;
2910 
2911 	/*
2912 	 * If someone grew the LUN, the replacement may be too small.
2913 	 */
2914 	if (error == EOVERFLOW || error == EBUSY)
2915 		expected_error = error;
2916 
2917 	/* XXX workaround 6690467 */
2918 	if (error != expected_error && expected_error != EBUSY) {
2919 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2920 		    "returned %d, expected %d",
2921 		    oldpath, oldsize, newpath,
2922 		    newsize, replacing, error, expected_error);
2923 	}
2924 
2925 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2926 }
2927 
2928 /*
2929  * Callback function which expands the physical size of the vdev.
2930  */
2931 vdev_t *
2932 grow_vdev(vdev_t *vd, void *arg)
2933 {
2934 	spa_t *spa = vd->vdev_spa;
2935 	size_t *newsize = arg;
2936 	size_t fsize;
2937 	int fd;
2938 
2939 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2940 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2941 
2942 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2943 		return (vd);
2944 
2945 	fsize = lseek(fd, 0, SEEK_END);
2946 	(void) ftruncate(fd, *newsize);
2947 
2948 	if (ztest_opts.zo_verbose >= 6) {
2949 		(void) printf("%s grew from %lu to %lu bytes\n",
2950 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2951 	}
2952 	(void) close(fd);
2953 	return (NULL);
2954 }
2955 
2956 /*
2957  * Callback function which expands a given vdev by calling vdev_online().
2958  */
2959 /* ARGSUSED */
2960 vdev_t *
2961 online_vdev(vdev_t *vd, void *arg)
2962 {
2963 	spa_t *spa = vd->vdev_spa;
2964 	vdev_t *tvd = vd->vdev_top;
2965 	uint64_t guid = vd->vdev_guid;
2966 	uint64_t generation = spa->spa_config_generation + 1;
2967 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2968 	int error;
2969 
2970 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2971 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2972 
2973 	/* Calling vdev_online will initialize the new metaslabs */
2974 	spa_config_exit(spa, SCL_STATE, spa);
2975 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2976 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2977 
2978 	/*
2979 	 * If vdev_online returned an error or the underlying vdev_open
2980 	 * failed then we abort the expand. The only way to know that
2981 	 * vdev_open fails is by checking the returned newstate.
2982 	 */
2983 	if (error || newstate != VDEV_STATE_HEALTHY) {
2984 		if (ztest_opts.zo_verbose >= 5) {
2985 			(void) printf("Unable to expand vdev, state %llu, "
2986 			    "error %d\n", (u_longlong_t)newstate, error);
2987 		}
2988 		return (vd);
2989 	}
2990 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2991 
2992 	/*
2993 	 * Since we dropped the lock we need to ensure that we're
2994 	 * still talking to the original vdev. It's possible this
2995 	 * vdev may have been detached/replaced while we were
2996 	 * trying to online it.
2997 	 */
2998 	if (generation != spa->spa_config_generation) {
2999 		if (ztest_opts.zo_verbose >= 5) {
3000 			(void) printf("vdev configuration has changed, "
3001 			    "guid %llu, state %llu, expected gen %llu, "
3002 			    "got gen %llu\n",
3003 			    (u_longlong_t)guid,
3004 			    (u_longlong_t)tvd->vdev_state,
3005 			    (u_longlong_t)generation,
3006 			    (u_longlong_t)spa->spa_config_generation);
3007 		}
3008 		return (vd);
3009 	}
3010 	return (NULL);
3011 }
3012 
3013 /*
3014  * Traverse the vdev tree calling the supplied function.
3015  * We continue to walk the tree until we either have walked all
3016  * children or we receive a non-NULL return from the callback.
3017  * If a NULL callback is passed, then we just return back the first
3018  * leaf vdev we encounter.
3019  */
3020 vdev_t *
3021 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3022 {
3023 	if (vd->vdev_ops->vdev_op_leaf) {
3024 		if (func == NULL)
3025 			return (vd);
3026 		else
3027 			return (func(vd, arg));
3028 	}
3029 
3030 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3031 		vdev_t *cvd = vd->vdev_child[c];
3032 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3033 			return (cvd);
3034 	}
3035 	return (NULL);
3036 }
3037 
3038 /*
3039  * Verify that dynamic LUN growth works as expected.
3040  */
3041 /* ARGSUSED */
3042 void
3043 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3044 {
3045 	spa_t *spa = ztest_spa;
3046 	vdev_t *vd, *tvd;
3047 	metaslab_class_t *mc;
3048 	metaslab_group_t *mg;
3049 	size_t psize, newsize;
3050 	uint64_t top;
3051 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3052 
3053 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3054 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3055 
3056 	top = ztest_random_vdev_top(spa, B_TRUE);
3057 
3058 	tvd = spa->spa_root_vdev->vdev_child[top];
3059 	mg = tvd->vdev_mg;
3060 	mc = mg->mg_class;
3061 	old_ms_count = tvd->vdev_ms_count;
3062 	old_class_space = metaslab_class_get_space(mc);
3063 
3064 	/*
3065 	 * Determine the size of the first leaf vdev associated with
3066 	 * our top-level device.
3067 	 */
3068 	vd = vdev_walk_tree(tvd, NULL, NULL);
3069 	ASSERT3P(vd, !=, NULL);
3070 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3071 
3072 	psize = vd->vdev_psize;
3073 
3074 	/*
3075 	 * We only try to expand the vdev if it's healthy, less than 4x its
3076 	 * original size, and it has a valid psize.
3077 	 */
3078 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3079 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3080 		spa_config_exit(spa, SCL_STATE, spa);
3081 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3082 		return;
3083 	}
3084 	ASSERT(psize > 0);
3085 	newsize = psize + psize / 8;
3086 	ASSERT3U(newsize, >, psize);
3087 
3088 	if (ztest_opts.zo_verbose >= 6) {
3089 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3090 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3091 	}
3092 
3093 	/*
3094 	 * Growing the vdev is a two step process:
3095 	 *	1). expand the physical size (i.e. relabel)
3096 	 *	2). online the vdev to create the new metaslabs
3097 	 */
3098 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3099 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3100 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3101 		if (ztest_opts.zo_verbose >= 5) {
3102 			(void) printf("Could not expand LUN because "
3103 			    "the vdev configuration changed.\n");
3104 		}
3105 		spa_config_exit(spa, SCL_STATE, spa);
3106 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3107 		return;
3108 	}
3109 
3110 	spa_config_exit(spa, SCL_STATE, spa);
3111 
3112 	/*
3113 	 * Expanding the LUN will update the config asynchronously,
3114 	 * thus we must wait for the async thread to complete any
3115 	 * pending tasks before proceeding.
3116 	 */
3117 	for (;;) {
3118 		boolean_t done;
3119 		mutex_enter(&spa->spa_async_lock);
3120 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3121 		mutex_exit(&spa->spa_async_lock);
3122 		if (done)
3123 			break;
3124 		txg_wait_synced(spa_get_dsl(spa), 0);
3125 		(void) poll(NULL, 0, 100);
3126 	}
3127 
3128 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3129 
3130 	tvd = spa->spa_root_vdev->vdev_child[top];
3131 	new_ms_count = tvd->vdev_ms_count;
3132 	new_class_space = metaslab_class_get_space(mc);
3133 
3134 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3135 		if (ztest_opts.zo_verbose >= 5) {
3136 			(void) printf("Could not verify LUN expansion due to "
3137 			    "intervening vdev offline or remove.\n");
3138 		}
3139 		spa_config_exit(spa, SCL_STATE, spa);
3140 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3141 		return;
3142 	}
3143 
3144 	/*
3145 	 * Make sure we were able to grow the vdev.
3146 	 */
3147 	if (new_ms_count <= old_ms_count)
3148 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3149 		    old_ms_count, new_ms_count);
3150 
3151 	/*
3152 	 * Make sure we were able to grow the pool.
3153 	 */
3154 	if (new_class_space <= old_class_space)
3155 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3156 		    old_class_space, new_class_space);
3157 
3158 	if (ztest_opts.zo_verbose >= 5) {
3159 		char oldnumbuf[6], newnumbuf[6];
3160 
3161 		nicenum(old_class_space, oldnumbuf);
3162 		nicenum(new_class_space, newnumbuf);
3163 		(void) printf("%s grew from %s to %s\n",
3164 		    spa->spa_name, oldnumbuf, newnumbuf);
3165 	}
3166 
3167 	spa_config_exit(spa, SCL_STATE, spa);
3168 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3169 }
3170 
3171 /*
3172  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3173  */
3174 /* ARGSUSED */
3175 static void
3176 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3177 {
3178 	/*
3179 	 * Create the objects common to all ztest datasets.
3180 	 */
3181 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3182 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3183 }
3184 
3185 static int
3186 ztest_dataset_create(char *dsname)
3187 {
3188 	uint64_t zilset = ztest_random(100);
3189 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3190 	    ztest_objset_create_cb, NULL);
3191 
3192 	if (err || zilset < 80)
3193 		return (err);
3194 
3195 	if (ztest_opts.zo_verbose >= 6)
3196 		(void) printf("Setting dataset %s to sync always\n", dsname);
3197 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3198 	    ZFS_SYNC_ALWAYS, B_FALSE));
3199 }
3200 
3201 /* ARGSUSED */
3202 static int
3203 ztest_objset_destroy_cb(const char *name, void *arg)
3204 {
3205 	objset_t *os;
3206 	dmu_object_info_t doi;
3207 	int error;
3208 
3209 	/*
3210 	 * Verify that the dataset contains a directory object.
3211 	 */
3212 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3213 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3214 	if (error != ENOENT) {
3215 		/* We could have crashed in the middle of destroying it */
3216 		ASSERT0(error);
3217 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3218 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3219 	}
3220 	dmu_objset_disown(os, FTAG);
3221 
3222 	/*
3223 	 * Destroy the dataset.
3224 	 */
3225 	if (strchr(name, '@') != NULL) {
3226 		VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3227 	} else {
3228 		VERIFY0(dsl_destroy_head(name));
3229 	}
3230 	return (0);
3231 }
3232 
3233 static boolean_t
3234 ztest_snapshot_create(char *osname, uint64_t id)
3235 {
3236 	char snapname[MAXNAMELEN];
3237 	int error;
3238 
3239 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3240 
3241 	error = dmu_objset_snapshot_one(osname, snapname);
3242 	if (error == ENOSPC) {
3243 		ztest_record_enospc(FTAG);
3244 		return (B_FALSE);
3245 	}
3246 	if (error != 0 && error != EEXIST) {
3247 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3248 		    snapname, error);
3249 	}
3250 	return (B_TRUE);
3251 }
3252 
3253 static boolean_t
3254 ztest_snapshot_destroy(char *osname, uint64_t id)
3255 {
3256 	char snapname[MAXNAMELEN];
3257 	int error;
3258 
3259 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3260 	    (u_longlong_t)id);
3261 
3262 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3263 	if (error != 0 && error != ENOENT)
3264 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3265 	return (B_TRUE);
3266 }
3267 
3268 /* ARGSUSED */
3269 void
3270 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3271 {
3272 	ztest_ds_t zdtmp;
3273 	int iters;
3274 	int error;
3275 	objset_t *os, *os2;
3276 	char name[MAXNAMELEN];
3277 	zilog_t *zilog;
3278 
3279 	(void) rw_rdlock(&ztest_name_lock);
3280 
3281 	(void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3282 	    ztest_opts.zo_pool, (u_longlong_t)id);
3283 
3284 	/*
3285 	 * If this dataset exists from a previous run, process its replay log
3286 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3287 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3288 	 */
3289 	if (ztest_random(2) == 0 &&
3290 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3291 		ztest_zd_init(&zdtmp, NULL, os);
3292 		zil_replay(os, &zdtmp, ztest_replay_vector);
3293 		ztest_zd_fini(&zdtmp);
3294 		dmu_objset_disown(os, FTAG);
3295 	}
3296 
3297 	/*
3298 	 * There may be an old instance of the dataset we're about to
3299 	 * create lying around from a previous run.  If so, destroy it
3300 	 * and all of its snapshots.
3301 	 */
3302 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3303 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3304 
3305 	/*
3306 	 * Verify that the destroyed dataset is no longer in the namespace.
3307 	 */
3308 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3309 	    FTAG, &os));
3310 
3311 	/*
3312 	 * Verify that we can create a new dataset.
3313 	 */
3314 	error = ztest_dataset_create(name);
3315 	if (error) {
3316 		if (error == ENOSPC) {
3317 			ztest_record_enospc(FTAG);
3318 			(void) rw_unlock(&ztest_name_lock);
3319 			return;
3320 		}
3321 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3322 	}
3323 
3324 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3325 
3326 	ztest_zd_init(&zdtmp, NULL, os);
3327 
3328 	/*
3329 	 * Open the intent log for it.
3330 	 */
3331 	zilog = zil_open(os, ztest_get_data);
3332 
3333 	/*
3334 	 * Put some objects in there, do a little I/O to them,
3335 	 * and randomly take a couple of snapshots along the way.
3336 	 */
3337 	iters = ztest_random(5);
3338 	for (int i = 0; i < iters; i++) {
3339 		ztest_dmu_object_alloc_free(&zdtmp, id);
3340 		if (ztest_random(iters) == 0)
3341 			(void) ztest_snapshot_create(name, i);
3342 	}
3343 
3344 	/*
3345 	 * Verify that we cannot create an existing dataset.
3346 	 */
3347 	VERIFY3U(EEXIST, ==,
3348 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3349 
3350 	/*
3351 	 * Verify that we can hold an objset that is also owned.
3352 	 */
3353 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3354 	dmu_objset_rele(os2, FTAG);
3355 
3356 	/*
3357 	 * Verify that we cannot own an objset that is already owned.
3358 	 */
3359 	VERIFY3U(EBUSY, ==,
3360 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3361 
3362 	zil_close(zilog);
3363 	dmu_objset_disown(os, FTAG);
3364 	ztest_zd_fini(&zdtmp);
3365 
3366 	(void) rw_unlock(&ztest_name_lock);
3367 }
3368 
3369 /*
3370  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3371  */
3372 void
3373 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3374 {
3375 	(void) rw_rdlock(&ztest_name_lock);
3376 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3377 	(void) ztest_snapshot_create(zd->zd_name, id);
3378 	(void) rw_unlock(&ztest_name_lock);
3379 }
3380 
3381 /*
3382  * Cleanup non-standard snapshots and clones.
3383  */
3384 void
3385 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3386 {
3387 	char snap1name[MAXNAMELEN];
3388 	char clone1name[MAXNAMELEN];
3389 	char snap2name[MAXNAMELEN];
3390 	char clone2name[MAXNAMELEN];
3391 	char snap3name[MAXNAMELEN];
3392 	int error;
3393 
3394 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3395 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3396 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3397 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3398 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3399 
3400 	error = dsl_destroy_head(clone2name);
3401 	if (error && error != ENOENT)
3402 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3403 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3404 	if (error && error != ENOENT)
3405 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3406 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3407 	if (error && error != ENOENT)
3408 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3409 	error = dsl_destroy_head(clone1name);
3410 	if (error && error != ENOENT)
3411 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3412 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3413 	if (error && error != ENOENT)
3414 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3415 }
3416 
3417 /*
3418  * Verify dsl_dataset_promote handles EBUSY
3419  */
3420 void
3421 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3422 {
3423 	objset_t *os;
3424 	char snap1name[MAXNAMELEN];
3425 	char clone1name[MAXNAMELEN];
3426 	char snap2name[MAXNAMELEN];
3427 	char clone2name[MAXNAMELEN];
3428 	char snap3name[MAXNAMELEN];
3429 	char *osname = zd->zd_name;
3430 	int error;
3431 
3432 	(void) rw_rdlock(&ztest_name_lock);
3433 
3434 	ztest_dsl_dataset_cleanup(osname, id);
3435 
3436 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3437 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3438 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3439 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3440 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3441 
3442 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3443 	if (error && error != EEXIST) {
3444 		if (error == ENOSPC) {
3445 			ztest_record_enospc(FTAG);
3446 			goto out;
3447 		}
3448 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3449 	}
3450 
3451 	error = dmu_objset_clone(clone1name, snap1name);
3452 	if (error) {
3453 		if (error == ENOSPC) {
3454 			ztest_record_enospc(FTAG);
3455 			goto out;
3456 		}
3457 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3458 	}
3459 
3460 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3461 	if (error && error != EEXIST) {
3462 		if (error == ENOSPC) {
3463 			ztest_record_enospc(FTAG);
3464 			goto out;
3465 		}
3466 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3467 	}
3468 
3469 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3470 	if (error && error != EEXIST) {
3471 		if (error == ENOSPC) {
3472 			ztest_record_enospc(FTAG);
3473 			goto out;
3474 		}
3475 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3476 	}
3477 
3478 	error = dmu_objset_clone(clone2name, snap3name);
3479 	if (error) {
3480 		if (error == ENOSPC) {
3481 			ztest_record_enospc(FTAG);
3482 			goto out;
3483 		}
3484 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3485 	}
3486 
3487 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3488 	if (error)
3489 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3490 	error = dsl_dataset_promote(clone2name, NULL);
3491 	if (error == ENOSPC) {
3492 		dmu_objset_disown(os, FTAG);
3493 		ztest_record_enospc(FTAG);
3494 		goto out;
3495 	}
3496 	if (error != EBUSY)
3497 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3498 		    error);
3499 	dmu_objset_disown(os, FTAG);
3500 
3501 out:
3502 	ztest_dsl_dataset_cleanup(osname, id);
3503 
3504 	(void) rw_unlock(&ztest_name_lock);
3505 }
3506 
3507 /*
3508  * Verify that dmu_object_{alloc,free} work as expected.
3509  */
3510 void
3511 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3512 {
3513 	ztest_od_t od[4];
3514 	int batchsize = sizeof (od) / sizeof (od[0]);
3515 
3516 	for (int b = 0; b < batchsize; b++)
3517 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3518 
3519 	/*
3520 	 * Destroy the previous batch of objects, create a new batch,
3521 	 * and do some I/O on the new objects.
3522 	 */
3523 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3524 		return;
3525 
3526 	while (ztest_random(4 * batchsize) != 0)
3527 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3528 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3529 }
3530 
3531 /*
3532  * Verify that dmu_{read,write} work as expected.
3533  */
3534 void
3535 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3536 {
3537 	objset_t *os = zd->zd_os;
3538 	ztest_od_t od[2];
3539 	dmu_tx_t *tx;
3540 	int i, freeit, error;
3541 	uint64_t n, s, txg;
3542 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3543 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3544 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3545 	uint64_t regions = 997;
3546 	uint64_t stride = 123456789ULL;
3547 	uint64_t width = 40;
3548 	int free_percent = 5;
3549 
3550 	/*
3551 	 * This test uses two objects, packobj and bigobj, that are always
3552 	 * updated together (i.e. in the same tx) so that their contents are
3553 	 * in sync and can be compared.  Their contents relate to each other
3554 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3555 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3556 	 * for any index n, there are three bufwads that should be identical:
3557 	 *
3558 	 *	packobj, at offset n * sizeof (bufwad_t)
3559 	 *	bigobj, at the head of the nth chunk
3560 	 *	bigobj, at the tail of the nth chunk
3561 	 *
3562 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3563 	 * and it doesn't have any relation to the object blocksize.
3564 	 * The only requirement is that it can hold at least two bufwads.
3565 	 *
3566 	 * Normally, we write the bufwad to each of these locations.
3567 	 * However, free_percent of the time we instead write zeroes to
3568 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3569 	 * bigobj to packobj, we can verify that the DMU is correctly
3570 	 * tracking which parts of an object are allocated and free,
3571 	 * and that the contents of the allocated blocks are correct.
3572 	 */
3573 
3574 	/*
3575 	 * Read the directory info.  If it's the first time, set things up.
3576 	 */
3577 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3578 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3579 
3580 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3581 		return;
3582 
3583 	bigobj = od[0].od_object;
3584 	packobj = od[1].od_object;
3585 	chunksize = od[0].od_gen;
3586 	ASSERT(chunksize == od[1].od_gen);
3587 
3588 	/*
3589 	 * Prefetch a random chunk of the big object.
3590 	 * Our aim here is to get some async reads in flight
3591 	 * for blocks that we may free below; the DMU should
3592 	 * handle this race correctly.
3593 	 */
3594 	n = ztest_random(regions) * stride + ztest_random(width);
3595 	s = 1 + ztest_random(2 * width - 1);
3596 	dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3597 
3598 	/*
3599 	 * Pick a random index and compute the offsets into packobj and bigobj.
3600 	 */
3601 	n = ztest_random(regions) * stride + ztest_random(width);
3602 	s = 1 + ztest_random(width - 1);
3603 
3604 	packoff = n * sizeof (bufwad_t);
3605 	packsize = s * sizeof (bufwad_t);
3606 
3607 	bigoff = n * chunksize;
3608 	bigsize = s * chunksize;
3609 
3610 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3611 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3612 
3613 	/*
3614 	 * free_percent of the time, free a range of bigobj rather than
3615 	 * overwriting it.
3616 	 */
3617 	freeit = (ztest_random(100) < free_percent);
3618 
3619 	/*
3620 	 * Read the current contents of our objects.
3621 	 */
3622 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3623 	    DMU_READ_PREFETCH);
3624 	ASSERT0(error);
3625 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3626 	    DMU_READ_PREFETCH);
3627 	ASSERT0(error);
3628 
3629 	/*
3630 	 * Get a tx for the mods to both packobj and bigobj.
3631 	 */
3632 	tx = dmu_tx_create(os);
3633 
3634 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3635 
3636 	if (freeit)
3637 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3638 	else
3639 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3640 
3641 	/* This accounts for setting the checksum/compression. */
3642 	dmu_tx_hold_bonus(tx, bigobj);
3643 
3644 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3645 	if (txg == 0) {
3646 		umem_free(packbuf, packsize);
3647 		umem_free(bigbuf, bigsize);
3648 		return;
3649 	}
3650 
3651 	enum zio_checksum cksum;
3652 	do {
3653 		cksum = (enum zio_checksum)
3654 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3655 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3656 	dmu_object_set_checksum(os, bigobj, cksum, tx);
3657 
3658 	enum zio_compress comp;
3659 	do {
3660 		comp = (enum zio_compress)
3661 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3662 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3663 	dmu_object_set_compress(os, bigobj, comp, tx);
3664 
3665 	/*
3666 	 * For each index from n to n + s, verify that the existing bufwad
3667 	 * in packobj matches the bufwads at the head and tail of the
3668 	 * corresponding chunk in bigobj.  Then update all three bufwads
3669 	 * with the new values we want to write out.
3670 	 */
3671 	for (i = 0; i < s; i++) {
3672 		/* LINTED */
3673 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3674 		/* LINTED */
3675 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3676 		/* LINTED */
3677 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3678 
3679 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3680 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3681 
3682 		if (pack->bw_txg > txg)
3683 			fatal(0, "future leak: got %llx, open txg is %llx",
3684 			    pack->bw_txg, txg);
3685 
3686 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3687 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3688 			    pack->bw_index, n, i);
3689 
3690 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3691 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3692 
3693 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3694 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3695 
3696 		if (freeit) {
3697 			bzero(pack, sizeof (bufwad_t));
3698 		} else {
3699 			pack->bw_index = n + i;
3700 			pack->bw_txg = txg;
3701 			pack->bw_data = 1 + ztest_random(-2ULL);
3702 		}
3703 		*bigH = *pack;
3704 		*bigT = *pack;
3705 	}
3706 
3707 	/*
3708 	 * We've verified all the old bufwads, and made new ones.
3709 	 * Now write them out.
3710 	 */
3711 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3712 
3713 	if (freeit) {
3714 		if (ztest_opts.zo_verbose >= 7) {
3715 			(void) printf("freeing offset %llx size %llx"
3716 			    " txg %llx\n",
3717 			    (u_longlong_t)bigoff,
3718 			    (u_longlong_t)bigsize,
3719 			    (u_longlong_t)txg);
3720 		}
3721 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3722 	} else {
3723 		if (ztest_opts.zo_verbose >= 7) {
3724 			(void) printf("writing offset %llx size %llx"
3725 			    " txg %llx\n",
3726 			    (u_longlong_t)bigoff,
3727 			    (u_longlong_t)bigsize,
3728 			    (u_longlong_t)txg);
3729 		}
3730 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3731 	}
3732 
3733 	dmu_tx_commit(tx);
3734 
3735 	/*
3736 	 * Sanity check the stuff we just wrote.
3737 	 */
3738 	{
3739 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3740 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3741 
3742 		VERIFY(0 == dmu_read(os, packobj, packoff,
3743 		    packsize, packcheck, DMU_READ_PREFETCH));
3744 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3745 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3746 
3747 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3748 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3749 
3750 		umem_free(packcheck, packsize);
3751 		umem_free(bigcheck, bigsize);
3752 	}
3753 
3754 	umem_free(packbuf, packsize);
3755 	umem_free(bigbuf, bigsize);
3756 }
3757 
3758 void
3759 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3760     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3761 {
3762 	uint64_t i;
3763 	bufwad_t *pack;
3764 	bufwad_t *bigH;
3765 	bufwad_t *bigT;
3766 
3767 	/*
3768 	 * For each index from n to n + s, verify that the existing bufwad
3769 	 * in packobj matches the bufwads at the head and tail of the
3770 	 * corresponding chunk in bigobj.  Then update all three bufwads
3771 	 * with the new values we want to write out.
3772 	 */
3773 	for (i = 0; i < s; i++) {
3774 		/* LINTED */
3775 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3776 		/* LINTED */
3777 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3778 		/* LINTED */
3779 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3780 
3781 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3782 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3783 
3784 		if (pack->bw_txg > txg)
3785 			fatal(0, "future leak: got %llx, open txg is %llx",
3786 			    pack->bw_txg, txg);
3787 
3788 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3789 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3790 			    pack->bw_index, n, i);
3791 
3792 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3793 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3794 
3795 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3796 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3797 
3798 		pack->bw_index = n + i;
3799 		pack->bw_txg = txg;
3800 		pack->bw_data = 1 + ztest_random(-2ULL);
3801 
3802 		*bigH = *pack;
3803 		*bigT = *pack;
3804 	}
3805 }
3806 
3807 void
3808 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3809 {
3810 	objset_t *os = zd->zd_os;
3811 	ztest_od_t od[2];
3812 	dmu_tx_t *tx;
3813 	uint64_t i;
3814 	int error;
3815 	uint64_t n, s, txg;
3816 	bufwad_t *packbuf, *bigbuf;
3817 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3818 	uint64_t blocksize = ztest_random_blocksize();
3819 	uint64_t chunksize = blocksize;
3820 	uint64_t regions = 997;
3821 	uint64_t stride = 123456789ULL;
3822 	uint64_t width = 9;
3823 	dmu_buf_t *bonus_db;
3824 	arc_buf_t **bigbuf_arcbufs;
3825 	dmu_object_info_t doi;
3826 
3827 	/*
3828 	 * This test uses two objects, packobj and bigobj, that are always
3829 	 * updated together (i.e. in the same tx) so that their contents are
3830 	 * in sync and can be compared.  Their contents relate to each other
3831 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3832 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3833 	 * for any index n, there are three bufwads that should be identical:
3834 	 *
3835 	 *	packobj, at offset n * sizeof (bufwad_t)
3836 	 *	bigobj, at the head of the nth chunk
3837 	 *	bigobj, at the tail of the nth chunk
3838 	 *
3839 	 * The chunk size is set equal to bigobj block size so that
3840 	 * dmu_assign_arcbuf() can be tested for object updates.
3841 	 */
3842 
3843 	/*
3844 	 * Read the directory info.  If it's the first time, set things up.
3845 	 */
3846 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3847 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3848 
3849 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3850 		return;
3851 
3852 	bigobj = od[0].od_object;
3853 	packobj = od[1].od_object;
3854 	blocksize = od[0].od_blocksize;
3855 	chunksize = blocksize;
3856 	ASSERT(chunksize == od[1].od_gen);
3857 
3858 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3859 	VERIFY(ISP2(doi.doi_data_block_size));
3860 	VERIFY(chunksize == doi.doi_data_block_size);
3861 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3862 
3863 	/*
3864 	 * Pick a random index and compute the offsets into packobj and bigobj.
3865 	 */
3866 	n = ztest_random(regions) * stride + ztest_random(width);
3867 	s = 1 + ztest_random(width - 1);
3868 
3869 	packoff = n * sizeof (bufwad_t);
3870 	packsize = s * sizeof (bufwad_t);
3871 
3872 	bigoff = n * chunksize;
3873 	bigsize = s * chunksize;
3874 
3875 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3876 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3877 
3878 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3879 
3880 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3881 
3882 	/*
3883 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3884 	 * Iteration 1 test zcopy to already referenced dbufs.
3885 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3886 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3887 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3888 	 * Iteration 5 test zcopy when it can't be done.
3889 	 * Iteration 6 one more zcopy write.
3890 	 */
3891 	for (i = 0; i < 7; i++) {
3892 		uint64_t j;
3893 		uint64_t off;
3894 
3895 		/*
3896 		 * In iteration 5 (i == 5) use arcbufs
3897 		 * that don't match bigobj blksz to test
3898 		 * dmu_assign_arcbuf() when it can't directly
3899 		 * assign an arcbuf to a dbuf.
3900 		 */
3901 		for (j = 0; j < s; j++) {
3902 			if (i != 5) {
3903 				bigbuf_arcbufs[j] =
3904 				    dmu_request_arcbuf(bonus_db, chunksize);
3905 			} else {
3906 				bigbuf_arcbufs[2 * j] =
3907 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3908 				bigbuf_arcbufs[2 * j + 1] =
3909 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3910 			}
3911 		}
3912 
3913 		/*
3914 		 * Get a tx for the mods to both packobj and bigobj.
3915 		 */
3916 		tx = dmu_tx_create(os);
3917 
3918 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3919 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3920 
3921 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3922 		if (txg == 0) {
3923 			umem_free(packbuf, packsize);
3924 			umem_free(bigbuf, bigsize);
3925 			for (j = 0; j < s; j++) {
3926 				if (i != 5) {
3927 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3928 				} else {
3929 					dmu_return_arcbuf(
3930 					    bigbuf_arcbufs[2 * j]);
3931 					dmu_return_arcbuf(
3932 					    bigbuf_arcbufs[2 * j + 1]);
3933 				}
3934 			}
3935 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3936 			dmu_buf_rele(bonus_db, FTAG);
3937 			return;
3938 		}
3939 
3940 		/*
3941 		 * 50% of the time don't read objects in the 1st iteration to
3942 		 * test dmu_assign_arcbuf() for the case when there're no
3943 		 * existing dbufs for the specified offsets.
3944 		 */
3945 		if (i != 0 || ztest_random(2) != 0) {
3946 			error = dmu_read(os, packobj, packoff,
3947 			    packsize, packbuf, DMU_READ_PREFETCH);
3948 			ASSERT0(error);
3949 			error = dmu_read(os, bigobj, bigoff, bigsize,
3950 			    bigbuf, DMU_READ_PREFETCH);
3951 			ASSERT0(error);
3952 		}
3953 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3954 		    n, chunksize, txg);
3955 
3956 		/*
3957 		 * We've verified all the old bufwads, and made new ones.
3958 		 * Now write them out.
3959 		 */
3960 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3961 		if (ztest_opts.zo_verbose >= 7) {
3962 			(void) printf("writing offset %llx size %llx"
3963 			    " txg %llx\n",
3964 			    (u_longlong_t)bigoff,
3965 			    (u_longlong_t)bigsize,
3966 			    (u_longlong_t)txg);
3967 		}
3968 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3969 			dmu_buf_t *dbt;
3970 			if (i != 5) {
3971 				bcopy((caddr_t)bigbuf + (off - bigoff),
3972 				    bigbuf_arcbufs[j]->b_data, chunksize);
3973 			} else {
3974 				bcopy((caddr_t)bigbuf + (off - bigoff),
3975 				    bigbuf_arcbufs[2 * j]->b_data,
3976 				    chunksize / 2);
3977 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3978 				    chunksize / 2,
3979 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3980 				    chunksize / 2);
3981 			}
3982 
3983 			if (i == 1) {
3984 				VERIFY(dmu_buf_hold(os, bigobj, off,
3985 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3986 			}
3987 			if (i != 5) {
3988 				dmu_assign_arcbuf(bonus_db, off,
3989 				    bigbuf_arcbufs[j], tx);
3990 			} else {
3991 				dmu_assign_arcbuf(bonus_db, off,
3992 				    bigbuf_arcbufs[2 * j], tx);
3993 				dmu_assign_arcbuf(bonus_db,
3994 				    off + chunksize / 2,
3995 				    bigbuf_arcbufs[2 * j + 1], tx);
3996 			}
3997 			if (i == 1) {
3998 				dmu_buf_rele(dbt, FTAG);
3999 			}
4000 		}
4001 		dmu_tx_commit(tx);
4002 
4003 		/*
4004 		 * Sanity check the stuff we just wrote.
4005 		 */
4006 		{
4007 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4008 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4009 
4010 			VERIFY(0 == dmu_read(os, packobj, packoff,
4011 			    packsize, packcheck, DMU_READ_PREFETCH));
4012 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4013 			    bigsize, bigcheck, DMU_READ_PREFETCH));
4014 
4015 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4016 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4017 
4018 			umem_free(packcheck, packsize);
4019 			umem_free(bigcheck, bigsize);
4020 		}
4021 		if (i == 2) {
4022 			txg_wait_open(dmu_objset_pool(os), 0);
4023 		} else if (i == 3) {
4024 			txg_wait_synced(dmu_objset_pool(os), 0);
4025 		}
4026 	}
4027 
4028 	dmu_buf_rele(bonus_db, FTAG);
4029 	umem_free(packbuf, packsize);
4030 	umem_free(bigbuf, bigsize);
4031 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4032 }
4033 
4034 /* ARGSUSED */
4035 void
4036 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4037 {
4038 	ztest_od_t od[1];
4039 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4040 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4041 
4042 	/*
4043 	 * Have multiple threads write to large offsets in an object
4044 	 * to verify that parallel writes to an object -- even to the
4045 	 * same blocks within the object -- doesn't cause any trouble.
4046 	 */
4047 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4048 
4049 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4050 		return;
4051 
4052 	while (ztest_random(10) != 0)
4053 		ztest_io(zd, od[0].od_object, offset);
4054 }
4055 
4056 void
4057 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4058 {
4059 	ztest_od_t od[1];
4060 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4061 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4062 	uint64_t count = ztest_random(20) + 1;
4063 	uint64_t blocksize = ztest_random_blocksize();
4064 	void *data;
4065 
4066 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4067 
4068 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4069 		return;
4070 
4071 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4072 		return;
4073 
4074 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4075 
4076 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4077 
4078 	while (ztest_random(count) != 0) {
4079 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4080 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4081 		    data) != 0)
4082 			break;
4083 		while (ztest_random(4) != 0)
4084 			ztest_io(zd, od[0].od_object, randoff);
4085 	}
4086 
4087 	umem_free(data, blocksize);
4088 }
4089 
4090 /*
4091  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4092  */
4093 #define	ZTEST_ZAP_MIN_INTS	1
4094 #define	ZTEST_ZAP_MAX_INTS	4
4095 #define	ZTEST_ZAP_MAX_PROPS	1000
4096 
4097 void
4098 ztest_zap(ztest_ds_t *zd, uint64_t id)
4099 {
4100 	objset_t *os = zd->zd_os;
4101 	ztest_od_t od[1];
4102 	uint64_t object;
4103 	uint64_t txg, last_txg;
4104 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4105 	uint64_t zl_ints, zl_intsize, prop;
4106 	int i, ints;
4107 	dmu_tx_t *tx;
4108 	char propname[100], txgname[100];
4109 	int error;
4110 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4111 
4112 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4113 
4114 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4115 		return;
4116 
4117 	object = od[0].od_object;
4118 
4119 	/*
4120 	 * Generate a known hash collision, and verify that
4121 	 * we can lookup and remove both entries.
4122 	 */
4123 	tx = dmu_tx_create(os);
4124 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4125 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4126 	if (txg == 0)
4127 		return;
4128 	for (i = 0; i < 2; i++) {
4129 		value[i] = i;
4130 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4131 		    1, &value[i], tx));
4132 	}
4133 	for (i = 0; i < 2; i++) {
4134 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4135 		    sizeof (uint64_t), 1, &value[i], tx));
4136 		VERIFY3U(0, ==,
4137 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4138 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4139 		ASSERT3U(zl_ints, ==, 1);
4140 	}
4141 	for (i = 0; i < 2; i++) {
4142 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4143 	}
4144 	dmu_tx_commit(tx);
4145 
4146 	/*
4147 	 * Generate a buch of random entries.
4148 	 */
4149 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4150 
4151 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4152 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4153 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4154 	bzero(value, sizeof (value));
4155 	last_txg = 0;
4156 
4157 	/*
4158 	 * If these zap entries already exist, validate their contents.
4159 	 */
4160 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4161 	if (error == 0) {
4162 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4163 		ASSERT3U(zl_ints, ==, 1);
4164 
4165 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4166 		    zl_ints, &last_txg) == 0);
4167 
4168 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4169 		    &zl_ints) == 0);
4170 
4171 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4172 		ASSERT3U(zl_ints, ==, ints);
4173 
4174 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4175 		    zl_ints, value) == 0);
4176 
4177 		for (i = 0; i < ints; i++) {
4178 			ASSERT3U(value[i], ==, last_txg + object + i);
4179 		}
4180 	} else {
4181 		ASSERT3U(error, ==, ENOENT);
4182 	}
4183 
4184 	/*
4185 	 * Atomically update two entries in our zap object.
4186 	 * The first is named txg_%llu, and contains the txg
4187 	 * in which the property was last updated.  The second
4188 	 * is named prop_%llu, and the nth element of its value
4189 	 * should be txg + object + n.
4190 	 */
4191 	tx = dmu_tx_create(os);
4192 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4193 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4194 	if (txg == 0)
4195 		return;
4196 
4197 	if (last_txg > txg)
4198 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4199 
4200 	for (i = 0; i < ints; i++)
4201 		value[i] = txg + object + i;
4202 
4203 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4204 	    1, &txg, tx));
4205 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4206 	    ints, value, tx));
4207 
4208 	dmu_tx_commit(tx);
4209 
4210 	/*
4211 	 * Remove a random pair of entries.
4212 	 */
4213 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4214 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4215 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4216 
4217 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4218 
4219 	if (error == ENOENT)
4220 		return;
4221 
4222 	ASSERT0(error);
4223 
4224 	tx = dmu_tx_create(os);
4225 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4226 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4227 	if (txg == 0)
4228 		return;
4229 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4230 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4231 	dmu_tx_commit(tx);
4232 }
4233 
4234 /*
4235  * Testcase to test the upgrading of a microzap to fatzap.
4236  */
4237 void
4238 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4239 {
4240 	objset_t *os = zd->zd_os;
4241 	ztest_od_t od[1];
4242 	uint64_t object, txg;
4243 
4244 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4245 
4246 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4247 		return;
4248 
4249 	object = od[0].od_object;
4250 
4251 	/*
4252 	 * Add entries to this ZAP and make sure it spills over
4253 	 * and gets upgraded to a fatzap. Also, since we are adding
4254 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4255 	 */
4256 	for (int i = 0; i < 2050; i++) {
4257 		char name[MAXNAMELEN];
4258 		uint64_t value = i;
4259 		dmu_tx_t *tx;
4260 		int error;
4261 
4262 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4263 		    id, value);
4264 
4265 		tx = dmu_tx_create(os);
4266 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4267 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4268 		if (txg == 0)
4269 			return;
4270 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4271 		    &value, tx);
4272 		ASSERT(error == 0 || error == EEXIST);
4273 		dmu_tx_commit(tx);
4274 	}
4275 }
4276 
4277 /* ARGSUSED */
4278 void
4279 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4280 {
4281 	objset_t *os = zd->zd_os;
4282 	ztest_od_t od[1];
4283 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4284 	dmu_tx_t *tx;
4285 	int i, namelen, error;
4286 	int micro = ztest_random(2);
4287 	char name[20], string_value[20];
4288 	void *data;
4289 
4290 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4291 
4292 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4293 		return;
4294 
4295 	object = od[0].od_object;
4296 
4297 	/*
4298 	 * Generate a random name of the form 'xxx.....' where each
4299 	 * x is a random printable character and the dots are dots.
4300 	 * There are 94 such characters, and the name length goes from
4301 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4302 	 */
4303 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4304 
4305 	for (i = 0; i < 3; i++)
4306 		name[i] = '!' + ztest_random('~' - '!' + 1);
4307 	for (; i < namelen - 1; i++)
4308 		name[i] = '.';
4309 	name[i] = '\0';
4310 
4311 	if ((namelen & 1) || micro) {
4312 		wsize = sizeof (txg);
4313 		wc = 1;
4314 		data = &txg;
4315 	} else {
4316 		wsize = 1;
4317 		wc = namelen;
4318 		data = string_value;
4319 	}
4320 
4321 	count = -1ULL;
4322 	VERIFY0(zap_count(os, object, &count));
4323 	ASSERT(count != -1ULL);
4324 
4325 	/*
4326 	 * Select an operation: length, lookup, add, update, remove.
4327 	 */
4328 	i = ztest_random(5);
4329 
4330 	if (i >= 2) {
4331 		tx = dmu_tx_create(os);
4332 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4333 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4334 		if (txg == 0)
4335 			return;
4336 		bcopy(name, string_value, namelen);
4337 	} else {
4338 		tx = NULL;
4339 		txg = 0;
4340 		bzero(string_value, namelen);
4341 	}
4342 
4343 	switch (i) {
4344 
4345 	case 0:
4346 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4347 		if (error == 0) {
4348 			ASSERT3U(wsize, ==, zl_wsize);
4349 			ASSERT3U(wc, ==, zl_wc);
4350 		} else {
4351 			ASSERT3U(error, ==, ENOENT);
4352 		}
4353 		break;
4354 
4355 	case 1:
4356 		error = zap_lookup(os, object, name, wsize, wc, data);
4357 		if (error == 0) {
4358 			if (data == string_value &&
4359 			    bcmp(name, data, namelen) != 0)
4360 				fatal(0, "name '%s' != val '%s' len %d",
4361 				    name, data, namelen);
4362 		} else {
4363 			ASSERT3U(error, ==, ENOENT);
4364 		}
4365 		break;
4366 
4367 	case 2:
4368 		error = zap_add(os, object, name, wsize, wc, data, tx);
4369 		ASSERT(error == 0 || error == EEXIST);
4370 		break;
4371 
4372 	case 3:
4373 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4374 		break;
4375 
4376 	case 4:
4377 		error = zap_remove(os, object, name, tx);
4378 		ASSERT(error == 0 || error == ENOENT);
4379 		break;
4380 	}
4381 
4382 	if (tx != NULL)
4383 		dmu_tx_commit(tx);
4384 }
4385 
4386 /*
4387  * Commit callback data.
4388  */
4389 typedef struct ztest_cb_data {
4390 	list_node_t		zcd_node;
4391 	uint64_t		zcd_txg;
4392 	int			zcd_expected_err;
4393 	boolean_t		zcd_added;
4394 	boolean_t		zcd_called;
4395 	spa_t			*zcd_spa;
4396 } ztest_cb_data_t;
4397 
4398 /* This is the actual commit callback function */
4399 static void
4400 ztest_commit_callback(void *arg, int error)
4401 {
4402 	ztest_cb_data_t *data = arg;
4403 	uint64_t synced_txg;
4404 
4405 	VERIFY(data != NULL);
4406 	VERIFY3S(data->zcd_expected_err, ==, error);
4407 	VERIFY(!data->zcd_called);
4408 
4409 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4410 	if (data->zcd_txg > synced_txg)
4411 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4412 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4413 		    synced_txg);
4414 
4415 	data->zcd_called = B_TRUE;
4416 
4417 	if (error == ECANCELED) {
4418 		ASSERT0(data->zcd_txg);
4419 		ASSERT(!data->zcd_added);
4420 
4421 		/*
4422 		 * The private callback data should be destroyed here, but
4423 		 * since we are going to check the zcd_called field after
4424 		 * dmu_tx_abort(), we will destroy it there.
4425 		 */
4426 		return;
4427 	}
4428 
4429 	/* Was this callback added to the global callback list? */
4430 	if (!data->zcd_added)
4431 		goto out;
4432 
4433 	ASSERT3U(data->zcd_txg, !=, 0);
4434 
4435 	/* Remove our callback from the list */
4436 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4437 	list_remove(&zcl.zcl_callbacks, data);
4438 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4439 
4440 out:
4441 	umem_free(data, sizeof (ztest_cb_data_t));
4442 }
4443 
4444 /* Allocate and initialize callback data structure */
4445 static ztest_cb_data_t *
4446 ztest_create_cb_data(objset_t *os, uint64_t txg)
4447 {
4448 	ztest_cb_data_t *cb_data;
4449 
4450 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4451 
4452 	cb_data->zcd_txg = txg;
4453 	cb_data->zcd_spa = dmu_objset_spa(os);
4454 
4455 	return (cb_data);
4456 }
4457 
4458 /*
4459  * If a number of txgs equal to this threshold have been created after a commit
4460  * callback has been registered but not called, then we assume there is an
4461  * implementation bug.
4462  */
4463 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4464 
4465 /*
4466  * Commit callback test.
4467  */
4468 void
4469 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4470 {
4471 	objset_t *os = zd->zd_os;
4472 	ztest_od_t od[1];
4473 	dmu_tx_t *tx;
4474 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4475 	uint64_t old_txg, txg;
4476 	int i, error;
4477 
4478 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4479 
4480 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4481 		return;
4482 
4483 	tx = dmu_tx_create(os);
4484 
4485 	cb_data[0] = ztest_create_cb_data(os, 0);
4486 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4487 
4488 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4489 
4490 	/* Every once in a while, abort the transaction on purpose */
4491 	if (ztest_random(100) == 0)
4492 		error = -1;
4493 
4494 	if (!error)
4495 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4496 
4497 	txg = error ? 0 : dmu_tx_get_txg(tx);
4498 
4499 	cb_data[0]->zcd_txg = txg;
4500 	cb_data[1] = ztest_create_cb_data(os, txg);
4501 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4502 
4503 	if (error) {
4504 		/*
4505 		 * It's not a strict requirement to call the registered
4506 		 * callbacks from inside dmu_tx_abort(), but that's what
4507 		 * it's supposed to happen in the current implementation
4508 		 * so we will check for that.
4509 		 */
4510 		for (i = 0; i < 2; i++) {
4511 			cb_data[i]->zcd_expected_err = ECANCELED;
4512 			VERIFY(!cb_data[i]->zcd_called);
4513 		}
4514 
4515 		dmu_tx_abort(tx);
4516 
4517 		for (i = 0; i < 2; i++) {
4518 			VERIFY(cb_data[i]->zcd_called);
4519 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4520 		}
4521 
4522 		return;
4523 	}
4524 
4525 	cb_data[2] = ztest_create_cb_data(os, txg);
4526 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4527 
4528 	/*
4529 	 * Read existing data to make sure there isn't a future leak.
4530 	 */
4531 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4532 	    &old_txg, DMU_READ_PREFETCH));
4533 
4534 	if (old_txg > txg)
4535 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4536 		    old_txg, txg);
4537 
4538 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4539 
4540 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4541 
4542 	/*
4543 	 * Since commit callbacks don't have any ordering requirement and since
4544 	 * it is theoretically possible for a commit callback to be called
4545 	 * after an arbitrary amount of time has elapsed since its txg has been
4546 	 * synced, it is difficult to reliably determine whether a commit
4547 	 * callback hasn't been called due to high load or due to a flawed
4548 	 * implementation.
4549 	 *
4550 	 * In practice, we will assume that if after a certain number of txgs a
4551 	 * commit callback hasn't been called, then most likely there's an
4552 	 * implementation bug..
4553 	 */
4554 	tmp_cb = list_head(&zcl.zcl_callbacks);
4555 	if (tmp_cb != NULL &&
4556 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4557 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4558 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4559 	}
4560 
4561 	/*
4562 	 * Let's find the place to insert our callbacks.
4563 	 *
4564 	 * Even though the list is ordered by txg, it is possible for the
4565 	 * insertion point to not be the end because our txg may already be
4566 	 * quiescing at this point and other callbacks in the open txg
4567 	 * (from other objsets) may have sneaked in.
4568 	 */
4569 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4570 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4571 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4572 
4573 	/* Add the 3 callbacks to the list */
4574 	for (i = 0; i < 3; i++) {
4575 		if (tmp_cb == NULL)
4576 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4577 		else
4578 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4579 			    cb_data[i]);
4580 
4581 		cb_data[i]->zcd_added = B_TRUE;
4582 		VERIFY(!cb_data[i]->zcd_called);
4583 
4584 		tmp_cb = cb_data[i];
4585 	}
4586 
4587 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4588 
4589 	dmu_tx_commit(tx);
4590 }
4591 
4592 /* ARGSUSED */
4593 void
4594 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4595 {
4596 	zfs_prop_t proplist[] = {
4597 		ZFS_PROP_CHECKSUM,
4598 		ZFS_PROP_COMPRESSION,
4599 		ZFS_PROP_COPIES,
4600 		ZFS_PROP_DEDUP
4601 	};
4602 
4603 	(void) rw_rdlock(&ztest_name_lock);
4604 
4605 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4606 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4607 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4608 
4609 	(void) rw_unlock(&ztest_name_lock);
4610 }
4611 
4612 /* ARGSUSED */
4613 void
4614 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4615 {
4616 	nvlist_t *props = NULL;
4617 
4618 	(void) rw_rdlock(&ztest_name_lock);
4619 
4620 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4621 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4622 
4623 	VERIFY0(spa_prop_get(ztest_spa, &props));
4624 
4625 	if (ztest_opts.zo_verbose >= 6)
4626 		dump_nvlist(props, 4);
4627 
4628 	nvlist_free(props);
4629 
4630 	(void) rw_unlock(&ztest_name_lock);
4631 }
4632 
4633 static int
4634 user_release_one(const char *snapname, const char *holdname)
4635 {
4636 	nvlist_t *snaps, *holds;
4637 	int error;
4638 
4639 	snaps = fnvlist_alloc();
4640 	holds = fnvlist_alloc();
4641 	fnvlist_add_boolean(holds, holdname);
4642 	fnvlist_add_nvlist(snaps, snapname, holds);
4643 	fnvlist_free(holds);
4644 	error = dsl_dataset_user_release(snaps, NULL);
4645 	fnvlist_free(snaps);
4646 	return (error);
4647 }
4648 
4649 /*
4650  * Test snapshot hold/release and deferred destroy.
4651  */
4652 void
4653 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4654 {
4655 	int error;
4656 	objset_t *os = zd->zd_os;
4657 	objset_t *origin;
4658 	char snapname[100];
4659 	char fullname[100];
4660 	char clonename[100];
4661 	char tag[100];
4662 	char osname[MAXNAMELEN];
4663 	nvlist_t *holds;
4664 
4665 	(void) rw_rdlock(&ztest_name_lock);
4666 
4667 	dmu_objset_name(os, osname);
4668 
4669 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4670 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4671 	(void) snprintf(clonename, sizeof (clonename),
4672 	    "%s/ch1_%llu", osname, id);
4673 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4674 
4675 	/*
4676 	 * Clean up from any previous run.
4677 	 */
4678 	error = dsl_destroy_head(clonename);
4679 	if (error != ENOENT)
4680 		ASSERT0(error);
4681 	error = user_release_one(fullname, tag);
4682 	if (error != ESRCH && error != ENOENT)
4683 		ASSERT0(error);
4684 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4685 	if (error != ENOENT)
4686 		ASSERT0(error);
4687 
4688 	/*
4689 	 * Create snapshot, clone it, mark snap for deferred destroy,
4690 	 * destroy clone, verify snap was also destroyed.
4691 	 */
4692 	error = dmu_objset_snapshot_one(osname, snapname);
4693 	if (error) {
4694 		if (error == ENOSPC) {
4695 			ztest_record_enospc("dmu_objset_snapshot");
4696 			goto out;
4697 		}
4698 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4699 	}
4700 
4701 	error = dmu_objset_clone(clonename, fullname);
4702 	if (error) {
4703 		if (error == ENOSPC) {
4704 			ztest_record_enospc("dmu_objset_clone");
4705 			goto out;
4706 		}
4707 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4708 	}
4709 
4710 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4711 	if (error) {
4712 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4713 		    fullname, error);
4714 	}
4715 
4716 	error = dsl_destroy_head(clonename);
4717 	if (error)
4718 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4719 
4720 	error = dmu_objset_hold(fullname, FTAG, &origin);
4721 	if (error != ENOENT)
4722 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4723 
4724 	/*
4725 	 * Create snapshot, add temporary hold, verify that we can't
4726 	 * destroy a held snapshot, mark for deferred destroy,
4727 	 * release hold, verify snapshot was destroyed.
4728 	 */
4729 	error = dmu_objset_snapshot_one(osname, snapname);
4730 	if (error) {
4731 		if (error == ENOSPC) {
4732 			ztest_record_enospc("dmu_objset_snapshot");
4733 			goto out;
4734 		}
4735 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4736 	}
4737 
4738 	holds = fnvlist_alloc();
4739 	fnvlist_add_string(holds, fullname, tag);
4740 	error = dsl_dataset_user_hold(holds, 0, NULL);
4741 	fnvlist_free(holds);
4742 
4743 	if (error == ENOSPC) {
4744 		ztest_record_enospc("dsl_dataset_user_hold");
4745 		goto out;
4746 	} else if (error) {
4747 		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4748 		    fullname, tag, error);
4749 	}
4750 
4751 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4752 	if (error != EBUSY) {
4753 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4754 		    fullname, error);
4755 	}
4756 
4757 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4758 	if (error) {
4759 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4760 		    fullname, error);
4761 	}
4762 
4763 	error = user_release_one(fullname, tag);
4764 	if (error)
4765 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4766 
4767 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4768 
4769 out:
4770 	(void) rw_unlock(&ztest_name_lock);
4771 }
4772 
4773 /*
4774  * Inject random faults into the on-disk data.
4775  */
4776 /* ARGSUSED */
4777 void
4778 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4779 {
4780 	ztest_shared_t *zs = ztest_shared;
4781 	spa_t *spa = ztest_spa;
4782 	int fd;
4783 	uint64_t offset;
4784 	uint64_t leaves;
4785 	uint64_t bad = 0x1990c0ffeedecade;
4786 	uint64_t top, leaf;
4787 	char path0[MAXPATHLEN];
4788 	char pathrand[MAXPATHLEN];
4789 	size_t fsize;
4790 	int bshift = SPA_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
4791 	int iters = 1000;
4792 	int maxfaults;
4793 	int mirror_save;
4794 	vdev_t *vd0 = NULL;
4795 	uint64_t guid0 = 0;
4796 	boolean_t islog = B_FALSE;
4797 
4798 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4799 	maxfaults = MAXFAULTS();
4800 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4801 	mirror_save = zs->zs_mirrors;
4802 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4803 
4804 	ASSERT(leaves >= 1);
4805 
4806 	/*
4807 	 * Grab the name lock as reader. There are some operations
4808 	 * which don't like to have their vdevs changed while
4809 	 * they are in progress (i.e. spa_change_guid). Those
4810 	 * operations will have grabbed the name lock as writer.
4811 	 */
4812 	(void) rw_rdlock(&ztest_name_lock);
4813 
4814 	/*
4815 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4816 	 */
4817 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4818 
4819 	if (ztest_random(2) == 0) {
4820 		/*
4821 		 * Inject errors on a normal data device or slog device.
4822 		 */
4823 		top = ztest_random_vdev_top(spa, B_TRUE);
4824 		leaf = ztest_random(leaves) + zs->zs_splits;
4825 
4826 		/*
4827 		 * Generate paths to the first leaf in this top-level vdev,
4828 		 * and to the random leaf we selected.  We'll induce transient
4829 		 * write failures and random online/offline activity on leaf 0,
4830 		 * and we'll write random garbage to the randomly chosen leaf.
4831 		 */
4832 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4833 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4834 		    top * leaves + zs->zs_splits);
4835 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4836 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4837 		    top * leaves + leaf);
4838 
4839 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4840 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4841 			islog = B_TRUE;
4842 
4843 		/*
4844 		 * If the top-level vdev needs to be resilvered
4845 		 * then we only allow faults on the device that is
4846 		 * resilvering.
4847 		 */
4848 		if (vd0 != NULL && maxfaults != 1 &&
4849 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4850 		    vd0->vdev_resilver_txg != 0)) {
4851 			/*
4852 			 * Make vd0 explicitly claim to be unreadable,
4853 			 * or unwriteable, or reach behind its back
4854 			 * and close the underlying fd.  We can do this if
4855 			 * maxfaults == 0 because we'll fail and reexecute,
4856 			 * and we can do it if maxfaults >= 2 because we'll
4857 			 * have enough redundancy.  If maxfaults == 1, the
4858 			 * combination of this with injection of random data
4859 			 * corruption below exceeds the pool's fault tolerance.
4860 			 */
4861 			vdev_file_t *vf = vd0->vdev_tsd;
4862 
4863 			if (vf != NULL && ztest_random(3) == 0) {
4864 				(void) close(vf->vf_vnode->v_fd);
4865 				vf->vf_vnode->v_fd = -1;
4866 			} else if (ztest_random(2) == 0) {
4867 				vd0->vdev_cant_read = B_TRUE;
4868 			} else {
4869 				vd0->vdev_cant_write = B_TRUE;
4870 			}
4871 			guid0 = vd0->vdev_guid;
4872 		}
4873 	} else {
4874 		/*
4875 		 * Inject errors on an l2cache device.
4876 		 */
4877 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4878 
4879 		if (sav->sav_count == 0) {
4880 			spa_config_exit(spa, SCL_STATE, FTAG);
4881 			(void) rw_unlock(&ztest_name_lock);
4882 			return;
4883 		}
4884 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4885 		guid0 = vd0->vdev_guid;
4886 		(void) strcpy(path0, vd0->vdev_path);
4887 		(void) strcpy(pathrand, vd0->vdev_path);
4888 
4889 		leaf = 0;
4890 		leaves = 1;
4891 		maxfaults = INT_MAX;	/* no limit on cache devices */
4892 	}
4893 
4894 	spa_config_exit(spa, SCL_STATE, FTAG);
4895 	(void) rw_unlock(&ztest_name_lock);
4896 
4897 	/*
4898 	 * If we can tolerate two or more faults, or we're dealing
4899 	 * with a slog, randomly online/offline vd0.
4900 	 */
4901 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4902 		if (ztest_random(10) < 6) {
4903 			int flags = (ztest_random(2) == 0 ?
4904 			    ZFS_OFFLINE_TEMPORARY : 0);
4905 
4906 			/*
4907 			 * We have to grab the zs_name_lock as writer to
4908 			 * prevent a race between offlining a slog and
4909 			 * destroying a dataset. Offlining the slog will
4910 			 * grab a reference on the dataset which may cause
4911 			 * dmu_objset_destroy() to fail with EBUSY thus
4912 			 * leaving the dataset in an inconsistent state.
4913 			 */
4914 			if (islog)
4915 				(void) rw_wrlock(&ztest_name_lock);
4916 
4917 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4918 
4919 			if (islog)
4920 				(void) rw_unlock(&ztest_name_lock);
4921 		} else {
4922 			/*
4923 			 * Ideally we would like to be able to randomly
4924 			 * call vdev_[on|off]line without holding locks
4925 			 * to force unpredictable failures but the side
4926 			 * effects of vdev_[on|off]line prevent us from
4927 			 * doing so. We grab the ztest_vdev_lock here to
4928 			 * prevent a race between injection testing and
4929 			 * aux_vdev removal.
4930 			 */
4931 			VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4932 			(void) vdev_online(spa, guid0, 0, NULL);
4933 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4934 		}
4935 	}
4936 
4937 	if (maxfaults == 0)
4938 		return;
4939 
4940 	/*
4941 	 * We have at least single-fault tolerance, so inject data corruption.
4942 	 */
4943 	fd = open(pathrand, O_RDWR);
4944 
4945 	if (fd == -1)	/* we hit a gap in the device namespace */
4946 		return;
4947 
4948 	fsize = lseek(fd, 0, SEEK_END);
4949 
4950 	while (--iters != 0) {
4951 		offset = ztest_random(fsize / (leaves << bshift)) *
4952 		    (leaves << bshift) + (leaf << bshift) +
4953 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4954 
4955 		if (offset >= fsize)
4956 			continue;
4957 
4958 		VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4959 		if (mirror_save != zs->zs_mirrors) {
4960 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4961 			(void) close(fd);
4962 			return;
4963 		}
4964 
4965 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4966 			fatal(1, "can't inject bad word at 0x%llx in %s",
4967 			    offset, pathrand);
4968 
4969 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4970 
4971 		if (ztest_opts.zo_verbose >= 7)
4972 			(void) printf("injected bad word into %s,"
4973 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4974 	}
4975 
4976 	(void) close(fd);
4977 }
4978 
4979 /*
4980  * Verify that DDT repair works as expected.
4981  */
4982 void
4983 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4984 {
4985 	ztest_shared_t *zs = ztest_shared;
4986 	spa_t *spa = ztest_spa;
4987 	objset_t *os = zd->zd_os;
4988 	ztest_od_t od[1];
4989 	uint64_t object, blocksize, txg, pattern, psize;
4990 	enum zio_checksum checksum = spa_dedup_checksum(spa);
4991 	dmu_buf_t *db;
4992 	dmu_tx_t *tx;
4993 	void *buf;
4994 	blkptr_t blk;
4995 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
4996 
4997 	blocksize = ztest_random_blocksize();
4998 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
4999 
5000 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5001 
5002 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5003 		return;
5004 
5005 	/*
5006 	 * Take the name lock as writer to prevent anyone else from changing
5007 	 * the pool and dataset properies we need to maintain during this test.
5008 	 */
5009 	(void) rw_wrlock(&ztest_name_lock);
5010 
5011 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5012 	    B_FALSE) != 0 ||
5013 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5014 	    B_FALSE) != 0) {
5015 		(void) rw_unlock(&ztest_name_lock);
5016 		return;
5017 	}
5018 
5019 	object = od[0].od_object;
5020 	blocksize = od[0].od_blocksize;
5021 	pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5022 
5023 	ASSERT(object != 0);
5024 
5025 	tx = dmu_tx_create(os);
5026 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5027 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5028 	if (txg == 0) {
5029 		(void) rw_unlock(&ztest_name_lock);
5030 		return;
5031 	}
5032 
5033 	/*
5034 	 * Write all the copies of our block.
5035 	 */
5036 	for (int i = 0; i < copies; i++) {
5037 		uint64_t offset = i * blocksize;
5038 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5039 		    DMU_READ_NO_PREFETCH);
5040 		if (error != 0) {
5041 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5042 			    os, (long long)object, (long long) offset, error);
5043 		}
5044 		ASSERT(db->db_offset == offset);
5045 		ASSERT(db->db_size == blocksize);
5046 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5047 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5048 		dmu_buf_will_fill(db, tx);
5049 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5050 		dmu_buf_rele(db, FTAG);
5051 	}
5052 
5053 	dmu_tx_commit(tx);
5054 	txg_wait_synced(spa_get_dsl(spa), txg);
5055 
5056 	/*
5057 	 * Find out what block we got.
5058 	 */
5059 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5060 	    DMU_READ_NO_PREFETCH));
5061 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5062 	dmu_buf_rele(db, FTAG);
5063 
5064 	/*
5065 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5066 	 */
5067 	psize = BP_GET_PSIZE(&blk);
5068 	buf = zio_buf_alloc(psize);
5069 	ztest_pattern_set(buf, psize, ~pattern);
5070 
5071 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5072 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5073 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5074 
5075 	zio_buf_free(buf, psize);
5076 
5077 	(void) rw_unlock(&ztest_name_lock);
5078 }
5079 
5080 /*
5081  * Scrub the pool.
5082  */
5083 /* ARGSUSED */
5084 void
5085 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5086 {
5087 	spa_t *spa = ztest_spa;
5088 
5089 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5090 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5091 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5092 }
5093 
5094 /*
5095  * Change the guid for the pool.
5096  */
5097 /* ARGSUSED */
5098 void
5099 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5100 {
5101 	spa_t *spa = ztest_spa;
5102 	uint64_t orig, load;
5103 	int error;
5104 
5105 	orig = spa_guid(spa);
5106 	load = spa_load_guid(spa);
5107 
5108 	(void) rw_wrlock(&ztest_name_lock);
5109 	error = spa_change_guid(spa);
5110 	(void) rw_unlock(&ztest_name_lock);
5111 
5112 	if (error != 0)
5113 		return;
5114 
5115 	if (ztest_opts.zo_verbose >= 4) {
5116 		(void) printf("Changed guid old %llu -> %llu\n",
5117 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5118 	}
5119 
5120 	VERIFY3U(orig, !=, spa_guid(spa));
5121 	VERIFY3U(load, ==, spa_load_guid(spa));
5122 }
5123 
5124 /*
5125  * Rename the pool to a different name and then rename it back.
5126  */
5127 /* ARGSUSED */
5128 void
5129 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5130 {
5131 	char *oldname, *newname;
5132 	spa_t *spa;
5133 
5134 	(void) rw_wrlock(&ztest_name_lock);
5135 
5136 	oldname = ztest_opts.zo_pool;
5137 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5138 	(void) strcpy(newname, oldname);
5139 	(void) strcat(newname, "_tmp");
5140 
5141 	/*
5142 	 * Do the rename
5143 	 */
5144 	VERIFY3U(0, ==, spa_rename(oldname, newname));
5145 
5146 	/*
5147 	 * Try to open it under the old name, which shouldn't exist
5148 	 */
5149 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5150 
5151 	/*
5152 	 * Open it under the new name and make sure it's still the same spa_t.
5153 	 */
5154 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5155 
5156 	ASSERT(spa == ztest_spa);
5157 	spa_close(spa, FTAG);
5158 
5159 	/*
5160 	 * Rename it back to the original
5161 	 */
5162 	VERIFY3U(0, ==, spa_rename(newname, oldname));
5163 
5164 	/*
5165 	 * Make sure it can still be opened
5166 	 */
5167 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5168 
5169 	ASSERT(spa == ztest_spa);
5170 	spa_close(spa, FTAG);
5171 
5172 	umem_free(newname, strlen(newname) + 1);
5173 
5174 	(void) rw_unlock(&ztest_name_lock);
5175 }
5176 
5177 /*
5178  * Verify pool integrity by running zdb.
5179  */
5180 static void
5181 ztest_run_zdb(char *pool)
5182 {
5183 	int status;
5184 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5185 	char zbuf[1024];
5186 	char *bin;
5187 	char *ztest;
5188 	char *isa;
5189 	int isalen;
5190 	FILE *fp;
5191 
5192 	(void) realpath(getexecname(), zdb);
5193 
5194 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5195 	bin = strstr(zdb, "/usr/bin/");
5196 	ztest = strstr(bin, "/ztest");
5197 	isa = bin + 8;
5198 	isalen = ztest - isa;
5199 	isa = strdup(isa);
5200 	/* LINTED */
5201 	(void) sprintf(bin,
5202 	    "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5203 	    isalen,
5204 	    isa,
5205 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5206 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5207 	    spa_config_path,
5208 	    pool);
5209 	free(isa);
5210 
5211 	if (ztest_opts.zo_verbose >= 5)
5212 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5213 
5214 	fp = popen(zdb, "r");
5215 
5216 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5217 		if (ztest_opts.zo_verbose >= 3)
5218 			(void) printf("%s", zbuf);
5219 
5220 	status = pclose(fp);
5221 
5222 	if (status == 0)
5223 		return;
5224 
5225 	ztest_dump_core = 0;
5226 	if (WIFEXITED(status))
5227 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5228 	else
5229 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5230 }
5231 
5232 static void
5233 ztest_walk_pool_directory(char *header)
5234 {
5235 	spa_t *spa = NULL;
5236 
5237 	if (ztest_opts.zo_verbose >= 6)
5238 		(void) printf("%s\n", header);
5239 
5240 	mutex_enter(&spa_namespace_lock);
5241 	while ((spa = spa_next(spa)) != NULL)
5242 		if (ztest_opts.zo_verbose >= 6)
5243 			(void) printf("\t%s\n", spa_name(spa));
5244 	mutex_exit(&spa_namespace_lock);
5245 }
5246 
5247 static void
5248 ztest_spa_import_export(char *oldname, char *newname)
5249 {
5250 	nvlist_t *config, *newconfig;
5251 	uint64_t pool_guid;
5252 	spa_t *spa;
5253 	int error;
5254 
5255 	if (ztest_opts.zo_verbose >= 4) {
5256 		(void) printf("import/export: old = %s, new = %s\n",
5257 		    oldname, newname);
5258 	}
5259 
5260 	/*
5261 	 * Clean up from previous runs.
5262 	 */
5263 	(void) spa_destroy(newname);
5264 
5265 	/*
5266 	 * Get the pool's configuration and guid.
5267 	 */
5268 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5269 
5270 	/*
5271 	 * Kick off a scrub to tickle scrub/export races.
5272 	 */
5273 	if (ztest_random(2) == 0)
5274 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5275 
5276 	pool_guid = spa_guid(spa);
5277 	spa_close(spa, FTAG);
5278 
5279 	ztest_walk_pool_directory("pools before export");
5280 
5281 	/*
5282 	 * Export it.
5283 	 */
5284 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5285 
5286 	ztest_walk_pool_directory("pools after export");
5287 
5288 	/*
5289 	 * Try to import it.
5290 	 */
5291 	newconfig = spa_tryimport(config);
5292 	ASSERT(newconfig != NULL);
5293 	nvlist_free(newconfig);
5294 
5295 	/*
5296 	 * Import it under the new name.
5297 	 */
5298 	error = spa_import(newname, config, NULL, 0);
5299 	if (error != 0) {
5300 		dump_nvlist(config, 0);
5301 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5302 		    oldname, newname, error);
5303 	}
5304 
5305 	ztest_walk_pool_directory("pools after import");
5306 
5307 	/*
5308 	 * Try to import it again -- should fail with EEXIST.
5309 	 */
5310 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5311 
5312 	/*
5313 	 * Try to import it under a different name -- should fail with EEXIST.
5314 	 */
5315 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5316 
5317 	/*
5318 	 * Verify that the pool is no longer visible under the old name.
5319 	 */
5320 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5321 
5322 	/*
5323 	 * Verify that we can open and close the pool using the new name.
5324 	 */
5325 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5326 	ASSERT(pool_guid == spa_guid(spa));
5327 	spa_close(spa, FTAG);
5328 
5329 	nvlist_free(config);
5330 }
5331 
5332 static void
5333 ztest_resume(spa_t *spa)
5334 {
5335 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5336 		(void) printf("resuming from suspended state\n");
5337 	spa_vdev_state_enter(spa, SCL_NONE);
5338 	vdev_clear(spa, NULL);
5339 	(void) spa_vdev_state_exit(spa, NULL, 0);
5340 	(void) zio_resume(spa);
5341 }
5342 
5343 static void *
5344 ztest_resume_thread(void *arg)
5345 {
5346 	spa_t *spa = arg;
5347 
5348 	while (!ztest_exiting) {
5349 		if (spa_suspended(spa))
5350 			ztest_resume(spa);
5351 		(void) poll(NULL, 0, 100);
5352 	}
5353 	return (NULL);
5354 }
5355 
5356 static void *
5357 ztest_deadman_thread(void *arg)
5358 {
5359 	ztest_shared_t *zs = arg;
5360 	spa_t *spa = ztest_spa;
5361 	hrtime_t delta, total = 0;
5362 
5363 	for (;;) {
5364 		delta = zs->zs_thread_stop - zs->zs_thread_start +
5365 		    MSEC2NSEC(zfs_deadman_synctime_ms);
5366 
5367 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5368 
5369 		/*
5370 		 * If the pool is suspended then fail immediately. Otherwise,
5371 		 * check to see if the pool is making any progress. If
5372 		 * vdev_deadman() discovers that there hasn't been any recent
5373 		 * I/Os then it will end up aborting the tests.
5374 		 */
5375 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5376 			fatal(0, "aborting test after %llu seconds because "
5377 			    "pool has transitioned to a suspended state.",
5378 			    zfs_deadman_synctime_ms / 1000);
5379 			return (NULL);
5380 		}
5381 		vdev_deadman(spa->spa_root_vdev);
5382 
5383 		total += zfs_deadman_synctime_ms/1000;
5384 		(void) printf("ztest has been running for %lld seconds\n",
5385 		    total);
5386 	}
5387 }
5388 
5389 static void
5390 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5391 {
5392 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5393 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5394 	hrtime_t functime = gethrtime();
5395 
5396 	for (int i = 0; i < zi->zi_iters; i++)
5397 		zi->zi_func(zd, id);
5398 
5399 	functime = gethrtime() - functime;
5400 
5401 	atomic_add_64(&zc->zc_count, 1);
5402 	atomic_add_64(&zc->zc_time, functime);
5403 
5404 	if (ztest_opts.zo_verbose >= 4) {
5405 		Dl_info dli;
5406 		(void) dladdr((void *)zi->zi_func, &dli);
5407 		(void) printf("%6.2f sec in %s\n",
5408 		    (double)functime / NANOSEC, dli.dli_sname);
5409 	}
5410 }
5411 
5412 static void *
5413 ztest_thread(void *arg)
5414 {
5415 	int rand;
5416 	uint64_t id = (uintptr_t)arg;
5417 	ztest_shared_t *zs = ztest_shared;
5418 	uint64_t call_next;
5419 	hrtime_t now;
5420 	ztest_info_t *zi;
5421 	ztest_shared_callstate_t *zc;
5422 
5423 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5424 		/*
5425 		 * See if it's time to force a crash.
5426 		 */
5427 		if (now > zs->zs_thread_kill)
5428 			ztest_kill(zs);
5429 
5430 		/*
5431 		 * If we're getting ENOSPC with some regularity, stop.
5432 		 */
5433 		if (zs->zs_enospc_count > 10)
5434 			break;
5435 
5436 		/*
5437 		 * Pick a random function to execute.
5438 		 */
5439 		rand = ztest_random(ZTEST_FUNCS);
5440 		zi = &ztest_info[rand];
5441 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5442 		call_next = zc->zc_next;
5443 
5444 		if (now >= call_next &&
5445 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5446 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5447 			ztest_execute(rand, zi, id);
5448 		}
5449 	}
5450 
5451 	return (NULL);
5452 }
5453 
5454 static void
5455 ztest_dataset_name(char *dsname, char *pool, int d)
5456 {
5457 	(void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5458 }
5459 
5460 static void
5461 ztest_dataset_destroy(int d)
5462 {
5463 	char name[MAXNAMELEN];
5464 
5465 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5466 
5467 	if (ztest_opts.zo_verbose >= 3)
5468 		(void) printf("Destroying %s to free up space\n", name);
5469 
5470 	/*
5471 	 * Cleanup any non-standard clones and snapshots.  In general,
5472 	 * ztest thread t operates on dataset (t % zopt_datasets),
5473 	 * so there may be more than one thing to clean up.
5474 	 */
5475 	for (int t = d; t < ztest_opts.zo_threads;
5476 	    t += ztest_opts.zo_datasets) {
5477 		ztest_dsl_dataset_cleanup(name, t);
5478 	}
5479 
5480 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5481 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5482 }
5483 
5484 static void
5485 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5486 {
5487 	uint64_t usedobjs, dirobjs, scratch;
5488 
5489 	/*
5490 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5491 	 * Therefore, the number of objects in use should equal the
5492 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5493 	 * If not, we have an object leak.
5494 	 *
5495 	 * Note that we can only check this in ztest_dataset_open(),
5496 	 * when the open-context and syncing-context values agree.
5497 	 * That's because zap_count() returns the open-context value,
5498 	 * while dmu_objset_space() returns the rootbp fill count.
5499 	 */
5500 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5501 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5502 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5503 }
5504 
5505 static int
5506 ztest_dataset_open(int d)
5507 {
5508 	ztest_ds_t *zd = &ztest_ds[d];
5509 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5510 	objset_t *os;
5511 	zilog_t *zilog;
5512 	char name[MAXNAMELEN];
5513 	int error;
5514 
5515 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5516 
5517 	(void) rw_rdlock(&ztest_name_lock);
5518 
5519 	error = ztest_dataset_create(name);
5520 	if (error == ENOSPC) {
5521 		(void) rw_unlock(&ztest_name_lock);
5522 		ztest_record_enospc(FTAG);
5523 		return (error);
5524 	}
5525 	ASSERT(error == 0 || error == EEXIST);
5526 
5527 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5528 	(void) rw_unlock(&ztest_name_lock);
5529 
5530 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5531 
5532 	zilog = zd->zd_zilog;
5533 
5534 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5535 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5536 		fatal(0, "missing log records: claimed %llu < committed %llu",
5537 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5538 
5539 	ztest_dataset_dirobj_verify(zd);
5540 
5541 	zil_replay(os, zd, ztest_replay_vector);
5542 
5543 	ztest_dataset_dirobj_verify(zd);
5544 
5545 	if (ztest_opts.zo_verbose >= 6)
5546 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5547 		    zd->zd_name,
5548 		    (u_longlong_t)zilog->zl_parse_blk_count,
5549 		    (u_longlong_t)zilog->zl_parse_lr_count,
5550 		    (u_longlong_t)zilog->zl_replaying_seq);
5551 
5552 	zilog = zil_open(os, ztest_get_data);
5553 
5554 	if (zilog->zl_replaying_seq != 0 &&
5555 	    zilog->zl_replaying_seq < committed_seq)
5556 		fatal(0, "missing log records: replayed %llu < committed %llu",
5557 		    zilog->zl_replaying_seq, committed_seq);
5558 
5559 	return (0);
5560 }
5561 
5562 static void
5563 ztest_dataset_close(int d)
5564 {
5565 	ztest_ds_t *zd = &ztest_ds[d];
5566 
5567 	zil_close(zd->zd_zilog);
5568 	dmu_objset_disown(zd->zd_os, zd);
5569 
5570 	ztest_zd_fini(zd);
5571 }
5572 
5573 /*
5574  * Kick off threads to run tests on all datasets in parallel.
5575  */
5576 static void
5577 ztest_run(ztest_shared_t *zs)
5578 {
5579 	thread_t *tid;
5580 	spa_t *spa;
5581 	objset_t *os;
5582 	thread_t resume_tid;
5583 	int error;
5584 
5585 	ztest_exiting = B_FALSE;
5586 
5587 	/*
5588 	 * Initialize parent/child shared state.
5589 	 */
5590 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5591 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5592 
5593 	zs->zs_thread_start = gethrtime();
5594 	zs->zs_thread_stop =
5595 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5596 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5597 	zs->zs_thread_kill = zs->zs_thread_stop;
5598 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5599 		zs->zs_thread_kill -=
5600 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5601 	}
5602 
5603 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5604 
5605 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5606 	    offsetof(ztest_cb_data_t, zcd_node));
5607 
5608 	/*
5609 	 * Open our pool.
5610 	 */
5611 	kernel_init(FREAD | FWRITE);
5612 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5613 	spa->spa_debug = B_TRUE;
5614 	metaslab_preload_limit = ztest_random(20) + 1;
5615 	ztest_spa = spa;
5616 
5617 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5618 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
5619 	zs->zs_guid = dmu_objset_fsid_guid(os);
5620 	dmu_objset_disown(os, FTAG);
5621 
5622 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5623 
5624 	/*
5625 	 * We don't expect the pool to suspend unless maxfaults == 0,
5626 	 * in which case ztest_fault_inject() temporarily takes away
5627 	 * the only valid replica.
5628 	 */
5629 	if (MAXFAULTS() == 0)
5630 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5631 	else
5632 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5633 
5634 	/*
5635 	 * Create a thread to periodically resume suspended I/O.
5636 	 */
5637 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5638 	    &resume_tid) == 0);
5639 
5640 	/*
5641 	 * Create a deadman thread to abort() if we hang.
5642 	 */
5643 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5644 	    NULL) == 0);
5645 
5646 	/*
5647 	 * Verify that we can safely inquire about about any object,
5648 	 * whether it's allocated or not.  To make it interesting,
5649 	 * we probe a 5-wide window around each power of two.
5650 	 * This hits all edge cases, including zero and the max.
5651 	 */
5652 	for (int t = 0; t < 64; t++) {
5653 		for (int d = -5; d <= 5; d++) {
5654 			error = dmu_object_info(spa->spa_meta_objset,
5655 			    (1ULL << t) + d, NULL);
5656 			ASSERT(error == 0 || error == ENOENT ||
5657 			    error == EINVAL);
5658 		}
5659 	}
5660 
5661 	/*
5662 	 * If we got any ENOSPC errors on the previous run, destroy something.
5663 	 */
5664 	if (zs->zs_enospc_count != 0) {
5665 		int d = ztest_random(ztest_opts.zo_datasets);
5666 		ztest_dataset_destroy(d);
5667 	}
5668 	zs->zs_enospc_count = 0;
5669 
5670 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5671 	    UMEM_NOFAIL);
5672 
5673 	if (ztest_opts.zo_verbose >= 4)
5674 		(void) printf("starting main threads...\n");
5675 
5676 	/*
5677 	 * Kick off all the tests that run in parallel.
5678 	 */
5679 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
5680 		if (t < ztest_opts.zo_datasets &&
5681 		    ztest_dataset_open(t) != 0)
5682 			return;
5683 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5684 		    THR_BOUND, &tid[t]) == 0);
5685 	}
5686 
5687 	/*
5688 	 * Wait for all of the tests to complete.  We go in reverse order
5689 	 * so we don't close datasets while threads are still using them.
5690 	 */
5691 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5692 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5693 		if (t < ztest_opts.zo_datasets)
5694 			ztest_dataset_close(t);
5695 	}
5696 
5697 	txg_wait_synced(spa_get_dsl(spa), 0);
5698 
5699 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5700 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5701 	zfs_dbgmsg_print(FTAG);
5702 
5703 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5704 
5705 	/* Kill the resume thread */
5706 	ztest_exiting = B_TRUE;
5707 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5708 	ztest_resume(spa);
5709 
5710 	/*
5711 	 * Right before closing the pool, kick off a bunch of async I/O;
5712 	 * spa_close() should wait for it to complete.
5713 	 */
5714 	for (uint64_t object = 1; object < 50; object++)
5715 		dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5716 
5717 	spa_close(spa, FTAG);
5718 
5719 	/*
5720 	 * Verify that we can loop over all pools.
5721 	 */
5722 	mutex_enter(&spa_namespace_lock);
5723 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5724 		if (ztest_opts.zo_verbose > 3)
5725 			(void) printf("spa_next: found %s\n", spa_name(spa));
5726 	mutex_exit(&spa_namespace_lock);
5727 
5728 	/*
5729 	 * Verify that we can export the pool and reimport it under a
5730 	 * different name.
5731 	 */
5732 	if (ztest_random(2) == 0) {
5733 		char name[MAXNAMELEN];
5734 		(void) snprintf(name, MAXNAMELEN, "%s_import",
5735 		    ztest_opts.zo_pool);
5736 		ztest_spa_import_export(ztest_opts.zo_pool, name);
5737 		ztest_spa_import_export(name, ztest_opts.zo_pool);
5738 	}
5739 
5740 	kernel_fini();
5741 
5742 	list_destroy(&zcl.zcl_callbacks);
5743 
5744 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5745 
5746 	(void) rwlock_destroy(&ztest_name_lock);
5747 	(void) _mutex_destroy(&ztest_vdev_lock);
5748 }
5749 
5750 static void
5751 ztest_freeze(void)
5752 {
5753 	ztest_ds_t *zd = &ztest_ds[0];
5754 	spa_t *spa;
5755 	int numloops = 0;
5756 
5757 	if (ztest_opts.zo_verbose >= 3)
5758 		(void) printf("testing spa_freeze()...\n");
5759 
5760 	kernel_init(FREAD | FWRITE);
5761 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5762 	VERIFY3U(0, ==, ztest_dataset_open(0));
5763 	spa->spa_debug = B_TRUE;
5764 	ztest_spa = spa;
5765 
5766 	/*
5767 	 * Force the first log block to be transactionally allocated.
5768 	 * We have to do this before we freeze the pool -- otherwise
5769 	 * the log chain won't be anchored.
5770 	 */
5771 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5772 		ztest_dmu_object_alloc_free(zd, 0);
5773 		zil_commit(zd->zd_zilog, 0);
5774 	}
5775 
5776 	txg_wait_synced(spa_get_dsl(spa), 0);
5777 
5778 	/*
5779 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5780 	 * so that the only way to record changes from now on is the ZIL.
5781 	 */
5782 	spa_freeze(spa);
5783 
5784 	/*
5785 	 * Because it is hard to predict how much space a write will actually
5786 	 * require beforehand, we leave ourselves some fudge space to write over
5787 	 * capacity.
5788 	 */
5789 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5790 
5791 	/*
5792 	 * Run tests that generate log records but don't alter the pool config
5793 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5794 	 * We do a txg_wait_synced() after each iteration to force the txg
5795 	 * to increase well beyond the last synced value in the uberblock.
5796 	 * The ZIL should be OK with that.
5797 	 *
5798 	 * Run a random number of times less than zo_maxloops and ensure we do
5799 	 * not run out of space on the pool.
5800 	 */
5801 	while (ztest_random(10) != 0 &&
5802 	    numloops++ < ztest_opts.zo_maxloops &&
5803 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5804 		ztest_od_t od;
5805 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5806 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5807 		ztest_io(zd, od.od_object,
5808 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5809 		txg_wait_synced(spa_get_dsl(spa), 0);
5810 	}
5811 
5812 	/*
5813 	 * Commit all of the changes we just generated.
5814 	 */
5815 	zil_commit(zd->zd_zilog, 0);
5816 	txg_wait_synced(spa_get_dsl(spa), 0);
5817 
5818 	/*
5819 	 * Close our dataset and close the pool.
5820 	 */
5821 	ztest_dataset_close(0);
5822 	spa_close(spa, FTAG);
5823 	kernel_fini();
5824 
5825 	/*
5826 	 * Open and close the pool and dataset to induce log replay.
5827 	 */
5828 	kernel_init(FREAD | FWRITE);
5829 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5830 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5831 	VERIFY3U(0, ==, ztest_dataset_open(0));
5832 	ztest_dataset_close(0);
5833 
5834 	spa->spa_debug = B_TRUE;
5835 	ztest_spa = spa;
5836 	txg_wait_synced(spa_get_dsl(spa), 0);
5837 	ztest_reguid(NULL, 0);
5838 
5839 	spa_close(spa, FTAG);
5840 	kernel_fini();
5841 }
5842 
5843 void
5844 print_time(hrtime_t t, char *timebuf)
5845 {
5846 	hrtime_t s = t / NANOSEC;
5847 	hrtime_t m = s / 60;
5848 	hrtime_t h = m / 60;
5849 	hrtime_t d = h / 24;
5850 
5851 	s -= m * 60;
5852 	m -= h * 60;
5853 	h -= d * 24;
5854 
5855 	timebuf[0] = '\0';
5856 
5857 	if (d)
5858 		(void) sprintf(timebuf,
5859 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5860 	else if (h)
5861 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5862 	else if (m)
5863 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5864 	else
5865 		(void) sprintf(timebuf, "%llus", s);
5866 }
5867 
5868 static nvlist_t *
5869 make_random_props()
5870 {
5871 	nvlist_t *props;
5872 
5873 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5874 	if (ztest_random(2) == 0)
5875 		return (props);
5876 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5877 
5878 	return (props);
5879 }
5880 
5881 /*
5882  * Create a storage pool with the given name and initial vdev size.
5883  * Then test spa_freeze() functionality.
5884  */
5885 static void
5886 ztest_init(ztest_shared_t *zs)
5887 {
5888 	spa_t *spa;
5889 	nvlist_t *nvroot, *props;
5890 
5891 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5892 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5893 
5894 	kernel_init(FREAD | FWRITE);
5895 
5896 	/*
5897 	 * Create the storage pool.
5898 	 */
5899 	(void) spa_destroy(ztest_opts.zo_pool);
5900 	ztest_shared->zs_vdev_next_leaf = 0;
5901 	zs->zs_splits = 0;
5902 	zs->zs_mirrors = ztest_opts.zo_mirrors;
5903 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5904 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5905 	props = make_random_props();
5906 	for (int i = 0; i < SPA_FEATURES; i++) {
5907 		char buf[1024];
5908 		(void) snprintf(buf, sizeof (buf), "feature@%s",
5909 		    spa_feature_table[i].fi_uname);
5910 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5911 	}
5912 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5913 	nvlist_free(nvroot);
5914 
5915 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5916 	zs->zs_metaslab_sz =
5917 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5918 
5919 	spa_close(spa, FTAG);
5920 
5921 	kernel_fini();
5922 
5923 	ztest_run_zdb(ztest_opts.zo_pool);
5924 
5925 	ztest_freeze();
5926 
5927 	ztest_run_zdb(ztest_opts.zo_pool);
5928 
5929 	(void) rwlock_destroy(&ztest_name_lock);
5930 	(void) _mutex_destroy(&ztest_vdev_lock);
5931 }
5932 
5933 static void
5934 setup_data_fd(void)
5935 {
5936 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5937 
5938 	ztest_fd_data = mkstemp(ztest_name_data);
5939 	ASSERT3S(ztest_fd_data, >=, 0);
5940 	(void) unlink(ztest_name_data);
5941 }
5942 
5943 
5944 static int
5945 shared_data_size(ztest_shared_hdr_t *hdr)
5946 {
5947 	int size;
5948 
5949 	size = hdr->zh_hdr_size;
5950 	size += hdr->zh_opts_size;
5951 	size += hdr->zh_size;
5952 	size += hdr->zh_stats_size * hdr->zh_stats_count;
5953 	size += hdr->zh_ds_size * hdr->zh_ds_count;
5954 
5955 	return (size);
5956 }
5957 
5958 static void
5959 setup_hdr(void)
5960 {
5961 	int size;
5962 	ztest_shared_hdr_t *hdr;
5963 
5964 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5965 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5966 	ASSERT(hdr != MAP_FAILED);
5967 
5968 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5969 
5970 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5971 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5972 	hdr->zh_size = sizeof (ztest_shared_t);
5973 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5974 	hdr->zh_stats_count = ZTEST_FUNCS;
5975 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5976 	hdr->zh_ds_count = ztest_opts.zo_datasets;
5977 
5978 	size = shared_data_size(hdr);
5979 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5980 
5981 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5982 }
5983 
5984 static void
5985 setup_data(void)
5986 {
5987 	int size, offset;
5988 	ztest_shared_hdr_t *hdr;
5989 	uint8_t *buf;
5990 
5991 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5992 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5993 	ASSERT(hdr != MAP_FAILED);
5994 
5995 	size = shared_data_size(hdr);
5996 
5997 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5998 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5999 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6000 	ASSERT(hdr != MAP_FAILED);
6001 	buf = (uint8_t *)hdr;
6002 
6003 	offset = hdr->zh_hdr_size;
6004 	ztest_shared_opts = (void *)&buf[offset];
6005 	offset += hdr->zh_opts_size;
6006 	ztest_shared = (void *)&buf[offset];
6007 	offset += hdr->zh_size;
6008 	ztest_shared_callstate = (void *)&buf[offset];
6009 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
6010 	ztest_shared_ds = (void *)&buf[offset];
6011 }
6012 
6013 static boolean_t
6014 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6015 {
6016 	pid_t pid;
6017 	int status;
6018 	char *cmdbuf = NULL;
6019 
6020 	pid = fork();
6021 
6022 	if (cmd == NULL) {
6023 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6024 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6025 		cmd = cmdbuf;
6026 	}
6027 
6028 	if (pid == -1)
6029 		fatal(1, "fork failed");
6030 
6031 	if (pid == 0) {	/* child */
6032 		char *emptyargv[2] = { cmd, NULL };
6033 		char fd_data_str[12];
6034 
6035 		struct rlimit rl = { 1024, 1024 };
6036 		(void) setrlimit(RLIMIT_NOFILE, &rl);
6037 
6038 		(void) close(ztest_fd_rand);
6039 		VERIFY3U(11, >=,
6040 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6041 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6042 
6043 		(void) enable_extended_FILE_stdio(-1, -1);
6044 		if (libpath != NULL)
6045 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6046 		(void) execv(cmd, emptyargv);
6047 		ztest_dump_core = B_FALSE;
6048 		fatal(B_TRUE, "exec failed: %s", cmd);
6049 	}
6050 
6051 	if (cmdbuf != NULL) {
6052 		umem_free(cmdbuf, MAXPATHLEN);
6053 		cmd = NULL;
6054 	}
6055 
6056 	while (waitpid(pid, &status, 0) != pid)
6057 		continue;
6058 	if (statusp != NULL)
6059 		*statusp = status;
6060 
6061 	if (WIFEXITED(status)) {
6062 		if (WEXITSTATUS(status) != 0) {
6063 			(void) fprintf(stderr, "child exited with code %d\n",
6064 			    WEXITSTATUS(status));
6065 			exit(2);
6066 		}
6067 		return (B_FALSE);
6068 	} else if (WIFSIGNALED(status)) {
6069 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6070 			(void) fprintf(stderr, "child died with signal %d\n",
6071 			    WTERMSIG(status));
6072 			exit(3);
6073 		}
6074 		return (B_TRUE);
6075 	} else {
6076 		(void) fprintf(stderr, "something strange happened to child\n");
6077 		exit(4);
6078 		/* NOTREACHED */
6079 	}
6080 }
6081 
6082 static void
6083 ztest_run_init(void)
6084 {
6085 	ztest_shared_t *zs = ztest_shared;
6086 
6087 	ASSERT(ztest_opts.zo_init != 0);
6088 
6089 	/*
6090 	 * Blow away any existing copy of zpool.cache
6091 	 */
6092 	(void) remove(spa_config_path);
6093 
6094 	/*
6095 	 * Create and initialize our storage pool.
6096 	 */
6097 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
6098 		bzero(zs, sizeof (ztest_shared_t));
6099 		if (ztest_opts.zo_verbose >= 3 &&
6100 		    ztest_opts.zo_init != 1) {
6101 			(void) printf("ztest_init(), pass %d\n", i);
6102 		}
6103 		ztest_init(zs);
6104 	}
6105 }
6106 
6107 int
6108 main(int argc, char **argv)
6109 {
6110 	int kills = 0;
6111 	int iters = 0;
6112 	int older = 0;
6113 	int newer = 0;
6114 	ztest_shared_t *zs;
6115 	ztest_info_t *zi;
6116 	ztest_shared_callstate_t *zc;
6117 	char timebuf[100];
6118 	char numbuf[6];
6119 	spa_t *spa;
6120 	char *cmd;
6121 	boolean_t hasalt;
6122 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6123 
6124 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6125 
6126 	dprintf_setup(&argc, argv);
6127 	zfs_deadman_synctime_ms = 300000;
6128 
6129 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6130 	ASSERT3S(ztest_fd_rand, >=, 0);
6131 
6132 	if (!fd_data_str) {
6133 		process_options(argc, argv);
6134 
6135 		setup_data_fd();
6136 		setup_hdr();
6137 		setup_data();
6138 		bcopy(&ztest_opts, ztest_shared_opts,
6139 		    sizeof (*ztest_shared_opts));
6140 	} else {
6141 		ztest_fd_data = atoi(fd_data_str);
6142 		setup_data();
6143 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6144 	}
6145 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6146 
6147 	/* Override location of zpool.cache */
6148 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6149 	    ztest_opts.zo_dir), !=, -1);
6150 
6151 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6152 	    UMEM_NOFAIL);
6153 	zs = ztest_shared;
6154 
6155 	if (fd_data_str) {
6156 		metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6157 		metaslab_df_alloc_threshold =
6158 		    zs->zs_metaslab_df_alloc_threshold;
6159 
6160 		if (zs->zs_do_init)
6161 			ztest_run_init();
6162 		else
6163 			ztest_run(zs);
6164 		exit(0);
6165 	}
6166 
6167 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6168 
6169 	if (ztest_opts.zo_verbose >= 1) {
6170 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6171 		    " %llu seconds...\n",
6172 		    (u_longlong_t)ztest_opts.zo_vdevs,
6173 		    ztest_opts.zo_datasets,
6174 		    ztest_opts.zo_threads,
6175 		    (u_longlong_t)ztest_opts.zo_time);
6176 	}
6177 
6178 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6179 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6180 
6181 	zs->zs_do_init = B_TRUE;
6182 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6183 		if (ztest_opts.zo_verbose >= 1) {
6184 			(void) printf("Executing older ztest for "
6185 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6186 		}
6187 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6188 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6189 	} else {
6190 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6191 	}
6192 	zs->zs_do_init = B_FALSE;
6193 
6194 	zs->zs_proc_start = gethrtime();
6195 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6196 
6197 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6198 		zi = &ztest_info[f];
6199 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6200 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6201 			zc->zc_next = UINT64_MAX;
6202 		else
6203 			zc->zc_next = zs->zs_proc_start +
6204 			    ztest_random(2 * zi->zi_interval[0] + 1);
6205 	}
6206 
6207 	/*
6208 	 * Run the tests in a loop.  These tests include fault injection
6209 	 * to verify that self-healing data works, and forced crashes
6210 	 * to verify that we never lose on-disk consistency.
6211 	 */
6212 	while (gethrtime() < zs->zs_proc_stop) {
6213 		int status;
6214 		boolean_t killed;
6215 
6216 		/*
6217 		 * Initialize the workload counters for each function.
6218 		 */
6219 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6220 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6221 			zc->zc_count = 0;
6222 			zc->zc_time = 0;
6223 		}
6224 
6225 		/* Set the allocation switch size */
6226 		zs->zs_metaslab_df_alloc_threshold =
6227 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6228 
6229 		if (!hasalt || ztest_random(2) == 0) {
6230 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6231 				(void) printf("Executing newer ztest: %s\n",
6232 				    cmd);
6233 			}
6234 			newer++;
6235 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6236 		} else {
6237 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6238 				(void) printf("Executing older ztest: %s\n",
6239 				    ztest_opts.zo_alt_ztest);
6240 			}
6241 			older++;
6242 			killed = exec_child(ztest_opts.zo_alt_ztest,
6243 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6244 		}
6245 
6246 		if (killed)
6247 			kills++;
6248 		iters++;
6249 
6250 		if (ztest_opts.zo_verbose >= 1) {
6251 			hrtime_t now = gethrtime();
6252 
6253 			now = MIN(now, zs->zs_proc_stop);
6254 			print_time(zs->zs_proc_stop - now, timebuf);
6255 			nicenum(zs->zs_space, numbuf);
6256 
6257 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6258 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6259 			    iters,
6260 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6261 			    (u_longlong_t)zs->zs_enospc_count,
6262 			    100.0 * zs->zs_alloc / zs->zs_space,
6263 			    numbuf,
6264 			    100.0 * (now - zs->zs_proc_start) /
6265 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6266 		}
6267 
6268 		if (ztest_opts.zo_verbose >= 2) {
6269 			(void) printf("\nWorkload summary:\n\n");
6270 			(void) printf("%7s %9s   %s\n",
6271 			    "Calls", "Time", "Function");
6272 			(void) printf("%7s %9s   %s\n",
6273 			    "-----", "----", "--------");
6274 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6275 				Dl_info dli;
6276 
6277 				zi = &ztest_info[f];
6278 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6279 				print_time(zc->zc_time, timebuf);
6280 				(void) dladdr((void *)zi->zi_func, &dli);
6281 				(void) printf("%7llu %9s   %s\n",
6282 				    (u_longlong_t)zc->zc_count, timebuf,
6283 				    dli.dli_sname);
6284 			}
6285 			(void) printf("\n");
6286 		}
6287 
6288 		/*
6289 		 * It's possible that we killed a child during a rename test,
6290 		 * in which case we'll have a 'ztest_tmp' pool lying around
6291 		 * instead of 'ztest'.  Do a blind rename in case this happened.
6292 		 */
6293 		kernel_init(FREAD);
6294 		if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6295 			spa_close(spa, FTAG);
6296 		} else {
6297 			char tmpname[MAXNAMELEN];
6298 			kernel_fini();
6299 			kernel_init(FREAD | FWRITE);
6300 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6301 			    ztest_opts.zo_pool);
6302 			(void) spa_rename(tmpname, ztest_opts.zo_pool);
6303 		}
6304 		kernel_fini();
6305 
6306 		ztest_run_zdb(ztest_opts.zo_pool);
6307 	}
6308 
6309 	if (ztest_opts.zo_verbose >= 1) {
6310 		if (hasalt) {
6311 			(void) printf("%d runs of older ztest: %s\n", older,
6312 			    ztest_opts.zo_alt_ztest);
6313 			(void) printf("%d runs of newer ztest: %s\n", newer,
6314 			    cmd);
6315 		}
6316 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6317 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6318 	}
6319 
6320 	umem_free(cmd, MAXNAMELEN);
6321 
6322 	return (0);
6323 }
6324