xref: /illumos-gate/usr/src/cmd/ztest/ztest.c (revision 406fc5100dac8d225a315a6def6be8d628f34e24)
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, 2015 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  */
28 
29 /*
30  * The objective of this program is to provide a DMU/ZAP/SPA stress test
31  * that runs entirely in userland, is easy to use, and easy to extend.
32  *
33  * The overall design of the ztest program is as follows:
34  *
35  * (1) For each major functional area (e.g. adding vdevs to a pool,
36  *     creating and destroying datasets, reading and writing objects, etc)
37  *     we have a simple routine to test that functionality.  These
38  *     individual routines do not have to do anything "stressful".
39  *
40  * (2) We turn these simple functionality tests into a stress test by
41  *     running them all in parallel, with as many threads as desired,
42  *     and spread across as many datasets, objects, and vdevs as desired.
43  *
44  * (3) While all this is happening, we inject faults into the pool to
45  *     verify that self-healing data really works.
46  *
47  * (4) Every time we open a dataset, we change its checksum and compression
48  *     functions.  Thus even individual objects vary from block to block
49  *     in which checksum they use and whether they're compressed.
50  *
51  * (5) To verify that we never lose on-disk consistency after a crash,
52  *     we run the entire test in a child of the main process.
53  *     At random times, the child self-immolates with a SIGKILL.
54  *     This is the software equivalent of pulling the power cord.
55  *     The parent then runs the test again, using the existing
56  *     storage pool, as many times as desired. If backwards compatibility
57  *     testing is enabled ztest will sometimes run the "older" version
58  *     of ztest after a SIGKILL.
59  *
60  * (6) To verify that we don't have future leaks or temporal incursions,
61  *     many of the functional tests record the transaction group number
62  *     as part of their data.  When reading old data, they verify that
63  *     the transaction group number is less than the current, open txg.
64  *     If you add a new test, please do this if applicable.
65  *
66  * When run with no arguments, ztest runs for about five minutes and
67  * produces no output if successful.  To get a little bit of information,
68  * specify -V.  To get more information, specify -VV, and so on.
69  *
70  * To turn this into an overnight stress test, use -T to specify run time.
71  *
72  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
73  * to increase the pool capacity, fanout, and overall stress level.
74  *
75  * Use the -k option to set the desired frequency of kills.
76  *
77  * When ztest invokes itself it passes all relevant information through a
78  * temporary file which is mmap-ed in the child process. This allows shared
79  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
80  * stored at offset 0 of this file and contains information on the size and
81  * number of shared structures in the file. The information stored in this file
82  * must remain backwards compatible with older versions of ztest so that
83  * ztest can invoke them during backwards compatibility testing (-B).
84  */
85 
86 #include <sys/zfs_context.h>
87 #include <sys/spa.h>
88 #include <sys/dmu.h>
89 #include <sys/txg.h>
90 #include <sys/dbuf.h>
91 #include <sys/zap.h>
92 #include <sys/dmu_objset.h>
93 #include <sys/poll.h>
94 #include <sys/stat.h>
95 #include <sys/time.h>
96 #include <sys/wait.h>
97 #include <sys/mman.h>
98 #include <sys/resource.h>
99 #include <sys/zio.h>
100 #include <sys/zil.h>
101 #include <sys/zil_impl.h>
102 #include <sys/vdev_impl.h>
103 #include <sys/vdev_file.h>
104 #include <sys/spa_impl.h>
105 #include <sys/metaslab_impl.h>
106 #include <sys/dsl_prop.h>
107 #include <sys/dsl_dataset.h>
108 #include <sys/dsl_destroy.h>
109 #include <sys/dsl_scan.h>
110 #include <sys/zio_checksum.h>
111 #include <sys/refcount.h>
112 #include <sys/zfeature.h>
113 #include <sys/dsl_userhold.h>
114 #include <stdio.h>
115 #include <stdio_ext.h>
116 #include <stdlib.h>
117 #include <unistd.h>
118 #include <signal.h>
119 #include <umem.h>
120 #include <dlfcn.h>
121 #include <ctype.h>
122 #include <math.h>
123 #include <sys/fs/zfs.h>
124 #include <libnvpair.h>
125 
126 static int ztest_fd_data = -1;
127 static int ztest_fd_rand = -1;
128 
129 typedef struct ztest_shared_hdr {
130 	uint64_t	zh_hdr_size;
131 	uint64_t	zh_opts_size;
132 	uint64_t	zh_size;
133 	uint64_t	zh_stats_size;
134 	uint64_t	zh_stats_count;
135 	uint64_t	zh_ds_size;
136 	uint64_t	zh_ds_count;
137 } ztest_shared_hdr_t;
138 
139 static ztest_shared_hdr_t *ztest_shared_hdr;
140 
141 typedef struct ztest_shared_opts {
142 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
143 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
144 	char zo_alt_ztest[MAXNAMELEN];
145 	char zo_alt_libpath[MAXNAMELEN];
146 	uint64_t zo_vdevs;
147 	uint64_t zo_vdevtime;
148 	size_t zo_vdev_size;
149 	int zo_ashift;
150 	int zo_mirrors;
151 	int zo_raidz;
152 	int zo_raidz_parity;
153 	int zo_datasets;
154 	int zo_threads;
155 	uint64_t zo_passtime;
156 	uint64_t zo_killrate;
157 	int zo_verbose;
158 	int zo_init;
159 	uint64_t zo_time;
160 	uint64_t zo_maxloops;
161 	uint64_t zo_metaslab_gang_bang;
162 } ztest_shared_opts_t;
163 
164 static const ztest_shared_opts_t ztest_opts_defaults = {
165 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
166 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
167 	.zo_alt_ztest = { '\0' },
168 	.zo_alt_libpath = { '\0' },
169 	.zo_vdevs = 5,
170 	.zo_ashift = SPA_MINBLOCKSHIFT,
171 	.zo_mirrors = 2,
172 	.zo_raidz = 4,
173 	.zo_raidz_parity = 1,
174 	.zo_vdev_size = SPA_MINDEVSIZE * 2,
175 	.zo_datasets = 7,
176 	.zo_threads = 23,
177 	.zo_passtime = 60,		/* 60 seconds */
178 	.zo_killrate = 70,		/* 70% kill rate */
179 	.zo_verbose = 0,
180 	.zo_init = 1,
181 	.zo_time = 300,			/* 5 minutes */
182 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
183 	.zo_metaslab_gang_bang = 32 << 10
184 };
185 
186 extern uint64_t metaslab_gang_bang;
187 extern uint64_t metaslab_df_alloc_threshold;
188 extern uint64_t zfs_deadman_synctime_ms;
189 extern int metaslab_preload_limit;
190 
191 static ztest_shared_opts_t *ztest_shared_opts;
192 static ztest_shared_opts_t ztest_opts;
193 
194 typedef struct ztest_shared_ds {
195 	uint64_t	zd_seq;
196 } ztest_shared_ds_t;
197 
198 static ztest_shared_ds_t *ztest_shared_ds;
199 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
200 
201 #define	BT_MAGIC	0x123456789abcdefULL
202 #define	MAXFAULTS() \
203 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
204 
205 enum ztest_io_type {
206 	ZTEST_IO_WRITE_TAG,
207 	ZTEST_IO_WRITE_PATTERN,
208 	ZTEST_IO_WRITE_ZEROES,
209 	ZTEST_IO_TRUNCATE,
210 	ZTEST_IO_SETATTR,
211 	ZTEST_IO_REWRITE,
212 	ZTEST_IO_TYPES
213 };
214 
215 typedef struct ztest_block_tag {
216 	uint64_t	bt_magic;
217 	uint64_t	bt_objset;
218 	uint64_t	bt_object;
219 	uint64_t	bt_offset;
220 	uint64_t	bt_gen;
221 	uint64_t	bt_txg;
222 	uint64_t	bt_crtxg;
223 } ztest_block_tag_t;
224 
225 typedef struct bufwad {
226 	uint64_t	bw_index;
227 	uint64_t	bw_txg;
228 	uint64_t	bw_data;
229 } bufwad_t;
230 
231 /*
232  * XXX -- fix zfs range locks to be generic so we can use them here.
233  */
234 typedef enum {
235 	RL_READER,
236 	RL_WRITER,
237 	RL_APPEND
238 } rl_type_t;
239 
240 typedef struct rll {
241 	void		*rll_writer;
242 	int		rll_readers;
243 	mutex_t		rll_lock;
244 	cond_t		rll_cv;
245 } rll_t;
246 
247 typedef struct rl {
248 	uint64_t	rl_object;
249 	uint64_t	rl_offset;
250 	uint64_t	rl_size;
251 	rll_t		*rl_lock;
252 } rl_t;
253 
254 #define	ZTEST_RANGE_LOCKS	64
255 #define	ZTEST_OBJECT_LOCKS	64
256 
257 /*
258  * Object descriptor.  Used as a template for object lookup/create/remove.
259  */
260 typedef struct ztest_od {
261 	uint64_t	od_dir;
262 	uint64_t	od_object;
263 	dmu_object_type_t od_type;
264 	dmu_object_type_t od_crtype;
265 	uint64_t	od_blocksize;
266 	uint64_t	od_crblocksize;
267 	uint64_t	od_gen;
268 	uint64_t	od_crgen;
269 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
270 } ztest_od_t;
271 
272 /*
273  * Per-dataset state.
274  */
275 typedef struct ztest_ds {
276 	ztest_shared_ds_t *zd_shared;
277 	objset_t	*zd_os;
278 	rwlock_t	zd_zilog_lock;
279 	zilog_t		*zd_zilog;
280 	ztest_od_t	*zd_od;		/* debugging aid */
281 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
282 	mutex_t		zd_dirobj_lock;
283 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
284 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
285 } ztest_ds_t;
286 
287 /*
288  * Per-iteration state.
289  */
290 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
291 
292 typedef struct ztest_info {
293 	ztest_func_t	*zi_func;	/* test function */
294 	uint64_t	zi_iters;	/* iterations per execution */
295 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
296 } ztest_info_t;
297 
298 typedef struct ztest_shared_callstate {
299 	uint64_t	zc_count;	/* per-pass count */
300 	uint64_t	zc_time;	/* per-pass time */
301 	uint64_t	zc_next;	/* next time to call this function */
302 } ztest_shared_callstate_t;
303 
304 static ztest_shared_callstate_t *ztest_shared_callstate;
305 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
306 
307 /*
308  * Note: these aren't static because we want dladdr() to work.
309  */
310 ztest_func_t ztest_dmu_read_write;
311 ztest_func_t ztest_dmu_write_parallel;
312 ztest_func_t ztest_dmu_object_alloc_free;
313 ztest_func_t ztest_dmu_commit_callbacks;
314 ztest_func_t ztest_zap;
315 ztest_func_t ztest_zap_parallel;
316 ztest_func_t ztest_zil_commit;
317 ztest_func_t ztest_zil_remount;
318 ztest_func_t ztest_dmu_read_write_zcopy;
319 ztest_func_t ztest_dmu_objset_create_destroy;
320 ztest_func_t ztest_dmu_prealloc;
321 ztest_func_t ztest_fzap;
322 ztest_func_t ztest_dmu_snapshot_create_destroy;
323 ztest_func_t ztest_dsl_prop_get_set;
324 ztest_func_t ztest_spa_prop_get_set;
325 ztest_func_t ztest_spa_create_destroy;
326 ztest_func_t ztest_fault_inject;
327 ztest_func_t ztest_ddt_repair;
328 ztest_func_t ztest_dmu_snapshot_hold;
329 ztest_func_t ztest_spa_rename;
330 ztest_func_t ztest_scrub;
331 ztest_func_t ztest_dsl_dataset_promote_busy;
332 ztest_func_t ztest_vdev_attach_detach;
333 ztest_func_t ztest_vdev_LUN_growth;
334 ztest_func_t ztest_vdev_add_remove;
335 ztest_func_t ztest_vdev_aux_add_remove;
336 ztest_func_t ztest_split_pool;
337 ztest_func_t ztest_reguid;
338 ztest_func_t ztest_spa_upgrade;
339 
340 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
341 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
342 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
343 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
344 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
345 
346 ztest_info_t ztest_info[] = {
347 	{ ztest_dmu_read_write,			1,	&zopt_always	},
348 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
349 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
350 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
351 	{ ztest_zap,				30,	&zopt_always	},
352 	{ ztest_zap_parallel,			100,	&zopt_always	},
353 	{ ztest_split_pool,			1,	&zopt_always	},
354 	{ ztest_zil_commit,			1,	&zopt_incessant	},
355 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
356 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
357 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
358 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
359 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
360 #if 0
361 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
362 #endif
363 	{ ztest_fzap,				1,	&zopt_sometimes	},
364 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
365 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
366 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
367 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
368 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
369 	{ ztest_reguid,				1,	&zopt_rarely	},
370 	{ ztest_spa_rename,			1,	&zopt_rarely	},
371 	{ ztest_scrub,				1,	&zopt_rarely	},
372 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
373 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
374 	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
375 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
376 	{ ztest_vdev_add_remove,		1,
377 	    &ztest_opts.zo_vdevtime				},
378 	{ ztest_vdev_aux_add_remove,		1,
379 	    &ztest_opts.zo_vdevtime				},
380 };
381 
382 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
383 
384 /*
385  * The following struct is used to hold a list of uncalled commit callbacks.
386  * The callbacks are ordered by txg number.
387  */
388 typedef struct ztest_cb_list {
389 	mutex_t	zcl_callbacks_lock;
390 	list_t	zcl_callbacks;
391 } ztest_cb_list_t;
392 
393 /*
394  * Stuff we need to share writably between parent and child.
395  */
396 typedef struct ztest_shared {
397 	boolean_t	zs_do_init;
398 	hrtime_t	zs_proc_start;
399 	hrtime_t	zs_proc_stop;
400 	hrtime_t	zs_thread_start;
401 	hrtime_t	zs_thread_stop;
402 	hrtime_t	zs_thread_kill;
403 	uint64_t	zs_enospc_count;
404 	uint64_t	zs_vdev_next_leaf;
405 	uint64_t	zs_vdev_aux;
406 	uint64_t	zs_alloc;
407 	uint64_t	zs_space;
408 	uint64_t	zs_splits;
409 	uint64_t	zs_mirrors;
410 	uint64_t	zs_metaslab_sz;
411 	uint64_t	zs_metaslab_df_alloc_threshold;
412 	uint64_t	zs_guid;
413 } ztest_shared_t;
414 
415 #define	ID_PARALLEL	-1ULL
416 
417 static char ztest_dev_template[] = "%s/%s.%llua";
418 static char ztest_aux_template[] = "%s/%s.%s.%llu";
419 ztest_shared_t *ztest_shared;
420 
421 static spa_t *ztest_spa = NULL;
422 static ztest_ds_t *ztest_ds;
423 
424 static mutex_t ztest_vdev_lock;
425 
426 /*
427  * The ztest_name_lock protects the pool and dataset namespace used by
428  * the individual tests. To modify the namespace, consumers must grab
429  * this lock as writer. Grabbing the lock as reader will ensure that the
430  * namespace does not change while the lock is held.
431  */
432 static rwlock_t ztest_name_lock;
433 
434 static boolean_t ztest_dump_core = B_TRUE;
435 static boolean_t ztest_exiting;
436 
437 /* Global commit callback list */
438 static ztest_cb_list_t zcl;
439 
440 enum ztest_object {
441 	ZTEST_META_DNODE = 0,
442 	ZTEST_DIROBJ,
443 	ZTEST_OBJECTS
444 };
445 
446 static void usage(boolean_t) __NORETURN;
447 
448 /*
449  * These libumem hooks provide a reasonable set of defaults for the allocator's
450  * debugging facilities.
451  */
452 const char *
453 _umem_debug_init()
454 {
455 	return ("default,verbose"); /* $UMEM_DEBUG setting */
456 }
457 
458 const char *
459 _umem_logging_init(void)
460 {
461 	return ("fail,contents"); /* $UMEM_LOGGING setting */
462 }
463 
464 #define	FATAL_MSG_SZ	1024
465 
466 char *fatal_msg;
467 
468 static void
469 fatal(int do_perror, char *message, ...)
470 {
471 	va_list args;
472 	int save_errno = errno;
473 	char buf[FATAL_MSG_SZ];
474 
475 	(void) fflush(stdout);
476 
477 	va_start(args, message);
478 	(void) sprintf(buf, "ztest: ");
479 	/* LINTED */
480 	(void) vsprintf(buf + strlen(buf), message, args);
481 	va_end(args);
482 	if (do_perror) {
483 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
484 		    ": %s", strerror(save_errno));
485 	}
486 	(void) fprintf(stderr, "%s\n", buf);
487 	fatal_msg = buf;			/* to ease debugging */
488 	if (ztest_dump_core)
489 		abort();
490 	exit(3);
491 }
492 
493 static int
494 str2shift(const char *buf)
495 {
496 	const char *ends = "BKMGTPEZ";
497 	int i;
498 
499 	if (buf[0] == '\0')
500 		return (0);
501 	for (i = 0; i < strlen(ends); i++) {
502 		if (toupper(buf[0]) == ends[i])
503 			break;
504 	}
505 	if (i == strlen(ends)) {
506 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
507 		    buf);
508 		usage(B_FALSE);
509 	}
510 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
511 		return (10*i);
512 	}
513 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
514 	usage(B_FALSE);
515 	/* NOTREACHED */
516 }
517 
518 static uint64_t
519 nicenumtoull(const char *buf)
520 {
521 	char *end;
522 	uint64_t val;
523 
524 	val = strtoull(buf, &end, 0);
525 	if (end == buf) {
526 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
527 		usage(B_FALSE);
528 	} else if (end[0] == '.') {
529 		double fval = strtod(buf, &end);
530 		fval *= pow(2, str2shift(end));
531 		if (fval > UINT64_MAX) {
532 			(void) fprintf(stderr, "ztest: value too large: %s\n",
533 			    buf);
534 			usage(B_FALSE);
535 		}
536 		val = (uint64_t)fval;
537 	} else {
538 		int shift = str2shift(end);
539 		if (shift >= 64 || (val << shift) >> shift != val) {
540 			(void) fprintf(stderr, "ztest: value too large: %s\n",
541 			    buf);
542 			usage(B_FALSE);
543 		}
544 		val <<= shift;
545 	}
546 	return (val);
547 }
548 
549 static void
550 usage(boolean_t requested)
551 {
552 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
553 
554 	char nice_vdev_size[10];
555 	char nice_gang_bang[10];
556 	FILE *fp = requested ? stdout : stderr;
557 
558 	nicenum(zo->zo_vdev_size, nice_vdev_size);
559 	nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
560 
561 	(void) fprintf(fp, "Usage: %s\n"
562 	    "\t[-v vdevs (default: %llu)]\n"
563 	    "\t[-s size_of_each_vdev (default: %s)]\n"
564 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
565 	    "\t[-m mirror_copies (default: %d)]\n"
566 	    "\t[-r raidz_disks (default: %d)]\n"
567 	    "\t[-R raidz_parity (default: %d)]\n"
568 	    "\t[-d datasets (default: %d)]\n"
569 	    "\t[-t threads (default: %d)]\n"
570 	    "\t[-g gang_block_threshold (default: %s)]\n"
571 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
572 	    "\t[-k kill_percentage (default: %llu%%)]\n"
573 	    "\t[-p pool_name (default: %s)]\n"
574 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
575 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
576 	    "\t[-E] use existing pool instead of creating new one\n"
577 	    "\t[-T time (default: %llu sec)] total run time\n"
578 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
579 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
580 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
581 	    "\t[-h] (print help)\n"
582 	    "",
583 	    zo->zo_pool,
584 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
585 	    nice_vdev_size,				/* -s */
586 	    zo->zo_ashift,				/* -a */
587 	    zo->zo_mirrors,				/* -m */
588 	    zo->zo_raidz,				/* -r */
589 	    zo->zo_raidz_parity,			/* -R */
590 	    zo->zo_datasets,				/* -d */
591 	    zo->zo_threads,				/* -t */
592 	    nice_gang_bang,				/* -g */
593 	    zo->zo_init,				/* -i */
594 	    (u_longlong_t)zo->zo_killrate,		/* -k */
595 	    zo->zo_pool,				/* -p */
596 	    zo->zo_dir,					/* -f */
597 	    (u_longlong_t)zo->zo_time,			/* -T */
598 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
599 	    (u_longlong_t)zo->zo_passtime);
600 	exit(requested ? 0 : 1);
601 }
602 
603 static void
604 process_options(int argc, char **argv)
605 {
606 	char *path;
607 	ztest_shared_opts_t *zo = &ztest_opts;
608 
609 	int opt;
610 	uint64_t value;
611 	char altdir[MAXNAMELEN] = { 0 };
612 
613 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
614 
615 	while ((opt = getopt(argc, argv,
616 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
617 		value = 0;
618 		switch (opt) {
619 		case 'v':
620 		case 's':
621 		case 'a':
622 		case 'm':
623 		case 'r':
624 		case 'R':
625 		case 'd':
626 		case 't':
627 		case 'g':
628 		case 'i':
629 		case 'k':
630 		case 'T':
631 		case 'P':
632 		case 'F':
633 			value = nicenumtoull(optarg);
634 		}
635 		switch (opt) {
636 		case 'v':
637 			zo->zo_vdevs = value;
638 			break;
639 		case 's':
640 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
641 			break;
642 		case 'a':
643 			zo->zo_ashift = value;
644 			break;
645 		case 'm':
646 			zo->zo_mirrors = value;
647 			break;
648 		case 'r':
649 			zo->zo_raidz = MAX(1, value);
650 			break;
651 		case 'R':
652 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
653 			break;
654 		case 'd':
655 			zo->zo_datasets = MAX(1, value);
656 			break;
657 		case 't':
658 			zo->zo_threads = MAX(1, value);
659 			break;
660 		case 'g':
661 			zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
662 			    value);
663 			break;
664 		case 'i':
665 			zo->zo_init = value;
666 			break;
667 		case 'k':
668 			zo->zo_killrate = value;
669 			break;
670 		case 'p':
671 			(void) strlcpy(zo->zo_pool, optarg,
672 			    sizeof (zo->zo_pool));
673 			break;
674 		case 'f':
675 			path = realpath(optarg, NULL);
676 			if (path == NULL) {
677 				(void) fprintf(stderr, "error: %s: %s\n",
678 				    optarg, strerror(errno));
679 				usage(B_FALSE);
680 			} else {
681 				(void) strlcpy(zo->zo_dir, path,
682 				    sizeof (zo->zo_dir));
683 			}
684 			break;
685 		case 'V':
686 			zo->zo_verbose++;
687 			break;
688 		case 'E':
689 			zo->zo_init = 0;
690 			break;
691 		case 'T':
692 			zo->zo_time = value;
693 			break;
694 		case 'P':
695 			zo->zo_passtime = MAX(1, value);
696 			break;
697 		case 'F':
698 			zo->zo_maxloops = MAX(1, value);
699 			break;
700 		case 'B':
701 			(void) strlcpy(altdir, optarg, sizeof (altdir));
702 			break;
703 		case 'h':
704 			usage(B_TRUE);
705 			break;
706 		case '?':
707 		default:
708 			usage(B_FALSE);
709 			break;
710 		}
711 	}
712 
713 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
714 
715 	zo->zo_vdevtime =
716 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
717 	    UINT64_MAX >> 2);
718 
719 	if (strlen(altdir) > 0) {
720 		char *cmd;
721 		char *realaltdir;
722 		char *bin;
723 		char *ztest;
724 		char *isa;
725 		int isalen;
726 
727 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
728 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
729 
730 		VERIFY(NULL != realpath(getexecname(), cmd));
731 		if (0 != access(altdir, F_OK)) {
732 			ztest_dump_core = B_FALSE;
733 			fatal(B_TRUE, "invalid alternate ztest path: %s",
734 			    altdir);
735 		}
736 		VERIFY(NULL != realpath(altdir, realaltdir));
737 
738 		/*
739 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
740 		 * We want to extract <isa> to determine if we should use
741 		 * 32 or 64 bit binaries.
742 		 */
743 		bin = strstr(cmd, "/usr/bin/");
744 		ztest = strstr(bin, "/ztest");
745 		isa = bin + 9;
746 		isalen = ztest - isa;
747 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
748 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
749 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
750 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
751 
752 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
753 			ztest_dump_core = B_FALSE;
754 			fatal(B_TRUE, "invalid alternate ztest: %s",
755 			    zo->zo_alt_ztest);
756 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
757 			ztest_dump_core = B_FALSE;
758 			fatal(B_TRUE, "invalid alternate lib directory %s",
759 			    zo->zo_alt_libpath);
760 		}
761 
762 		umem_free(cmd, MAXPATHLEN);
763 		umem_free(realaltdir, MAXPATHLEN);
764 	}
765 }
766 
767 static void
768 ztest_kill(ztest_shared_t *zs)
769 {
770 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
771 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
772 
773 	/*
774 	 * Before we kill off ztest, make sure that the config is updated.
775 	 * See comment above spa_config_sync().
776 	 */
777 	mutex_enter(&spa_namespace_lock);
778 	spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
779 	mutex_exit(&spa_namespace_lock);
780 
781 	zfs_dbgmsg_print(FTAG);
782 	(void) kill(getpid(), SIGKILL);
783 }
784 
785 static uint64_t
786 ztest_random(uint64_t range)
787 {
788 	uint64_t r;
789 
790 	ASSERT3S(ztest_fd_rand, >=, 0);
791 
792 	if (range == 0)
793 		return (0);
794 
795 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
796 		fatal(1, "short read from /dev/urandom");
797 
798 	return (r % range);
799 }
800 
801 /* ARGSUSED */
802 static void
803 ztest_record_enospc(const char *s)
804 {
805 	ztest_shared->zs_enospc_count++;
806 }
807 
808 static uint64_t
809 ztest_get_ashift(void)
810 {
811 	if (ztest_opts.zo_ashift == 0)
812 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
813 	return (ztest_opts.zo_ashift);
814 }
815 
816 static nvlist_t *
817 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
818 {
819 	char pathbuf[MAXPATHLEN];
820 	uint64_t vdev;
821 	nvlist_t *file;
822 
823 	if (ashift == 0)
824 		ashift = ztest_get_ashift();
825 
826 	if (path == NULL) {
827 		path = pathbuf;
828 
829 		if (aux != NULL) {
830 			vdev = ztest_shared->zs_vdev_aux;
831 			(void) snprintf(path, sizeof (pathbuf),
832 			    ztest_aux_template, ztest_opts.zo_dir,
833 			    pool == NULL ? ztest_opts.zo_pool : pool,
834 			    aux, vdev);
835 		} else {
836 			vdev = ztest_shared->zs_vdev_next_leaf++;
837 			(void) snprintf(path, sizeof (pathbuf),
838 			    ztest_dev_template, ztest_opts.zo_dir,
839 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
840 		}
841 	}
842 
843 	if (size != 0) {
844 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
845 		if (fd == -1)
846 			fatal(1, "can't open %s", path);
847 		if (ftruncate(fd, size) != 0)
848 			fatal(1, "can't ftruncate %s", path);
849 		(void) close(fd);
850 	}
851 
852 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
853 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
854 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
855 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
856 
857 	return (file);
858 }
859 
860 static nvlist_t *
861 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
862     uint64_t ashift, int r)
863 {
864 	nvlist_t *raidz, **child;
865 	int c;
866 
867 	if (r < 2)
868 		return (make_vdev_file(path, aux, pool, size, ashift));
869 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
870 
871 	for (c = 0; c < r; c++)
872 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
873 
874 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
875 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
876 	    VDEV_TYPE_RAIDZ) == 0);
877 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
878 	    ztest_opts.zo_raidz_parity) == 0);
879 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
880 	    child, r) == 0);
881 
882 	for (c = 0; c < r; c++)
883 		nvlist_free(child[c]);
884 
885 	umem_free(child, r * sizeof (nvlist_t *));
886 
887 	return (raidz);
888 }
889 
890 static nvlist_t *
891 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
892     uint64_t ashift, int r, int m)
893 {
894 	nvlist_t *mirror, **child;
895 	int c;
896 
897 	if (m < 1)
898 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
899 
900 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
901 
902 	for (c = 0; c < m; c++)
903 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
904 
905 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
906 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
907 	    VDEV_TYPE_MIRROR) == 0);
908 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
909 	    child, m) == 0);
910 
911 	for (c = 0; c < m; c++)
912 		nvlist_free(child[c]);
913 
914 	umem_free(child, m * sizeof (nvlist_t *));
915 
916 	return (mirror);
917 }
918 
919 static nvlist_t *
920 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
921     int log, int r, int m, int t)
922 {
923 	nvlist_t *root, **child;
924 	int c;
925 
926 	ASSERT(t > 0);
927 
928 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
929 
930 	for (c = 0; c < t; c++) {
931 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
932 		    r, m);
933 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
934 		    log) == 0);
935 	}
936 
937 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
938 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
939 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
940 	    child, t) == 0);
941 
942 	for (c = 0; c < t; c++)
943 		nvlist_free(child[c]);
944 
945 	umem_free(child, t * sizeof (nvlist_t *));
946 
947 	return (root);
948 }
949 
950 /*
951  * Find a random spa version. Returns back a random spa version in the
952  * range [initial_version, SPA_VERSION_FEATURES].
953  */
954 static uint64_t
955 ztest_random_spa_version(uint64_t initial_version)
956 {
957 	uint64_t version = initial_version;
958 
959 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
960 		version = version +
961 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
962 	}
963 
964 	if (version > SPA_VERSION_BEFORE_FEATURES)
965 		version = SPA_VERSION_FEATURES;
966 
967 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
968 	return (version);
969 }
970 
971 static int
972 ztest_random_blocksize(void)
973 {
974 	uint64_t block_shift;
975 	/*
976 	 * Choose a block size >= the ashift.
977 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
978 	 */
979 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
980 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
981 		maxbs = 20;
982 	block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
983 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
984 }
985 
986 static int
987 ztest_random_ibshift(void)
988 {
989 	return (DN_MIN_INDBLKSHIFT +
990 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
991 }
992 
993 static uint64_t
994 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
995 {
996 	uint64_t top;
997 	vdev_t *rvd = spa->spa_root_vdev;
998 	vdev_t *tvd;
999 
1000 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1001 
1002 	do {
1003 		top = ztest_random(rvd->vdev_children);
1004 		tvd = rvd->vdev_child[top];
1005 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1006 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1007 
1008 	return (top);
1009 }
1010 
1011 static uint64_t
1012 ztest_random_dsl_prop(zfs_prop_t prop)
1013 {
1014 	uint64_t value;
1015 
1016 	do {
1017 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1018 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1019 
1020 	return (value);
1021 }
1022 
1023 static int
1024 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1025     boolean_t inherit)
1026 {
1027 	const char *propname = zfs_prop_to_name(prop);
1028 	const char *valname;
1029 	char setpoint[MAXPATHLEN];
1030 	uint64_t curval;
1031 	int error;
1032 
1033 	error = dsl_prop_set_int(osname, propname,
1034 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1035 
1036 	if (error == ENOSPC) {
1037 		ztest_record_enospc(FTAG);
1038 		return (error);
1039 	}
1040 	ASSERT0(error);
1041 
1042 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1043 
1044 	if (ztest_opts.zo_verbose >= 6) {
1045 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1046 		(void) printf("%s %s = %s at '%s'\n",
1047 		    osname, propname, valname, setpoint);
1048 	}
1049 
1050 	return (error);
1051 }
1052 
1053 static int
1054 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1055 {
1056 	spa_t *spa = ztest_spa;
1057 	nvlist_t *props = NULL;
1058 	int error;
1059 
1060 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1061 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1062 
1063 	error = spa_prop_set(spa, props);
1064 
1065 	nvlist_free(props);
1066 
1067 	if (error == ENOSPC) {
1068 		ztest_record_enospc(FTAG);
1069 		return (error);
1070 	}
1071 	ASSERT0(error);
1072 
1073 	return (error);
1074 }
1075 
1076 static void
1077 ztest_rll_init(rll_t *rll)
1078 {
1079 	rll->rll_writer = NULL;
1080 	rll->rll_readers = 0;
1081 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1082 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1083 }
1084 
1085 static void
1086 ztest_rll_destroy(rll_t *rll)
1087 {
1088 	ASSERT(rll->rll_writer == NULL);
1089 	ASSERT(rll->rll_readers == 0);
1090 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1091 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
1092 }
1093 
1094 static void
1095 ztest_rll_lock(rll_t *rll, rl_type_t type)
1096 {
1097 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1098 
1099 	if (type == RL_READER) {
1100 		while (rll->rll_writer != NULL)
1101 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1102 		rll->rll_readers++;
1103 	} else {
1104 		while (rll->rll_writer != NULL || rll->rll_readers)
1105 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1106 		rll->rll_writer = curthread;
1107 	}
1108 
1109 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1110 }
1111 
1112 static void
1113 ztest_rll_unlock(rll_t *rll)
1114 {
1115 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1116 
1117 	if (rll->rll_writer) {
1118 		ASSERT(rll->rll_readers == 0);
1119 		rll->rll_writer = NULL;
1120 	} else {
1121 		ASSERT(rll->rll_readers != 0);
1122 		ASSERT(rll->rll_writer == NULL);
1123 		rll->rll_readers--;
1124 	}
1125 
1126 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1127 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1128 
1129 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1130 }
1131 
1132 static void
1133 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1134 {
1135 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1136 
1137 	ztest_rll_lock(rll, type);
1138 }
1139 
1140 static void
1141 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1142 {
1143 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1144 
1145 	ztest_rll_unlock(rll);
1146 }
1147 
1148 static rl_t *
1149 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1150     uint64_t size, rl_type_t type)
1151 {
1152 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1153 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1154 	rl_t *rl;
1155 
1156 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1157 	rl->rl_object = object;
1158 	rl->rl_offset = offset;
1159 	rl->rl_size = size;
1160 	rl->rl_lock = rll;
1161 
1162 	ztest_rll_lock(rll, type);
1163 
1164 	return (rl);
1165 }
1166 
1167 static void
1168 ztest_range_unlock(rl_t *rl)
1169 {
1170 	rll_t *rll = rl->rl_lock;
1171 
1172 	ztest_rll_unlock(rll);
1173 
1174 	umem_free(rl, sizeof (*rl));
1175 }
1176 
1177 static void
1178 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1179 {
1180 	zd->zd_os = os;
1181 	zd->zd_zilog = dmu_objset_zil(os);
1182 	zd->zd_shared = szd;
1183 	dmu_objset_name(os, zd->zd_name);
1184 
1185 	if (zd->zd_shared != NULL)
1186 		zd->zd_shared->zd_seq = 0;
1187 
1188 	VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1189 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1190 
1191 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1192 		ztest_rll_init(&zd->zd_object_lock[l]);
1193 
1194 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1195 		ztest_rll_init(&zd->zd_range_lock[l]);
1196 }
1197 
1198 static void
1199 ztest_zd_fini(ztest_ds_t *zd)
1200 {
1201 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1202 
1203 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1204 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1205 
1206 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1207 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1208 }
1209 
1210 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1211 
1212 static uint64_t
1213 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1214 {
1215 	uint64_t txg;
1216 	int error;
1217 
1218 	/*
1219 	 * Attempt to assign tx to some transaction group.
1220 	 */
1221 	error = dmu_tx_assign(tx, txg_how);
1222 	if (error) {
1223 		if (error == ERESTART) {
1224 			ASSERT(txg_how == TXG_NOWAIT);
1225 			dmu_tx_wait(tx);
1226 		} else {
1227 			ASSERT3U(error, ==, ENOSPC);
1228 			ztest_record_enospc(tag);
1229 		}
1230 		dmu_tx_abort(tx);
1231 		return (0);
1232 	}
1233 	txg = dmu_tx_get_txg(tx);
1234 	ASSERT(txg != 0);
1235 	return (txg);
1236 }
1237 
1238 static void
1239 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1240 {
1241 	uint64_t *ip = buf;
1242 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1243 
1244 	while (ip < ip_end)
1245 		*ip++ = value;
1246 }
1247 
1248 static boolean_t
1249 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1250 {
1251 	uint64_t *ip = buf;
1252 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1253 	uint64_t diff = 0;
1254 
1255 	while (ip < ip_end)
1256 		diff |= (value - *ip++);
1257 
1258 	return (diff == 0);
1259 }
1260 
1261 static void
1262 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1263     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1264 {
1265 	bt->bt_magic = BT_MAGIC;
1266 	bt->bt_objset = dmu_objset_id(os);
1267 	bt->bt_object = object;
1268 	bt->bt_offset = offset;
1269 	bt->bt_gen = gen;
1270 	bt->bt_txg = txg;
1271 	bt->bt_crtxg = crtxg;
1272 }
1273 
1274 static void
1275 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1276     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1277 {
1278 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1279 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1280 	ASSERT3U(bt->bt_object, ==, object);
1281 	ASSERT3U(bt->bt_offset, ==, offset);
1282 	ASSERT3U(bt->bt_gen, <=, gen);
1283 	ASSERT3U(bt->bt_txg, <=, txg);
1284 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1285 }
1286 
1287 static ztest_block_tag_t *
1288 ztest_bt_bonus(dmu_buf_t *db)
1289 {
1290 	dmu_object_info_t doi;
1291 	ztest_block_tag_t *bt;
1292 
1293 	dmu_object_info_from_db(db, &doi);
1294 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1295 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1296 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1297 
1298 	return (bt);
1299 }
1300 
1301 /*
1302  * ZIL logging ops
1303  */
1304 
1305 #define	lrz_type	lr_mode
1306 #define	lrz_blocksize	lr_uid
1307 #define	lrz_ibshift	lr_gid
1308 #define	lrz_bonustype	lr_rdev
1309 #define	lrz_bonuslen	lr_crtime[1]
1310 
1311 static void
1312 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1313 {
1314 	char *name = (void *)(lr + 1);		/* name follows lr */
1315 	size_t namesize = strlen(name) + 1;
1316 	itx_t *itx;
1317 
1318 	if (zil_replaying(zd->zd_zilog, tx))
1319 		return;
1320 
1321 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1322 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1323 	    sizeof (*lr) + namesize - sizeof (lr_t));
1324 
1325 	zil_itx_assign(zd->zd_zilog, itx, tx);
1326 }
1327 
1328 static void
1329 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1330 {
1331 	char *name = (void *)(lr + 1);		/* name follows lr */
1332 	size_t namesize = strlen(name) + 1;
1333 	itx_t *itx;
1334 
1335 	if (zil_replaying(zd->zd_zilog, tx))
1336 		return;
1337 
1338 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1339 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1340 	    sizeof (*lr) + namesize - sizeof (lr_t));
1341 
1342 	itx->itx_oid = object;
1343 	zil_itx_assign(zd->zd_zilog, itx, tx);
1344 }
1345 
1346 static void
1347 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1348 {
1349 	itx_t *itx;
1350 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1351 
1352 	if (zil_replaying(zd->zd_zilog, tx))
1353 		return;
1354 
1355 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1356 		write_state = WR_INDIRECT;
1357 
1358 	itx = zil_itx_create(TX_WRITE,
1359 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1360 
1361 	if (write_state == WR_COPIED &&
1362 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1363 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1364 		zil_itx_destroy(itx);
1365 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1366 		write_state = WR_NEED_COPY;
1367 	}
1368 	itx->itx_private = zd;
1369 	itx->itx_wr_state = write_state;
1370 	itx->itx_sync = (ztest_random(8) == 0);
1371 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1372 
1373 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1374 	    sizeof (*lr) - sizeof (lr_t));
1375 
1376 	zil_itx_assign(zd->zd_zilog, itx, tx);
1377 }
1378 
1379 static void
1380 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1381 {
1382 	itx_t *itx;
1383 
1384 	if (zil_replaying(zd->zd_zilog, tx))
1385 		return;
1386 
1387 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1388 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1389 	    sizeof (*lr) - sizeof (lr_t));
1390 
1391 	itx->itx_sync = B_FALSE;
1392 	zil_itx_assign(zd->zd_zilog, itx, tx);
1393 }
1394 
1395 static void
1396 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1397 {
1398 	itx_t *itx;
1399 
1400 	if (zil_replaying(zd->zd_zilog, tx))
1401 		return;
1402 
1403 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1404 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1405 	    sizeof (*lr) - sizeof (lr_t));
1406 
1407 	itx->itx_sync = B_FALSE;
1408 	zil_itx_assign(zd->zd_zilog, itx, tx);
1409 }
1410 
1411 /*
1412  * ZIL replay ops
1413  */
1414 static int
1415 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1416 {
1417 	char *name = (void *)(lr + 1);		/* name follows lr */
1418 	objset_t *os = zd->zd_os;
1419 	ztest_block_tag_t *bbt;
1420 	dmu_buf_t *db;
1421 	dmu_tx_t *tx;
1422 	uint64_t txg;
1423 	int error = 0;
1424 
1425 	if (byteswap)
1426 		byteswap_uint64_array(lr, sizeof (*lr));
1427 
1428 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1429 	ASSERT(name[0] != '\0');
1430 
1431 	tx = dmu_tx_create(os);
1432 
1433 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1434 
1435 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1436 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1437 	} else {
1438 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1439 	}
1440 
1441 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1442 	if (txg == 0)
1443 		return (ENOSPC);
1444 
1445 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1446 
1447 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1448 		if (lr->lr_foid == 0) {
1449 			lr->lr_foid = zap_create(os,
1450 			    lr->lrz_type, lr->lrz_bonustype,
1451 			    lr->lrz_bonuslen, tx);
1452 		} else {
1453 			error = zap_create_claim(os, lr->lr_foid,
1454 			    lr->lrz_type, lr->lrz_bonustype,
1455 			    lr->lrz_bonuslen, tx);
1456 		}
1457 	} else {
1458 		if (lr->lr_foid == 0) {
1459 			lr->lr_foid = dmu_object_alloc(os,
1460 			    lr->lrz_type, 0, lr->lrz_bonustype,
1461 			    lr->lrz_bonuslen, tx);
1462 		} else {
1463 			error = dmu_object_claim(os, lr->lr_foid,
1464 			    lr->lrz_type, 0, lr->lrz_bonustype,
1465 			    lr->lrz_bonuslen, tx);
1466 		}
1467 	}
1468 
1469 	if (error) {
1470 		ASSERT3U(error, ==, EEXIST);
1471 		ASSERT(zd->zd_zilog->zl_replay);
1472 		dmu_tx_commit(tx);
1473 		return (error);
1474 	}
1475 
1476 	ASSERT(lr->lr_foid != 0);
1477 
1478 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1479 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1480 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1481 
1482 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1483 	bbt = ztest_bt_bonus(db);
1484 	dmu_buf_will_dirty(db, tx);
1485 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1486 	dmu_buf_rele(db, FTAG);
1487 
1488 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1489 	    &lr->lr_foid, tx));
1490 
1491 	(void) ztest_log_create(zd, tx, lr);
1492 
1493 	dmu_tx_commit(tx);
1494 
1495 	return (0);
1496 }
1497 
1498 static int
1499 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1500 {
1501 	char *name = (void *)(lr + 1);		/* name follows lr */
1502 	objset_t *os = zd->zd_os;
1503 	dmu_object_info_t doi;
1504 	dmu_tx_t *tx;
1505 	uint64_t object, txg;
1506 
1507 	if (byteswap)
1508 		byteswap_uint64_array(lr, sizeof (*lr));
1509 
1510 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1511 	ASSERT(name[0] != '\0');
1512 
1513 	VERIFY3U(0, ==,
1514 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1515 	ASSERT(object != 0);
1516 
1517 	ztest_object_lock(zd, object, RL_WRITER);
1518 
1519 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1520 
1521 	tx = dmu_tx_create(os);
1522 
1523 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1524 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1525 
1526 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1527 	if (txg == 0) {
1528 		ztest_object_unlock(zd, object);
1529 		return (ENOSPC);
1530 	}
1531 
1532 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1533 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1534 	} else {
1535 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1536 	}
1537 
1538 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1539 
1540 	(void) ztest_log_remove(zd, tx, lr, object);
1541 
1542 	dmu_tx_commit(tx);
1543 
1544 	ztest_object_unlock(zd, object);
1545 
1546 	return (0);
1547 }
1548 
1549 static int
1550 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1551 {
1552 	objset_t *os = zd->zd_os;
1553 	void *data = lr + 1;			/* data follows lr */
1554 	uint64_t offset, length;
1555 	ztest_block_tag_t *bt = data;
1556 	ztest_block_tag_t *bbt;
1557 	uint64_t gen, txg, lrtxg, crtxg;
1558 	dmu_object_info_t doi;
1559 	dmu_tx_t *tx;
1560 	dmu_buf_t *db;
1561 	arc_buf_t *abuf = NULL;
1562 	rl_t *rl;
1563 
1564 	if (byteswap)
1565 		byteswap_uint64_array(lr, sizeof (*lr));
1566 
1567 	offset = lr->lr_offset;
1568 	length = lr->lr_length;
1569 
1570 	/* If it's a dmu_sync() block, write the whole block */
1571 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1572 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1573 		if (length < blocksize) {
1574 			offset -= offset % blocksize;
1575 			length = blocksize;
1576 		}
1577 	}
1578 
1579 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1580 		byteswap_uint64_array(bt, sizeof (*bt));
1581 
1582 	if (bt->bt_magic != BT_MAGIC)
1583 		bt = NULL;
1584 
1585 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1586 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1587 
1588 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1589 
1590 	dmu_object_info_from_db(db, &doi);
1591 
1592 	bbt = ztest_bt_bonus(db);
1593 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1594 	gen = bbt->bt_gen;
1595 	crtxg = bbt->bt_crtxg;
1596 	lrtxg = lr->lr_common.lrc_txg;
1597 
1598 	tx = dmu_tx_create(os);
1599 
1600 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1601 
1602 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1603 	    P2PHASE(offset, length) == 0)
1604 		abuf = dmu_request_arcbuf(db, length);
1605 
1606 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1607 	if (txg == 0) {
1608 		if (abuf != NULL)
1609 			dmu_return_arcbuf(abuf);
1610 		dmu_buf_rele(db, FTAG);
1611 		ztest_range_unlock(rl);
1612 		ztest_object_unlock(zd, lr->lr_foid);
1613 		return (ENOSPC);
1614 	}
1615 
1616 	if (bt != NULL) {
1617 		/*
1618 		 * Usually, verify the old data before writing new data --
1619 		 * but not always, because we also want to verify correct
1620 		 * behavior when the data was not recently read into cache.
1621 		 */
1622 		ASSERT(offset % doi.doi_data_block_size == 0);
1623 		if (ztest_random(4) != 0) {
1624 			int prefetch = ztest_random(2) ?
1625 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1626 			ztest_block_tag_t rbt;
1627 
1628 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1629 			    sizeof (rbt), &rbt, prefetch) == 0);
1630 			if (rbt.bt_magic == BT_MAGIC) {
1631 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1632 				    offset, gen, txg, crtxg);
1633 			}
1634 		}
1635 
1636 		/*
1637 		 * Writes can appear to be newer than the bonus buffer because
1638 		 * the ztest_get_data() callback does a dmu_read() of the
1639 		 * open-context data, which may be different than the data
1640 		 * as it was when the write was generated.
1641 		 */
1642 		if (zd->zd_zilog->zl_replay) {
1643 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1644 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1645 			    bt->bt_crtxg);
1646 		}
1647 
1648 		/*
1649 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1650 		 * so that all of the usual ASSERTs will work.
1651 		 */
1652 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1653 	}
1654 
1655 	if (abuf == NULL) {
1656 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1657 	} else {
1658 		bcopy(data, abuf->b_data, length);
1659 		dmu_assign_arcbuf(db, offset, abuf, tx);
1660 	}
1661 
1662 	(void) ztest_log_write(zd, tx, lr);
1663 
1664 	dmu_buf_rele(db, FTAG);
1665 
1666 	dmu_tx_commit(tx);
1667 
1668 	ztest_range_unlock(rl);
1669 	ztest_object_unlock(zd, lr->lr_foid);
1670 
1671 	return (0);
1672 }
1673 
1674 static int
1675 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1676 {
1677 	objset_t *os = zd->zd_os;
1678 	dmu_tx_t *tx;
1679 	uint64_t txg;
1680 	rl_t *rl;
1681 
1682 	if (byteswap)
1683 		byteswap_uint64_array(lr, sizeof (*lr));
1684 
1685 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1686 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1687 	    RL_WRITER);
1688 
1689 	tx = dmu_tx_create(os);
1690 
1691 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1692 
1693 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1694 	if (txg == 0) {
1695 		ztest_range_unlock(rl);
1696 		ztest_object_unlock(zd, lr->lr_foid);
1697 		return (ENOSPC);
1698 	}
1699 
1700 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1701 	    lr->lr_length, tx) == 0);
1702 
1703 	(void) ztest_log_truncate(zd, tx, lr);
1704 
1705 	dmu_tx_commit(tx);
1706 
1707 	ztest_range_unlock(rl);
1708 	ztest_object_unlock(zd, lr->lr_foid);
1709 
1710 	return (0);
1711 }
1712 
1713 static int
1714 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1715 {
1716 	objset_t *os = zd->zd_os;
1717 	dmu_tx_t *tx;
1718 	dmu_buf_t *db;
1719 	ztest_block_tag_t *bbt;
1720 	uint64_t txg, lrtxg, crtxg;
1721 
1722 	if (byteswap)
1723 		byteswap_uint64_array(lr, sizeof (*lr));
1724 
1725 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1726 
1727 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1728 
1729 	tx = dmu_tx_create(os);
1730 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1731 
1732 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1733 	if (txg == 0) {
1734 		dmu_buf_rele(db, FTAG);
1735 		ztest_object_unlock(zd, lr->lr_foid);
1736 		return (ENOSPC);
1737 	}
1738 
1739 	bbt = ztest_bt_bonus(db);
1740 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1741 	crtxg = bbt->bt_crtxg;
1742 	lrtxg = lr->lr_common.lrc_txg;
1743 
1744 	if (zd->zd_zilog->zl_replay) {
1745 		ASSERT(lr->lr_size != 0);
1746 		ASSERT(lr->lr_mode != 0);
1747 		ASSERT(lrtxg != 0);
1748 	} else {
1749 		/*
1750 		 * Randomly change the size and increment the generation.
1751 		 */
1752 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1753 		    sizeof (*bbt);
1754 		lr->lr_mode = bbt->bt_gen + 1;
1755 		ASSERT(lrtxg == 0);
1756 	}
1757 
1758 	/*
1759 	 * Verify that the current bonus buffer is not newer than our txg.
1760 	 */
1761 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1762 	    MAX(txg, lrtxg), crtxg);
1763 
1764 	dmu_buf_will_dirty(db, tx);
1765 
1766 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1767 	ASSERT3U(lr->lr_size, <=, db->db_size);
1768 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1769 	bbt = ztest_bt_bonus(db);
1770 
1771 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1772 
1773 	dmu_buf_rele(db, FTAG);
1774 
1775 	(void) ztest_log_setattr(zd, tx, lr);
1776 
1777 	dmu_tx_commit(tx);
1778 
1779 	ztest_object_unlock(zd, lr->lr_foid);
1780 
1781 	return (0);
1782 }
1783 
1784 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1785 	NULL,			/* 0 no such transaction type */
1786 	ztest_replay_create,	/* TX_CREATE */
1787 	NULL,			/* TX_MKDIR */
1788 	NULL,			/* TX_MKXATTR */
1789 	NULL,			/* TX_SYMLINK */
1790 	ztest_replay_remove,	/* TX_REMOVE */
1791 	NULL,			/* TX_RMDIR */
1792 	NULL,			/* TX_LINK */
1793 	NULL,			/* TX_RENAME */
1794 	ztest_replay_write,	/* TX_WRITE */
1795 	ztest_replay_truncate,	/* TX_TRUNCATE */
1796 	ztest_replay_setattr,	/* TX_SETATTR */
1797 	NULL,			/* TX_ACL */
1798 	NULL,			/* TX_CREATE_ACL */
1799 	NULL,			/* TX_CREATE_ATTR */
1800 	NULL,			/* TX_CREATE_ACL_ATTR */
1801 	NULL,			/* TX_MKDIR_ACL */
1802 	NULL,			/* TX_MKDIR_ATTR */
1803 	NULL,			/* TX_MKDIR_ACL_ATTR */
1804 	NULL,			/* TX_WRITE2 */
1805 };
1806 
1807 /*
1808  * ZIL get_data callbacks
1809  */
1810 
1811 static void
1812 ztest_get_done(zgd_t *zgd, int error)
1813 {
1814 	ztest_ds_t *zd = zgd->zgd_private;
1815 	uint64_t object = zgd->zgd_rl->rl_object;
1816 
1817 	if (zgd->zgd_db)
1818 		dmu_buf_rele(zgd->zgd_db, zgd);
1819 
1820 	ztest_range_unlock(zgd->zgd_rl);
1821 	ztest_object_unlock(zd, object);
1822 
1823 	if (error == 0 && zgd->zgd_bp)
1824 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1825 
1826 	umem_free(zgd, sizeof (*zgd));
1827 }
1828 
1829 static int
1830 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1831 {
1832 	ztest_ds_t *zd = arg;
1833 	objset_t *os = zd->zd_os;
1834 	uint64_t object = lr->lr_foid;
1835 	uint64_t offset = lr->lr_offset;
1836 	uint64_t size = lr->lr_length;
1837 	blkptr_t *bp = &lr->lr_blkptr;
1838 	uint64_t txg = lr->lr_common.lrc_txg;
1839 	uint64_t crtxg;
1840 	dmu_object_info_t doi;
1841 	dmu_buf_t *db;
1842 	zgd_t *zgd;
1843 	int error;
1844 
1845 	ztest_object_lock(zd, object, RL_READER);
1846 	error = dmu_bonus_hold(os, object, FTAG, &db);
1847 	if (error) {
1848 		ztest_object_unlock(zd, object);
1849 		return (error);
1850 	}
1851 
1852 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1853 
1854 	if (crtxg == 0 || crtxg > txg) {
1855 		dmu_buf_rele(db, FTAG);
1856 		ztest_object_unlock(zd, object);
1857 		return (ENOENT);
1858 	}
1859 
1860 	dmu_object_info_from_db(db, &doi);
1861 	dmu_buf_rele(db, FTAG);
1862 	db = NULL;
1863 
1864 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1865 	zgd->zgd_zilog = zd->zd_zilog;
1866 	zgd->zgd_private = zd;
1867 
1868 	if (buf != NULL) {	/* immediate write */
1869 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1870 		    RL_READER);
1871 
1872 		error = dmu_read(os, object, offset, size, buf,
1873 		    DMU_READ_NO_PREFETCH);
1874 		ASSERT(error == 0);
1875 	} else {
1876 		size = doi.doi_data_block_size;
1877 		if (ISP2(size)) {
1878 			offset = P2ALIGN(offset, size);
1879 		} else {
1880 			ASSERT(offset < size);
1881 			offset = 0;
1882 		}
1883 
1884 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1885 		    RL_READER);
1886 
1887 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1888 		    DMU_READ_NO_PREFETCH);
1889 
1890 		if (error == 0) {
1891 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1892 			if (obp) {
1893 				ASSERT(BP_IS_HOLE(bp));
1894 				*bp = *obp;
1895 			}
1896 
1897 			zgd->zgd_db = db;
1898 			zgd->zgd_bp = bp;
1899 
1900 			ASSERT(db->db_offset == offset);
1901 			ASSERT(db->db_size == size);
1902 
1903 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1904 			    ztest_get_done, zgd);
1905 
1906 			if (error == 0)
1907 				return (0);
1908 		}
1909 	}
1910 
1911 	ztest_get_done(zgd, error);
1912 
1913 	return (error);
1914 }
1915 
1916 static void *
1917 ztest_lr_alloc(size_t lrsize, char *name)
1918 {
1919 	char *lr;
1920 	size_t namesize = name ? strlen(name) + 1 : 0;
1921 
1922 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1923 
1924 	if (name)
1925 		bcopy(name, lr + lrsize, namesize);
1926 
1927 	return (lr);
1928 }
1929 
1930 void
1931 ztest_lr_free(void *lr, size_t lrsize, char *name)
1932 {
1933 	size_t namesize = name ? strlen(name) + 1 : 0;
1934 
1935 	umem_free(lr, lrsize + namesize);
1936 }
1937 
1938 /*
1939  * Lookup a bunch of objects.  Returns the number of objects not found.
1940  */
1941 static int
1942 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1943 {
1944 	int missing = 0;
1945 	int error;
1946 
1947 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1948 
1949 	for (int i = 0; i < count; i++, od++) {
1950 		od->od_object = 0;
1951 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1952 		    sizeof (uint64_t), 1, &od->od_object);
1953 		if (error) {
1954 			ASSERT(error == ENOENT);
1955 			ASSERT(od->od_object == 0);
1956 			missing++;
1957 		} else {
1958 			dmu_buf_t *db;
1959 			ztest_block_tag_t *bbt;
1960 			dmu_object_info_t doi;
1961 
1962 			ASSERT(od->od_object != 0);
1963 			ASSERT(missing == 0);	/* there should be no gaps */
1964 
1965 			ztest_object_lock(zd, od->od_object, RL_READER);
1966 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1967 			    od->od_object, FTAG, &db));
1968 			dmu_object_info_from_db(db, &doi);
1969 			bbt = ztest_bt_bonus(db);
1970 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1971 			od->od_type = doi.doi_type;
1972 			od->od_blocksize = doi.doi_data_block_size;
1973 			od->od_gen = bbt->bt_gen;
1974 			dmu_buf_rele(db, FTAG);
1975 			ztest_object_unlock(zd, od->od_object);
1976 		}
1977 	}
1978 
1979 	return (missing);
1980 }
1981 
1982 static int
1983 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1984 {
1985 	int missing = 0;
1986 
1987 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1988 
1989 	for (int i = 0; i < count; i++, od++) {
1990 		if (missing) {
1991 			od->od_object = 0;
1992 			missing++;
1993 			continue;
1994 		}
1995 
1996 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1997 
1998 		lr->lr_doid = od->od_dir;
1999 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2000 		lr->lrz_type = od->od_crtype;
2001 		lr->lrz_blocksize = od->od_crblocksize;
2002 		lr->lrz_ibshift = ztest_random_ibshift();
2003 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2004 		lr->lrz_bonuslen = dmu_bonus_max();
2005 		lr->lr_gen = od->od_crgen;
2006 		lr->lr_crtime[0] = time(NULL);
2007 
2008 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2009 			ASSERT(missing == 0);
2010 			od->od_object = 0;
2011 			missing++;
2012 		} else {
2013 			od->od_object = lr->lr_foid;
2014 			od->od_type = od->od_crtype;
2015 			od->od_blocksize = od->od_crblocksize;
2016 			od->od_gen = od->od_crgen;
2017 			ASSERT(od->od_object != 0);
2018 		}
2019 
2020 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2021 	}
2022 
2023 	return (missing);
2024 }
2025 
2026 static int
2027 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2028 {
2029 	int missing = 0;
2030 	int error;
2031 
2032 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2033 
2034 	od += count - 1;
2035 
2036 	for (int i = count - 1; i >= 0; i--, od--) {
2037 		if (missing) {
2038 			missing++;
2039 			continue;
2040 		}
2041 
2042 		/*
2043 		 * No object was found.
2044 		 */
2045 		if (od->od_object == 0)
2046 			continue;
2047 
2048 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2049 
2050 		lr->lr_doid = od->od_dir;
2051 
2052 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2053 			ASSERT3U(error, ==, ENOSPC);
2054 			missing++;
2055 		} else {
2056 			od->od_object = 0;
2057 		}
2058 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2059 	}
2060 
2061 	return (missing);
2062 }
2063 
2064 static int
2065 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2066     void *data)
2067 {
2068 	lr_write_t *lr;
2069 	int error;
2070 
2071 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2072 
2073 	lr->lr_foid = object;
2074 	lr->lr_offset = offset;
2075 	lr->lr_length = size;
2076 	lr->lr_blkoff = 0;
2077 	BP_ZERO(&lr->lr_blkptr);
2078 
2079 	bcopy(data, lr + 1, size);
2080 
2081 	error = ztest_replay_write(zd, lr, B_FALSE);
2082 
2083 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2084 
2085 	return (error);
2086 }
2087 
2088 static int
2089 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2090 {
2091 	lr_truncate_t *lr;
2092 	int error;
2093 
2094 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2095 
2096 	lr->lr_foid = object;
2097 	lr->lr_offset = offset;
2098 	lr->lr_length = size;
2099 
2100 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2101 
2102 	ztest_lr_free(lr, sizeof (*lr), NULL);
2103 
2104 	return (error);
2105 }
2106 
2107 static int
2108 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2109 {
2110 	lr_setattr_t *lr;
2111 	int error;
2112 
2113 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2114 
2115 	lr->lr_foid = object;
2116 	lr->lr_size = 0;
2117 	lr->lr_mode = 0;
2118 
2119 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2120 
2121 	ztest_lr_free(lr, sizeof (*lr), NULL);
2122 
2123 	return (error);
2124 }
2125 
2126 static void
2127 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2128 {
2129 	objset_t *os = zd->zd_os;
2130 	dmu_tx_t *tx;
2131 	uint64_t txg;
2132 	rl_t *rl;
2133 
2134 	txg_wait_synced(dmu_objset_pool(os), 0);
2135 
2136 	ztest_object_lock(zd, object, RL_READER);
2137 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2138 
2139 	tx = dmu_tx_create(os);
2140 
2141 	dmu_tx_hold_write(tx, object, offset, size);
2142 
2143 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2144 
2145 	if (txg != 0) {
2146 		dmu_prealloc(os, object, offset, size, tx);
2147 		dmu_tx_commit(tx);
2148 		txg_wait_synced(dmu_objset_pool(os), txg);
2149 	} else {
2150 		(void) dmu_free_long_range(os, object, offset, size);
2151 	}
2152 
2153 	ztest_range_unlock(rl);
2154 	ztest_object_unlock(zd, object);
2155 }
2156 
2157 static void
2158 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2159 {
2160 	int err;
2161 	ztest_block_tag_t wbt;
2162 	dmu_object_info_t doi;
2163 	enum ztest_io_type io_type;
2164 	uint64_t blocksize;
2165 	void *data;
2166 
2167 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2168 	blocksize = doi.doi_data_block_size;
2169 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2170 
2171 	/*
2172 	 * Pick an i/o type at random, biased toward writing block tags.
2173 	 */
2174 	io_type = ztest_random(ZTEST_IO_TYPES);
2175 	if (ztest_random(2) == 0)
2176 		io_type = ZTEST_IO_WRITE_TAG;
2177 
2178 	(void) rw_rdlock(&zd->zd_zilog_lock);
2179 
2180 	switch (io_type) {
2181 
2182 	case ZTEST_IO_WRITE_TAG:
2183 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2184 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2185 		break;
2186 
2187 	case ZTEST_IO_WRITE_PATTERN:
2188 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2189 		if (ztest_random(2) == 0) {
2190 			/*
2191 			 * Induce fletcher2 collisions to ensure that
2192 			 * zio_ddt_collision() detects and resolves them
2193 			 * when using fletcher2-verify for deduplication.
2194 			 */
2195 			((uint64_t *)data)[0] ^= 1ULL << 63;
2196 			((uint64_t *)data)[4] ^= 1ULL << 63;
2197 		}
2198 		(void) ztest_write(zd, object, offset, blocksize, data);
2199 		break;
2200 
2201 	case ZTEST_IO_WRITE_ZEROES:
2202 		bzero(data, blocksize);
2203 		(void) ztest_write(zd, object, offset, blocksize, data);
2204 		break;
2205 
2206 	case ZTEST_IO_TRUNCATE:
2207 		(void) ztest_truncate(zd, object, offset, blocksize);
2208 		break;
2209 
2210 	case ZTEST_IO_SETATTR:
2211 		(void) ztest_setattr(zd, object);
2212 		break;
2213 
2214 	case ZTEST_IO_REWRITE:
2215 		(void) rw_rdlock(&ztest_name_lock);
2216 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2217 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2218 		    B_FALSE);
2219 		VERIFY(err == 0 || err == ENOSPC);
2220 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2221 		    ZFS_PROP_COMPRESSION,
2222 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2223 		    B_FALSE);
2224 		VERIFY(err == 0 || err == ENOSPC);
2225 		(void) rw_unlock(&ztest_name_lock);
2226 
2227 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2228 		    DMU_READ_NO_PREFETCH));
2229 
2230 		(void) ztest_write(zd, object, offset, blocksize, data);
2231 		break;
2232 	}
2233 
2234 	(void) rw_unlock(&zd->zd_zilog_lock);
2235 
2236 	umem_free(data, blocksize);
2237 }
2238 
2239 /*
2240  * Initialize an object description template.
2241  */
2242 static void
2243 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2244     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2245 {
2246 	od->od_dir = ZTEST_DIROBJ;
2247 	od->od_object = 0;
2248 
2249 	od->od_crtype = type;
2250 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2251 	od->od_crgen = gen;
2252 
2253 	od->od_type = DMU_OT_NONE;
2254 	od->od_blocksize = 0;
2255 	od->od_gen = 0;
2256 
2257 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2258 	    tag, (int64_t)id, index);
2259 }
2260 
2261 /*
2262  * Lookup or create the objects for a test using the od template.
2263  * If the objects do not all exist, or if 'remove' is specified,
2264  * remove any existing objects and create new ones.  Otherwise,
2265  * use the existing objects.
2266  */
2267 static int
2268 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2269 {
2270 	int count = size / sizeof (*od);
2271 	int rv = 0;
2272 
2273 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2274 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2275 	    (ztest_remove(zd, od, count) != 0 ||
2276 	    ztest_create(zd, od, count) != 0))
2277 		rv = -1;
2278 	zd->zd_od = od;
2279 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2280 
2281 	return (rv);
2282 }
2283 
2284 /* ARGSUSED */
2285 void
2286 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2287 {
2288 	zilog_t *zilog = zd->zd_zilog;
2289 
2290 	(void) rw_rdlock(&zd->zd_zilog_lock);
2291 
2292 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2293 
2294 	/*
2295 	 * Remember the committed values in zd, which is in parent/child
2296 	 * shared memory.  If we die, the next iteration of ztest_run()
2297 	 * will verify that the log really does contain this record.
2298 	 */
2299 	mutex_enter(&zilog->zl_lock);
2300 	ASSERT(zd->zd_shared != NULL);
2301 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2302 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2303 	mutex_exit(&zilog->zl_lock);
2304 
2305 	(void) rw_unlock(&zd->zd_zilog_lock);
2306 }
2307 
2308 /*
2309  * This function is designed to simulate the operations that occur during a
2310  * mount/unmount operation.  We hold the dataset across these operations in an
2311  * attempt to expose any implicit assumptions about ZIL management.
2312  */
2313 /* ARGSUSED */
2314 void
2315 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2316 {
2317 	objset_t *os = zd->zd_os;
2318 
2319 	/*
2320 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2321 	 * updating the zil (i.e. adding in-memory log records) and the
2322 	 * zd_zilog_lock to block any I/O.
2323 	 */
2324 	VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2325 	(void) rw_wrlock(&zd->zd_zilog_lock);
2326 
2327 	/* zfsvfs_teardown() */
2328 	zil_close(zd->zd_zilog);
2329 
2330 	/* zfsvfs_setup() */
2331 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2332 	zil_replay(os, zd, ztest_replay_vector);
2333 
2334 	(void) rw_unlock(&zd->zd_zilog_lock);
2335 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2336 }
2337 
2338 /*
2339  * Verify that we can't destroy an active pool, create an existing pool,
2340  * or create a pool with a bad vdev spec.
2341  */
2342 /* ARGSUSED */
2343 void
2344 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2345 {
2346 	ztest_shared_opts_t *zo = &ztest_opts;
2347 	spa_t *spa;
2348 	nvlist_t *nvroot;
2349 
2350 	/*
2351 	 * Attempt to create using a bad file.
2352 	 */
2353 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2354 	VERIFY3U(ENOENT, ==,
2355 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2356 	nvlist_free(nvroot);
2357 
2358 	/*
2359 	 * Attempt to create using a bad mirror.
2360 	 */
2361 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2362 	VERIFY3U(ENOENT, ==,
2363 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2364 	nvlist_free(nvroot);
2365 
2366 	/*
2367 	 * Attempt to create an existing pool.  It shouldn't matter
2368 	 * what's in the nvroot; we should fail with EEXIST.
2369 	 */
2370 	(void) rw_rdlock(&ztest_name_lock);
2371 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2372 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2373 	nvlist_free(nvroot);
2374 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2375 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2376 	spa_close(spa, FTAG);
2377 
2378 	(void) rw_unlock(&ztest_name_lock);
2379 }
2380 
2381 /* ARGSUSED */
2382 void
2383 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2384 {
2385 	spa_t *spa;
2386 	uint64_t initial_version = SPA_VERSION_INITIAL;
2387 	uint64_t version, newversion;
2388 	nvlist_t *nvroot, *props;
2389 	char *name;
2390 
2391 	VERIFY0(mutex_lock(&ztest_vdev_lock));
2392 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2393 
2394 	/*
2395 	 * Clean up from previous runs.
2396 	 */
2397 	(void) spa_destroy(name);
2398 
2399 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2400 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2401 
2402 	/*
2403 	 * If we're configuring a RAIDZ device then make sure that the
2404 	 * the initial version is capable of supporting that feature.
2405 	 */
2406 	switch (ztest_opts.zo_raidz_parity) {
2407 	case 0:
2408 	case 1:
2409 		initial_version = SPA_VERSION_INITIAL;
2410 		break;
2411 	case 2:
2412 		initial_version = SPA_VERSION_RAIDZ2;
2413 		break;
2414 	case 3:
2415 		initial_version = SPA_VERSION_RAIDZ3;
2416 		break;
2417 	}
2418 
2419 	/*
2420 	 * Create a pool with a spa version that can be upgraded. Pick
2421 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2422 	 */
2423 	do {
2424 		version = ztest_random_spa_version(initial_version);
2425 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2426 
2427 	props = fnvlist_alloc();
2428 	fnvlist_add_uint64(props,
2429 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2430 	VERIFY0(spa_create(name, nvroot, props, NULL));
2431 	fnvlist_free(nvroot);
2432 	fnvlist_free(props);
2433 
2434 	VERIFY0(spa_open(name, &spa, FTAG));
2435 	VERIFY3U(spa_version(spa), ==, version);
2436 	newversion = ztest_random_spa_version(version + 1);
2437 
2438 	if (ztest_opts.zo_verbose >= 4) {
2439 		(void) printf("upgrading spa version from %llu to %llu\n",
2440 		    (u_longlong_t)version, (u_longlong_t)newversion);
2441 	}
2442 
2443 	spa_upgrade(spa, newversion);
2444 	VERIFY3U(spa_version(spa), >, version);
2445 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2446 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2447 	spa_close(spa, FTAG);
2448 
2449 	strfree(name);
2450 	VERIFY0(mutex_unlock(&ztest_vdev_lock));
2451 }
2452 
2453 static vdev_t *
2454 vdev_lookup_by_path(vdev_t *vd, const char *path)
2455 {
2456 	vdev_t *mvd;
2457 
2458 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2459 		return (vd);
2460 
2461 	for (int c = 0; c < vd->vdev_children; c++)
2462 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2463 		    NULL)
2464 			return (mvd);
2465 
2466 	return (NULL);
2467 }
2468 
2469 /*
2470  * Find the first available hole which can be used as a top-level.
2471  */
2472 int
2473 find_vdev_hole(spa_t *spa)
2474 {
2475 	vdev_t *rvd = spa->spa_root_vdev;
2476 	int c;
2477 
2478 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2479 
2480 	for (c = 0; c < rvd->vdev_children; c++) {
2481 		vdev_t *cvd = rvd->vdev_child[c];
2482 
2483 		if (cvd->vdev_ishole)
2484 			break;
2485 	}
2486 	return (c);
2487 }
2488 
2489 /*
2490  * Verify that vdev_add() works as expected.
2491  */
2492 /* ARGSUSED */
2493 void
2494 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2495 {
2496 	ztest_shared_t *zs = ztest_shared;
2497 	spa_t *spa = ztest_spa;
2498 	uint64_t leaves;
2499 	uint64_t guid;
2500 	nvlist_t *nvroot;
2501 	int error;
2502 
2503 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2504 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2505 
2506 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2507 
2508 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2509 
2510 	/*
2511 	 * If we have slogs then remove them 1/4 of the time.
2512 	 */
2513 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2514 		/*
2515 		 * Grab the guid from the head of the log class rotor.
2516 		 */
2517 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2518 
2519 		spa_config_exit(spa, SCL_VDEV, FTAG);
2520 
2521 		/*
2522 		 * We have to grab the zs_name_lock as writer to
2523 		 * prevent a race between removing a slog (dmu_objset_find)
2524 		 * and destroying a dataset. Removing the slog will
2525 		 * grab a reference on the dataset which may cause
2526 		 * dmu_objset_destroy() to fail with EBUSY thus
2527 		 * leaving the dataset in an inconsistent state.
2528 		 */
2529 		VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2530 		error = spa_vdev_remove(spa, guid, B_FALSE);
2531 		VERIFY(rw_unlock(&ztest_name_lock) == 0);
2532 
2533 		if (error && error != EEXIST)
2534 			fatal(0, "spa_vdev_remove() = %d", error);
2535 	} else {
2536 		spa_config_exit(spa, SCL_VDEV, FTAG);
2537 
2538 		/*
2539 		 * Make 1/4 of the devices be log devices.
2540 		 */
2541 		nvroot = make_vdev_root(NULL, NULL, NULL,
2542 		    ztest_opts.zo_vdev_size, 0,
2543 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2544 		    zs->zs_mirrors, 1);
2545 
2546 		error = spa_vdev_add(spa, nvroot);
2547 		nvlist_free(nvroot);
2548 
2549 		if (error == ENOSPC)
2550 			ztest_record_enospc("spa_vdev_add");
2551 		else if (error != 0)
2552 			fatal(0, "spa_vdev_add() = %d", error);
2553 	}
2554 
2555 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2556 }
2557 
2558 /*
2559  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2560  */
2561 /* ARGSUSED */
2562 void
2563 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2564 {
2565 	ztest_shared_t *zs = ztest_shared;
2566 	spa_t *spa = ztest_spa;
2567 	vdev_t *rvd = spa->spa_root_vdev;
2568 	spa_aux_vdev_t *sav;
2569 	char *aux;
2570 	uint64_t guid = 0;
2571 	int error;
2572 
2573 	if (ztest_random(2) == 0) {
2574 		sav = &spa->spa_spares;
2575 		aux = ZPOOL_CONFIG_SPARES;
2576 	} else {
2577 		sav = &spa->spa_l2cache;
2578 		aux = ZPOOL_CONFIG_L2CACHE;
2579 	}
2580 
2581 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2582 
2583 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2584 
2585 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2586 		/*
2587 		 * Pick a random device to remove.
2588 		 */
2589 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2590 	} else {
2591 		/*
2592 		 * Find an unused device we can add.
2593 		 */
2594 		zs->zs_vdev_aux = 0;
2595 		for (;;) {
2596 			char path[MAXPATHLEN];
2597 			int c;
2598 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2599 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2600 			    zs->zs_vdev_aux);
2601 			for (c = 0; c < sav->sav_count; c++)
2602 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2603 				    path) == 0)
2604 					break;
2605 			if (c == sav->sav_count &&
2606 			    vdev_lookup_by_path(rvd, path) == NULL)
2607 				break;
2608 			zs->zs_vdev_aux++;
2609 		}
2610 	}
2611 
2612 	spa_config_exit(spa, SCL_VDEV, FTAG);
2613 
2614 	if (guid == 0) {
2615 		/*
2616 		 * Add a new device.
2617 		 */
2618 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2619 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2620 		error = spa_vdev_add(spa, nvroot);
2621 		if (error != 0)
2622 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2623 		nvlist_free(nvroot);
2624 	} else {
2625 		/*
2626 		 * Remove an existing device.  Sometimes, dirty its
2627 		 * vdev state first to make sure we handle removal
2628 		 * of devices that have pending state changes.
2629 		 */
2630 		if (ztest_random(2) == 0)
2631 			(void) vdev_online(spa, guid, 0, NULL);
2632 
2633 		error = spa_vdev_remove(spa, guid, B_FALSE);
2634 		if (error != 0 && error != EBUSY)
2635 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2636 	}
2637 
2638 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2639 }
2640 
2641 /*
2642  * split a pool if it has mirror tlvdevs
2643  */
2644 /* ARGSUSED */
2645 void
2646 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2647 {
2648 	ztest_shared_t *zs = ztest_shared;
2649 	spa_t *spa = ztest_spa;
2650 	vdev_t *rvd = spa->spa_root_vdev;
2651 	nvlist_t *tree, **child, *config, *split, **schild;
2652 	uint_t c, children, schildren = 0, lastlogid = 0;
2653 	int error = 0;
2654 
2655 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2656 
2657 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2658 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2659 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2660 		return;
2661 	}
2662 
2663 	/* clean up the old pool, if any */
2664 	(void) spa_destroy("splitp");
2665 
2666 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2667 
2668 	/* generate a config from the existing config */
2669 	mutex_enter(&spa->spa_props_lock);
2670 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2671 	    &tree) == 0);
2672 	mutex_exit(&spa->spa_props_lock);
2673 
2674 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2675 	    &children) == 0);
2676 
2677 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2678 	for (c = 0; c < children; c++) {
2679 		vdev_t *tvd = rvd->vdev_child[c];
2680 		nvlist_t **mchild;
2681 		uint_t mchildren;
2682 
2683 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2684 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2685 			    0) == 0);
2686 			VERIFY(nvlist_add_string(schild[schildren],
2687 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2688 			VERIFY(nvlist_add_uint64(schild[schildren],
2689 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2690 			if (lastlogid == 0)
2691 				lastlogid = schildren;
2692 			++schildren;
2693 			continue;
2694 		}
2695 		lastlogid = 0;
2696 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2697 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2698 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2699 	}
2700 
2701 	/* OK, create a config that can be used to split */
2702 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2703 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2704 	    VDEV_TYPE_ROOT) == 0);
2705 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2706 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2707 
2708 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2709 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2710 
2711 	for (c = 0; c < schildren; c++)
2712 		nvlist_free(schild[c]);
2713 	free(schild);
2714 	nvlist_free(split);
2715 
2716 	spa_config_exit(spa, SCL_VDEV, FTAG);
2717 
2718 	(void) rw_wrlock(&ztest_name_lock);
2719 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2720 	(void) rw_unlock(&ztest_name_lock);
2721 
2722 	nvlist_free(config);
2723 
2724 	if (error == 0) {
2725 		(void) printf("successful split - results:\n");
2726 		mutex_enter(&spa_namespace_lock);
2727 		show_pool_stats(spa);
2728 		show_pool_stats(spa_lookup("splitp"));
2729 		mutex_exit(&spa_namespace_lock);
2730 		++zs->zs_splits;
2731 		--zs->zs_mirrors;
2732 	}
2733 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2734 
2735 }
2736 
2737 /*
2738  * Verify that we can attach and detach devices.
2739  */
2740 /* ARGSUSED */
2741 void
2742 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2743 {
2744 	ztest_shared_t *zs = ztest_shared;
2745 	spa_t *spa = ztest_spa;
2746 	spa_aux_vdev_t *sav = &spa->spa_spares;
2747 	vdev_t *rvd = spa->spa_root_vdev;
2748 	vdev_t *oldvd, *newvd, *pvd;
2749 	nvlist_t *root;
2750 	uint64_t leaves;
2751 	uint64_t leaf, top;
2752 	uint64_t ashift = ztest_get_ashift();
2753 	uint64_t oldguid, pguid;
2754 	uint64_t oldsize, newsize;
2755 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2756 	int replacing;
2757 	int oldvd_has_siblings = B_FALSE;
2758 	int newvd_is_spare = B_FALSE;
2759 	int oldvd_is_log;
2760 	int error, expected_error;
2761 
2762 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2763 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2764 
2765 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2766 
2767 	/*
2768 	 * Decide whether to do an attach or a replace.
2769 	 */
2770 	replacing = ztest_random(2);
2771 
2772 	/*
2773 	 * Pick a random top-level vdev.
2774 	 */
2775 	top = ztest_random_vdev_top(spa, B_TRUE);
2776 
2777 	/*
2778 	 * Pick a random leaf within it.
2779 	 */
2780 	leaf = ztest_random(leaves);
2781 
2782 	/*
2783 	 * Locate this vdev.
2784 	 */
2785 	oldvd = rvd->vdev_child[top];
2786 	if (zs->zs_mirrors >= 1) {
2787 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2788 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2789 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2790 	}
2791 	if (ztest_opts.zo_raidz > 1) {
2792 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2793 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2794 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2795 	}
2796 
2797 	/*
2798 	 * If we're already doing an attach or replace, oldvd may be a
2799 	 * mirror vdev -- in which case, pick a random child.
2800 	 */
2801 	while (oldvd->vdev_children != 0) {
2802 		oldvd_has_siblings = B_TRUE;
2803 		ASSERT(oldvd->vdev_children >= 2);
2804 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2805 	}
2806 
2807 	oldguid = oldvd->vdev_guid;
2808 	oldsize = vdev_get_min_asize(oldvd);
2809 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2810 	(void) strcpy(oldpath, oldvd->vdev_path);
2811 	pvd = oldvd->vdev_parent;
2812 	pguid = pvd->vdev_guid;
2813 
2814 	/*
2815 	 * If oldvd has siblings, then half of the time, detach it.
2816 	 */
2817 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2818 		spa_config_exit(spa, SCL_VDEV, FTAG);
2819 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2820 		if (error != 0 && error != ENODEV && error != EBUSY &&
2821 		    error != ENOTSUP)
2822 			fatal(0, "detach (%s) returned %d", oldpath, error);
2823 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2824 		return;
2825 	}
2826 
2827 	/*
2828 	 * For the new vdev, choose with equal probability between the two
2829 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2830 	 */
2831 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2832 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2833 		newvd_is_spare = B_TRUE;
2834 		(void) strcpy(newpath, newvd->vdev_path);
2835 	} else {
2836 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2837 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2838 		    top * leaves + leaf);
2839 		if (ztest_random(2) == 0)
2840 			newpath[strlen(newpath) - 1] = 'b';
2841 		newvd = vdev_lookup_by_path(rvd, newpath);
2842 	}
2843 
2844 	if (newvd) {
2845 		newsize = vdev_get_min_asize(newvd);
2846 	} else {
2847 		/*
2848 		 * Make newsize a little bigger or smaller than oldsize.
2849 		 * If it's smaller, the attach should fail.
2850 		 * If it's larger, and we're doing a replace,
2851 		 * we should get dynamic LUN growth when we're done.
2852 		 */
2853 		newsize = 10 * oldsize / (9 + ztest_random(3));
2854 	}
2855 
2856 	/*
2857 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2858 	 * unless it's a replace; in that case any non-replacing parent is OK.
2859 	 *
2860 	 * If newvd is already part of the pool, it should fail with EBUSY.
2861 	 *
2862 	 * If newvd is too small, it should fail with EOVERFLOW.
2863 	 */
2864 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2865 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2866 	    pvd->vdev_ops == &vdev_replacing_ops ||
2867 	    pvd->vdev_ops == &vdev_spare_ops))
2868 		expected_error = ENOTSUP;
2869 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2870 		expected_error = ENOTSUP;
2871 	else if (newvd == oldvd)
2872 		expected_error = replacing ? 0 : EBUSY;
2873 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2874 		expected_error = EBUSY;
2875 	else if (newsize < oldsize)
2876 		expected_error = EOVERFLOW;
2877 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2878 		expected_error = EDOM;
2879 	else
2880 		expected_error = 0;
2881 
2882 	spa_config_exit(spa, SCL_VDEV, FTAG);
2883 
2884 	/*
2885 	 * Build the nvlist describing newpath.
2886 	 */
2887 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2888 	    ashift, 0, 0, 0, 1);
2889 
2890 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2891 
2892 	nvlist_free(root);
2893 
2894 	/*
2895 	 * If our parent was the replacing vdev, but the replace completed,
2896 	 * then instead of failing with ENOTSUP we may either succeed,
2897 	 * fail with ENODEV, or fail with EOVERFLOW.
2898 	 */
2899 	if (expected_error == ENOTSUP &&
2900 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2901 		expected_error = error;
2902 
2903 	/*
2904 	 * If someone grew the LUN, the replacement may be too small.
2905 	 */
2906 	if (error == EOVERFLOW || error == EBUSY)
2907 		expected_error = error;
2908 
2909 	/* XXX workaround 6690467 */
2910 	if (error != expected_error && expected_error != EBUSY) {
2911 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2912 		    "returned %d, expected %d",
2913 		    oldpath, oldsize, newpath,
2914 		    newsize, replacing, error, expected_error);
2915 	}
2916 
2917 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2918 }
2919 
2920 /*
2921  * Callback function which expands the physical size of the vdev.
2922  */
2923 vdev_t *
2924 grow_vdev(vdev_t *vd, void *arg)
2925 {
2926 	spa_t *spa = vd->vdev_spa;
2927 	size_t *newsize = arg;
2928 	size_t fsize;
2929 	int fd;
2930 
2931 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2932 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2933 
2934 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2935 		return (vd);
2936 
2937 	fsize = lseek(fd, 0, SEEK_END);
2938 	(void) ftruncate(fd, *newsize);
2939 
2940 	if (ztest_opts.zo_verbose >= 6) {
2941 		(void) printf("%s grew from %lu to %lu bytes\n",
2942 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2943 	}
2944 	(void) close(fd);
2945 	return (NULL);
2946 }
2947 
2948 /*
2949  * Callback function which expands a given vdev by calling vdev_online().
2950  */
2951 /* ARGSUSED */
2952 vdev_t *
2953 online_vdev(vdev_t *vd, void *arg)
2954 {
2955 	spa_t *spa = vd->vdev_spa;
2956 	vdev_t *tvd = vd->vdev_top;
2957 	uint64_t guid = vd->vdev_guid;
2958 	uint64_t generation = spa->spa_config_generation + 1;
2959 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2960 	int error;
2961 
2962 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2963 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2964 
2965 	/* Calling vdev_online will initialize the new metaslabs */
2966 	spa_config_exit(spa, SCL_STATE, spa);
2967 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2968 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2969 
2970 	/*
2971 	 * If vdev_online returned an error or the underlying vdev_open
2972 	 * failed then we abort the expand. The only way to know that
2973 	 * vdev_open fails is by checking the returned newstate.
2974 	 */
2975 	if (error || newstate != VDEV_STATE_HEALTHY) {
2976 		if (ztest_opts.zo_verbose >= 5) {
2977 			(void) printf("Unable to expand vdev, state %llu, "
2978 			    "error %d\n", (u_longlong_t)newstate, error);
2979 		}
2980 		return (vd);
2981 	}
2982 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2983 
2984 	/*
2985 	 * Since we dropped the lock we need to ensure that we're
2986 	 * still talking to the original vdev. It's possible this
2987 	 * vdev may have been detached/replaced while we were
2988 	 * trying to online it.
2989 	 */
2990 	if (generation != spa->spa_config_generation) {
2991 		if (ztest_opts.zo_verbose >= 5) {
2992 			(void) printf("vdev configuration has changed, "
2993 			    "guid %llu, state %llu, expected gen %llu, "
2994 			    "got gen %llu\n",
2995 			    (u_longlong_t)guid,
2996 			    (u_longlong_t)tvd->vdev_state,
2997 			    (u_longlong_t)generation,
2998 			    (u_longlong_t)spa->spa_config_generation);
2999 		}
3000 		return (vd);
3001 	}
3002 	return (NULL);
3003 }
3004 
3005 /*
3006  * Traverse the vdev tree calling the supplied function.
3007  * We continue to walk the tree until we either have walked all
3008  * children or we receive a non-NULL return from the callback.
3009  * If a NULL callback is passed, then we just return back the first
3010  * leaf vdev we encounter.
3011  */
3012 vdev_t *
3013 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3014 {
3015 	if (vd->vdev_ops->vdev_op_leaf) {
3016 		if (func == NULL)
3017 			return (vd);
3018 		else
3019 			return (func(vd, arg));
3020 	}
3021 
3022 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3023 		vdev_t *cvd = vd->vdev_child[c];
3024 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3025 			return (cvd);
3026 	}
3027 	return (NULL);
3028 }
3029 
3030 /*
3031  * Verify that dynamic LUN growth works as expected.
3032  */
3033 /* ARGSUSED */
3034 void
3035 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3036 {
3037 	spa_t *spa = ztest_spa;
3038 	vdev_t *vd, *tvd;
3039 	metaslab_class_t *mc;
3040 	metaslab_group_t *mg;
3041 	size_t psize, newsize;
3042 	uint64_t top;
3043 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3044 
3045 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3046 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3047 
3048 	top = ztest_random_vdev_top(spa, B_TRUE);
3049 
3050 	tvd = spa->spa_root_vdev->vdev_child[top];
3051 	mg = tvd->vdev_mg;
3052 	mc = mg->mg_class;
3053 	old_ms_count = tvd->vdev_ms_count;
3054 	old_class_space = metaslab_class_get_space(mc);
3055 
3056 	/*
3057 	 * Determine the size of the first leaf vdev associated with
3058 	 * our top-level device.
3059 	 */
3060 	vd = vdev_walk_tree(tvd, NULL, NULL);
3061 	ASSERT3P(vd, !=, NULL);
3062 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3063 
3064 	psize = vd->vdev_psize;
3065 
3066 	/*
3067 	 * We only try to expand the vdev if it's healthy, less than 4x its
3068 	 * original size, and it has a valid psize.
3069 	 */
3070 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3071 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3072 		spa_config_exit(spa, SCL_STATE, spa);
3073 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3074 		return;
3075 	}
3076 	ASSERT(psize > 0);
3077 	newsize = psize + psize / 8;
3078 	ASSERT3U(newsize, >, psize);
3079 
3080 	if (ztest_opts.zo_verbose >= 6) {
3081 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3082 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3083 	}
3084 
3085 	/*
3086 	 * Growing the vdev is a two step process:
3087 	 *	1). expand the physical size (i.e. relabel)
3088 	 *	2). online the vdev to create the new metaslabs
3089 	 */
3090 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3091 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3092 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3093 		if (ztest_opts.zo_verbose >= 5) {
3094 			(void) printf("Could not expand LUN because "
3095 			    "the vdev configuration changed.\n");
3096 		}
3097 		spa_config_exit(spa, SCL_STATE, spa);
3098 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3099 		return;
3100 	}
3101 
3102 	spa_config_exit(spa, SCL_STATE, spa);
3103 
3104 	/*
3105 	 * Expanding the LUN will update the config asynchronously,
3106 	 * thus we must wait for the async thread to complete any
3107 	 * pending tasks before proceeding.
3108 	 */
3109 	for (;;) {
3110 		boolean_t done;
3111 		mutex_enter(&spa->spa_async_lock);
3112 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3113 		mutex_exit(&spa->spa_async_lock);
3114 		if (done)
3115 			break;
3116 		txg_wait_synced(spa_get_dsl(spa), 0);
3117 		(void) poll(NULL, 0, 100);
3118 	}
3119 
3120 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3121 
3122 	tvd = spa->spa_root_vdev->vdev_child[top];
3123 	new_ms_count = tvd->vdev_ms_count;
3124 	new_class_space = metaslab_class_get_space(mc);
3125 
3126 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3127 		if (ztest_opts.zo_verbose >= 5) {
3128 			(void) printf("Could not verify LUN expansion due to "
3129 			    "intervening vdev offline or remove.\n");
3130 		}
3131 		spa_config_exit(spa, SCL_STATE, spa);
3132 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3133 		return;
3134 	}
3135 
3136 	/*
3137 	 * Make sure we were able to grow the vdev.
3138 	 */
3139 	if (new_ms_count <= old_ms_count)
3140 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3141 		    old_ms_count, new_ms_count);
3142 
3143 	/*
3144 	 * Make sure we were able to grow the pool.
3145 	 */
3146 	if (new_class_space <= old_class_space)
3147 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3148 		    old_class_space, new_class_space);
3149 
3150 	if (ztest_opts.zo_verbose >= 5) {
3151 		char oldnumbuf[6], newnumbuf[6];
3152 
3153 		nicenum(old_class_space, oldnumbuf);
3154 		nicenum(new_class_space, newnumbuf);
3155 		(void) printf("%s grew from %s to %s\n",
3156 		    spa->spa_name, oldnumbuf, newnumbuf);
3157 	}
3158 
3159 	spa_config_exit(spa, SCL_STATE, spa);
3160 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3161 }
3162 
3163 /*
3164  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3165  */
3166 /* ARGSUSED */
3167 static void
3168 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3169 {
3170 	/*
3171 	 * Create the objects common to all ztest datasets.
3172 	 */
3173 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3174 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3175 }
3176 
3177 static int
3178 ztest_dataset_create(char *dsname)
3179 {
3180 	uint64_t zilset = ztest_random(100);
3181 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3182 	    ztest_objset_create_cb, NULL);
3183 
3184 	if (err || zilset < 80)
3185 		return (err);
3186 
3187 	if (ztest_opts.zo_verbose >= 6)
3188 		(void) printf("Setting dataset %s to sync always\n", dsname);
3189 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3190 	    ZFS_SYNC_ALWAYS, B_FALSE));
3191 }
3192 
3193 /* ARGSUSED */
3194 static int
3195 ztest_objset_destroy_cb(const char *name, void *arg)
3196 {
3197 	objset_t *os;
3198 	dmu_object_info_t doi;
3199 	int error;
3200 
3201 	/*
3202 	 * Verify that the dataset contains a directory object.
3203 	 */
3204 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3205 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3206 	if (error != ENOENT) {
3207 		/* We could have crashed in the middle of destroying it */
3208 		ASSERT0(error);
3209 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3210 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3211 	}
3212 	dmu_objset_disown(os, FTAG);
3213 
3214 	/*
3215 	 * Destroy the dataset.
3216 	 */
3217 	if (strchr(name, '@') != NULL) {
3218 		VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3219 	} else {
3220 		VERIFY0(dsl_destroy_head(name));
3221 	}
3222 	return (0);
3223 }
3224 
3225 static boolean_t
3226 ztest_snapshot_create(char *osname, uint64_t id)
3227 {
3228 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3229 	int error;
3230 
3231 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3232 
3233 	error = dmu_objset_snapshot_one(osname, snapname);
3234 	if (error == ENOSPC) {
3235 		ztest_record_enospc(FTAG);
3236 		return (B_FALSE);
3237 	}
3238 	if (error != 0 && error != EEXIST) {
3239 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3240 		    snapname, error);
3241 	}
3242 	return (B_TRUE);
3243 }
3244 
3245 static boolean_t
3246 ztest_snapshot_destroy(char *osname, uint64_t id)
3247 {
3248 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3249 	int error;
3250 
3251 	(void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3252 	    (u_longlong_t)id);
3253 
3254 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3255 	if (error != 0 && error != ENOENT)
3256 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3257 	return (B_TRUE);
3258 }
3259 
3260 /* ARGSUSED */
3261 void
3262 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3263 {
3264 	ztest_ds_t zdtmp;
3265 	int iters;
3266 	int error;
3267 	objset_t *os, *os2;
3268 	char name[ZFS_MAX_DATASET_NAME_LEN];
3269 	zilog_t *zilog;
3270 
3271 	(void) rw_rdlock(&ztest_name_lock);
3272 
3273 	(void) snprintf(name, sizeof (name), "%s/temp_%llu",
3274 	    ztest_opts.zo_pool, (u_longlong_t)id);
3275 
3276 	/*
3277 	 * If this dataset exists from a previous run, process its replay log
3278 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3279 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3280 	 */
3281 	if (ztest_random(2) == 0 &&
3282 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3283 		ztest_zd_init(&zdtmp, NULL, os);
3284 		zil_replay(os, &zdtmp, ztest_replay_vector);
3285 		ztest_zd_fini(&zdtmp);
3286 		dmu_objset_disown(os, FTAG);
3287 	}
3288 
3289 	/*
3290 	 * There may be an old instance of the dataset we're about to
3291 	 * create lying around from a previous run.  If so, destroy it
3292 	 * and all of its snapshots.
3293 	 */
3294 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3295 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3296 
3297 	/*
3298 	 * Verify that the destroyed dataset is no longer in the namespace.
3299 	 */
3300 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3301 	    FTAG, &os));
3302 
3303 	/*
3304 	 * Verify that we can create a new dataset.
3305 	 */
3306 	error = ztest_dataset_create(name);
3307 	if (error) {
3308 		if (error == ENOSPC) {
3309 			ztest_record_enospc(FTAG);
3310 			(void) rw_unlock(&ztest_name_lock);
3311 			return;
3312 		}
3313 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3314 	}
3315 
3316 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3317 
3318 	ztest_zd_init(&zdtmp, NULL, os);
3319 
3320 	/*
3321 	 * Open the intent log for it.
3322 	 */
3323 	zilog = zil_open(os, ztest_get_data);
3324 
3325 	/*
3326 	 * Put some objects in there, do a little I/O to them,
3327 	 * and randomly take a couple of snapshots along the way.
3328 	 */
3329 	iters = ztest_random(5);
3330 	for (int i = 0; i < iters; i++) {
3331 		ztest_dmu_object_alloc_free(&zdtmp, id);
3332 		if (ztest_random(iters) == 0)
3333 			(void) ztest_snapshot_create(name, i);
3334 	}
3335 
3336 	/*
3337 	 * Verify that we cannot create an existing dataset.
3338 	 */
3339 	VERIFY3U(EEXIST, ==,
3340 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3341 
3342 	/*
3343 	 * Verify that we can hold an objset that is also owned.
3344 	 */
3345 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3346 	dmu_objset_rele(os2, FTAG);
3347 
3348 	/*
3349 	 * Verify that we cannot own an objset that is already owned.
3350 	 */
3351 	VERIFY3U(EBUSY, ==,
3352 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3353 
3354 	zil_close(zilog);
3355 	dmu_objset_disown(os, FTAG);
3356 	ztest_zd_fini(&zdtmp);
3357 
3358 	(void) rw_unlock(&ztest_name_lock);
3359 }
3360 
3361 /*
3362  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3363  */
3364 void
3365 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3366 {
3367 	(void) rw_rdlock(&ztest_name_lock);
3368 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3369 	(void) ztest_snapshot_create(zd->zd_name, id);
3370 	(void) rw_unlock(&ztest_name_lock);
3371 }
3372 
3373 /*
3374  * Cleanup non-standard snapshots and clones.
3375  */
3376 void
3377 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3378 {
3379 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3380 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3381 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3382 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3383 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3384 	int error;
3385 
3386 	(void) snprintf(snap1name, sizeof (snap1name),
3387 	    "%s@s1_%llu", osname, id);
3388 	(void) snprintf(clone1name, sizeof (clone1name),
3389 	    "%s/c1_%llu", osname, id);
3390 	(void) snprintf(snap2name, sizeof (snap2name),
3391 	    "%s@s2_%llu", clone1name, id);
3392 	(void) snprintf(clone2name, sizeof (clone2name),
3393 	    "%s/c2_%llu", osname, id);
3394 	(void) snprintf(snap3name, sizeof (snap3name),
3395 	    "%s@s3_%llu", clone1name, id);
3396 
3397 	error = dsl_destroy_head(clone2name);
3398 	if (error && error != ENOENT)
3399 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3400 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3401 	if (error && error != ENOENT)
3402 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3403 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3404 	if (error && error != ENOENT)
3405 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3406 	error = dsl_destroy_head(clone1name);
3407 	if (error && error != ENOENT)
3408 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3409 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3410 	if (error && error != ENOENT)
3411 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3412 }
3413 
3414 /*
3415  * Verify dsl_dataset_promote handles EBUSY
3416  */
3417 void
3418 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3419 {
3420 	objset_t *os;
3421 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3422 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3423 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3424 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3425 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3426 	char *osname = zd->zd_name;
3427 	int error;
3428 
3429 	(void) rw_rdlock(&ztest_name_lock);
3430 
3431 	ztest_dsl_dataset_cleanup(osname, id);
3432 
3433 	(void) snprintf(snap1name, sizeof (snap1name),
3434 	    "%s@s1_%llu", osname, id);
3435 	(void) snprintf(clone1name, sizeof (clone1name),
3436 	    "%s/c1_%llu", osname, id);
3437 	(void) snprintf(snap2name, sizeof (snap2name),
3438 	    "%s@s2_%llu", clone1name, id);
3439 	(void) snprintf(clone2name, sizeof (clone2name),
3440 	    "%s/c2_%llu", osname, id);
3441 	(void) snprintf(snap3name, sizeof (snap3name),
3442 	    "%s@s3_%llu", clone1name, id);
3443 
3444 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3445 	if (error && error != EEXIST) {
3446 		if (error == ENOSPC) {
3447 			ztest_record_enospc(FTAG);
3448 			goto out;
3449 		}
3450 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3451 	}
3452 
3453 	error = dmu_objset_clone(clone1name, snap1name);
3454 	if (error) {
3455 		if (error == ENOSPC) {
3456 			ztest_record_enospc(FTAG);
3457 			goto out;
3458 		}
3459 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3460 	}
3461 
3462 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3463 	if (error && error != EEXIST) {
3464 		if (error == ENOSPC) {
3465 			ztest_record_enospc(FTAG);
3466 			goto out;
3467 		}
3468 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3469 	}
3470 
3471 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3472 	if (error && error != EEXIST) {
3473 		if (error == ENOSPC) {
3474 			ztest_record_enospc(FTAG);
3475 			goto out;
3476 		}
3477 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3478 	}
3479 
3480 	error = dmu_objset_clone(clone2name, snap3name);
3481 	if (error) {
3482 		if (error == ENOSPC) {
3483 			ztest_record_enospc(FTAG);
3484 			goto out;
3485 		}
3486 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3487 	}
3488 
3489 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3490 	if (error)
3491 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3492 	error = dsl_dataset_promote(clone2name, NULL);
3493 	if (error == ENOSPC) {
3494 		dmu_objset_disown(os, FTAG);
3495 		ztest_record_enospc(FTAG);
3496 		goto out;
3497 	}
3498 	if (error != EBUSY)
3499 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3500 		    error);
3501 	dmu_objset_disown(os, FTAG);
3502 
3503 out:
3504 	ztest_dsl_dataset_cleanup(osname, id);
3505 
3506 	(void) rw_unlock(&ztest_name_lock);
3507 }
3508 
3509 /*
3510  * Verify that dmu_object_{alloc,free} work as expected.
3511  */
3512 void
3513 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3514 {
3515 	ztest_od_t od[4];
3516 	int batchsize = sizeof (od) / sizeof (od[0]);
3517 
3518 	for (int b = 0; b < batchsize; b++)
3519 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3520 
3521 	/*
3522 	 * Destroy the previous batch of objects, create a new batch,
3523 	 * and do some I/O on the new objects.
3524 	 */
3525 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3526 		return;
3527 
3528 	while (ztest_random(4 * batchsize) != 0)
3529 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3530 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3531 }
3532 
3533 /*
3534  * Verify that dmu_{read,write} work as expected.
3535  */
3536 void
3537 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3538 {
3539 	objset_t *os = zd->zd_os;
3540 	ztest_od_t od[2];
3541 	dmu_tx_t *tx;
3542 	int i, freeit, error;
3543 	uint64_t n, s, txg;
3544 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3545 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3546 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3547 	uint64_t regions = 997;
3548 	uint64_t stride = 123456789ULL;
3549 	uint64_t width = 40;
3550 	int free_percent = 5;
3551 
3552 	/*
3553 	 * This test uses two objects, packobj and bigobj, that are always
3554 	 * updated together (i.e. in the same tx) so that their contents are
3555 	 * in sync and can be compared.  Their contents relate to each other
3556 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3557 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3558 	 * for any index n, there are three bufwads that should be identical:
3559 	 *
3560 	 *	packobj, at offset n * sizeof (bufwad_t)
3561 	 *	bigobj, at the head of the nth chunk
3562 	 *	bigobj, at the tail of the nth chunk
3563 	 *
3564 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3565 	 * and it doesn't have any relation to the object blocksize.
3566 	 * The only requirement is that it can hold at least two bufwads.
3567 	 *
3568 	 * Normally, we write the bufwad to each of these locations.
3569 	 * However, free_percent of the time we instead write zeroes to
3570 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3571 	 * bigobj to packobj, we can verify that the DMU is correctly
3572 	 * tracking which parts of an object are allocated and free,
3573 	 * and that the contents of the allocated blocks are correct.
3574 	 */
3575 
3576 	/*
3577 	 * Read the directory info.  If it's the first time, set things up.
3578 	 */
3579 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3580 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3581 
3582 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3583 		return;
3584 
3585 	bigobj = od[0].od_object;
3586 	packobj = od[1].od_object;
3587 	chunksize = od[0].od_gen;
3588 	ASSERT(chunksize == od[1].od_gen);
3589 
3590 	/*
3591 	 * Prefetch a random chunk of the big object.
3592 	 * Our aim here is to get some async reads in flight
3593 	 * for blocks that we may free below; the DMU should
3594 	 * handle this race correctly.
3595 	 */
3596 	n = ztest_random(regions) * stride + ztest_random(width);
3597 	s = 1 + ztest_random(2 * width - 1);
3598 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3599 	    ZIO_PRIORITY_SYNC_READ);
3600 
3601 	/*
3602 	 * Pick a random index and compute the offsets into packobj and bigobj.
3603 	 */
3604 	n = ztest_random(regions) * stride + ztest_random(width);
3605 	s = 1 + ztest_random(width - 1);
3606 
3607 	packoff = n * sizeof (bufwad_t);
3608 	packsize = s * sizeof (bufwad_t);
3609 
3610 	bigoff = n * chunksize;
3611 	bigsize = s * chunksize;
3612 
3613 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3614 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3615 
3616 	/*
3617 	 * free_percent of the time, free a range of bigobj rather than
3618 	 * overwriting it.
3619 	 */
3620 	freeit = (ztest_random(100) < free_percent);
3621 
3622 	/*
3623 	 * Read the current contents of our objects.
3624 	 */
3625 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3626 	    DMU_READ_PREFETCH);
3627 	ASSERT0(error);
3628 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3629 	    DMU_READ_PREFETCH);
3630 	ASSERT0(error);
3631 
3632 	/*
3633 	 * Get a tx for the mods to both packobj and bigobj.
3634 	 */
3635 	tx = dmu_tx_create(os);
3636 
3637 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3638 
3639 	if (freeit)
3640 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3641 	else
3642 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3643 
3644 	/* This accounts for setting the checksum/compression. */
3645 	dmu_tx_hold_bonus(tx, bigobj);
3646 
3647 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3648 	if (txg == 0) {
3649 		umem_free(packbuf, packsize);
3650 		umem_free(bigbuf, bigsize);
3651 		return;
3652 	}
3653 
3654 	enum zio_checksum cksum;
3655 	do {
3656 		cksum = (enum zio_checksum)
3657 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3658 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3659 	dmu_object_set_checksum(os, bigobj, cksum, tx);
3660 
3661 	enum zio_compress comp;
3662 	do {
3663 		comp = (enum zio_compress)
3664 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3665 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3666 	dmu_object_set_compress(os, bigobj, comp, tx);
3667 
3668 	/*
3669 	 * For each index from n to n + s, verify that the existing bufwad
3670 	 * in packobj matches the bufwads at the head and tail of the
3671 	 * corresponding chunk in bigobj.  Then update all three bufwads
3672 	 * with the new values we want to write out.
3673 	 */
3674 	for (i = 0; i < s; i++) {
3675 		/* LINTED */
3676 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3677 		/* LINTED */
3678 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3679 		/* LINTED */
3680 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3681 
3682 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3683 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3684 
3685 		if (pack->bw_txg > txg)
3686 			fatal(0, "future leak: got %llx, open txg is %llx",
3687 			    pack->bw_txg, txg);
3688 
3689 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3690 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3691 			    pack->bw_index, n, i);
3692 
3693 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3694 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3695 
3696 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3697 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3698 
3699 		if (freeit) {
3700 			bzero(pack, sizeof (bufwad_t));
3701 		} else {
3702 			pack->bw_index = n + i;
3703 			pack->bw_txg = txg;
3704 			pack->bw_data = 1 + ztest_random(-2ULL);
3705 		}
3706 		*bigH = *pack;
3707 		*bigT = *pack;
3708 	}
3709 
3710 	/*
3711 	 * We've verified all the old bufwads, and made new ones.
3712 	 * Now write them out.
3713 	 */
3714 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3715 
3716 	if (freeit) {
3717 		if (ztest_opts.zo_verbose >= 7) {
3718 			(void) printf("freeing offset %llx size %llx"
3719 			    " txg %llx\n",
3720 			    (u_longlong_t)bigoff,
3721 			    (u_longlong_t)bigsize,
3722 			    (u_longlong_t)txg);
3723 		}
3724 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3725 	} else {
3726 		if (ztest_opts.zo_verbose >= 7) {
3727 			(void) printf("writing offset %llx size %llx"
3728 			    " txg %llx\n",
3729 			    (u_longlong_t)bigoff,
3730 			    (u_longlong_t)bigsize,
3731 			    (u_longlong_t)txg);
3732 		}
3733 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3734 	}
3735 
3736 	dmu_tx_commit(tx);
3737 
3738 	/*
3739 	 * Sanity check the stuff we just wrote.
3740 	 */
3741 	{
3742 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3743 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3744 
3745 		VERIFY(0 == dmu_read(os, packobj, packoff,
3746 		    packsize, packcheck, DMU_READ_PREFETCH));
3747 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3748 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3749 
3750 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3751 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3752 
3753 		umem_free(packcheck, packsize);
3754 		umem_free(bigcheck, bigsize);
3755 	}
3756 
3757 	umem_free(packbuf, packsize);
3758 	umem_free(bigbuf, bigsize);
3759 }
3760 
3761 void
3762 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3763     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3764 {
3765 	uint64_t i;
3766 	bufwad_t *pack;
3767 	bufwad_t *bigH;
3768 	bufwad_t *bigT;
3769 
3770 	/*
3771 	 * For each index from n to n + s, verify that the existing bufwad
3772 	 * in packobj matches the bufwads at the head and tail of the
3773 	 * corresponding chunk in bigobj.  Then update all three bufwads
3774 	 * with the new values we want to write out.
3775 	 */
3776 	for (i = 0; i < s; i++) {
3777 		/* LINTED */
3778 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3779 		/* LINTED */
3780 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3781 		/* LINTED */
3782 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3783 
3784 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3785 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3786 
3787 		if (pack->bw_txg > txg)
3788 			fatal(0, "future leak: got %llx, open txg is %llx",
3789 			    pack->bw_txg, txg);
3790 
3791 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3792 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3793 			    pack->bw_index, n, i);
3794 
3795 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3796 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3797 
3798 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3799 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3800 
3801 		pack->bw_index = n + i;
3802 		pack->bw_txg = txg;
3803 		pack->bw_data = 1 + ztest_random(-2ULL);
3804 
3805 		*bigH = *pack;
3806 		*bigT = *pack;
3807 	}
3808 }
3809 
3810 void
3811 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3812 {
3813 	objset_t *os = zd->zd_os;
3814 	ztest_od_t od[2];
3815 	dmu_tx_t *tx;
3816 	uint64_t i;
3817 	int error;
3818 	uint64_t n, s, txg;
3819 	bufwad_t *packbuf, *bigbuf;
3820 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3821 	uint64_t blocksize = ztest_random_blocksize();
3822 	uint64_t chunksize = blocksize;
3823 	uint64_t regions = 997;
3824 	uint64_t stride = 123456789ULL;
3825 	uint64_t width = 9;
3826 	dmu_buf_t *bonus_db;
3827 	arc_buf_t **bigbuf_arcbufs;
3828 	dmu_object_info_t doi;
3829 
3830 	/*
3831 	 * This test uses two objects, packobj and bigobj, that are always
3832 	 * updated together (i.e. in the same tx) so that their contents are
3833 	 * in sync and can be compared.  Their contents relate to each other
3834 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3835 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3836 	 * for any index n, there are three bufwads that should be identical:
3837 	 *
3838 	 *	packobj, at offset n * sizeof (bufwad_t)
3839 	 *	bigobj, at the head of the nth chunk
3840 	 *	bigobj, at the tail of the nth chunk
3841 	 *
3842 	 * The chunk size is set equal to bigobj block size so that
3843 	 * dmu_assign_arcbuf() can be tested for object updates.
3844 	 */
3845 
3846 	/*
3847 	 * Read the directory info.  If it's the first time, set things up.
3848 	 */
3849 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3850 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3851 
3852 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3853 		return;
3854 
3855 	bigobj = od[0].od_object;
3856 	packobj = od[1].od_object;
3857 	blocksize = od[0].od_blocksize;
3858 	chunksize = blocksize;
3859 	ASSERT(chunksize == od[1].od_gen);
3860 
3861 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3862 	VERIFY(ISP2(doi.doi_data_block_size));
3863 	VERIFY(chunksize == doi.doi_data_block_size);
3864 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3865 
3866 	/*
3867 	 * Pick a random index and compute the offsets into packobj and bigobj.
3868 	 */
3869 	n = ztest_random(regions) * stride + ztest_random(width);
3870 	s = 1 + ztest_random(width - 1);
3871 
3872 	packoff = n * sizeof (bufwad_t);
3873 	packsize = s * sizeof (bufwad_t);
3874 
3875 	bigoff = n * chunksize;
3876 	bigsize = s * chunksize;
3877 
3878 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3879 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3880 
3881 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3882 
3883 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3884 
3885 	/*
3886 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3887 	 * Iteration 1 test zcopy to already referenced dbufs.
3888 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3889 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3890 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3891 	 * Iteration 5 test zcopy when it can't be done.
3892 	 * Iteration 6 one more zcopy write.
3893 	 */
3894 	for (i = 0; i < 7; i++) {
3895 		uint64_t j;
3896 		uint64_t off;
3897 
3898 		/*
3899 		 * In iteration 5 (i == 5) use arcbufs
3900 		 * that don't match bigobj blksz to test
3901 		 * dmu_assign_arcbuf() when it can't directly
3902 		 * assign an arcbuf to a dbuf.
3903 		 */
3904 		for (j = 0; j < s; j++) {
3905 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3906 				bigbuf_arcbufs[j] =
3907 				    dmu_request_arcbuf(bonus_db, chunksize);
3908 			} else {
3909 				bigbuf_arcbufs[2 * j] =
3910 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3911 				bigbuf_arcbufs[2 * j + 1] =
3912 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3913 			}
3914 		}
3915 
3916 		/*
3917 		 * Get a tx for the mods to both packobj and bigobj.
3918 		 */
3919 		tx = dmu_tx_create(os);
3920 
3921 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3922 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3923 
3924 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3925 		if (txg == 0) {
3926 			umem_free(packbuf, packsize);
3927 			umem_free(bigbuf, bigsize);
3928 			for (j = 0; j < s; j++) {
3929 				if (i != 5 ||
3930 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
3931 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3932 				} else {
3933 					dmu_return_arcbuf(
3934 					    bigbuf_arcbufs[2 * j]);
3935 					dmu_return_arcbuf(
3936 					    bigbuf_arcbufs[2 * j + 1]);
3937 				}
3938 			}
3939 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3940 			dmu_buf_rele(bonus_db, FTAG);
3941 			return;
3942 		}
3943 
3944 		/*
3945 		 * 50% of the time don't read objects in the 1st iteration to
3946 		 * test dmu_assign_arcbuf() for the case when there're no
3947 		 * existing dbufs for the specified offsets.
3948 		 */
3949 		if (i != 0 || ztest_random(2) != 0) {
3950 			error = dmu_read(os, packobj, packoff,
3951 			    packsize, packbuf, DMU_READ_PREFETCH);
3952 			ASSERT0(error);
3953 			error = dmu_read(os, bigobj, bigoff, bigsize,
3954 			    bigbuf, DMU_READ_PREFETCH);
3955 			ASSERT0(error);
3956 		}
3957 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3958 		    n, chunksize, txg);
3959 
3960 		/*
3961 		 * We've verified all the old bufwads, and made new ones.
3962 		 * Now write them out.
3963 		 */
3964 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3965 		if (ztest_opts.zo_verbose >= 7) {
3966 			(void) printf("writing offset %llx size %llx"
3967 			    " txg %llx\n",
3968 			    (u_longlong_t)bigoff,
3969 			    (u_longlong_t)bigsize,
3970 			    (u_longlong_t)txg);
3971 		}
3972 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3973 			dmu_buf_t *dbt;
3974 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3975 				bcopy((caddr_t)bigbuf + (off - bigoff),
3976 				    bigbuf_arcbufs[j]->b_data, chunksize);
3977 			} else {
3978 				bcopy((caddr_t)bigbuf + (off - bigoff),
3979 				    bigbuf_arcbufs[2 * j]->b_data,
3980 				    chunksize / 2);
3981 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3982 				    chunksize / 2,
3983 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3984 				    chunksize / 2);
3985 			}
3986 
3987 			if (i == 1) {
3988 				VERIFY(dmu_buf_hold(os, bigobj, off,
3989 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3990 			}
3991 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3992 				dmu_assign_arcbuf(bonus_db, off,
3993 				    bigbuf_arcbufs[j], tx);
3994 			} else {
3995 				dmu_assign_arcbuf(bonus_db, off,
3996 				    bigbuf_arcbufs[2 * j], tx);
3997 				dmu_assign_arcbuf(bonus_db,
3998 				    off + chunksize / 2,
3999 				    bigbuf_arcbufs[2 * j + 1], tx);
4000 			}
4001 			if (i == 1) {
4002 				dmu_buf_rele(dbt, FTAG);
4003 			}
4004 		}
4005 		dmu_tx_commit(tx);
4006 
4007 		/*
4008 		 * Sanity check the stuff we just wrote.
4009 		 */
4010 		{
4011 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4012 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4013 
4014 			VERIFY(0 == dmu_read(os, packobj, packoff,
4015 			    packsize, packcheck, DMU_READ_PREFETCH));
4016 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4017 			    bigsize, bigcheck, DMU_READ_PREFETCH));
4018 
4019 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4020 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4021 
4022 			umem_free(packcheck, packsize);
4023 			umem_free(bigcheck, bigsize);
4024 		}
4025 		if (i == 2) {
4026 			txg_wait_open(dmu_objset_pool(os), 0);
4027 		} else if (i == 3) {
4028 			txg_wait_synced(dmu_objset_pool(os), 0);
4029 		}
4030 	}
4031 
4032 	dmu_buf_rele(bonus_db, FTAG);
4033 	umem_free(packbuf, packsize);
4034 	umem_free(bigbuf, bigsize);
4035 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4036 }
4037 
4038 /* ARGSUSED */
4039 void
4040 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4041 {
4042 	ztest_od_t od[1];
4043 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4044 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4045 
4046 	/*
4047 	 * Have multiple threads write to large offsets in an object
4048 	 * to verify that parallel writes to an object -- even to the
4049 	 * same blocks within the object -- doesn't cause any trouble.
4050 	 */
4051 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4052 
4053 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4054 		return;
4055 
4056 	while (ztest_random(10) != 0)
4057 		ztest_io(zd, od[0].od_object, offset);
4058 }
4059 
4060 void
4061 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4062 {
4063 	ztest_od_t od[1];
4064 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4065 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4066 	uint64_t count = ztest_random(20) + 1;
4067 	uint64_t blocksize = ztest_random_blocksize();
4068 	void *data;
4069 
4070 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4071 
4072 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4073 		return;
4074 
4075 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4076 		return;
4077 
4078 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4079 
4080 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4081 
4082 	while (ztest_random(count) != 0) {
4083 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4084 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4085 		    data) != 0)
4086 			break;
4087 		while (ztest_random(4) != 0)
4088 			ztest_io(zd, od[0].od_object, randoff);
4089 	}
4090 
4091 	umem_free(data, blocksize);
4092 }
4093 
4094 /*
4095  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4096  */
4097 #define	ZTEST_ZAP_MIN_INTS	1
4098 #define	ZTEST_ZAP_MAX_INTS	4
4099 #define	ZTEST_ZAP_MAX_PROPS	1000
4100 
4101 void
4102 ztest_zap(ztest_ds_t *zd, uint64_t id)
4103 {
4104 	objset_t *os = zd->zd_os;
4105 	ztest_od_t od[1];
4106 	uint64_t object;
4107 	uint64_t txg, last_txg;
4108 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4109 	uint64_t zl_ints, zl_intsize, prop;
4110 	int i, ints;
4111 	dmu_tx_t *tx;
4112 	char propname[100], txgname[100];
4113 	int error;
4114 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4115 
4116 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4117 
4118 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4119 		return;
4120 
4121 	object = od[0].od_object;
4122 
4123 	/*
4124 	 * Generate a known hash collision, and verify that
4125 	 * we can lookup and remove both entries.
4126 	 */
4127 	tx = dmu_tx_create(os);
4128 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4129 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4130 	if (txg == 0)
4131 		return;
4132 	for (i = 0; i < 2; i++) {
4133 		value[i] = i;
4134 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4135 		    1, &value[i], tx));
4136 	}
4137 	for (i = 0; i < 2; i++) {
4138 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4139 		    sizeof (uint64_t), 1, &value[i], tx));
4140 		VERIFY3U(0, ==,
4141 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4142 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4143 		ASSERT3U(zl_ints, ==, 1);
4144 	}
4145 	for (i = 0; i < 2; i++) {
4146 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4147 	}
4148 	dmu_tx_commit(tx);
4149 
4150 	/*
4151 	 * Generate a buch of random entries.
4152 	 */
4153 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4154 
4155 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4156 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4157 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4158 	bzero(value, sizeof (value));
4159 	last_txg = 0;
4160 
4161 	/*
4162 	 * If these zap entries already exist, validate their contents.
4163 	 */
4164 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4165 	if (error == 0) {
4166 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4167 		ASSERT3U(zl_ints, ==, 1);
4168 
4169 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4170 		    zl_ints, &last_txg) == 0);
4171 
4172 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4173 		    &zl_ints) == 0);
4174 
4175 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4176 		ASSERT3U(zl_ints, ==, ints);
4177 
4178 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4179 		    zl_ints, value) == 0);
4180 
4181 		for (i = 0; i < ints; i++) {
4182 			ASSERT3U(value[i], ==, last_txg + object + i);
4183 		}
4184 	} else {
4185 		ASSERT3U(error, ==, ENOENT);
4186 	}
4187 
4188 	/*
4189 	 * Atomically update two entries in our zap object.
4190 	 * The first is named txg_%llu, and contains the txg
4191 	 * in which the property was last updated.  The second
4192 	 * is named prop_%llu, and the nth element of its value
4193 	 * should be txg + object + n.
4194 	 */
4195 	tx = dmu_tx_create(os);
4196 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4197 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4198 	if (txg == 0)
4199 		return;
4200 
4201 	if (last_txg > txg)
4202 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4203 
4204 	for (i = 0; i < ints; i++)
4205 		value[i] = txg + object + i;
4206 
4207 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4208 	    1, &txg, tx));
4209 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4210 	    ints, value, tx));
4211 
4212 	dmu_tx_commit(tx);
4213 
4214 	/*
4215 	 * Remove a random pair of entries.
4216 	 */
4217 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4218 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4219 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4220 
4221 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4222 
4223 	if (error == ENOENT)
4224 		return;
4225 
4226 	ASSERT0(error);
4227 
4228 	tx = dmu_tx_create(os);
4229 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4230 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4231 	if (txg == 0)
4232 		return;
4233 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4234 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4235 	dmu_tx_commit(tx);
4236 }
4237 
4238 /*
4239  * Testcase to test the upgrading of a microzap to fatzap.
4240  */
4241 void
4242 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4243 {
4244 	objset_t *os = zd->zd_os;
4245 	ztest_od_t od[1];
4246 	uint64_t object, txg;
4247 
4248 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4249 
4250 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4251 		return;
4252 
4253 	object = od[0].od_object;
4254 
4255 	/*
4256 	 * Add entries to this ZAP and make sure it spills over
4257 	 * and gets upgraded to a fatzap. Also, since we are adding
4258 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4259 	 */
4260 	for (int i = 0; i < 2050; i++) {
4261 		char name[ZFS_MAX_DATASET_NAME_LEN];
4262 		uint64_t value = i;
4263 		dmu_tx_t *tx;
4264 		int error;
4265 
4266 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4267 		    id, value);
4268 
4269 		tx = dmu_tx_create(os);
4270 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4271 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4272 		if (txg == 0)
4273 			return;
4274 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4275 		    &value, tx);
4276 		ASSERT(error == 0 || error == EEXIST);
4277 		dmu_tx_commit(tx);
4278 	}
4279 }
4280 
4281 /* ARGSUSED */
4282 void
4283 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4284 {
4285 	objset_t *os = zd->zd_os;
4286 	ztest_od_t od[1];
4287 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4288 	dmu_tx_t *tx;
4289 	int i, namelen, error;
4290 	int micro = ztest_random(2);
4291 	char name[20], string_value[20];
4292 	void *data;
4293 
4294 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4295 
4296 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4297 		return;
4298 
4299 	object = od[0].od_object;
4300 
4301 	/*
4302 	 * Generate a random name of the form 'xxx.....' where each
4303 	 * x is a random printable character and the dots are dots.
4304 	 * There are 94 such characters, and the name length goes from
4305 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4306 	 */
4307 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4308 
4309 	for (i = 0; i < 3; i++)
4310 		name[i] = '!' + ztest_random('~' - '!' + 1);
4311 	for (; i < namelen - 1; i++)
4312 		name[i] = '.';
4313 	name[i] = '\0';
4314 
4315 	if ((namelen & 1) || micro) {
4316 		wsize = sizeof (txg);
4317 		wc = 1;
4318 		data = &txg;
4319 	} else {
4320 		wsize = 1;
4321 		wc = namelen;
4322 		data = string_value;
4323 	}
4324 
4325 	count = -1ULL;
4326 	VERIFY0(zap_count(os, object, &count));
4327 	ASSERT(count != -1ULL);
4328 
4329 	/*
4330 	 * Select an operation: length, lookup, add, update, remove.
4331 	 */
4332 	i = ztest_random(5);
4333 
4334 	if (i >= 2) {
4335 		tx = dmu_tx_create(os);
4336 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4337 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4338 		if (txg == 0)
4339 			return;
4340 		bcopy(name, string_value, namelen);
4341 	} else {
4342 		tx = NULL;
4343 		txg = 0;
4344 		bzero(string_value, namelen);
4345 	}
4346 
4347 	switch (i) {
4348 
4349 	case 0:
4350 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4351 		if (error == 0) {
4352 			ASSERT3U(wsize, ==, zl_wsize);
4353 			ASSERT3U(wc, ==, zl_wc);
4354 		} else {
4355 			ASSERT3U(error, ==, ENOENT);
4356 		}
4357 		break;
4358 
4359 	case 1:
4360 		error = zap_lookup(os, object, name, wsize, wc, data);
4361 		if (error == 0) {
4362 			if (data == string_value &&
4363 			    bcmp(name, data, namelen) != 0)
4364 				fatal(0, "name '%s' != val '%s' len %d",
4365 				    name, data, namelen);
4366 		} else {
4367 			ASSERT3U(error, ==, ENOENT);
4368 		}
4369 		break;
4370 
4371 	case 2:
4372 		error = zap_add(os, object, name, wsize, wc, data, tx);
4373 		ASSERT(error == 0 || error == EEXIST);
4374 		break;
4375 
4376 	case 3:
4377 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4378 		break;
4379 
4380 	case 4:
4381 		error = zap_remove(os, object, name, tx);
4382 		ASSERT(error == 0 || error == ENOENT);
4383 		break;
4384 	}
4385 
4386 	if (tx != NULL)
4387 		dmu_tx_commit(tx);
4388 }
4389 
4390 /*
4391  * Commit callback data.
4392  */
4393 typedef struct ztest_cb_data {
4394 	list_node_t		zcd_node;
4395 	uint64_t		zcd_txg;
4396 	int			zcd_expected_err;
4397 	boolean_t		zcd_added;
4398 	boolean_t		zcd_called;
4399 	spa_t			*zcd_spa;
4400 } ztest_cb_data_t;
4401 
4402 /* This is the actual commit callback function */
4403 static void
4404 ztest_commit_callback(void *arg, int error)
4405 {
4406 	ztest_cb_data_t *data = arg;
4407 	uint64_t synced_txg;
4408 
4409 	VERIFY(data != NULL);
4410 	VERIFY3S(data->zcd_expected_err, ==, error);
4411 	VERIFY(!data->zcd_called);
4412 
4413 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4414 	if (data->zcd_txg > synced_txg)
4415 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4416 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4417 		    synced_txg);
4418 
4419 	data->zcd_called = B_TRUE;
4420 
4421 	if (error == ECANCELED) {
4422 		ASSERT0(data->zcd_txg);
4423 		ASSERT(!data->zcd_added);
4424 
4425 		/*
4426 		 * The private callback data should be destroyed here, but
4427 		 * since we are going to check the zcd_called field after
4428 		 * dmu_tx_abort(), we will destroy it there.
4429 		 */
4430 		return;
4431 	}
4432 
4433 	/* Was this callback added to the global callback list? */
4434 	if (!data->zcd_added)
4435 		goto out;
4436 
4437 	ASSERT3U(data->zcd_txg, !=, 0);
4438 
4439 	/* Remove our callback from the list */
4440 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4441 	list_remove(&zcl.zcl_callbacks, data);
4442 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4443 
4444 out:
4445 	umem_free(data, sizeof (ztest_cb_data_t));
4446 }
4447 
4448 /* Allocate and initialize callback data structure */
4449 static ztest_cb_data_t *
4450 ztest_create_cb_data(objset_t *os, uint64_t txg)
4451 {
4452 	ztest_cb_data_t *cb_data;
4453 
4454 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4455 
4456 	cb_data->zcd_txg = txg;
4457 	cb_data->zcd_spa = dmu_objset_spa(os);
4458 
4459 	return (cb_data);
4460 }
4461 
4462 /*
4463  * If a number of txgs equal to this threshold have been created after a commit
4464  * callback has been registered but not called, then we assume there is an
4465  * implementation bug.
4466  */
4467 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4468 
4469 /*
4470  * Commit callback test.
4471  */
4472 void
4473 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4474 {
4475 	objset_t *os = zd->zd_os;
4476 	ztest_od_t od[1];
4477 	dmu_tx_t *tx;
4478 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4479 	uint64_t old_txg, txg;
4480 	int i, error;
4481 
4482 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4483 
4484 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4485 		return;
4486 
4487 	tx = dmu_tx_create(os);
4488 
4489 	cb_data[0] = ztest_create_cb_data(os, 0);
4490 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4491 
4492 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4493 
4494 	/* Every once in a while, abort the transaction on purpose */
4495 	if (ztest_random(100) == 0)
4496 		error = -1;
4497 
4498 	if (!error)
4499 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4500 
4501 	txg = error ? 0 : dmu_tx_get_txg(tx);
4502 
4503 	cb_data[0]->zcd_txg = txg;
4504 	cb_data[1] = ztest_create_cb_data(os, txg);
4505 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4506 
4507 	if (error) {
4508 		/*
4509 		 * It's not a strict requirement to call the registered
4510 		 * callbacks from inside dmu_tx_abort(), but that's what
4511 		 * it's supposed to happen in the current implementation
4512 		 * so we will check for that.
4513 		 */
4514 		for (i = 0; i < 2; i++) {
4515 			cb_data[i]->zcd_expected_err = ECANCELED;
4516 			VERIFY(!cb_data[i]->zcd_called);
4517 		}
4518 
4519 		dmu_tx_abort(tx);
4520 
4521 		for (i = 0; i < 2; i++) {
4522 			VERIFY(cb_data[i]->zcd_called);
4523 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4524 		}
4525 
4526 		return;
4527 	}
4528 
4529 	cb_data[2] = ztest_create_cb_data(os, txg);
4530 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4531 
4532 	/*
4533 	 * Read existing data to make sure there isn't a future leak.
4534 	 */
4535 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4536 	    &old_txg, DMU_READ_PREFETCH));
4537 
4538 	if (old_txg > txg)
4539 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4540 		    old_txg, txg);
4541 
4542 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4543 
4544 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4545 
4546 	/*
4547 	 * Since commit callbacks don't have any ordering requirement and since
4548 	 * it is theoretically possible for a commit callback to be called
4549 	 * after an arbitrary amount of time has elapsed since its txg has been
4550 	 * synced, it is difficult to reliably determine whether a commit
4551 	 * callback hasn't been called due to high load or due to a flawed
4552 	 * implementation.
4553 	 *
4554 	 * In practice, we will assume that if after a certain number of txgs a
4555 	 * commit callback hasn't been called, then most likely there's an
4556 	 * implementation bug..
4557 	 */
4558 	tmp_cb = list_head(&zcl.zcl_callbacks);
4559 	if (tmp_cb != NULL &&
4560 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4561 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4562 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4563 	}
4564 
4565 	/*
4566 	 * Let's find the place to insert our callbacks.
4567 	 *
4568 	 * Even though the list is ordered by txg, it is possible for the
4569 	 * insertion point to not be the end because our txg may already be
4570 	 * quiescing at this point and other callbacks in the open txg
4571 	 * (from other objsets) may have sneaked in.
4572 	 */
4573 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4574 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4575 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4576 
4577 	/* Add the 3 callbacks to the list */
4578 	for (i = 0; i < 3; i++) {
4579 		if (tmp_cb == NULL)
4580 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4581 		else
4582 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4583 			    cb_data[i]);
4584 
4585 		cb_data[i]->zcd_added = B_TRUE;
4586 		VERIFY(!cb_data[i]->zcd_called);
4587 
4588 		tmp_cb = cb_data[i];
4589 	}
4590 
4591 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4592 
4593 	dmu_tx_commit(tx);
4594 }
4595 
4596 /* ARGSUSED */
4597 void
4598 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4599 {
4600 	zfs_prop_t proplist[] = {
4601 		ZFS_PROP_CHECKSUM,
4602 		ZFS_PROP_COMPRESSION,
4603 		ZFS_PROP_COPIES,
4604 		ZFS_PROP_DEDUP
4605 	};
4606 
4607 	(void) rw_rdlock(&ztest_name_lock);
4608 
4609 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4610 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4611 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4612 
4613 	(void) rw_unlock(&ztest_name_lock);
4614 }
4615 
4616 /* ARGSUSED */
4617 void
4618 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4619 {
4620 	nvlist_t *props = NULL;
4621 
4622 	(void) rw_rdlock(&ztest_name_lock);
4623 
4624 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4625 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4626 
4627 	VERIFY0(spa_prop_get(ztest_spa, &props));
4628 
4629 	if (ztest_opts.zo_verbose >= 6)
4630 		dump_nvlist(props, 4);
4631 
4632 	nvlist_free(props);
4633 
4634 	(void) rw_unlock(&ztest_name_lock);
4635 }
4636 
4637 static int
4638 user_release_one(const char *snapname, const char *holdname)
4639 {
4640 	nvlist_t *snaps, *holds;
4641 	int error;
4642 
4643 	snaps = fnvlist_alloc();
4644 	holds = fnvlist_alloc();
4645 	fnvlist_add_boolean(holds, holdname);
4646 	fnvlist_add_nvlist(snaps, snapname, holds);
4647 	fnvlist_free(holds);
4648 	error = dsl_dataset_user_release(snaps, NULL);
4649 	fnvlist_free(snaps);
4650 	return (error);
4651 }
4652 
4653 /*
4654  * Test snapshot hold/release and deferred destroy.
4655  */
4656 void
4657 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4658 {
4659 	int error;
4660 	objset_t *os = zd->zd_os;
4661 	objset_t *origin;
4662 	char snapname[100];
4663 	char fullname[100];
4664 	char clonename[100];
4665 	char tag[100];
4666 	char osname[ZFS_MAX_DATASET_NAME_LEN];
4667 	nvlist_t *holds;
4668 
4669 	(void) rw_rdlock(&ztest_name_lock);
4670 
4671 	dmu_objset_name(os, osname);
4672 
4673 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4674 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4675 	(void) snprintf(clonename, sizeof (clonename),
4676 	    "%s/ch1_%llu", osname, id);
4677 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4678 
4679 	/*
4680 	 * Clean up from any previous run.
4681 	 */
4682 	error = dsl_destroy_head(clonename);
4683 	if (error != ENOENT)
4684 		ASSERT0(error);
4685 	error = user_release_one(fullname, tag);
4686 	if (error != ESRCH && error != ENOENT)
4687 		ASSERT0(error);
4688 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4689 	if (error != ENOENT)
4690 		ASSERT0(error);
4691 
4692 	/*
4693 	 * Create snapshot, clone it, mark snap for deferred destroy,
4694 	 * destroy clone, verify snap was also destroyed.
4695 	 */
4696 	error = dmu_objset_snapshot_one(osname, snapname);
4697 	if (error) {
4698 		if (error == ENOSPC) {
4699 			ztest_record_enospc("dmu_objset_snapshot");
4700 			goto out;
4701 		}
4702 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4703 	}
4704 
4705 	error = dmu_objset_clone(clonename, fullname);
4706 	if (error) {
4707 		if (error == ENOSPC) {
4708 			ztest_record_enospc("dmu_objset_clone");
4709 			goto out;
4710 		}
4711 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4712 	}
4713 
4714 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4715 	if (error) {
4716 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4717 		    fullname, error);
4718 	}
4719 
4720 	error = dsl_destroy_head(clonename);
4721 	if (error)
4722 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4723 
4724 	error = dmu_objset_hold(fullname, FTAG, &origin);
4725 	if (error != ENOENT)
4726 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4727 
4728 	/*
4729 	 * Create snapshot, add temporary hold, verify that we can't
4730 	 * destroy a held snapshot, mark for deferred destroy,
4731 	 * release hold, verify snapshot was destroyed.
4732 	 */
4733 	error = dmu_objset_snapshot_one(osname, snapname);
4734 	if (error) {
4735 		if (error == ENOSPC) {
4736 			ztest_record_enospc("dmu_objset_snapshot");
4737 			goto out;
4738 		}
4739 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4740 	}
4741 
4742 	holds = fnvlist_alloc();
4743 	fnvlist_add_string(holds, fullname, tag);
4744 	error = dsl_dataset_user_hold(holds, 0, NULL);
4745 	fnvlist_free(holds);
4746 
4747 	if (error == ENOSPC) {
4748 		ztest_record_enospc("dsl_dataset_user_hold");
4749 		goto out;
4750 	} else if (error) {
4751 		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4752 		    fullname, tag, error);
4753 	}
4754 
4755 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4756 	if (error != EBUSY) {
4757 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4758 		    fullname, error);
4759 	}
4760 
4761 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4762 	if (error) {
4763 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4764 		    fullname, error);
4765 	}
4766 
4767 	error = user_release_one(fullname, tag);
4768 	if (error)
4769 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4770 
4771 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4772 
4773 out:
4774 	(void) rw_unlock(&ztest_name_lock);
4775 }
4776 
4777 /*
4778  * Inject random faults into the on-disk data.
4779  */
4780 /* ARGSUSED */
4781 void
4782 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4783 {
4784 	ztest_shared_t *zs = ztest_shared;
4785 	spa_t *spa = ztest_spa;
4786 	int fd;
4787 	uint64_t offset;
4788 	uint64_t leaves;
4789 	uint64_t bad = 0x1990c0ffeedecade;
4790 	uint64_t top, leaf;
4791 	char path0[MAXPATHLEN];
4792 	char pathrand[MAXPATHLEN];
4793 	size_t fsize;
4794 	int bshift = SPA_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
4795 	int iters = 1000;
4796 	int maxfaults;
4797 	int mirror_save;
4798 	vdev_t *vd0 = NULL;
4799 	uint64_t guid0 = 0;
4800 	boolean_t islog = B_FALSE;
4801 
4802 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4803 	maxfaults = MAXFAULTS();
4804 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4805 	mirror_save = zs->zs_mirrors;
4806 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4807 
4808 	ASSERT(leaves >= 1);
4809 
4810 	/*
4811 	 * Grab the name lock as reader. There are some operations
4812 	 * which don't like to have their vdevs changed while
4813 	 * they are in progress (i.e. spa_change_guid). Those
4814 	 * operations will have grabbed the name lock as writer.
4815 	 */
4816 	(void) rw_rdlock(&ztest_name_lock);
4817 
4818 	/*
4819 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4820 	 */
4821 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4822 
4823 	if (ztest_random(2) == 0) {
4824 		/*
4825 		 * Inject errors on a normal data device or slog device.
4826 		 */
4827 		top = ztest_random_vdev_top(spa, B_TRUE);
4828 		leaf = ztest_random(leaves) + zs->zs_splits;
4829 
4830 		/*
4831 		 * Generate paths to the first leaf in this top-level vdev,
4832 		 * and to the random leaf we selected.  We'll induce transient
4833 		 * write failures and random online/offline activity on leaf 0,
4834 		 * and we'll write random garbage to the randomly chosen leaf.
4835 		 */
4836 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4837 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4838 		    top * leaves + zs->zs_splits);
4839 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4840 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4841 		    top * leaves + leaf);
4842 
4843 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4844 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4845 			islog = B_TRUE;
4846 
4847 		/*
4848 		 * If the top-level vdev needs to be resilvered
4849 		 * then we only allow faults on the device that is
4850 		 * resilvering.
4851 		 */
4852 		if (vd0 != NULL && maxfaults != 1 &&
4853 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4854 		    vd0->vdev_resilver_txg != 0)) {
4855 			/*
4856 			 * Make vd0 explicitly claim to be unreadable,
4857 			 * or unwriteable, or reach behind its back
4858 			 * and close the underlying fd.  We can do this if
4859 			 * maxfaults == 0 because we'll fail and reexecute,
4860 			 * and we can do it if maxfaults >= 2 because we'll
4861 			 * have enough redundancy.  If maxfaults == 1, the
4862 			 * combination of this with injection of random data
4863 			 * corruption below exceeds the pool's fault tolerance.
4864 			 */
4865 			vdev_file_t *vf = vd0->vdev_tsd;
4866 
4867 			if (vf != NULL && ztest_random(3) == 0) {
4868 				(void) close(vf->vf_vnode->v_fd);
4869 				vf->vf_vnode->v_fd = -1;
4870 			} else if (ztest_random(2) == 0) {
4871 				vd0->vdev_cant_read = B_TRUE;
4872 			} else {
4873 				vd0->vdev_cant_write = B_TRUE;
4874 			}
4875 			guid0 = vd0->vdev_guid;
4876 		}
4877 	} else {
4878 		/*
4879 		 * Inject errors on an l2cache device.
4880 		 */
4881 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4882 
4883 		if (sav->sav_count == 0) {
4884 			spa_config_exit(spa, SCL_STATE, FTAG);
4885 			(void) rw_unlock(&ztest_name_lock);
4886 			return;
4887 		}
4888 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4889 		guid0 = vd0->vdev_guid;
4890 		(void) strcpy(path0, vd0->vdev_path);
4891 		(void) strcpy(pathrand, vd0->vdev_path);
4892 
4893 		leaf = 0;
4894 		leaves = 1;
4895 		maxfaults = INT_MAX;	/* no limit on cache devices */
4896 	}
4897 
4898 	spa_config_exit(spa, SCL_STATE, FTAG);
4899 	(void) rw_unlock(&ztest_name_lock);
4900 
4901 	/*
4902 	 * If we can tolerate two or more faults, or we're dealing
4903 	 * with a slog, randomly online/offline vd0.
4904 	 */
4905 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4906 		if (ztest_random(10) < 6) {
4907 			int flags = (ztest_random(2) == 0 ?
4908 			    ZFS_OFFLINE_TEMPORARY : 0);
4909 
4910 			/*
4911 			 * We have to grab the zs_name_lock as writer to
4912 			 * prevent a race between offlining a slog and
4913 			 * destroying a dataset. Offlining the slog will
4914 			 * grab a reference on the dataset which may cause
4915 			 * dmu_objset_destroy() to fail with EBUSY thus
4916 			 * leaving the dataset in an inconsistent state.
4917 			 */
4918 			if (islog)
4919 				(void) rw_wrlock(&ztest_name_lock);
4920 
4921 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4922 
4923 			if (islog)
4924 				(void) rw_unlock(&ztest_name_lock);
4925 		} else {
4926 			/*
4927 			 * Ideally we would like to be able to randomly
4928 			 * call vdev_[on|off]line without holding locks
4929 			 * to force unpredictable failures but the side
4930 			 * effects of vdev_[on|off]line prevent us from
4931 			 * doing so. We grab the ztest_vdev_lock here to
4932 			 * prevent a race between injection testing and
4933 			 * aux_vdev removal.
4934 			 */
4935 			VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4936 			(void) vdev_online(spa, guid0, 0, NULL);
4937 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4938 		}
4939 	}
4940 
4941 	if (maxfaults == 0)
4942 		return;
4943 
4944 	/*
4945 	 * We have at least single-fault tolerance, so inject data corruption.
4946 	 */
4947 	fd = open(pathrand, O_RDWR);
4948 
4949 	if (fd == -1)	/* we hit a gap in the device namespace */
4950 		return;
4951 
4952 	fsize = lseek(fd, 0, SEEK_END);
4953 
4954 	while (--iters != 0) {
4955 		/*
4956 		 * The offset must be chosen carefully to ensure that
4957 		 * we do not inject a given logical block with errors
4958 		 * on two different leaf devices, because ZFS can not
4959 		 * tolerate that (if maxfaults==1).
4960 		 *
4961 		 * We divide each leaf into chunks of size
4962 		 * (# leaves * SPA_MAXBLOCKSIZE * 4).  Within each chunk
4963 		 * there is a series of ranges to which we can inject errors.
4964 		 * Each range can accept errors on only a single leaf vdev.
4965 		 * The error injection ranges are separated by ranges
4966 		 * which we will not inject errors on any device (DMZs).
4967 		 * Each DMZ must be large enough such that a single block
4968 		 * can not straddle it, so that a single block can not be
4969 		 * a target in two different injection ranges (on different
4970 		 * leaf vdevs).
4971 		 *
4972 		 * For example, with 3 leaves, each chunk looks like:
4973 		 *    0 to  32M: injection range for leaf 0
4974 		 *  32M to  64M: DMZ - no injection allowed
4975 		 *  64M to  96M: injection range for leaf 1
4976 		 *  96M to 128M: DMZ - no injection allowed
4977 		 * 128M to 160M: injection range for leaf 2
4978 		 * 160M to 192M: DMZ - no injection allowed
4979 		 */
4980 		offset = ztest_random(fsize / (leaves << bshift)) *
4981 		    (leaves << bshift) + (leaf << bshift) +
4982 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4983 
4984 		if (offset >= fsize)
4985 			continue;
4986 
4987 		VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4988 		if (mirror_save != zs->zs_mirrors) {
4989 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4990 			(void) close(fd);
4991 			return;
4992 		}
4993 
4994 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4995 			fatal(1, "can't inject bad word at 0x%llx in %s",
4996 			    offset, pathrand);
4997 
4998 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4999 
5000 		if (ztest_opts.zo_verbose >= 7)
5001 			(void) printf("injected bad word into %s,"
5002 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5003 	}
5004 
5005 	(void) close(fd);
5006 }
5007 
5008 /*
5009  * Verify that DDT repair works as expected.
5010  */
5011 void
5012 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5013 {
5014 	ztest_shared_t *zs = ztest_shared;
5015 	spa_t *spa = ztest_spa;
5016 	objset_t *os = zd->zd_os;
5017 	ztest_od_t od[1];
5018 	uint64_t object, blocksize, txg, pattern, psize;
5019 	enum zio_checksum checksum = spa_dedup_checksum(spa);
5020 	dmu_buf_t *db;
5021 	dmu_tx_t *tx;
5022 	void *buf;
5023 	blkptr_t blk;
5024 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
5025 
5026 	blocksize = ztest_random_blocksize();
5027 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
5028 
5029 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5030 
5031 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5032 		return;
5033 
5034 	/*
5035 	 * Take the name lock as writer to prevent anyone else from changing
5036 	 * the pool and dataset properies we need to maintain during this test.
5037 	 */
5038 	(void) rw_wrlock(&ztest_name_lock);
5039 
5040 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5041 	    B_FALSE) != 0 ||
5042 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5043 	    B_FALSE) != 0) {
5044 		(void) rw_unlock(&ztest_name_lock);
5045 		return;
5046 	}
5047 
5048 	object = od[0].od_object;
5049 	blocksize = od[0].od_blocksize;
5050 	pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5051 
5052 	ASSERT(object != 0);
5053 
5054 	tx = dmu_tx_create(os);
5055 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5056 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5057 	if (txg == 0) {
5058 		(void) rw_unlock(&ztest_name_lock);
5059 		return;
5060 	}
5061 
5062 	/*
5063 	 * Write all the copies of our block.
5064 	 */
5065 	for (int i = 0; i < copies; i++) {
5066 		uint64_t offset = i * blocksize;
5067 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5068 		    DMU_READ_NO_PREFETCH);
5069 		if (error != 0) {
5070 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5071 			    os, (long long)object, (long long) offset, error);
5072 		}
5073 		ASSERT(db->db_offset == offset);
5074 		ASSERT(db->db_size == blocksize);
5075 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5076 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5077 		dmu_buf_will_fill(db, tx);
5078 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5079 		dmu_buf_rele(db, FTAG);
5080 	}
5081 
5082 	dmu_tx_commit(tx);
5083 	txg_wait_synced(spa_get_dsl(spa), txg);
5084 
5085 	/*
5086 	 * Find out what block we got.
5087 	 */
5088 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5089 	    DMU_READ_NO_PREFETCH));
5090 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5091 	dmu_buf_rele(db, FTAG);
5092 
5093 	/*
5094 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5095 	 */
5096 	psize = BP_GET_PSIZE(&blk);
5097 	buf = zio_buf_alloc(psize);
5098 	ztest_pattern_set(buf, psize, ~pattern);
5099 
5100 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5101 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5102 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5103 
5104 	zio_buf_free(buf, psize);
5105 
5106 	(void) rw_unlock(&ztest_name_lock);
5107 }
5108 
5109 /*
5110  * Scrub the pool.
5111  */
5112 /* ARGSUSED */
5113 void
5114 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5115 {
5116 	spa_t *spa = ztest_spa;
5117 
5118 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5119 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5120 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5121 }
5122 
5123 /*
5124  * Change the guid for the pool.
5125  */
5126 /* ARGSUSED */
5127 void
5128 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5129 {
5130 	spa_t *spa = ztest_spa;
5131 	uint64_t orig, load;
5132 	int error;
5133 
5134 	orig = spa_guid(spa);
5135 	load = spa_load_guid(spa);
5136 
5137 	(void) rw_wrlock(&ztest_name_lock);
5138 	error = spa_change_guid(spa);
5139 	(void) rw_unlock(&ztest_name_lock);
5140 
5141 	if (error != 0)
5142 		return;
5143 
5144 	if (ztest_opts.zo_verbose >= 4) {
5145 		(void) printf("Changed guid old %llu -> %llu\n",
5146 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5147 	}
5148 
5149 	VERIFY3U(orig, !=, spa_guid(spa));
5150 	VERIFY3U(load, ==, spa_load_guid(spa));
5151 }
5152 
5153 /*
5154  * Rename the pool to a different name and then rename it back.
5155  */
5156 /* ARGSUSED */
5157 void
5158 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5159 {
5160 	char *oldname, *newname;
5161 	spa_t *spa;
5162 
5163 	(void) rw_wrlock(&ztest_name_lock);
5164 
5165 	oldname = ztest_opts.zo_pool;
5166 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5167 	(void) strcpy(newname, oldname);
5168 	(void) strcat(newname, "_tmp");
5169 
5170 	/*
5171 	 * Do the rename
5172 	 */
5173 	VERIFY3U(0, ==, spa_rename(oldname, newname));
5174 
5175 	/*
5176 	 * Try to open it under the old name, which shouldn't exist
5177 	 */
5178 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5179 
5180 	/*
5181 	 * Open it under the new name and make sure it's still the same spa_t.
5182 	 */
5183 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5184 
5185 	ASSERT(spa == ztest_spa);
5186 	spa_close(spa, FTAG);
5187 
5188 	/*
5189 	 * Rename it back to the original
5190 	 */
5191 	VERIFY3U(0, ==, spa_rename(newname, oldname));
5192 
5193 	/*
5194 	 * Make sure it can still be opened
5195 	 */
5196 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5197 
5198 	ASSERT(spa == ztest_spa);
5199 	spa_close(spa, FTAG);
5200 
5201 	umem_free(newname, strlen(newname) + 1);
5202 
5203 	(void) rw_unlock(&ztest_name_lock);
5204 }
5205 
5206 /*
5207  * Verify pool integrity by running zdb.
5208  */
5209 static void
5210 ztest_run_zdb(char *pool)
5211 {
5212 	int status;
5213 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5214 	char zbuf[1024];
5215 	char *bin;
5216 	char *ztest;
5217 	char *isa;
5218 	int isalen;
5219 	FILE *fp;
5220 
5221 	(void) realpath(getexecname(), zdb);
5222 
5223 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5224 	bin = strstr(zdb, "/usr/bin/");
5225 	ztest = strstr(bin, "/ztest");
5226 	isa = bin + 8;
5227 	isalen = ztest - isa;
5228 	isa = strdup(isa);
5229 	/* LINTED */
5230 	(void) sprintf(bin,
5231 	    "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5232 	    isalen,
5233 	    isa,
5234 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5235 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5236 	    spa_config_path,
5237 	    pool);
5238 	free(isa);
5239 
5240 	if (ztest_opts.zo_verbose >= 5)
5241 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5242 
5243 	fp = popen(zdb, "r");
5244 
5245 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5246 		if (ztest_opts.zo_verbose >= 3)
5247 			(void) printf("%s", zbuf);
5248 
5249 	status = pclose(fp);
5250 
5251 	if (status == 0)
5252 		return;
5253 
5254 	ztest_dump_core = 0;
5255 	if (WIFEXITED(status))
5256 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5257 	else
5258 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5259 }
5260 
5261 static void
5262 ztest_walk_pool_directory(char *header)
5263 {
5264 	spa_t *spa = NULL;
5265 
5266 	if (ztest_opts.zo_verbose >= 6)
5267 		(void) printf("%s\n", header);
5268 
5269 	mutex_enter(&spa_namespace_lock);
5270 	while ((spa = spa_next(spa)) != NULL)
5271 		if (ztest_opts.zo_verbose >= 6)
5272 			(void) printf("\t%s\n", spa_name(spa));
5273 	mutex_exit(&spa_namespace_lock);
5274 }
5275 
5276 static void
5277 ztest_spa_import_export(char *oldname, char *newname)
5278 {
5279 	nvlist_t *config, *newconfig;
5280 	uint64_t pool_guid;
5281 	spa_t *spa;
5282 	int error;
5283 
5284 	if (ztest_opts.zo_verbose >= 4) {
5285 		(void) printf("import/export: old = %s, new = %s\n",
5286 		    oldname, newname);
5287 	}
5288 
5289 	/*
5290 	 * Clean up from previous runs.
5291 	 */
5292 	(void) spa_destroy(newname);
5293 
5294 	/*
5295 	 * Get the pool's configuration and guid.
5296 	 */
5297 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5298 
5299 	/*
5300 	 * Kick off a scrub to tickle scrub/export races.
5301 	 */
5302 	if (ztest_random(2) == 0)
5303 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5304 
5305 	pool_guid = spa_guid(spa);
5306 	spa_close(spa, FTAG);
5307 
5308 	ztest_walk_pool_directory("pools before export");
5309 
5310 	/*
5311 	 * Export it.
5312 	 */
5313 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5314 
5315 	ztest_walk_pool_directory("pools after export");
5316 
5317 	/*
5318 	 * Try to import it.
5319 	 */
5320 	newconfig = spa_tryimport(config);
5321 	ASSERT(newconfig != NULL);
5322 	nvlist_free(newconfig);
5323 
5324 	/*
5325 	 * Import it under the new name.
5326 	 */
5327 	error = spa_import(newname, config, NULL, 0);
5328 	if (error != 0) {
5329 		dump_nvlist(config, 0);
5330 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5331 		    oldname, newname, error);
5332 	}
5333 
5334 	ztest_walk_pool_directory("pools after import");
5335 
5336 	/*
5337 	 * Try to import it again -- should fail with EEXIST.
5338 	 */
5339 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5340 
5341 	/*
5342 	 * Try to import it under a different name -- should fail with EEXIST.
5343 	 */
5344 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5345 
5346 	/*
5347 	 * Verify that the pool is no longer visible under the old name.
5348 	 */
5349 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5350 
5351 	/*
5352 	 * Verify that we can open and close the pool using the new name.
5353 	 */
5354 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5355 	ASSERT(pool_guid == spa_guid(spa));
5356 	spa_close(spa, FTAG);
5357 
5358 	nvlist_free(config);
5359 }
5360 
5361 static void
5362 ztest_resume(spa_t *spa)
5363 {
5364 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5365 		(void) printf("resuming from suspended state\n");
5366 	spa_vdev_state_enter(spa, SCL_NONE);
5367 	vdev_clear(spa, NULL);
5368 	(void) spa_vdev_state_exit(spa, NULL, 0);
5369 	(void) zio_resume(spa);
5370 }
5371 
5372 static void *
5373 ztest_resume_thread(void *arg)
5374 {
5375 	spa_t *spa = arg;
5376 
5377 	while (!ztest_exiting) {
5378 		if (spa_suspended(spa))
5379 			ztest_resume(spa);
5380 		(void) poll(NULL, 0, 100);
5381 	}
5382 	return (NULL);
5383 }
5384 
5385 static void *
5386 ztest_deadman_thread(void *arg)
5387 {
5388 	ztest_shared_t *zs = arg;
5389 	spa_t *spa = ztest_spa;
5390 	hrtime_t delta, total = 0;
5391 
5392 	for (;;) {
5393 		delta = zs->zs_thread_stop - zs->zs_thread_start +
5394 		    MSEC2NSEC(zfs_deadman_synctime_ms);
5395 
5396 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5397 
5398 		/*
5399 		 * If the pool is suspended then fail immediately. Otherwise,
5400 		 * check to see if the pool is making any progress. If
5401 		 * vdev_deadman() discovers that there hasn't been any recent
5402 		 * I/Os then it will end up aborting the tests.
5403 		 */
5404 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5405 			fatal(0, "aborting test after %llu seconds because "
5406 			    "pool has transitioned to a suspended state.",
5407 			    zfs_deadman_synctime_ms / 1000);
5408 			return (NULL);
5409 		}
5410 		vdev_deadman(spa->spa_root_vdev);
5411 
5412 		total += zfs_deadman_synctime_ms/1000;
5413 		(void) printf("ztest has been running for %lld seconds\n",
5414 		    total);
5415 	}
5416 }
5417 
5418 static void
5419 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5420 {
5421 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5422 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5423 	hrtime_t functime = gethrtime();
5424 
5425 	for (int i = 0; i < zi->zi_iters; i++)
5426 		zi->zi_func(zd, id);
5427 
5428 	functime = gethrtime() - functime;
5429 
5430 	atomic_add_64(&zc->zc_count, 1);
5431 	atomic_add_64(&zc->zc_time, functime);
5432 
5433 	if (ztest_opts.zo_verbose >= 4) {
5434 		Dl_info dli;
5435 		(void) dladdr((void *)zi->zi_func, &dli);
5436 		(void) printf("%6.2f sec in %s\n",
5437 		    (double)functime / NANOSEC, dli.dli_sname);
5438 	}
5439 }
5440 
5441 static void *
5442 ztest_thread(void *arg)
5443 {
5444 	int rand;
5445 	uint64_t id = (uintptr_t)arg;
5446 	ztest_shared_t *zs = ztest_shared;
5447 	uint64_t call_next;
5448 	hrtime_t now;
5449 	ztest_info_t *zi;
5450 	ztest_shared_callstate_t *zc;
5451 
5452 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5453 		/*
5454 		 * See if it's time to force a crash.
5455 		 */
5456 		if (now > zs->zs_thread_kill)
5457 			ztest_kill(zs);
5458 
5459 		/*
5460 		 * If we're getting ENOSPC with some regularity, stop.
5461 		 */
5462 		if (zs->zs_enospc_count > 10)
5463 			break;
5464 
5465 		/*
5466 		 * Pick a random function to execute.
5467 		 */
5468 		rand = ztest_random(ZTEST_FUNCS);
5469 		zi = &ztest_info[rand];
5470 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5471 		call_next = zc->zc_next;
5472 
5473 		if (now >= call_next &&
5474 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5475 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5476 			ztest_execute(rand, zi, id);
5477 		}
5478 	}
5479 
5480 	return (NULL);
5481 }
5482 
5483 static void
5484 ztest_dataset_name(char *dsname, char *pool, int d)
5485 {
5486 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5487 }
5488 
5489 static void
5490 ztest_dataset_destroy(int d)
5491 {
5492 	char name[ZFS_MAX_DATASET_NAME_LEN];
5493 
5494 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5495 
5496 	if (ztest_opts.zo_verbose >= 3)
5497 		(void) printf("Destroying %s to free up space\n", name);
5498 
5499 	/*
5500 	 * Cleanup any non-standard clones and snapshots.  In general,
5501 	 * ztest thread t operates on dataset (t % zopt_datasets),
5502 	 * so there may be more than one thing to clean up.
5503 	 */
5504 	for (int t = d; t < ztest_opts.zo_threads;
5505 	    t += ztest_opts.zo_datasets) {
5506 		ztest_dsl_dataset_cleanup(name, t);
5507 	}
5508 
5509 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5510 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5511 }
5512 
5513 static void
5514 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5515 {
5516 	uint64_t usedobjs, dirobjs, scratch;
5517 
5518 	/*
5519 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5520 	 * Therefore, the number of objects in use should equal the
5521 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5522 	 * If not, we have an object leak.
5523 	 *
5524 	 * Note that we can only check this in ztest_dataset_open(),
5525 	 * when the open-context and syncing-context values agree.
5526 	 * That's because zap_count() returns the open-context value,
5527 	 * while dmu_objset_space() returns the rootbp fill count.
5528 	 */
5529 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5530 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5531 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5532 }
5533 
5534 static int
5535 ztest_dataset_open(int d)
5536 {
5537 	ztest_ds_t *zd = &ztest_ds[d];
5538 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5539 	objset_t *os;
5540 	zilog_t *zilog;
5541 	char name[ZFS_MAX_DATASET_NAME_LEN];
5542 	int error;
5543 
5544 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5545 
5546 	(void) rw_rdlock(&ztest_name_lock);
5547 
5548 	error = ztest_dataset_create(name);
5549 	if (error == ENOSPC) {
5550 		(void) rw_unlock(&ztest_name_lock);
5551 		ztest_record_enospc(FTAG);
5552 		return (error);
5553 	}
5554 	ASSERT(error == 0 || error == EEXIST);
5555 
5556 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5557 	(void) rw_unlock(&ztest_name_lock);
5558 
5559 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5560 
5561 	zilog = zd->zd_zilog;
5562 
5563 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5564 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5565 		fatal(0, "missing log records: claimed %llu < committed %llu",
5566 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5567 
5568 	ztest_dataset_dirobj_verify(zd);
5569 
5570 	zil_replay(os, zd, ztest_replay_vector);
5571 
5572 	ztest_dataset_dirobj_verify(zd);
5573 
5574 	if (ztest_opts.zo_verbose >= 6)
5575 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5576 		    zd->zd_name,
5577 		    (u_longlong_t)zilog->zl_parse_blk_count,
5578 		    (u_longlong_t)zilog->zl_parse_lr_count,
5579 		    (u_longlong_t)zilog->zl_replaying_seq);
5580 
5581 	zilog = zil_open(os, ztest_get_data);
5582 
5583 	if (zilog->zl_replaying_seq != 0 &&
5584 	    zilog->zl_replaying_seq < committed_seq)
5585 		fatal(0, "missing log records: replayed %llu < committed %llu",
5586 		    zilog->zl_replaying_seq, committed_seq);
5587 
5588 	return (0);
5589 }
5590 
5591 static void
5592 ztest_dataset_close(int d)
5593 {
5594 	ztest_ds_t *zd = &ztest_ds[d];
5595 
5596 	zil_close(zd->zd_zilog);
5597 	dmu_objset_disown(zd->zd_os, zd);
5598 
5599 	ztest_zd_fini(zd);
5600 }
5601 
5602 /*
5603  * Kick off threads to run tests on all datasets in parallel.
5604  */
5605 static void
5606 ztest_run(ztest_shared_t *zs)
5607 {
5608 	thread_t *tid;
5609 	spa_t *spa;
5610 	objset_t *os;
5611 	thread_t resume_tid;
5612 	int error;
5613 
5614 	ztest_exiting = B_FALSE;
5615 
5616 	/*
5617 	 * Initialize parent/child shared state.
5618 	 */
5619 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5620 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5621 
5622 	zs->zs_thread_start = gethrtime();
5623 	zs->zs_thread_stop =
5624 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5625 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5626 	zs->zs_thread_kill = zs->zs_thread_stop;
5627 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5628 		zs->zs_thread_kill -=
5629 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5630 	}
5631 
5632 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5633 
5634 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5635 	    offsetof(ztest_cb_data_t, zcd_node));
5636 
5637 	/*
5638 	 * Open our pool.
5639 	 */
5640 	kernel_init(FREAD | FWRITE);
5641 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5642 	spa->spa_debug = B_TRUE;
5643 	metaslab_preload_limit = ztest_random(20) + 1;
5644 	ztest_spa = spa;
5645 
5646 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5647 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
5648 	zs->zs_guid = dmu_objset_fsid_guid(os);
5649 	dmu_objset_disown(os, FTAG);
5650 
5651 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5652 
5653 	/*
5654 	 * We don't expect the pool to suspend unless maxfaults == 0,
5655 	 * in which case ztest_fault_inject() temporarily takes away
5656 	 * the only valid replica.
5657 	 */
5658 	if (MAXFAULTS() == 0)
5659 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5660 	else
5661 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5662 
5663 	/*
5664 	 * Create a thread to periodically resume suspended I/O.
5665 	 */
5666 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5667 	    &resume_tid) == 0);
5668 
5669 	/*
5670 	 * Create a deadman thread to abort() if we hang.
5671 	 */
5672 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5673 	    NULL) == 0);
5674 
5675 	/*
5676 	 * Verify that we can safely inquire about about any object,
5677 	 * whether it's allocated or not.  To make it interesting,
5678 	 * we probe a 5-wide window around each power of two.
5679 	 * This hits all edge cases, including zero and the max.
5680 	 */
5681 	for (int t = 0; t < 64; t++) {
5682 		for (int d = -5; d <= 5; d++) {
5683 			error = dmu_object_info(spa->spa_meta_objset,
5684 			    (1ULL << t) + d, NULL);
5685 			ASSERT(error == 0 || error == ENOENT ||
5686 			    error == EINVAL);
5687 		}
5688 	}
5689 
5690 	/*
5691 	 * If we got any ENOSPC errors on the previous run, destroy something.
5692 	 */
5693 	if (zs->zs_enospc_count != 0) {
5694 		int d = ztest_random(ztest_opts.zo_datasets);
5695 		ztest_dataset_destroy(d);
5696 	}
5697 	zs->zs_enospc_count = 0;
5698 
5699 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5700 	    UMEM_NOFAIL);
5701 
5702 	if (ztest_opts.zo_verbose >= 4)
5703 		(void) printf("starting main threads...\n");
5704 
5705 	/*
5706 	 * Kick off all the tests that run in parallel.
5707 	 */
5708 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
5709 		if (t < ztest_opts.zo_datasets &&
5710 		    ztest_dataset_open(t) != 0)
5711 			return;
5712 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5713 		    THR_BOUND, &tid[t]) == 0);
5714 	}
5715 
5716 	/*
5717 	 * Wait for all of the tests to complete.  We go in reverse order
5718 	 * so we don't close datasets while threads are still using them.
5719 	 */
5720 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5721 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5722 		if (t < ztest_opts.zo_datasets)
5723 			ztest_dataset_close(t);
5724 	}
5725 
5726 	txg_wait_synced(spa_get_dsl(spa), 0);
5727 
5728 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5729 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5730 	zfs_dbgmsg_print(FTAG);
5731 
5732 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5733 
5734 	/* Kill the resume thread */
5735 	ztest_exiting = B_TRUE;
5736 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5737 	ztest_resume(spa);
5738 
5739 	/*
5740 	 * Right before closing the pool, kick off a bunch of async I/O;
5741 	 * spa_close() should wait for it to complete.
5742 	 */
5743 	for (uint64_t object = 1; object < 50; object++) {
5744 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5745 		    ZIO_PRIORITY_SYNC_READ);
5746 	}
5747 
5748 	spa_close(spa, FTAG);
5749 
5750 	/*
5751 	 * Verify that we can loop over all pools.
5752 	 */
5753 	mutex_enter(&spa_namespace_lock);
5754 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5755 		if (ztest_opts.zo_verbose > 3)
5756 			(void) printf("spa_next: found %s\n", spa_name(spa));
5757 	mutex_exit(&spa_namespace_lock);
5758 
5759 	/*
5760 	 * Verify that we can export the pool and reimport it under a
5761 	 * different name.
5762 	 */
5763 	if (ztest_random(2) == 0) {
5764 		char name[ZFS_MAX_DATASET_NAME_LEN];
5765 		(void) snprintf(name, sizeof (name), "%s_import",
5766 		    ztest_opts.zo_pool);
5767 		ztest_spa_import_export(ztest_opts.zo_pool, name);
5768 		ztest_spa_import_export(name, ztest_opts.zo_pool);
5769 	}
5770 
5771 	kernel_fini();
5772 
5773 	list_destroy(&zcl.zcl_callbacks);
5774 
5775 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5776 
5777 	(void) rwlock_destroy(&ztest_name_lock);
5778 	(void) _mutex_destroy(&ztest_vdev_lock);
5779 }
5780 
5781 static void
5782 ztest_freeze(void)
5783 {
5784 	ztest_ds_t *zd = &ztest_ds[0];
5785 	spa_t *spa;
5786 	int numloops = 0;
5787 
5788 	if (ztest_opts.zo_verbose >= 3)
5789 		(void) printf("testing spa_freeze()...\n");
5790 
5791 	kernel_init(FREAD | FWRITE);
5792 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5793 	VERIFY3U(0, ==, ztest_dataset_open(0));
5794 	spa->spa_debug = B_TRUE;
5795 	ztest_spa = spa;
5796 
5797 	/*
5798 	 * Force the first log block to be transactionally allocated.
5799 	 * We have to do this before we freeze the pool -- otherwise
5800 	 * the log chain won't be anchored.
5801 	 */
5802 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5803 		ztest_dmu_object_alloc_free(zd, 0);
5804 		zil_commit(zd->zd_zilog, 0);
5805 	}
5806 
5807 	txg_wait_synced(spa_get_dsl(spa), 0);
5808 
5809 	/*
5810 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5811 	 * so that the only way to record changes from now on is the ZIL.
5812 	 */
5813 	spa_freeze(spa);
5814 
5815 	/*
5816 	 * Because it is hard to predict how much space a write will actually
5817 	 * require beforehand, we leave ourselves some fudge space to write over
5818 	 * capacity.
5819 	 */
5820 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5821 
5822 	/*
5823 	 * Run tests that generate log records but don't alter the pool config
5824 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5825 	 * We do a txg_wait_synced() after each iteration to force the txg
5826 	 * to increase well beyond the last synced value in the uberblock.
5827 	 * The ZIL should be OK with that.
5828 	 *
5829 	 * Run a random number of times less than zo_maxloops and ensure we do
5830 	 * not run out of space on the pool.
5831 	 */
5832 	while (ztest_random(10) != 0 &&
5833 	    numloops++ < ztest_opts.zo_maxloops &&
5834 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5835 		ztest_od_t od;
5836 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5837 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5838 		ztest_io(zd, od.od_object,
5839 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5840 		txg_wait_synced(spa_get_dsl(spa), 0);
5841 	}
5842 
5843 	/*
5844 	 * Commit all of the changes we just generated.
5845 	 */
5846 	zil_commit(zd->zd_zilog, 0);
5847 	txg_wait_synced(spa_get_dsl(spa), 0);
5848 
5849 	/*
5850 	 * Close our dataset and close the pool.
5851 	 */
5852 	ztest_dataset_close(0);
5853 	spa_close(spa, FTAG);
5854 	kernel_fini();
5855 
5856 	/*
5857 	 * Open and close the pool and dataset to induce log replay.
5858 	 */
5859 	kernel_init(FREAD | FWRITE);
5860 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5861 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5862 	VERIFY3U(0, ==, ztest_dataset_open(0));
5863 	ztest_dataset_close(0);
5864 
5865 	spa->spa_debug = B_TRUE;
5866 	ztest_spa = spa;
5867 	txg_wait_synced(spa_get_dsl(spa), 0);
5868 	ztest_reguid(NULL, 0);
5869 
5870 	spa_close(spa, FTAG);
5871 	kernel_fini();
5872 }
5873 
5874 void
5875 print_time(hrtime_t t, char *timebuf)
5876 {
5877 	hrtime_t s = t / NANOSEC;
5878 	hrtime_t m = s / 60;
5879 	hrtime_t h = m / 60;
5880 	hrtime_t d = h / 24;
5881 
5882 	s -= m * 60;
5883 	m -= h * 60;
5884 	h -= d * 24;
5885 
5886 	timebuf[0] = '\0';
5887 
5888 	if (d)
5889 		(void) sprintf(timebuf,
5890 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5891 	else if (h)
5892 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5893 	else if (m)
5894 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5895 	else
5896 		(void) sprintf(timebuf, "%llus", s);
5897 }
5898 
5899 static nvlist_t *
5900 make_random_props()
5901 {
5902 	nvlist_t *props;
5903 
5904 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5905 	if (ztest_random(2) == 0)
5906 		return (props);
5907 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5908 
5909 	return (props);
5910 }
5911 
5912 /*
5913  * Create a storage pool with the given name and initial vdev size.
5914  * Then test spa_freeze() functionality.
5915  */
5916 static void
5917 ztest_init(ztest_shared_t *zs)
5918 {
5919 	spa_t *spa;
5920 	nvlist_t *nvroot, *props;
5921 
5922 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5923 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5924 
5925 	kernel_init(FREAD | FWRITE);
5926 
5927 	/*
5928 	 * Create the storage pool.
5929 	 */
5930 	(void) spa_destroy(ztest_opts.zo_pool);
5931 	ztest_shared->zs_vdev_next_leaf = 0;
5932 	zs->zs_splits = 0;
5933 	zs->zs_mirrors = ztest_opts.zo_mirrors;
5934 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5935 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5936 	props = make_random_props();
5937 	for (int i = 0; i < SPA_FEATURES; i++) {
5938 		char buf[1024];
5939 		(void) snprintf(buf, sizeof (buf), "feature@%s",
5940 		    spa_feature_table[i].fi_uname);
5941 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5942 	}
5943 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5944 	nvlist_free(nvroot);
5945 	nvlist_free(props);
5946 
5947 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5948 	zs->zs_metaslab_sz =
5949 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5950 
5951 	spa_close(spa, FTAG);
5952 
5953 	kernel_fini();
5954 
5955 	ztest_run_zdb(ztest_opts.zo_pool);
5956 
5957 	ztest_freeze();
5958 
5959 	ztest_run_zdb(ztest_opts.zo_pool);
5960 
5961 	(void) rwlock_destroy(&ztest_name_lock);
5962 	(void) _mutex_destroy(&ztest_vdev_lock);
5963 }
5964 
5965 static void
5966 setup_data_fd(void)
5967 {
5968 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5969 
5970 	ztest_fd_data = mkstemp(ztest_name_data);
5971 	ASSERT3S(ztest_fd_data, >=, 0);
5972 	(void) unlink(ztest_name_data);
5973 }
5974 
5975 
5976 static int
5977 shared_data_size(ztest_shared_hdr_t *hdr)
5978 {
5979 	int size;
5980 
5981 	size = hdr->zh_hdr_size;
5982 	size += hdr->zh_opts_size;
5983 	size += hdr->zh_size;
5984 	size += hdr->zh_stats_size * hdr->zh_stats_count;
5985 	size += hdr->zh_ds_size * hdr->zh_ds_count;
5986 
5987 	return (size);
5988 }
5989 
5990 static void
5991 setup_hdr(void)
5992 {
5993 	int size;
5994 	ztest_shared_hdr_t *hdr;
5995 
5996 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5997 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5998 	ASSERT(hdr != MAP_FAILED);
5999 
6000 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6001 
6002 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6003 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6004 	hdr->zh_size = sizeof (ztest_shared_t);
6005 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6006 	hdr->zh_stats_count = ZTEST_FUNCS;
6007 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6008 	hdr->zh_ds_count = ztest_opts.zo_datasets;
6009 
6010 	size = shared_data_size(hdr);
6011 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6012 
6013 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6014 }
6015 
6016 static void
6017 setup_data(void)
6018 {
6019 	int size, offset;
6020 	ztest_shared_hdr_t *hdr;
6021 	uint8_t *buf;
6022 
6023 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6024 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6025 	ASSERT(hdr != MAP_FAILED);
6026 
6027 	size = shared_data_size(hdr);
6028 
6029 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6030 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6031 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6032 	ASSERT(hdr != MAP_FAILED);
6033 	buf = (uint8_t *)hdr;
6034 
6035 	offset = hdr->zh_hdr_size;
6036 	ztest_shared_opts = (void *)&buf[offset];
6037 	offset += hdr->zh_opts_size;
6038 	ztest_shared = (void *)&buf[offset];
6039 	offset += hdr->zh_size;
6040 	ztest_shared_callstate = (void *)&buf[offset];
6041 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
6042 	ztest_shared_ds = (void *)&buf[offset];
6043 }
6044 
6045 static boolean_t
6046 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6047 {
6048 	pid_t pid;
6049 	int status;
6050 	char *cmdbuf = NULL;
6051 
6052 	pid = fork();
6053 
6054 	if (cmd == NULL) {
6055 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6056 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6057 		cmd = cmdbuf;
6058 	}
6059 
6060 	if (pid == -1)
6061 		fatal(1, "fork failed");
6062 
6063 	if (pid == 0) {	/* child */
6064 		char *emptyargv[2] = { cmd, NULL };
6065 		char fd_data_str[12];
6066 
6067 		struct rlimit rl = { 1024, 1024 };
6068 		(void) setrlimit(RLIMIT_NOFILE, &rl);
6069 
6070 		(void) close(ztest_fd_rand);
6071 		VERIFY3U(11, >=,
6072 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6073 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6074 
6075 		(void) enable_extended_FILE_stdio(-1, -1);
6076 		if (libpath != NULL)
6077 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6078 		(void) execv(cmd, emptyargv);
6079 		ztest_dump_core = B_FALSE;
6080 		fatal(B_TRUE, "exec failed: %s", cmd);
6081 	}
6082 
6083 	if (cmdbuf != NULL) {
6084 		umem_free(cmdbuf, MAXPATHLEN);
6085 		cmd = NULL;
6086 	}
6087 
6088 	while (waitpid(pid, &status, 0) != pid)
6089 		continue;
6090 	if (statusp != NULL)
6091 		*statusp = status;
6092 
6093 	if (WIFEXITED(status)) {
6094 		if (WEXITSTATUS(status) != 0) {
6095 			(void) fprintf(stderr, "child exited with code %d\n",
6096 			    WEXITSTATUS(status));
6097 			exit(2);
6098 		}
6099 		return (B_FALSE);
6100 	} else if (WIFSIGNALED(status)) {
6101 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6102 			(void) fprintf(stderr, "child died with signal %d\n",
6103 			    WTERMSIG(status));
6104 			exit(3);
6105 		}
6106 		return (B_TRUE);
6107 	} else {
6108 		(void) fprintf(stderr, "something strange happened to child\n");
6109 		exit(4);
6110 		/* NOTREACHED */
6111 	}
6112 }
6113 
6114 static void
6115 ztest_run_init(void)
6116 {
6117 	ztest_shared_t *zs = ztest_shared;
6118 
6119 	ASSERT(ztest_opts.zo_init != 0);
6120 
6121 	/*
6122 	 * Blow away any existing copy of zpool.cache
6123 	 */
6124 	(void) remove(spa_config_path);
6125 
6126 	/*
6127 	 * Create and initialize our storage pool.
6128 	 */
6129 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
6130 		bzero(zs, sizeof (ztest_shared_t));
6131 		if (ztest_opts.zo_verbose >= 3 &&
6132 		    ztest_opts.zo_init != 1) {
6133 			(void) printf("ztest_init(), pass %d\n", i);
6134 		}
6135 		ztest_init(zs);
6136 	}
6137 }
6138 
6139 int
6140 main(int argc, char **argv)
6141 {
6142 	int kills = 0;
6143 	int iters = 0;
6144 	int older = 0;
6145 	int newer = 0;
6146 	ztest_shared_t *zs;
6147 	ztest_info_t *zi;
6148 	ztest_shared_callstate_t *zc;
6149 	char timebuf[100];
6150 	char numbuf[6];
6151 	spa_t *spa;
6152 	char *cmd;
6153 	boolean_t hasalt;
6154 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6155 
6156 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6157 
6158 	dprintf_setup(&argc, argv);
6159 	zfs_deadman_synctime_ms = 300000;
6160 
6161 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6162 	ASSERT3S(ztest_fd_rand, >=, 0);
6163 
6164 	if (!fd_data_str) {
6165 		process_options(argc, argv);
6166 
6167 		setup_data_fd();
6168 		setup_hdr();
6169 		setup_data();
6170 		bcopy(&ztest_opts, ztest_shared_opts,
6171 		    sizeof (*ztest_shared_opts));
6172 	} else {
6173 		ztest_fd_data = atoi(fd_data_str);
6174 		setup_data();
6175 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6176 	}
6177 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6178 
6179 	/* Override location of zpool.cache */
6180 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6181 	    ztest_opts.zo_dir), !=, -1);
6182 
6183 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6184 	    UMEM_NOFAIL);
6185 	zs = ztest_shared;
6186 
6187 	if (fd_data_str) {
6188 		metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6189 		metaslab_df_alloc_threshold =
6190 		    zs->zs_metaslab_df_alloc_threshold;
6191 
6192 		if (zs->zs_do_init)
6193 			ztest_run_init();
6194 		else
6195 			ztest_run(zs);
6196 		exit(0);
6197 	}
6198 
6199 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6200 
6201 	if (ztest_opts.zo_verbose >= 1) {
6202 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6203 		    " %llu seconds...\n",
6204 		    (u_longlong_t)ztest_opts.zo_vdevs,
6205 		    ztest_opts.zo_datasets,
6206 		    ztest_opts.zo_threads,
6207 		    (u_longlong_t)ztest_opts.zo_time);
6208 	}
6209 
6210 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6211 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6212 
6213 	zs->zs_do_init = B_TRUE;
6214 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6215 		if (ztest_opts.zo_verbose >= 1) {
6216 			(void) printf("Executing older ztest for "
6217 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6218 		}
6219 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6220 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6221 	} else {
6222 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6223 	}
6224 	zs->zs_do_init = B_FALSE;
6225 
6226 	zs->zs_proc_start = gethrtime();
6227 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6228 
6229 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6230 		zi = &ztest_info[f];
6231 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6232 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6233 			zc->zc_next = UINT64_MAX;
6234 		else
6235 			zc->zc_next = zs->zs_proc_start +
6236 			    ztest_random(2 * zi->zi_interval[0] + 1);
6237 	}
6238 
6239 	/*
6240 	 * Run the tests in a loop.  These tests include fault injection
6241 	 * to verify that self-healing data works, and forced crashes
6242 	 * to verify that we never lose on-disk consistency.
6243 	 */
6244 	while (gethrtime() < zs->zs_proc_stop) {
6245 		int status;
6246 		boolean_t killed;
6247 
6248 		/*
6249 		 * Initialize the workload counters for each function.
6250 		 */
6251 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6252 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6253 			zc->zc_count = 0;
6254 			zc->zc_time = 0;
6255 		}
6256 
6257 		/* Set the allocation switch size */
6258 		zs->zs_metaslab_df_alloc_threshold =
6259 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6260 
6261 		if (!hasalt || ztest_random(2) == 0) {
6262 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6263 				(void) printf("Executing newer ztest: %s\n",
6264 				    cmd);
6265 			}
6266 			newer++;
6267 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6268 		} else {
6269 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6270 				(void) printf("Executing older ztest: %s\n",
6271 				    ztest_opts.zo_alt_ztest);
6272 			}
6273 			older++;
6274 			killed = exec_child(ztest_opts.zo_alt_ztest,
6275 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6276 		}
6277 
6278 		if (killed)
6279 			kills++;
6280 		iters++;
6281 
6282 		if (ztest_opts.zo_verbose >= 1) {
6283 			hrtime_t now = gethrtime();
6284 
6285 			now = MIN(now, zs->zs_proc_stop);
6286 			print_time(zs->zs_proc_stop - now, timebuf);
6287 			nicenum(zs->zs_space, numbuf);
6288 
6289 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6290 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6291 			    iters,
6292 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6293 			    (u_longlong_t)zs->zs_enospc_count,
6294 			    100.0 * zs->zs_alloc / zs->zs_space,
6295 			    numbuf,
6296 			    100.0 * (now - zs->zs_proc_start) /
6297 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6298 		}
6299 
6300 		if (ztest_opts.zo_verbose >= 2) {
6301 			(void) printf("\nWorkload summary:\n\n");
6302 			(void) printf("%7s %9s   %s\n",
6303 			    "Calls", "Time", "Function");
6304 			(void) printf("%7s %9s   %s\n",
6305 			    "-----", "----", "--------");
6306 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6307 				Dl_info dli;
6308 
6309 				zi = &ztest_info[f];
6310 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6311 				print_time(zc->zc_time, timebuf);
6312 				(void) dladdr((void *)zi->zi_func, &dli);
6313 				(void) printf("%7llu %9s   %s\n",
6314 				    (u_longlong_t)zc->zc_count, timebuf,
6315 				    dli.dli_sname);
6316 			}
6317 			(void) printf("\n");
6318 		}
6319 
6320 		/*
6321 		 * It's possible that we killed a child during a rename test,
6322 		 * in which case we'll have a 'ztest_tmp' pool lying around
6323 		 * instead of 'ztest'.  Do a blind rename in case this happened.
6324 		 */
6325 		kernel_init(FREAD);
6326 		if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6327 			spa_close(spa, FTAG);
6328 		} else {
6329 			char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6330 			kernel_fini();
6331 			kernel_init(FREAD | FWRITE);
6332 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6333 			    ztest_opts.zo_pool);
6334 			(void) spa_rename(tmpname, ztest_opts.zo_pool);
6335 		}
6336 		kernel_fini();
6337 
6338 		ztest_run_zdb(ztest_opts.zo_pool);
6339 	}
6340 
6341 	if (ztest_opts.zo_verbose >= 1) {
6342 		if (hasalt) {
6343 			(void) printf("%d runs of older ztest: %s\n", older,
6344 			    ztest_opts.zo_alt_ztest);
6345 			(void) printf("%d runs of newer ztest: %s\n", newer,
6346 			    cmd);
6347 		}
6348 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6349 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6350 	}
6351 
6352 	umem_free(cmd, MAXNAMELEN);
6353 
6354 	return (0);
6355 }
6356