xref: /titanic_51/usr/src/cmd/ztest/ztest.c (revision 1410cb930a3e26032c59c6835837a28c47366b3c)
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) 2013 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  */
27 
28 /*
29  * The objective of this program is to provide a DMU/ZAP/SPA stress test
30  * that runs entirely in userland, is easy to use, and easy to extend.
31  *
32  * The overall design of the ztest program is as follows:
33  *
34  * (1) For each major functional area (e.g. adding vdevs to a pool,
35  *     creating and destroying datasets, reading and writing objects, etc)
36  *     we have a simple routine to test that functionality.  These
37  *     individual routines do not have to do anything "stressful".
38  *
39  * (2) We turn these simple functionality tests into a stress test by
40  *     running them all in parallel, with as many threads as desired,
41  *     and spread across as many datasets, objects, and vdevs as desired.
42  *
43  * (3) While all this is happening, we inject faults into the pool to
44  *     verify that self-healing data really works.
45  *
46  * (4) Every time we open a dataset, we change its checksum and compression
47  *     functions.  Thus even individual objects vary from block to block
48  *     in which checksum they use and whether they're compressed.
49  *
50  * (5) To verify that we never lose on-disk consistency after a crash,
51  *     we run the entire test in a child of the main process.
52  *     At random times, the child self-immolates with a SIGKILL.
53  *     This is the software equivalent of pulling the power cord.
54  *     The parent then runs the test again, using the existing
55  *     storage pool, as many times as desired. If backwards compatability
56  *     testing is enabled ztest will sometimes run the "older" version
57  *     of ztest after a SIGKILL.
58  *
59  * (6) To verify that we don't have future leaks or temporal incursions,
60  *     many of the functional tests record the transaction group number
61  *     as part of their data.  When reading old data, they verify that
62  *     the transaction group number is less than the current, open txg.
63  *     If you add a new test, please do this if applicable.
64  *
65  * When run with no arguments, ztest runs for about five minutes and
66  * produces no output if successful.  To get a little bit of information,
67  * specify -V.  To get more information, specify -VV, and so on.
68  *
69  * To turn this into an overnight stress test, use -T to specify run time.
70  *
71  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
72  * to increase the pool capacity, fanout, and overall stress level.
73  *
74  * Use the -k option to set the desired frequency of kills.
75  *
76  * When ztest invokes itself it passes all relevant information through a
77  * temporary file which is mmap-ed in the child process. This allows shared
78  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
79  * stored at offset 0 of this file and contains information on the size and
80  * number of shared structures in the file. The information stored in this file
81  * must remain backwards compatible with older versions of ztest so that
82  * ztest can invoke them during backwards compatibility testing (-B).
83  */
84 
85 #include <sys/zfs_context.h>
86 #include <sys/spa.h>
87 #include <sys/dmu.h>
88 #include <sys/txg.h>
89 #include <sys/dbuf.h>
90 #include <sys/zap.h>
91 #include <sys/dmu_objset.h>
92 #include <sys/poll.h>
93 #include <sys/stat.h>
94 #include <sys/time.h>
95 #include <sys/wait.h>
96 #include <sys/mman.h>
97 #include <sys/resource.h>
98 #include <sys/zio.h>
99 #include <sys/zil.h>
100 #include <sys/zil_impl.h>
101 #include <sys/vdev_impl.h>
102 #include <sys/vdev_file.h>
103 #include <sys/spa_impl.h>
104 #include <sys/metaslab_impl.h>
105 #include <sys/dsl_prop.h>
106 #include <sys/dsl_dataset.h>
107 #include <sys/dsl_destroy.h>
108 #include <sys/dsl_scan.h>
109 #include <sys/zio_checksum.h>
110 #include <sys/refcount.h>
111 #include <sys/zfeature.h>
112 #include <sys/dsl_userhold.h>
113 #include <stdio.h>
114 #include <stdio_ext.h>
115 #include <stdlib.h>
116 #include <unistd.h>
117 #include <signal.h>
118 #include <umem.h>
119 #include <dlfcn.h>
120 #include <ctype.h>
121 #include <math.h>
122 #include <sys/fs/zfs.h>
123 #include <libnvpair.h>
124 
125 static int ztest_fd_data = -1;
126 static int ztest_fd_rand = -1;
127 
128 typedef struct ztest_shared_hdr {
129 	uint64_t	zh_hdr_size;
130 	uint64_t	zh_opts_size;
131 	uint64_t	zh_size;
132 	uint64_t	zh_stats_size;
133 	uint64_t	zh_stats_count;
134 	uint64_t	zh_ds_size;
135 	uint64_t	zh_ds_count;
136 } ztest_shared_hdr_t;
137 
138 static ztest_shared_hdr_t *ztest_shared_hdr;
139 
140 typedef struct ztest_shared_opts {
141 	char zo_pool[MAXNAMELEN];
142 	char zo_dir[MAXNAMELEN];
143 	char zo_alt_ztest[MAXNAMELEN];
144 	char zo_alt_libpath[MAXNAMELEN];
145 	uint64_t zo_vdevs;
146 	uint64_t zo_vdevtime;
147 	size_t zo_vdev_size;
148 	int zo_ashift;
149 	int zo_mirrors;
150 	int zo_raidz;
151 	int zo_raidz_parity;
152 	int zo_datasets;
153 	int zo_threads;
154 	uint64_t zo_passtime;
155 	uint64_t zo_killrate;
156 	int zo_verbose;
157 	int zo_init;
158 	uint64_t zo_time;
159 	uint64_t zo_maxloops;
160 	uint64_t zo_metaslab_gang_bang;
161 } ztest_shared_opts_t;
162 
163 static const ztest_shared_opts_t ztest_opts_defaults = {
164 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
165 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
166 	.zo_alt_ztest = { '\0' },
167 	.zo_alt_libpath = { '\0' },
168 	.zo_vdevs = 5,
169 	.zo_ashift = SPA_MINBLOCKSHIFT,
170 	.zo_mirrors = 2,
171 	.zo_raidz = 4,
172 	.zo_raidz_parity = 1,
173 	.zo_vdev_size = SPA_MINDEVSIZE,
174 	.zo_datasets = 7,
175 	.zo_threads = 23,
176 	.zo_passtime = 60,		/* 60 seconds */
177 	.zo_killrate = 70,		/* 70% kill rate */
178 	.zo_verbose = 0,
179 	.zo_init = 1,
180 	.zo_time = 300,			/* 5 minutes */
181 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
182 	.zo_metaslab_gang_bang = 32 << 10
183 };
184 
185 extern uint64_t metaslab_gang_bang;
186 extern uint64_t metaslab_df_alloc_threshold;
187 extern uint64_t zfs_deadman_synctime_ms;
188 
189 static ztest_shared_opts_t *ztest_shared_opts;
190 static ztest_shared_opts_t ztest_opts;
191 
192 typedef struct ztest_shared_ds {
193 	uint64_t	zd_seq;
194 } ztest_shared_ds_t;
195 
196 static ztest_shared_ds_t *ztest_shared_ds;
197 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
198 
199 #define	BT_MAGIC	0x123456789abcdefULL
200 #define	MAXFAULTS() \
201 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
202 
203 enum ztest_io_type {
204 	ZTEST_IO_WRITE_TAG,
205 	ZTEST_IO_WRITE_PATTERN,
206 	ZTEST_IO_WRITE_ZEROES,
207 	ZTEST_IO_TRUNCATE,
208 	ZTEST_IO_SETATTR,
209 	ZTEST_IO_REWRITE,
210 	ZTEST_IO_TYPES
211 };
212 
213 typedef struct ztest_block_tag {
214 	uint64_t	bt_magic;
215 	uint64_t	bt_objset;
216 	uint64_t	bt_object;
217 	uint64_t	bt_offset;
218 	uint64_t	bt_gen;
219 	uint64_t	bt_txg;
220 	uint64_t	bt_crtxg;
221 } ztest_block_tag_t;
222 
223 typedef struct bufwad {
224 	uint64_t	bw_index;
225 	uint64_t	bw_txg;
226 	uint64_t	bw_data;
227 } bufwad_t;
228 
229 /*
230  * XXX -- fix zfs range locks to be generic so we can use them here.
231  */
232 typedef enum {
233 	RL_READER,
234 	RL_WRITER,
235 	RL_APPEND
236 } rl_type_t;
237 
238 typedef struct rll {
239 	void		*rll_writer;
240 	int		rll_readers;
241 	mutex_t		rll_lock;
242 	cond_t		rll_cv;
243 } rll_t;
244 
245 typedef struct rl {
246 	uint64_t	rl_object;
247 	uint64_t	rl_offset;
248 	uint64_t	rl_size;
249 	rll_t		*rl_lock;
250 } rl_t;
251 
252 #define	ZTEST_RANGE_LOCKS	64
253 #define	ZTEST_OBJECT_LOCKS	64
254 
255 /*
256  * Object descriptor.  Used as a template for object lookup/create/remove.
257  */
258 typedef struct ztest_od {
259 	uint64_t	od_dir;
260 	uint64_t	od_object;
261 	dmu_object_type_t od_type;
262 	dmu_object_type_t od_crtype;
263 	uint64_t	od_blocksize;
264 	uint64_t	od_crblocksize;
265 	uint64_t	od_gen;
266 	uint64_t	od_crgen;
267 	char		od_name[MAXNAMELEN];
268 } ztest_od_t;
269 
270 /*
271  * Per-dataset state.
272  */
273 typedef struct ztest_ds {
274 	ztest_shared_ds_t *zd_shared;
275 	objset_t	*zd_os;
276 	rwlock_t	zd_zilog_lock;
277 	zilog_t		*zd_zilog;
278 	ztest_od_t	*zd_od;		/* debugging aid */
279 	char		zd_name[MAXNAMELEN];
280 	mutex_t		zd_dirobj_lock;
281 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
282 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
283 } ztest_ds_t;
284 
285 /*
286  * Per-iteration state.
287  */
288 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
289 
290 typedef struct ztest_info {
291 	ztest_func_t	*zi_func;	/* test function */
292 	uint64_t	zi_iters;	/* iterations per execution */
293 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
294 } ztest_info_t;
295 
296 typedef struct ztest_shared_callstate {
297 	uint64_t	zc_count;	/* per-pass count */
298 	uint64_t	zc_time;	/* per-pass time */
299 	uint64_t	zc_next;	/* next time to call this function */
300 } ztest_shared_callstate_t;
301 
302 static ztest_shared_callstate_t *ztest_shared_callstate;
303 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
304 
305 /*
306  * Note: these aren't static because we want dladdr() to work.
307  */
308 ztest_func_t ztest_dmu_read_write;
309 ztest_func_t ztest_dmu_write_parallel;
310 ztest_func_t ztest_dmu_object_alloc_free;
311 ztest_func_t ztest_dmu_commit_callbacks;
312 ztest_func_t ztest_zap;
313 ztest_func_t ztest_zap_parallel;
314 ztest_func_t ztest_zil_commit;
315 ztest_func_t ztest_zil_remount;
316 ztest_func_t ztest_dmu_read_write_zcopy;
317 ztest_func_t ztest_dmu_objset_create_destroy;
318 ztest_func_t ztest_dmu_prealloc;
319 ztest_func_t ztest_fzap;
320 ztest_func_t ztest_dmu_snapshot_create_destroy;
321 ztest_func_t ztest_dsl_prop_get_set;
322 ztest_func_t ztest_spa_prop_get_set;
323 ztest_func_t ztest_spa_create_destroy;
324 ztest_func_t ztest_fault_inject;
325 ztest_func_t ztest_ddt_repair;
326 ztest_func_t ztest_dmu_snapshot_hold;
327 ztest_func_t ztest_spa_rename;
328 ztest_func_t ztest_scrub;
329 ztest_func_t ztest_dsl_dataset_promote_busy;
330 ztest_func_t ztest_vdev_attach_detach;
331 ztest_func_t ztest_vdev_LUN_growth;
332 ztest_func_t ztest_vdev_add_remove;
333 ztest_func_t ztest_vdev_aux_add_remove;
334 ztest_func_t ztest_split_pool;
335 ztest_func_t ztest_reguid;
336 ztest_func_t ztest_spa_upgrade;
337 
338 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
339 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
340 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
341 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
342 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
343 
344 ztest_info_t ztest_info[] = {
345 	{ ztest_dmu_read_write,			1,	&zopt_always	},
346 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
347 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
348 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
349 	{ ztest_zap,				30,	&zopt_always	},
350 	{ ztest_zap_parallel,			100,	&zopt_always	},
351 	{ ztest_split_pool,			1,	&zopt_always	},
352 	{ ztest_zil_commit,			1,	&zopt_incessant	},
353 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
354 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
355 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
356 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
357 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
358 #if 0
359 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
360 #endif
361 	{ ztest_fzap,				1,	&zopt_sometimes	},
362 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
363 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
364 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
365 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
366 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
367 	{ ztest_reguid,				1,	&zopt_rarely	},
368 	{ ztest_spa_rename,			1,	&zopt_rarely	},
369 	{ ztest_scrub,				1,	&zopt_rarely	},
370 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
371 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
372 	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
373 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
374 	{ ztest_vdev_add_remove,		1,
375 	    &ztest_opts.zo_vdevtime				},
376 	{ ztest_vdev_aux_add_remove,		1,
377 	    &ztest_opts.zo_vdevtime				},
378 };
379 
380 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
381 
382 /*
383  * The following struct is used to hold a list of uncalled commit callbacks.
384  * The callbacks are ordered by txg number.
385  */
386 typedef struct ztest_cb_list {
387 	mutex_t	zcl_callbacks_lock;
388 	list_t	zcl_callbacks;
389 } ztest_cb_list_t;
390 
391 /*
392  * Stuff we need to share writably between parent and child.
393  */
394 typedef struct ztest_shared {
395 	boolean_t	zs_do_init;
396 	hrtime_t	zs_proc_start;
397 	hrtime_t	zs_proc_stop;
398 	hrtime_t	zs_thread_start;
399 	hrtime_t	zs_thread_stop;
400 	hrtime_t	zs_thread_kill;
401 	uint64_t	zs_enospc_count;
402 	uint64_t	zs_vdev_next_leaf;
403 	uint64_t	zs_vdev_aux;
404 	uint64_t	zs_alloc;
405 	uint64_t	zs_space;
406 	uint64_t	zs_splits;
407 	uint64_t	zs_mirrors;
408 	uint64_t	zs_metaslab_sz;
409 	uint64_t	zs_metaslab_df_alloc_threshold;
410 	uint64_t	zs_guid;
411 } ztest_shared_t;
412 
413 #define	ID_PARALLEL	-1ULL
414 
415 static char ztest_dev_template[] = "%s/%s.%llua";
416 static char ztest_aux_template[] = "%s/%s.%s.%llu";
417 ztest_shared_t *ztest_shared;
418 
419 static spa_t *ztest_spa = NULL;
420 static ztest_ds_t *ztest_ds;
421 
422 static mutex_t ztest_vdev_lock;
423 
424 /*
425  * The ztest_name_lock protects the pool and dataset namespace used by
426  * the individual tests. To modify the namespace, consumers must grab
427  * this lock as writer. Grabbing the lock as reader will ensure that the
428  * namespace does not change while the lock is held.
429  */
430 static rwlock_t ztest_name_lock;
431 
432 static boolean_t ztest_dump_core = B_TRUE;
433 static boolean_t ztest_exiting;
434 
435 /* Global commit callback list */
436 static ztest_cb_list_t zcl;
437 
438 enum ztest_object {
439 	ZTEST_META_DNODE = 0,
440 	ZTEST_DIROBJ,
441 	ZTEST_OBJECTS
442 };
443 
444 static void usage(boolean_t) __NORETURN;
445 
446 /*
447  * These libumem hooks provide a reasonable set of defaults for the allocator's
448  * debugging facilities.
449  */
450 const char *
451 _umem_debug_init()
452 {
453 	return ("default,verbose"); /* $UMEM_DEBUG setting */
454 }
455 
456 const char *
457 _umem_logging_init(void)
458 {
459 	return ("fail,contents"); /* $UMEM_LOGGING setting */
460 }
461 
462 #define	FATAL_MSG_SZ	1024
463 
464 char *fatal_msg;
465 
466 static void
467 fatal(int do_perror, char *message, ...)
468 {
469 	va_list args;
470 	int save_errno = errno;
471 	char buf[FATAL_MSG_SZ];
472 
473 	(void) fflush(stdout);
474 
475 	va_start(args, message);
476 	(void) sprintf(buf, "ztest: ");
477 	/* LINTED */
478 	(void) vsprintf(buf + strlen(buf), message, args);
479 	va_end(args);
480 	if (do_perror) {
481 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
482 		    ": %s", strerror(save_errno));
483 	}
484 	(void) fprintf(stderr, "%s\n", buf);
485 	fatal_msg = buf;			/* to ease debugging */
486 	if (ztest_dump_core)
487 		abort();
488 	exit(3);
489 }
490 
491 static int
492 str2shift(const char *buf)
493 {
494 	const char *ends = "BKMGTPEZ";
495 	int i;
496 
497 	if (buf[0] == '\0')
498 		return (0);
499 	for (i = 0; i < strlen(ends); i++) {
500 		if (toupper(buf[0]) == ends[i])
501 			break;
502 	}
503 	if (i == strlen(ends)) {
504 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
505 		    buf);
506 		usage(B_FALSE);
507 	}
508 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
509 		return (10*i);
510 	}
511 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
512 	usage(B_FALSE);
513 	/* NOTREACHED */
514 }
515 
516 static uint64_t
517 nicenumtoull(const char *buf)
518 {
519 	char *end;
520 	uint64_t val;
521 
522 	val = strtoull(buf, &end, 0);
523 	if (end == buf) {
524 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
525 		usage(B_FALSE);
526 	} else if (end[0] == '.') {
527 		double fval = strtod(buf, &end);
528 		fval *= pow(2, str2shift(end));
529 		if (fval > UINT64_MAX) {
530 			(void) fprintf(stderr, "ztest: value too large: %s\n",
531 			    buf);
532 			usage(B_FALSE);
533 		}
534 		val = (uint64_t)fval;
535 	} else {
536 		int shift = str2shift(end);
537 		if (shift >= 64 || (val << shift) >> shift != val) {
538 			(void) fprintf(stderr, "ztest: value too large: %s\n",
539 			    buf);
540 			usage(B_FALSE);
541 		}
542 		val <<= shift;
543 	}
544 	return (val);
545 }
546 
547 static void
548 usage(boolean_t requested)
549 {
550 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
551 
552 	char nice_vdev_size[10];
553 	char nice_gang_bang[10];
554 	FILE *fp = requested ? stdout : stderr;
555 
556 	nicenum(zo->zo_vdev_size, nice_vdev_size);
557 	nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
558 
559 	(void) fprintf(fp, "Usage: %s\n"
560 	    "\t[-v vdevs (default: %llu)]\n"
561 	    "\t[-s size_of_each_vdev (default: %s)]\n"
562 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
563 	    "\t[-m mirror_copies (default: %d)]\n"
564 	    "\t[-r raidz_disks (default: %d)]\n"
565 	    "\t[-R raidz_parity (default: %d)]\n"
566 	    "\t[-d datasets (default: %d)]\n"
567 	    "\t[-t threads (default: %d)]\n"
568 	    "\t[-g gang_block_threshold (default: %s)]\n"
569 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
570 	    "\t[-k kill_percentage (default: %llu%%)]\n"
571 	    "\t[-p pool_name (default: %s)]\n"
572 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
573 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
574 	    "\t[-E] use existing pool instead of creating new one\n"
575 	    "\t[-T time (default: %llu sec)] total run time\n"
576 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
577 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
578 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
579 	    "\t[-h] (print help)\n"
580 	    "",
581 	    zo->zo_pool,
582 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
583 	    nice_vdev_size,				/* -s */
584 	    zo->zo_ashift,				/* -a */
585 	    zo->zo_mirrors,				/* -m */
586 	    zo->zo_raidz,				/* -r */
587 	    zo->zo_raidz_parity,			/* -R */
588 	    zo->zo_datasets,				/* -d */
589 	    zo->zo_threads,				/* -t */
590 	    nice_gang_bang,				/* -g */
591 	    zo->zo_init,				/* -i */
592 	    (u_longlong_t)zo->zo_killrate,		/* -k */
593 	    zo->zo_pool,				/* -p */
594 	    zo->zo_dir,					/* -f */
595 	    (u_longlong_t)zo->zo_time,			/* -T */
596 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
597 	    (u_longlong_t)zo->zo_passtime);
598 	exit(requested ? 0 : 1);
599 }
600 
601 static void
602 process_options(int argc, char **argv)
603 {
604 	char *path;
605 	ztest_shared_opts_t *zo = &ztest_opts;
606 
607 	int opt;
608 	uint64_t value;
609 	char altdir[MAXNAMELEN] = { 0 };
610 
611 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
612 
613 	while ((opt = getopt(argc, argv,
614 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
615 		value = 0;
616 		switch (opt) {
617 		case 'v':
618 		case 's':
619 		case 'a':
620 		case 'm':
621 		case 'r':
622 		case 'R':
623 		case 'd':
624 		case 't':
625 		case 'g':
626 		case 'i':
627 		case 'k':
628 		case 'T':
629 		case 'P':
630 		case 'F':
631 			value = nicenumtoull(optarg);
632 		}
633 		switch (opt) {
634 		case 'v':
635 			zo->zo_vdevs = value;
636 			break;
637 		case 's':
638 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
639 			break;
640 		case 'a':
641 			zo->zo_ashift = value;
642 			break;
643 		case 'm':
644 			zo->zo_mirrors = value;
645 			break;
646 		case 'r':
647 			zo->zo_raidz = MAX(1, value);
648 			break;
649 		case 'R':
650 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
651 			break;
652 		case 'd':
653 			zo->zo_datasets = MAX(1, value);
654 			break;
655 		case 't':
656 			zo->zo_threads = MAX(1, value);
657 			break;
658 		case 'g':
659 			zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
660 			    value);
661 			break;
662 		case 'i':
663 			zo->zo_init = value;
664 			break;
665 		case 'k':
666 			zo->zo_killrate = value;
667 			break;
668 		case 'p':
669 			(void) strlcpy(zo->zo_pool, optarg,
670 			    sizeof (zo->zo_pool));
671 			break;
672 		case 'f':
673 			path = realpath(optarg, NULL);
674 			if (path == NULL) {
675 				(void) fprintf(stderr, "error: %s: %s\n",
676 				    optarg, strerror(errno));
677 				usage(B_FALSE);
678 			} else {
679 				(void) strlcpy(zo->zo_dir, path,
680 				    sizeof (zo->zo_dir));
681 			}
682 			break;
683 		case 'V':
684 			zo->zo_verbose++;
685 			break;
686 		case 'E':
687 			zo->zo_init = 0;
688 			break;
689 		case 'T':
690 			zo->zo_time = value;
691 			break;
692 		case 'P':
693 			zo->zo_passtime = MAX(1, value);
694 			break;
695 		case 'F':
696 			zo->zo_maxloops = MAX(1, value);
697 			break;
698 		case 'B':
699 			(void) strlcpy(altdir, optarg, sizeof (altdir));
700 			break;
701 		case 'h':
702 			usage(B_TRUE);
703 			break;
704 		case '?':
705 		default:
706 			usage(B_FALSE);
707 			break;
708 		}
709 	}
710 
711 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
712 
713 	zo->zo_vdevtime =
714 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
715 	    UINT64_MAX >> 2);
716 
717 	if (strlen(altdir) > 0) {
718 		char *cmd;
719 		char *realaltdir;
720 		char *bin;
721 		char *ztest;
722 		char *isa;
723 		int isalen;
724 
725 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
726 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
727 
728 		VERIFY(NULL != realpath(getexecname(), cmd));
729 		if (0 != access(altdir, F_OK)) {
730 			ztest_dump_core = B_FALSE;
731 			fatal(B_TRUE, "invalid alternate ztest path: %s",
732 			    altdir);
733 		}
734 		VERIFY(NULL != realpath(altdir, realaltdir));
735 
736 		/*
737 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
738 		 * We want to extract <isa> to determine if we should use
739 		 * 32 or 64 bit binaries.
740 		 */
741 		bin = strstr(cmd, "/usr/bin/");
742 		ztest = strstr(bin, "/ztest");
743 		isa = bin + 9;
744 		isalen = ztest - isa;
745 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
746 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
747 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
748 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
749 
750 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
751 			ztest_dump_core = B_FALSE;
752 			fatal(B_TRUE, "invalid alternate ztest: %s",
753 			    zo->zo_alt_ztest);
754 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
755 			ztest_dump_core = B_FALSE;
756 			fatal(B_TRUE, "invalid alternate lib directory %s",
757 			    zo->zo_alt_libpath);
758 		}
759 
760 		umem_free(cmd, MAXPATHLEN);
761 		umem_free(realaltdir, MAXPATHLEN);
762 	}
763 }
764 
765 static void
766 ztest_kill(ztest_shared_t *zs)
767 {
768 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
769 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
770 
771 	/*
772 	 * Before we kill off ztest, make sure that the config is updated.
773 	 * See comment above spa_config_sync().
774 	 */
775 	mutex_enter(&spa_namespace_lock);
776 	spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
777 	mutex_exit(&spa_namespace_lock);
778 
779 	zfs_dbgmsg_print(FTAG);
780 	(void) kill(getpid(), SIGKILL);
781 }
782 
783 static uint64_t
784 ztest_random(uint64_t range)
785 {
786 	uint64_t r;
787 
788 	ASSERT3S(ztest_fd_rand, >=, 0);
789 
790 	if (range == 0)
791 		return (0);
792 
793 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
794 		fatal(1, "short read from /dev/urandom");
795 
796 	return (r % range);
797 }
798 
799 /* ARGSUSED */
800 static void
801 ztest_record_enospc(const char *s)
802 {
803 	ztest_shared->zs_enospc_count++;
804 }
805 
806 static uint64_t
807 ztest_get_ashift(void)
808 {
809 	if (ztest_opts.zo_ashift == 0)
810 		return (SPA_MINBLOCKSHIFT + ztest_random(3));
811 	return (ztest_opts.zo_ashift);
812 }
813 
814 static nvlist_t *
815 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
816 {
817 	char pathbuf[MAXPATHLEN];
818 	uint64_t vdev;
819 	nvlist_t *file;
820 
821 	if (ashift == 0)
822 		ashift = ztest_get_ashift();
823 
824 	if (path == NULL) {
825 		path = pathbuf;
826 
827 		if (aux != NULL) {
828 			vdev = ztest_shared->zs_vdev_aux;
829 			(void) snprintf(path, sizeof (pathbuf),
830 			    ztest_aux_template, ztest_opts.zo_dir,
831 			    pool == NULL ? ztest_opts.zo_pool : pool,
832 			    aux, vdev);
833 		} else {
834 			vdev = ztest_shared->zs_vdev_next_leaf++;
835 			(void) snprintf(path, sizeof (pathbuf),
836 			    ztest_dev_template, ztest_opts.zo_dir,
837 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
838 		}
839 	}
840 
841 	if (size != 0) {
842 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
843 		if (fd == -1)
844 			fatal(1, "can't open %s", path);
845 		if (ftruncate(fd, size) != 0)
846 			fatal(1, "can't ftruncate %s", path);
847 		(void) close(fd);
848 	}
849 
850 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
851 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
852 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
853 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
854 
855 	return (file);
856 }
857 
858 static nvlist_t *
859 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
860     uint64_t ashift, int r)
861 {
862 	nvlist_t *raidz, **child;
863 	int c;
864 
865 	if (r < 2)
866 		return (make_vdev_file(path, aux, pool, size, ashift));
867 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
868 
869 	for (c = 0; c < r; c++)
870 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
871 
872 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
873 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
874 	    VDEV_TYPE_RAIDZ) == 0);
875 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
876 	    ztest_opts.zo_raidz_parity) == 0);
877 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
878 	    child, r) == 0);
879 
880 	for (c = 0; c < r; c++)
881 		nvlist_free(child[c]);
882 
883 	umem_free(child, r * sizeof (nvlist_t *));
884 
885 	return (raidz);
886 }
887 
888 static nvlist_t *
889 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
890     uint64_t ashift, int r, int m)
891 {
892 	nvlist_t *mirror, **child;
893 	int c;
894 
895 	if (m < 1)
896 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
897 
898 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
899 
900 	for (c = 0; c < m; c++)
901 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
902 
903 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
904 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
905 	    VDEV_TYPE_MIRROR) == 0);
906 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
907 	    child, m) == 0);
908 
909 	for (c = 0; c < m; c++)
910 		nvlist_free(child[c]);
911 
912 	umem_free(child, m * sizeof (nvlist_t *));
913 
914 	return (mirror);
915 }
916 
917 static nvlist_t *
918 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
919     int log, int r, int m, int t)
920 {
921 	nvlist_t *root, **child;
922 	int c;
923 
924 	ASSERT(t > 0);
925 
926 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
927 
928 	for (c = 0; c < t; c++) {
929 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
930 		    r, m);
931 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
932 		    log) == 0);
933 	}
934 
935 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
936 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
937 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
938 	    child, t) == 0);
939 
940 	for (c = 0; c < t; c++)
941 		nvlist_free(child[c]);
942 
943 	umem_free(child, t * sizeof (nvlist_t *));
944 
945 	return (root);
946 }
947 
948 /*
949  * Find a random spa version. Returns back a random spa version in the
950  * range [initial_version, SPA_VERSION_FEATURES].
951  */
952 static uint64_t
953 ztest_random_spa_version(uint64_t initial_version)
954 {
955 	uint64_t version = initial_version;
956 
957 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
958 		version = version +
959 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
960 	}
961 
962 	if (version > SPA_VERSION_BEFORE_FEATURES)
963 		version = SPA_VERSION_FEATURES;
964 
965 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
966 	return (version);
967 }
968 
969 static int
970 ztest_random_blocksize(void)
971 {
972 	return (1 << (SPA_MINBLOCKSHIFT +
973 	    ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
974 }
975 
976 static int
977 ztest_random_ibshift(void)
978 {
979 	return (DN_MIN_INDBLKSHIFT +
980 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
981 }
982 
983 static uint64_t
984 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
985 {
986 	uint64_t top;
987 	vdev_t *rvd = spa->spa_root_vdev;
988 	vdev_t *tvd;
989 
990 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
991 
992 	do {
993 		top = ztest_random(rvd->vdev_children);
994 		tvd = rvd->vdev_child[top];
995 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
996 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
997 
998 	return (top);
999 }
1000 
1001 static uint64_t
1002 ztest_random_dsl_prop(zfs_prop_t prop)
1003 {
1004 	uint64_t value;
1005 
1006 	do {
1007 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1008 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1009 
1010 	return (value);
1011 }
1012 
1013 static int
1014 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1015     boolean_t inherit)
1016 {
1017 	const char *propname = zfs_prop_to_name(prop);
1018 	const char *valname;
1019 	char setpoint[MAXPATHLEN];
1020 	uint64_t curval;
1021 	int error;
1022 
1023 	error = dsl_prop_set_int(osname, propname,
1024 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1025 
1026 	if (error == ENOSPC) {
1027 		ztest_record_enospc(FTAG);
1028 		return (error);
1029 	}
1030 	ASSERT0(error);
1031 
1032 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1033 
1034 	if (ztest_opts.zo_verbose >= 6) {
1035 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1036 		(void) printf("%s %s = %s at '%s'\n",
1037 		    osname, propname, valname, setpoint);
1038 	}
1039 
1040 	return (error);
1041 }
1042 
1043 static int
1044 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1045 {
1046 	spa_t *spa = ztest_spa;
1047 	nvlist_t *props = NULL;
1048 	int error;
1049 
1050 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1051 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1052 
1053 	error = spa_prop_set(spa, props);
1054 
1055 	nvlist_free(props);
1056 
1057 	if (error == ENOSPC) {
1058 		ztest_record_enospc(FTAG);
1059 		return (error);
1060 	}
1061 	ASSERT0(error);
1062 
1063 	return (error);
1064 }
1065 
1066 static void
1067 ztest_rll_init(rll_t *rll)
1068 {
1069 	rll->rll_writer = NULL;
1070 	rll->rll_readers = 0;
1071 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1072 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1073 }
1074 
1075 static void
1076 ztest_rll_destroy(rll_t *rll)
1077 {
1078 	ASSERT(rll->rll_writer == NULL);
1079 	ASSERT(rll->rll_readers == 0);
1080 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1081 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
1082 }
1083 
1084 static void
1085 ztest_rll_lock(rll_t *rll, rl_type_t type)
1086 {
1087 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1088 
1089 	if (type == RL_READER) {
1090 		while (rll->rll_writer != NULL)
1091 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1092 		rll->rll_readers++;
1093 	} else {
1094 		while (rll->rll_writer != NULL || rll->rll_readers)
1095 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1096 		rll->rll_writer = curthread;
1097 	}
1098 
1099 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1100 }
1101 
1102 static void
1103 ztest_rll_unlock(rll_t *rll)
1104 {
1105 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1106 
1107 	if (rll->rll_writer) {
1108 		ASSERT(rll->rll_readers == 0);
1109 		rll->rll_writer = NULL;
1110 	} else {
1111 		ASSERT(rll->rll_readers != 0);
1112 		ASSERT(rll->rll_writer == NULL);
1113 		rll->rll_readers--;
1114 	}
1115 
1116 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1117 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1118 
1119 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1120 }
1121 
1122 static void
1123 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1124 {
1125 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1126 
1127 	ztest_rll_lock(rll, type);
1128 }
1129 
1130 static void
1131 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1132 {
1133 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1134 
1135 	ztest_rll_unlock(rll);
1136 }
1137 
1138 static rl_t *
1139 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1140     uint64_t size, rl_type_t type)
1141 {
1142 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1143 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1144 	rl_t *rl;
1145 
1146 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1147 	rl->rl_object = object;
1148 	rl->rl_offset = offset;
1149 	rl->rl_size = size;
1150 	rl->rl_lock = rll;
1151 
1152 	ztest_rll_lock(rll, type);
1153 
1154 	return (rl);
1155 }
1156 
1157 static void
1158 ztest_range_unlock(rl_t *rl)
1159 {
1160 	rll_t *rll = rl->rl_lock;
1161 
1162 	ztest_rll_unlock(rll);
1163 
1164 	umem_free(rl, sizeof (*rl));
1165 }
1166 
1167 static void
1168 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1169 {
1170 	zd->zd_os = os;
1171 	zd->zd_zilog = dmu_objset_zil(os);
1172 	zd->zd_shared = szd;
1173 	dmu_objset_name(os, zd->zd_name);
1174 
1175 	if (zd->zd_shared != NULL)
1176 		zd->zd_shared->zd_seq = 0;
1177 
1178 	VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1179 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1180 
1181 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1182 		ztest_rll_init(&zd->zd_object_lock[l]);
1183 
1184 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1185 		ztest_rll_init(&zd->zd_range_lock[l]);
1186 }
1187 
1188 static void
1189 ztest_zd_fini(ztest_ds_t *zd)
1190 {
1191 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1192 
1193 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1194 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1195 
1196 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1197 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1198 }
1199 
1200 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1201 
1202 static uint64_t
1203 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1204 {
1205 	uint64_t txg;
1206 	int error;
1207 
1208 	/*
1209 	 * Attempt to assign tx to some transaction group.
1210 	 */
1211 	error = dmu_tx_assign(tx, txg_how);
1212 	if (error) {
1213 		if (error == ERESTART) {
1214 			ASSERT(txg_how == TXG_NOWAIT);
1215 			dmu_tx_wait(tx);
1216 		} else {
1217 			ASSERT3U(error, ==, ENOSPC);
1218 			ztest_record_enospc(tag);
1219 		}
1220 		dmu_tx_abort(tx);
1221 		return (0);
1222 	}
1223 	txg = dmu_tx_get_txg(tx);
1224 	ASSERT(txg != 0);
1225 	return (txg);
1226 }
1227 
1228 static void
1229 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1230 {
1231 	uint64_t *ip = buf;
1232 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1233 
1234 	while (ip < ip_end)
1235 		*ip++ = value;
1236 }
1237 
1238 static boolean_t
1239 ztest_pattern_match(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 	uint64_t diff = 0;
1244 
1245 	while (ip < ip_end)
1246 		diff |= (value - *ip++);
1247 
1248 	return (diff == 0);
1249 }
1250 
1251 static void
1252 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1253     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1254 {
1255 	bt->bt_magic = BT_MAGIC;
1256 	bt->bt_objset = dmu_objset_id(os);
1257 	bt->bt_object = object;
1258 	bt->bt_offset = offset;
1259 	bt->bt_gen = gen;
1260 	bt->bt_txg = txg;
1261 	bt->bt_crtxg = crtxg;
1262 }
1263 
1264 static void
1265 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1266     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1267 {
1268 	ASSERT(bt->bt_magic == BT_MAGIC);
1269 	ASSERT(bt->bt_objset == dmu_objset_id(os));
1270 	ASSERT(bt->bt_object == object);
1271 	ASSERT(bt->bt_offset == offset);
1272 	ASSERT(bt->bt_gen <= gen);
1273 	ASSERT(bt->bt_txg <= txg);
1274 	ASSERT(bt->bt_crtxg == crtxg);
1275 }
1276 
1277 static ztest_block_tag_t *
1278 ztest_bt_bonus(dmu_buf_t *db)
1279 {
1280 	dmu_object_info_t doi;
1281 	ztest_block_tag_t *bt;
1282 
1283 	dmu_object_info_from_db(db, &doi);
1284 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1285 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1286 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1287 
1288 	return (bt);
1289 }
1290 
1291 /*
1292  * ZIL logging ops
1293  */
1294 
1295 #define	lrz_type	lr_mode
1296 #define	lrz_blocksize	lr_uid
1297 #define	lrz_ibshift	lr_gid
1298 #define	lrz_bonustype	lr_rdev
1299 #define	lrz_bonuslen	lr_crtime[1]
1300 
1301 static void
1302 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1303 {
1304 	char *name = (void *)(lr + 1);		/* name follows lr */
1305 	size_t namesize = strlen(name) + 1;
1306 	itx_t *itx;
1307 
1308 	if (zil_replaying(zd->zd_zilog, tx))
1309 		return;
1310 
1311 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1312 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1313 	    sizeof (*lr) + namesize - sizeof (lr_t));
1314 
1315 	zil_itx_assign(zd->zd_zilog, itx, tx);
1316 }
1317 
1318 static void
1319 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1320 {
1321 	char *name = (void *)(lr + 1);		/* name follows lr */
1322 	size_t namesize = strlen(name) + 1;
1323 	itx_t *itx;
1324 
1325 	if (zil_replaying(zd->zd_zilog, tx))
1326 		return;
1327 
1328 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1329 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1330 	    sizeof (*lr) + namesize - sizeof (lr_t));
1331 
1332 	itx->itx_oid = object;
1333 	zil_itx_assign(zd->zd_zilog, itx, tx);
1334 }
1335 
1336 static void
1337 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1338 {
1339 	itx_t *itx;
1340 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1341 
1342 	if (zil_replaying(zd->zd_zilog, tx))
1343 		return;
1344 
1345 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1346 		write_state = WR_INDIRECT;
1347 
1348 	itx = zil_itx_create(TX_WRITE,
1349 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1350 
1351 	if (write_state == WR_COPIED &&
1352 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1353 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1354 		zil_itx_destroy(itx);
1355 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1356 		write_state = WR_NEED_COPY;
1357 	}
1358 	itx->itx_private = zd;
1359 	itx->itx_wr_state = write_state;
1360 	itx->itx_sync = (ztest_random(8) == 0);
1361 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1362 
1363 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1364 	    sizeof (*lr) - sizeof (lr_t));
1365 
1366 	zil_itx_assign(zd->zd_zilog, itx, tx);
1367 }
1368 
1369 static void
1370 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1371 {
1372 	itx_t *itx;
1373 
1374 	if (zil_replaying(zd->zd_zilog, tx))
1375 		return;
1376 
1377 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1378 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1379 	    sizeof (*lr) - sizeof (lr_t));
1380 
1381 	itx->itx_sync = B_FALSE;
1382 	zil_itx_assign(zd->zd_zilog, itx, tx);
1383 }
1384 
1385 static void
1386 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1387 {
1388 	itx_t *itx;
1389 
1390 	if (zil_replaying(zd->zd_zilog, tx))
1391 		return;
1392 
1393 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1394 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1395 	    sizeof (*lr) - sizeof (lr_t));
1396 
1397 	itx->itx_sync = B_FALSE;
1398 	zil_itx_assign(zd->zd_zilog, itx, tx);
1399 }
1400 
1401 /*
1402  * ZIL replay ops
1403  */
1404 static int
1405 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1406 {
1407 	char *name = (void *)(lr + 1);		/* name follows lr */
1408 	objset_t *os = zd->zd_os;
1409 	ztest_block_tag_t *bbt;
1410 	dmu_buf_t *db;
1411 	dmu_tx_t *tx;
1412 	uint64_t txg;
1413 	int error = 0;
1414 
1415 	if (byteswap)
1416 		byteswap_uint64_array(lr, sizeof (*lr));
1417 
1418 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1419 	ASSERT(name[0] != '\0');
1420 
1421 	tx = dmu_tx_create(os);
1422 
1423 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1424 
1425 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1426 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1427 	} else {
1428 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1429 	}
1430 
1431 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1432 	if (txg == 0)
1433 		return (ENOSPC);
1434 
1435 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1436 
1437 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1438 		if (lr->lr_foid == 0) {
1439 			lr->lr_foid = zap_create(os,
1440 			    lr->lrz_type, lr->lrz_bonustype,
1441 			    lr->lrz_bonuslen, tx);
1442 		} else {
1443 			error = zap_create_claim(os, lr->lr_foid,
1444 			    lr->lrz_type, lr->lrz_bonustype,
1445 			    lr->lrz_bonuslen, tx);
1446 		}
1447 	} else {
1448 		if (lr->lr_foid == 0) {
1449 			lr->lr_foid = dmu_object_alloc(os,
1450 			    lr->lrz_type, 0, lr->lrz_bonustype,
1451 			    lr->lrz_bonuslen, tx);
1452 		} else {
1453 			error = dmu_object_claim(os, lr->lr_foid,
1454 			    lr->lrz_type, 0, lr->lrz_bonustype,
1455 			    lr->lrz_bonuslen, tx);
1456 		}
1457 	}
1458 
1459 	if (error) {
1460 		ASSERT3U(error, ==, EEXIST);
1461 		ASSERT(zd->zd_zilog->zl_replay);
1462 		dmu_tx_commit(tx);
1463 		return (error);
1464 	}
1465 
1466 	ASSERT(lr->lr_foid != 0);
1467 
1468 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1469 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1470 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1471 
1472 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1473 	bbt = ztest_bt_bonus(db);
1474 	dmu_buf_will_dirty(db, tx);
1475 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1476 	dmu_buf_rele(db, FTAG);
1477 
1478 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1479 	    &lr->lr_foid, tx));
1480 
1481 	(void) ztest_log_create(zd, tx, lr);
1482 
1483 	dmu_tx_commit(tx);
1484 
1485 	return (0);
1486 }
1487 
1488 static int
1489 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1490 {
1491 	char *name = (void *)(lr + 1);		/* name follows lr */
1492 	objset_t *os = zd->zd_os;
1493 	dmu_object_info_t doi;
1494 	dmu_tx_t *tx;
1495 	uint64_t object, txg;
1496 
1497 	if (byteswap)
1498 		byteswap_uint64_array(lr, sizeof (*lr));
1499 
1500 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1501 	ASSERT(name[0] != '\0');
1502 
1503 	VERIFY3U(0, ==,
1504 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1505 	ASSERT(object != 0);
1506 
1507 	ztest_object_lock(zd, object, RL_WRITER);
1508 
1509 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1510 
1511 	tx = dmu_tx_create(os);
1512 
1513 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1514 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1515 
1516 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1517 	if (txg == 0) {
1518 		ztest_object_unlock(zd, object);
1519 		return (ENOSPC);
1520 	}
1521 
1522 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1523 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1524 	} else {
1525 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1526 	}
1527 
1528 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1529 
1530 	(void) ztest_log_remove(zd, tx, lr, object);
1531 
1532 	dmu_tx_commit(tx);
1533 
1534 	ztest_object_unlock(zd, object);
1535 
1536 	return (0);
1537 }
1538 
1539 static int
1540 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1541 {
1542 	objset_t *os = zd->zd_os;
1543 	void *data = lr + 1;			/* data follows lr */
1544 	uint64_t offset, length;
1545 	ztest_block_tag_t *bt = data;
1546 	ztest_block_tag_t *bbt;
1547 	uint64_t gen, txg, lrtxg, crtxg;
1548 	dmu_object_info_t doi;
1549 	dmu_tx_t *tx;
1550 	dmu_buf_t *db;
1551 	arc_buf_t *abuf = NULL;
1552 	rl_t *rl;
1553 
1554 	if (byteswap)
1555 		byteswap_uint64_array(lr, sizeof (*lr));
1556 
1557 	offset = lr->lr_offset;
1558 	length = lr->lr_length;
1559 
1560 	/* If it's a dmu_sync() block, write the whole block */
1561 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1562 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1563 		if (length < blocksize) {
1564 			offset -= offset % blocksize;
1565 			length = blocksize;
1566 		}
1567 	}
1568 
1569 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1570 		byteswap_uint64_array(bt, sizeof (*bt));
1571 
1572 	if (bt->bt_magic != BT_MAGIC)
1573 		bt = NULL;
1574 
1575 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1576 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1577 
1578 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1579 
1580 	dmu_object_info_from_db(db, &doi);
1581 
1582 	bbt = ztest_bt_bonus(db);
1583 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1584 	gen = bbt->bt_gen;
1585 	crtxg = bbt->bt_crtxg;
1586 	lrtxg = lr->lr_common.lrc_txg;
1587 
1588 	tx = dmu_tx_create(os);
1589 
1590 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1591 
1592 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1593 	    P2PHASE(offset, length) == 0)
1594 		abuf = dmu_request_arcbuf(db, length);
1595 
1596 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1597 	if (txg == 0) {
1598 		if (abuf != NULL)
1599 			dmu_return_arcbuf(abuf);
1600 		dmu_buf_rele(db, FTAG);
1601 		ztest_range_unlock(rl);
1602 		ztest_object_unlock(zd, lr->lr_foid);
1603 		return (ENOSPC);
1604 	}
1605 
1606 	if (bt != NULL) {
1607 		/*
1608 		 * Usually, verify the old data before writing new data --
1609 		 * but not always, because we also want to verify correct
1610 		 * behavior when the data was not recently read into cache.
1611 		 */
1612 		ASSERT(offset % doi.doi_data_block_size == 0);
1613 		if (ztest_random(4) != 0) {
1614 			int prefetch = ztest_random(2) ?
1615 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1616 			ztest_block_tag_t rbt;
1617 
1618 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1619 			    sizeof (rbt), &rbt, prefetch) == 0);
1620 			if (rbt.bt_magic == BT_MAGIC) {
1621 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1622 				    offset, gen, txg, crtxg);
1623 			}
1624 		}
1625 
1626 		/*
1627 		 * Writes can appear to be newer than the bonus buffer because
1628 		 * the ztest_get_data() callback does a dmu_read() of the
1629 		 * open-context data, which may be different than the data
1630 		 * as it was when the write was generated.
1631 		 */
1632 		if (zd->zd_zilog->zl_replay) {
1633 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1634 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1635 			    bt->bt_crtxg);
1636 		}
1637 
1638 		/*
1639 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1640 		 * so that all of the usual ASSERTs will work.
1641 		 */
1642 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1643 	}
1644 
1645 	if (abuf == NULL) {
1646 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1647 	} else {
1648 		bcopy(data, abuf->b_data, length);
1649 		dmu_assign_arcbuf(db, offset, abuf, tx);
1650 	}
1651 
1652 	(void) ztest_log_write(zd, tx, lr);
1653 
1654 	dmu_buf_rele(db, FTAG);
1655 
1656 	dmu_tx_commit(tx);
1657 
1658 	ztest_range_unlock(rl);
1659 	ztest_object_unlock(zd, lr->lr_foid);
1660 
1661 	return (0);
1662 }
1663 
1664 static int
1665 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1666 {
1667 	objset_t *os = zd->zd_os;
1668 	dmu_tx_t *tx;
1669 	uint64_t txg;
1670 	rl_t *rl;
1671 
1672 	if (byteswap)
1673 		byteswap_uint64_array(lr, sizeof (*lr));
1674 
1675 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1676 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1677 	    RL_WRITER);
1678 
1679 	tx = dmu_tx_create(os);
1680 
1681 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1682 
1683 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1684 	if (txg == 0) {
1685 		ztest_range_unlock(rl);
1686 		ztest_object_unlock(zd, lr->lr_foid);
1687 		return (ENOSPC);
1688 	}
1689 
1690 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1691 	    lr->lr_length, tx) == 0);
1692 
1693 	(void) ztest_log_truncate(zd, tx, lr);
1694 
1695 	dmu_tx_commit(tx);
1696 
1697 	ztest_range_unlock(rl);
1698 	ztest_object_unlock(zd, lr->lr_foid);
1699 
1700 	return (0);
1701 }
1702 
1703 static int
1704 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1705 {
1706 	objset_t *os = zd->zd_os;
1707 	dmu_tx_t *tx;
1708 	dmu_buf_t *db;
1709 	ztest_block_tag_t *bbt;
1710 	uint64_t txg, lrtxg, crtxg;
1711 
1712 	if (byteswap)
1713 		byteswap_uint64_array(lr, sizeof (*lr));
1714 
1715 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1716 
1717 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1718 
1719 	tx = dmu_tx_create(os);
1720 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1721 
1722 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1723 	if (txg == 0) {
1724 		dmu_buf_rele(db, FTAG);
1725 		ztest_object_unlock(zd, lr->lr_foid);
1726 		return (ENOSPC);
1727 	}
1728 
1729 	bbt = ztest_bt_bonus(db);
1730 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1731 	crtxg = bbt->bt_crtxg;
1732 	lrtxg = lr->lr_common.lrc_txg;
1733 
1734 	if (zd->zd_zilog->zl_replay) {
1735 		ASSERT(lr->lr_size != 0);
1736 		ASSERT(lr->lr_mode != 0);
1737 		ASSERT(lrtxg != 0);
1738 	} else {
1739 		/*
1740 		 * Randomly change the size and increment the generation.
1741 		 */
1742 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1743 		    sizeof (*bbt);
1744 		lr->lr_mode = bbt->bt_gen + 1;
1745 		ASSERT(lrtxg == 0);
1746 	}
1747 
1748 	/*
1749 	 * Verify that the current bonus buffer is not newer than our txg.
1750 	 */
1751 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1752 	    MAX(txg, lrtxg), crtxg);
1753 
1754 	dmu_buf_will_dirty(db, tx);
1755 
1756 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1757 	ASSERT3U(lr->lr_size, <=, db->db_size);
1758 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1759 	bbt = ztest_bt_bonus(db);
1760 
1761 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1762 
1763 	dmu_buf_rele(db, FTAG);
1764 
1765 	(void) ztest_log_setattr(zd, tx, lr);
1766 
1767 	dmu_tx_commit(tx);
1768 
1769 	ztest_object_unlock(zd, lr->lr_foid);
1770 
1771 	return (0);
1772 }
1773 
1774 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1775 	NULL,			/* 0 no such transaction type */
1776 	ztest_replay_create,	/* TX_CREATE */
1777 	NULL,			/* TX_MKDIR */
1778 	NULL,			/* TX_MKXATTR */
1779 	NULL,			/* TX_SYMLINK */
1780 	ztest_replay_remove,	/* TX_REMOVE */
1781 	NULL,			/* TX_RMDIR */
1782 	NULL,			/* TX_LINK */
1783 	NULL,			/* TX_RENAME */
1784 	ztest_replay_write,	/* TX_WRITE */
1785 	ztest_replay_truncate,	/* TX_TRUNCATE */
1786 	ztest_replay_setattr,	/* TX_SETATTR */
1787 	NULL,			/* TX_ACL */
1788 	NULL,			/* TX_CREATE_ACL */
1789 	NULL,			/* TX_CREATE_ATTR */
1790 	NULL,			/* TX_CREATE_ACL_ATTR */
1791 	NULL,			/* TX_MKDIR_ACL */
1792 	NULL,			/* TX_MKDIR_ATTR */
1793 	NULL,			/* TX_MKDIR_ACL_ATTR */
1794 	NULL,			/* TX_WRITE2 */
1795 };
1796 
1797 /*
1798  * ZIL get_data callbacks
1799  */
1800 
1801 static void
1802 ztest_get_done(zgd_t *zgd, int error)
1803 {
1804 	ztest_ds_t *zd = zgd->zgd_private;
1805 	uint64_t object = zgd->zgd_rl->rl_object;
1806 
1807 	if (zgd->zgd_db)
1808 		dmu_buf_rele(zgd->zgd_db, zgd);
1809 
1810 	ztest_range_unlock(zgd->zgd_rl);
1811 	ztest_object_unlock(zd, object);
1812 
1813 	if (error == 0 && zgd->zgd_bp)
1814 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1815 
1816 	umem_free(zgd, sizeof (*zgd));
1817 }
1818 
1819 static int
1820 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1821 {
1822 	ztest_ds_t *zd = arg;
1823 	objset_t *os = zd->zd_os;
1824 	uint64_t object = lr->lr_foid;
1825 	uint64_t offset = lr->lr_offset;
1826 	uint64_t size = lr->lr_length;
1827 	blkptr_t *bp = &lr->lr_blkptr;
1828 	uint64_t txg = lr->lr_common.lrc_txg;
1829 	uint64_t crtxg;
1830 	dmu_object_info_t doi;
1831 	dmu_buf_t *db;
1832 	zgd_t *zgd;
1833 	int error;
1834 
1835 	ztest_object_lock(zd, object, RL_READER);
1836 	error = dmu_bonus_hold(os, object, FTAG, &db);
1837 	if (error) {
1838 		ztest_object_unlock(zd, object);
1839 		return (error);
1840 	}
1841 
1842 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1843 
1844 	if (crtxg == 0 || crtxg > txg) {
1845 		dmu_buf_rele(db, FTAG);
1846 		ztest_object_unlock(zd, object);
1847 		return (ENOENT);
1848 	}
1849 
1850 	dmu_object_info_from_db(db, &doi);
1851 	dmu_buf_rele(db, FTAG);
1852 	db = NULL;
1853 
1854 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1855 	zgd->zgd_zilog = zd->zd_zilog;
1856 	zgd->zgd_private = zd;
1857 
1858 	if (buf != NULL) {	/* immediate write */
1859 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1860 		    RL_READER);
1861 
1862 		error = dmu_read(os, object, offset, size, buf,
1863 		    DMU_READ_NO_PREFETCH);
1864 		ASSERT(error == 0);
1865 	} else {
1866 		size = doi.doi_data_block_size;
1867 		if (ISP2(size)) {
1868 			offset = P2ALIGN(offset, size);
1869 		} else {
1870 			ASSERT(offset < size);
1871 			offset = 0;
1872 		}
1873 
1874 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1875 		    RL_READER);
1876 
1877 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1878 		    DMU_READ_NO_PREFETCH);
1879 
1880 		if (error == 0) {
1881 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1882 			if (obp) {
1883 				ASSERT(BP_IS_HOLE(bp));
1884 				*bp = *obp;
1885 			}
1886 
1887 			zgd->zgd_db = db;
1888 			zgd->zgd_bp = bp;
1889 
1890 			ASSERT(db->db_offset == offset);
1891 			ASSERT(db->db_size == size);
1892 
1893 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1894 			    ztest_get_done, zgd);
1895 
1896 			if (error == 0)
1897 				return (0);
1898 		}
1899 	}
1900 
1901 	ztest_get_done(zgd, error);
1902 
1903 	return (error);
1904 }
1905 
1906 static void *
1907 ztest_lr_alloc(size_t lrsize, char *name)
1908 {
1909 	char *lr;
1910 	size_t namesize = name ? strlen(name) + 1 : 0;
1911 
1912 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1913 
1914 	if (name)
1915 		bcopy(name, lr + lrsize, namesize);
1916 
1917 	return (lr);
1918 }
1919 
1920 void
1921 ztest_lr_free(void *lr, size_t lrsize, char *name)
1922 {
1923 	size_t namesize = name ? strlen(name) + 1 : 0;
1924 
1925 	umem_free(lr, lrsize + namesize);
1926 }
1927 
1928 /*
1929  * Lookup a bunch of objects.  Returns the number of objects not found.
1930  */
1931 static int
1932 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1933 {
1934 	int missing = 0;
1935 	int error;
1936 
1937 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1938 
1939 	for (int i = 0; i < count; i++, od++) {
1940 		od->od_object = 0;
1941 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1942 		    sizeof (uint64_t), 1, &od->od_object);
1943 		if (error) {
1944 			ASSERT(error == ENOENT);
1945 			ASSERT(od->od_object == 0);
1946 			missing++;
1947 		} else {
1948 			dmu_buf_t *db;
1949 			ztest_block_tag_t *bbt;
1950 			dmu_object_info_t doi;
1951 
1952 			ASSERT(od->od_object != 0);
1953 			ASSERT(missing == 0);	/* there should be no gaps */
1954 
1955 			ztest_object_lock(zd, od->od_object, RL_READER);
1956 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1957 			    od->od_object, FTAG, &db));
1958 			dmu_object_info_from_db(db, &doi);
1959 			bbt = ztest_bt_bonus(db);
1960 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1961 			od->od_type = doi.doi_type;
1962 			od->od_blocksize = doi.doi_data_block_size;
1963 			od->od_gen = bbt->bt_gen;
1964 			dmu_buf_rele(db, FTAG);
1965 			ztest_object_unlock(zd, od->od_object);
1966 		}
1967 	}
1968 
1969 	return (missing);
1970 }
1971 
1972 static int
1973 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1974 {
1975 	int missing = 0;
1976 
1977 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1978 
1979 	for (int i = 0; i < count; i++, od++) {
1980 		if (missing) {
1981 			od->od_object = 0;
1982 			missing++;
1983 			continue;
1984 		}
1985 
1986 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1987 
1988 		lr->lr_doid = od->od_dir;
1989 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
1990 		lr->lrz_type = od->od_crtype;
1991 		lr->lrz_blocksize = od->od_crblocksize;
1992 		lr->lrz_ibshift = ztest_random_ibshift();
1993 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1994 		lr->lrz_bonuslen = dmu_bonus_max();
1995 		lr->lr_gen = od->od_crgen;
1996 		lr->lr_crtime[0] = time(NULL);
1997 
1998 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1999 			ASSERT(missing == 0);
2000 			od->od_object = 0;
2001 			missing++;
2002 		} else {
2003 			od->od_object = lr->lr_foid;
2004 			od->od_type = od->od_crtype;
2005 			od->od_blocksize = od->od_crblocksize;
2006 			od->od_gen = od->od_crgen;
2007 			ASSERT(od->od_object != 0);
2008 		}
2009 
2010 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2011 	}
2012 
2013 	return (missing);
2014 }
2015 
2016 static int
2017 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2018 {
2019 	int missing = 0;
2020 	int error;
2021 
2022 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2023 
2024 	od += count - 1;
2025 
2026 	for (int i = count - 1; i >= 0; i--, od--) {
2027 		if (missing) {
2028 			missing++;
2029 			continue;
2030 		}
2031 
2032 		/*
2033 		 * No object was found.
2034 		 */
2035 		if (od->od_object == 0)
2036 			continue;
2037 
2038 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2039 
2040 		lr->lr_doid = od->od_dir;
2041 
2042 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2043 			ASSERT3U(error, ==, ENOSPC);
2044 			missing++;
2045 		} else {
2046 			od->od_object = 0;
2047 		}
2048 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2049 	}
2050 
2051 	return (missing);
2052 }
2053 
2054 static int
2055 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2056     void *data)
2057 {
2058 	lr_write_t *lr;
2059 	int error;
2060 
2061 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2062 
2063 	lr->lr_foid = object;
2064 	lr->lr_offset = offset;
2065 	lr->lr_length = size;
2066 	lr->lr_blkoff = 0;
2067 	BP_ZERO(&lr->lr_blkptr);
2068 
2069 	bcopy(data, lr + 1, size);
2070 
2071 	error = ztest_replay_write(zd, lr, B_FALSE);
2072 
2073 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2074 
2075 	return (error);
2076 }
2077 
2078 static int
2079 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2080 {
2081 	lr_truncate_t *lr;
2082 	int error;
2083 
2084 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2085 
2086 	lr->lr_foid = object;
2087 	lr->lr_offset = offset;
2088 	lr->lr_length = size;
2089 
2090 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2091 
2092 	ztest_lr_free(lr, sizeof (*lr), NULL);
2093 
2094 	return (error);
2095 }
2096 
2097 static int
2098 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2099 {
2100 	lr_setattr_t *lr;
2101 	int error;
2102 
2103 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2104 
2105 	lr->lr_foid = object;
2106 	lr->lr_size = 0;
2107 	lr->lr_mode = 0;
2108 
2109 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2110 
2111 	ztest_lr_free(lr, sizeof (*lr), NULL);
2112 
2113 	return (error);
2114 }
2115 
2116 static void
2117 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2118 {
2119 	objset_t *os = zd->zd_os;
2120 	dmu_tx_t *tx;
2121 	uint64_t txg;
2122 	rl_t *rl;
2123 
2124 	txg_wait_synced(dmu_objset_pool(os), 0);
2125 
2126 	ztest_object_lock(zd, object, RL_READER);
2127 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2128 
2129 	tx = dmu_tx_create(os);
2130 
2131 	dmu_tx_hold_write(tx, object, offset, size);
2132 
2133 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2134 
2135 	if (txg != 0) {
2136 		dmu_prealloc(os, object, offset, size, tx);
2137 		dmu_tx_commit(tx);
2138 		txg_wait_synced(dmu_objset_pool(os), txg);
2139 	} else {
2140 		(void) dmu_free_long_range(os, object, offset, size);
2141 	}
2142 
2143 	ztest_range_unlock(rl);
2144 	ztest_object_unlock(zd, object);
2145 }
2146 
2147 static void
2148 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2149 {
2150 	int err;
2151 	ztest_block_tag_t wbt;
2152 	dmu_object_info_t doi;
2153 	enum ztest_io_type io_type;
2154 	uint64_t blocksize;
2155 	void *data;
2156 
2157 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2158 	blocksize = doi.doi_data_block_size;
2159 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2160 
2161 	/*
2162 	 * Pick an i/o type at random, biased toward writing block tags.
2163 	 */
2164 	io_type = ztest_random(ZTEST_IO_TYPES);
2165 	if (ztest_random(2) == 0)
2166 		io_type = ZTEST_IO_WRITE_TAG;
2167 
2168 	(void) rw_rdlock(&zd->zd_zilog_lock);
2169 
2170 	switch (io_type) {
2171 
2172 	case ZTEST_IO_WRITE_TAG:
2173 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2174 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2175 		break;
2176 
2177 	case ZTEST_IO_WRITE_PATTERN:
2178 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2179 		if (ztest_random(2) == 0) {
2180 			/*
2181 			 * Induce fletcher2 collisions to ensure that
2182 			 * zio_ddt_collision() detects and resolves them
2183 			 * when using fletcher2-verify for deduplication.
2184 			 */
2185 			((uint64_t *)data)[0] ^= 1ULL << 63;
2186 			((uint64_t *)data)[4] ^= 1ULL << 63;
2187 		}
2188 		(void) ztest_write(zd, object, offset, blocksize, data);
2189 		break;
2190 
2191 	case ZTEST_IO_WRITE_ZEROES:
2192 		bzero(data, blocksize);
2193 		(void) ztest_write(zd, object, offset, blocksize, data);
2194 		break;
2195 
2196 	case ZTEST_IO_TRUNCATE:
2197 		(void) ztest_truncate(zd, object, offset, blocksize);
2198 		break;
2199 
2200 	case ZTEST_IO_SETATTR:
2201 		(void) ztest_setattr(zd, object);
2202 		break;
2203 
2204 	case ZTEST_IO_REWRITE:
2205 		(void) rw_rdlock(&ztest_name_lock);
2206 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2207 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2208 		    B_FALSE);
2209 		VERIFY(err == 0 || err == ENOSPC);
2210 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2211 		    ZFS_PROP_COMPRESSION,
2212 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2213 		    B_FALSE);
2214 		VERIFY(err == 0 || err == ENOSPC);
2215 		(void) rw_unlock(&ztest_name_lock);
2216 
2217 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2218 		    DMU_READ_NO_PREFETCH));
2219 
2220 		(void) ztest_write(zd, object, offset, blocksize, data);
2221 		break;
2222 	}
2223 
2224 	(void) rw_unlock(&zd->zd_zilog_lock);
2225 
2226 	umem_free(data, blocksize);
2227 }
2228 
2229 /*
2230  * Initialize an object description template.
2231  */
2232 static void
2233 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2234     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2235 {
2236 	od->od_dir = ZTEST_DIROBJ;
2237 	od->od_object = 0;
2238 
2239 	od->od_crtype = type;
2240 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2241 	od->od_crgen = gen;
2242 
2243 	od->od_type = DMU_OT_NONE;
2244 	od->od_blocksize = 0;
2245 	od->od_gen = 0;
2246 
2247 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2248 	    tag, (int64_t)id, index);
2249 }
2250 
2251 /*
2252  * Lookup or create the objects for a test using the od template.
2253  * If the objects do not all exist, or if 'remove' is specified,
2254  * remove any existing objects and create new ones.  Otherwise,
2255  * use the existing objects.
2256  */
2257 static int
2258 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2259 {
2260 	int count = size / sizeof (*od);
2261 	int rv = 0;
2262 
2263 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2264 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2265 	    (ztest_remove(zd, od, count) != 0 ||
2266 	    ztest_create(zd, od, count) != 0))
2267 		rv = -1;
2268 	zd->zd_od = od;
2269 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2270 
2271 	return (rv);
2272 }
2273 
2274 /* ARGSUSED */
2275 void
2276 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2277 {
2278 	zilog_t *zilog = zd->zd_zilog;
2279 
2280 	(void) rw_rdlock(&zd->zd_zilog_lock);
2281 
2282 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2283 
2284 	/*
2285 	 * Remember the committed values in zd, which is in parent/child
2286 	 * shared memory.  If we die, the next iteration of ztest_run()
2287 	 * will verify that the log really does contain this record.
2288 	 */
2289 	mutex_enter(&zilog->zl_lock);
2290 	ASSERT(zd->zd_shared != NULL);
2291 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2292 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2293 	mutex_exit(&zilog->zl_lock);
2294 
2295 	(void) rw_unlock(&zd->zd_zilog_lock);
2296 }
2297 
2298 /*
2299  * This function is designed to simulate the operations that occur during a
2300  * mount/unmount operation.  We hold the dataset across these operations in an
2301  * attempt to expose any implicit assumptions about ZIL management.
2302  */
2303 /* ARGSUSED */
2304 void
2305 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2306 {
2307 	objset_t *os = zd->zd_os;
2308 
2309 	/*
2310 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2311 	 * updating the zil (i.e. adding in-memory log records) and the
2312 	 * zd_zilog_lock to block any I/O.
2313 	 */
2314 	VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2315 	(void) rw_wrlock(&zd->zd_zilog_lock);
2316 
2317 	/* zfsvfs_teardown() */
2318 	zil_close(zd->zd_zilog);
2319 
2320 	/* zfsvfs_setup() */
2321 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2322 	zil_replay(os, zd, ztest_replay_vector);
2323 
2324 	(void) rw_unlock(&zd->zd_zilog_lock);
2325 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2326 }
2327 
2328 /*
2329  * Verify that we can't destroy an active pool, create an existing pool,
2330  * or create a pool with a bad vdev spec.
2331  */
2332 /* ARGSUSED */
2333 void
2334 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2335 {
2336 	ztest_shared_opts_t *zo = &ztest_opts;
2337 	spa_t *spa;
2338 	nvlist_t *nvroot;
2339 
2340 	/*
2341 	 * Attempt to create using a bad file.
2342 	 */
2343 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2344 	VERIFY3U(ENOENT, ==,
2345 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2346 	nvlist_free(nvroot);
2347 
2348 	/*
2349 	 * Attempt to create using a bad mirror.
2350 	 */
2351 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2352 	VERIFY3U(ENOENT, ==,
2353 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2354 	nvlist_free(nvroot);
2355 
2356 	/*
2357 	 * Attempt to create an existing pool.  It shouldn't matter
2358 	 * what's in the nvroot; we should fail with EEXIST.
2359 	 */
2360 	(void) rw_rdlock(&ztest_name_lock);
2361 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2362 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2363 	nvlist_free(nvroot);
2364 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2365 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2366 	spa_close(spa, FTAG);
2367 
2368 	(void) rw_unlock(&ztest_name_lock);
2369 }
2370 
2371 /* ARGSUSED */
2372 void
2373 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2374 {
2375 	spa_t *spa;
2376 	uint64_t initial_version = SPA_VERSION_INITIAL;
2377 	uint64_t version, newversion;
2378 	nvlist_t *nvroot, *props;
2379 	char *name;
2380 
2381 	VERIFY0(mutex_lock(&ztest_vdev_lock));
2382 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2383 
2384 	/*
2385 	 * Clean up from previous runs.
2386 	 */
2387 	(void) spa_destroy(name);
2388 
2389 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2390 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2391 
2392 	/*
2393 	 * If we're configuring a RAIDZ device then make sure that the
2394 	 * the initial version is capable of supporting that feature.
2395 	 */
2396 	switch (ztest_opts.zo_raidz_parity) {
2397 	case 0:
2398 	case 1:
2399 		initial_version = SPA_VERSION_INITIAL;
2400 		break;
2401 	case 2:
2402 		initial_version = SPA_VERSION_RAIDZ2;
2403 		break;
2404 	case 3:
2405 		initial_version = SPA_VERSION_RAIDZ3;
2406 		break;
2407 	}
2408 
2409 	/*
2410 	 * Create a pool with a spa version that can be upgraded. Pick
2411 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2412 	 */
2413 	do {
2414 		version = ztest_random_spa_version(initial_version);
2415 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2416 
2417 	props = fnvlist_alloc();
2418 	fnvlist_add_uint64(props,
2419 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2420 	VERIFY0(spa_create(name, nvroot, props, NULL));
2421 	fnvlist_free(nvroot);
2422 	fnvlist_free(props);
2423 
2424 	VERIFY0(spa_open(name, &spa, FTAG));
2425 	VERIFY3U(spa_version(spa), ==, version);
2426 	newversion = ztest_random_spa_version(version + 1);
2427 
2428 	if (ztest_opts.zo_verbose >= 4) {
2429 		(void) printf("upgrading spa version from %llu to %llu\n",
2430 		    (u_longlong_t)version, (u_longlong_t)newversion);
2431 	}
2432 
2433 	spa_upgrade(spa, newversion);
2434 	VERIFY3U(spa_version(spa), >, version);
2435 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2436 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2437 	spa_close(spa, FTAG);
2438 
2439 	strfree(name);
2440 	VERIFY0(mutex_unlock(&ztest_vdev_lock));
2441 }
2442 
2443 static vdev_t *
2444 vdev_lookup_by_path(vdev_t *vd, const char *path)
2445 {
2446 	vdev_t *mvd;
2447 
2448 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2449 		return (vd);
2450 
2451 	for (int c = 0; c < vd->vdev_children; c++)
2452 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2453 		    NULL)
2454 			return (mvd);
2455 
2456 	return (NULL);
2457 }
2458 
2459 /*
2460  * Find the first available hole which can be used as a top-level.
2461  */
2462 int
2463 find_vdev_hole(spa_t *spa)
2464 {
2465 	vdev_t *rvd = spa->spa_root_vdev;
2466 	int c;
2467 
2468 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2469 
2470 	for (c = 0; c < rvd->vdev_children; c++) {
2471 		vdev_t *cvd = rvd->vdev_child[c];
2472 
2473 		if (cvd->vdev_ishole)
2474 			break;
2475 	}
2476 	return (c);
2477 }
2478 
2479 /*
2480  * Verify that vdev_add() works as expected.
2481  */
2482 /* ARGSUSED */
2483 void
2484 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2485 {
2486 	ztest_shared_t *zs = ztest_shared;
2487 	spa_t *spa = ztest_spa;
2488 	uint64_t leaves;
2489 	uint64_t guid;
2490 	nvlist_t *nvroot;
2491 	int error;
2492 
2493 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2494 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2495 
2496 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2497 
2498 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2499 
2500 	/*
2501 	 * If we have slogs then remove them 1/4 of the time.
2502 	 */
2503 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2504 		/*
2505 		 * Grab the guid from the head of the log class rotor.
2506 		 */
2507 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2508 
2509 		spa_config_exit(spa, SCL_VDEV, FTAG);
2510 
2511 		/*
2512 		 * We have to grab the zs_name_lock as writer to
2513 		 * prevent a race between removing a slog (dmu_objset_find)
2514 		 * and destroying a dataset. Removing the slog will
2515 		 * grab a reference on the dataset which may cause
2516 		 * dmu_objset_destroy() to fail with EBUSY thus
2517 		 * leaving the dataset in an inconsistent state.
2518 		 */
2519 		VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2520 		error = spa_vdev_remove(spa, guid, B_FALSE);
2521 		VERIFY(rw_unlock(&ztest_name_lock) == 0);
2522 
2523 		if (error && error != EEXIST)
2524 			fatal(0, "spa_vdev_remove() = %d", error);
2525 	} else {
2526 		spa_config_exit(spa, SCL_VDEV, FTAG);
2527 
2528 		/*
2529 		 * Make 1/4 of the devices be log devices.
2530 		 */
2531 		nvroot = make_vdev_root(NULL, NULL, NULL,
2532 		    ztest_opts.zo_vdev_size, 0,
2533 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2534 		    zs->zs_mirrors, 1);
2535 
2536 		error = spa_vdev_add(spa, nvroot);
2537 		nvlist_free(nvroot);
2538 
2539 		if (error == ENOSPC)
2540 			ztest_record_enospc("spa_vdev_add");
2541 		else if (error != 0)
2542 			fatal(0, "spa_vdev_add() = %d", error);
2543 	}
2544 
2545 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2546 }
2547 
2548 /*
2549  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2550  */
2551 /* ARGSUSED */
2552 void
2553 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2554 {
2555 	ztest_shared_t *zs = ztest_shared;
2556 	spa_t *spa = ztest_spa;
2557 	vdev_t *rvd = spa->spa_root_vdev;
2558 	spa_aux_vdev_t *sav;
2559 	char *aux;
2560 	uint64_t guid = 0;
2561 	int error;
2562 
2563 	if (ztest_random(2) == 0) {
2564 		sav = &spa->spa_spares;
2565 		aux = ZPOOL_CONFIG_SPARES;
2566 	} else {
2567 		sav = &spa->spa_l2cache;
2568 		aux = ZPOOL_CONFIG_L2CACHE;
2569 	}
2570 
2571 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2572 
2573 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2574 
2575 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2576 		/*
2577 		 * Pick a random device to remove.
2578 		 */
2579 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2580 	} else {
2581 		/*
2582 		 * Find an unused device we can add.
2583 		 */
2584 		zs->zs_vdev_aux = 0;
2585 		for (;;) {
2586 			char path[MAXPATHLEN];
2587 			int c;
2588 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2589 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2590 			    zs->zs_vdev_aux);
2591 			for (c = 0; c < sav->sav_count; c++)
2592 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2593 				    path) == 0)
2594 					break;
2595 			if (c == sav->sav_count &&
2596 			    vdev_lookup_by_path(rvd, path) == NULL)
2597 				break;
2598 			zs->zs_vdev_aux++;
2599 		}
2600 	}
2601 
2602 	spa_config_exit(spa, SCL_VDEV, FTAG);
2603 
2604 	if (guid == 0) {
2605 		/*
2606 		 * Add a new device.
2607 		 */
2608 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2609 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2610 		error = spa_vdev_add(spa, nvroot);
2611 		if (error != 0)
2612 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2613 		nvlist_free(nvroot);
2614 	} else {
2615 		/*
2616 		 * Remove an existing device.  Sometimes, dirty its
2617 		 * vdev state first to make sure we handle removal
2618 		 * of devices that have pending state changes.
2619 		 */
2620 		if (ztest_random(2) == 0)
2621 			(void) vdev_online(spa, guid, 0, NULL);
2622 
2623 		error = spa_vdev_remove(spa, guid, B_FALSE);
2624 		if (error != 0 && error != EBUSY)
2625 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2626 	}
2627 
2628 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2629 }
2630 
2631 /*
2632  * split a pool if it has mirror tlvdevs
2633  */
2634 /* ARGSUSED */
2635 void
2636 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2637 {
2638 	ztest_shared_t *zs = ztest_shared;
2639 	spa_t *spa = ztest_spa;
2640 	vdev_t *rvd = spa->spa_root_vdev;
2641 	nvlist_t *tree, **child, *config, *split, **schild;
2642 	uint_t c, children, schildren = 0, lastlogid = 0;
2643 	int error = 0;
2644 
2645 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2646 
2647 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2648 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2649 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2650 		return;
2651 	}
2652 
2653 	/* clean up the old pool, if any */
2654 	(void) spa_destroy("splitp");
2655 
2656 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2657 
2658 	/* generate a config from the existing config */
2659 	mutex_enter(&spa->spa_props_lock);
2660 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2661 	    &tree) == 0);
2662 	mutex_exit(&spa->spa_props_lock);
2663 
2664 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2665 	    &children) == 0);
2666 
2667 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2668 	for (c = 0; c < children; c++) {
2669 		vdev_t *tvd = rvd->vdev_child[c];
2670 		nvlist_t **mchild;
2671 		uint_t mchildren;
2672 
2673 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2674 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2675 			    0) == 0);
2676 			VERIFY(nvlist_add_string(schild[schildren],
2677 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2678 			VERIFY(nvlist_add_uint64(schild[schildren],
2679 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2680 			if (lastlogid == 0)
2681 				lastlogid = schildren;
2682 			++schildren;
2683 			continue;
2684 		}
2685 		lastlogid = 0;
2686 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2687 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2688 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2689 	}
2690 
2691 	/* OK, create a config that can be used to split */
2692 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2693 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2694 	    VDEV_TYPE_ROOT) == 0);
2695 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2696 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2697 
2698 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2699 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2700 
2701 	for (c = 0; c < schildren; c++)
2702 		nvlist_free(schild[c]);
2703 	free(schild);
2704 	nvlist_free(split);
2705 
2706 	spa_config_exit(spa, SCL_VDEV, FTAG);
2707 
2708 	(void) rw_wrlock(&ztest_name_lock);
2709 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2710 	(void) rw_unlock(&ztest_name_lock);
2711 
2712 	nvlist_free(config);
2713 
2714 	if (error == 0) {
2715 		(void) printf("successful split - results:\n");
2716 		mutex_enter(&spa_namespace_lock);
2717 		show_pool_stats(spa);
2718 		show_pool_stats(spa_lookup("splitp"));
2719 		mutex_exit(&spa_namespace_lock);
2720 		++zs->zs_splits;
2721 		--zs->zs_mirrors;
2722 	}
2723 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2724 
2725 }
2726 
2727 /*
2728  * Verify that we can attach and detach devices.
2729  */
2730 /* ARGSUSED */
2731 void
2732 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2733 {
2734 	ztest_shared_t *zs = ztest_shared;
2735 	spa_t *spa = ztest_spa;
2736 	spa_aux_vdev_t *sav = &spa->spa_spares;
2737 	vdev_t *rvd = spa->spa_root_vdev;
2738 	vdev_t *oldvd, *newvd, *pvd;
2739 	nvlist_t *root;
2740 	uint64_t leaves;
2741 	uint64_t leaf, top;
2742 	uint64_t ashift = ztest_get_ashift();
2743 	uint64_t oldguid, pguid;
2744 	uint64_t oldsize, newsize;
2745 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2746 	int replacing;
2747 	int oldvd_has_siblings = B_FALSE;
2748 	int newvd_is_spare = B_FALSE;
2749 	int oldvd_is_log;
2750 	int error, expected_error;
2751 
2752 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2753 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2754 
2755 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2756 
2757 	/*
2758 	 * Decide whether to do an attach or a replace.
2759 	 */
2760 	replacing = ztest_random(2);
2761 
2762 	/*
2763 	 * Pick a random top-level vdev.
2764 	 */
2765 	top = ztest_random_vdev_top(spa, B_TRUE);
2766 
2767 	/*
2768 	 * Pick a random leaf within it.
2769 	 */
2770 	leaf = ztest_random(leaves);
2771 
2772 	/*
2773 	 * Locate this vdev.
2774 	 */
2775 	oldvd = rvd->vdev_child[top];
2776 	if (zs->zs_mirrors >= 1) {
2777 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2778 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2779 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2780 	}
2781 	if (ztest_opts.zo_raidz > 1) {
2782 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2783 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2784 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2785 	}
2786 
2787 	/*
2788 	 * If we're already doing an attach or replace, oldvd may be a
2789 	 * mirror vdev -- in which case, pick a random child.
2790 	 */
2791 	while (oldvd->vdev_children != 0) {
2792 		oldvd_has_siblings = B_TRUE;
2793 		ASSERT(oldvd->vdev_children >= 2);
2794 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2795 	}
2796 
2797 	oldguid = oldvd->vdev_guid;
2798 	oldsize = vdev_get_min_asize(oldvd);
2799 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2800 	(void) strcpy(oldpath, oldvd->vdev_path);
2801 	pvd = oldvd->vdev_parent;
2802 	pguid = pvd->vdev_guid;
2803 
2804 	/*
2805 	 * If oldvd has siblings, then half of the time, detach it.
2806 	 */
2807 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2808 		spa_config_exit(spa, SCL_VDEV, FTAG);
2809 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2810 		if (error != 0 && error != ENODEV && error != EBUSY &&
2811 		    error != ENOTSUP)
2812 			fatal(0, "detach (%s) returned %d", oldpath, error);
2813 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2814 		return;
2815 	}
2816 
2817 	/*
2818 	 * For the new vdev, choose with equal probability between the two
2819 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2820 	 */
2821 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2822 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2823 		newvd_is_spare = B_TRUE;
2824 		(void) strcpy(newpath, newvd->vdev_path);
2825 	} else {
2826 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2827 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2828 		    top * leaves + leaf);
2829 		if (ztest_random(2) == 0)
2830 			newpath[strlen(newpath) - 1] = 'b';
2831 		newvd = vdev_lookup_by_path(rvd, newpath);
2832 	}
2833 
2834 	if (newvd) {
2835 		newsize = vdev_get_min_asize(newvd);
2836 	} else {
2837 		/*
2838 		 * Make newsize a little bigger or smaller than oldsize.
2839 		 * If it's smaller, the attach should fail.
2840 		 * If it's larger, and we're doing a replace,
2841 		 * we should get dynamic LUN growth when we're done.
2842 		 */
2843 		newsize = 10 * oldsize / (9 + ztest_random(3));
2844 	}
2845 
2846 	/*
2847 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2848 	 * unless it's a replace; in that case any non-replacing parent is OK.
2849 	 *
2850 	 * If newvd is already part of the pool, it should fail with EBUSY.
2851 	 *
2852 	 * If newvd is too small, it should fail with EOVERFLOW.
2853 	 */
2854 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2855 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2856 	    pvd->vdev_ops == &vdev_replacing_ops ||
2857 	    pvd->vdev_ops == &vdev_spare_ops))
2858 		expected_error = ENOTSUP;
2859 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2860 		expected_error = ENOTSUP;
2861 	else if (newvd == oldvd)
2862 		expected_error = replacing ? 0 : EBUSY;
2863 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2864 		expected_error = EBUSY;
2865 	else if (newsize < oldsize)
2866 		expected_error = EOVERFLOW;
2867 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2868 		expected_error = EDOM;
2869 	else
2870 		expected_error = 0;
2871 
2872 	spa_config_exit(spa, SCL_VDEV, FTAG);
2873 
2874 	/*
2875 	 * Build the nvlist describing newpath.
2876 	 */
2877 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2878 	    ashift, 0, 0, 0, 1);
2879 
2880 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2881 
2882 	nvlist_free(root);
2883 
2884 	/*
2885 	 * If our parent was the replacing vdev, but the replace completed,
2886 	 * then instead of failing with ENOTSUP we may either succeed,
2887 	 * fail with ENODEV, or fail with EOVERFLOW.
2888 	 */
2889 	if (expected_error == ENOTSUP &&
2890 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2891 		expected_error = error;
2892 
2893 	/*
2894 	 * If someone grew the LUN, the replacement may be too small.
2895 	 */
2896 	if (error == EOVERFLOW || error == EBUSY)
2897 		expected_error = error;
2898 
2899 	/* XXX workaround 6690467 */
2900 	if (error != expected_error && expected_error != EBUSY) {
2901 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2902 		    "returned %d, expected %d",
2903 		    oldpath, oldsize, newpath,
2904 		    newsize, replacing, error, expected_error);
2905 	}
2906 
2907 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2908 }
2909 
2910 /*
2911  * Callback function which expands the physical size of the vdev.
2912  */
2913 vdev_t *
2914 grow_vdev(vdev_t *vd, void *arg)
2915 {
2916 	spa_t *spa = vd->vdev_spa;
2917 	size_t *newsize = arg;
2918 	size_t fsize;
2919 	int fd;
2920 
2921 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2922 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2923 
2924 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2925 		return (vd);
2926 
2927 	fsize = lseek(fd, 0, SEEK_END);
2928 	(void) ftruncate(fd, *newsize);
2929 
2930 	if (ztest_opts.zo_verbose >= 6) {
2931 		(void) printf("%s grew from %lu to %lu bytes\n",
2932 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2933 	}
2934 	(void) close(fd);
2935 	return (NULL);
2936 }
2937 
2938 /*
2939  * Callback function which expands a given vdev by calling vdev_online().
2940  */
2941 /* ARGSUSED */
2942 vdev_t *
2943 online_vdev(vdev_t *vd, void *arg)
2944 {
2945 	spa_t *spa = vd->vdev_spa;
2946 	vdev_t *tvd = vd->vdev_top;
2947 	uint64_t guid = vd->vdev_guid;
2948 	uint64_t generation = spa->spa_config_generation + 1;
2949 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2950 	int error;
2951 
2952 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2953 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2954 
2955 	/* Calling vdev_online will initialize the new metaslabs */
2956 	spa_config_exit(spa, SCL_STATE, spa);
2957 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2958 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2959 
2960 	/*
2961 	 * If vdev_online returned an error or the underlying vdev_open
2962 	 * failed then we abort the expand. The only way to know that
2963 	 * vdev_open fails is by checking the returned newstate.
2964 	 */
2965 	if (error || newstate != VDEV_STATE_HEALTHY) {
2966 		if (ztest_opts.zo_verbose >= 5) {
2967 			(void) printf("Unable to expand vdev, state %llu, "
2968 			    "error %d\n", (u_longlong_t)newstate, error);
2969 		}
2970 		return (vd);
2971 	}
2972 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2973 
2974 	/*
2975 	 * Since we dropped the lock we need to ensure that we're
2976 	 * still talking to the original vdev. It's possible this
2977 	 * vdev may have been detached/replaced while we were
2978 	 * trying to online it.
2979 	 */
2980 	if (generation != spa->spa_config_generation) {
2981 		if (ztest_opts.zo_verbose >= 5) {
2982 			(void) printf("vdev configuration has changed, "
2983 			    "guid %llu, state %llu, expected gen %llu, "
2984 			    "got gen %llu\n",
2985 			    (u_longlong_t)guid,
2986 			    (u_longlong_t)tvd->vdev_state,
2987 			    (u_longlong_t)generation,
2988 			    (u_longlong_t)spa->spa_config_generation);
2989 		}
2990 		return (vd);
2991 	}
2992 	return (NULL);
2993 }
2994 
2995 /*
2996  * Traverse the vdev tree calling the supplied function.
2997  * We continue to walk the tree until we either have walked all
2998  * children or we receive a non-NULL return from the callback.
2999  * If a NULL callback is passed, then we just return back the first
3000  * leaf vdev we encounter.
3001  */
3002 vdev_t *
3003 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3004 {
3005 	if (vd->vdev_ops->vdev_op_leaf) {
3006 		if (func == NULL)
3007 			return (vd);
3008 		else
3009 			return (func(vd, arg));
3010 	}
3011 
3012 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3013 		vdev_t *cvd = vd->vdev_child[c];
3014 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3015 			return (cvd);
3016 	}
3017 	return (NULL);
3018 }
3019 
3020 /*
3021  * Verify that dynamic LUN growth works as expected.
3022  */
3023 /* ARGSUSED */
3024 void
3025 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3026 {
3027 	spa_t *spa = ztest_spa;
3028 	vdev_t *vd, *tvd;
3029 	metaslab_class_t *mc;
3030 	metaslab_group_t *mg;
3031 	size_t psize, newsize;
3032 	uint64_t top;
3033 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3034 
3035 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3036 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3037 
3038 	top = ztest_random_vdev_top(spa, B_TRUE);
3039 
3040 	tvd = spa->spa_root_vdev->vdev_child[top];
3041 	mg = tvd->vdev_mg;
3042 	mc = mg->mg_class;
3043 	old_ms_count = tvd->vdev_ms_count;
3044 	old_class_space = metaslab_class_get_space(mc);
3045 
3046 	/*
3047 	 * Determine the size of the first leaf vdev associated with
3048 	 * our top-level device.
3049 	 */
3050 	vd = vdev_walk_tree(tvd, NULL, NULL);
3051 	ASSERT3P(vd, !=, NULL);
3052 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3053 
3054 	psize = vd->vdev_psize;
3055 
3056 	/*
3057 	 * We only try to expand the vdev if it's healthy, less than 4x its
3058 	 * original size, and it has a valid psize.
3059 	 */
3060 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3061 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3062 		spa_config_exit(spa, SCL_STATE, spa);
3063 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3064 		return;
3065 	}
3066 	ASSERT(psize > 0);
3067 	newsize = psize + psize / 8;
3068 	ASSERT3U(newsize, >, psize);
3069 
3070 	if (ztest_opts.zo_verbose >= 6) {
3071 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3072 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3073 	}
3074 
3075 	/*
3076 	 * Growing the vdev is a two step process:
3077 	 *	1). expand the physical size (i.e. relabel)
3078 	 *	2). online the vdev to create the new metaslabs
3079 	 */
3080 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3081 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3082 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3083 		if (ztest_opts.zo_verbose >= 5) {
3084 			(void) printf("Could not expand LUN because "
3085 			    "the vdev configuration changed.\n");
3086 		}
3087 		spa_config_exit(spa, SCL_STATE, spa);
3088 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3089 		return;
3090 	}
3091 
3092 	spa_config_exit(spa, SCL_STATE, spa);
3093 
3094 	/*
3095 	 * Expanding the LUN will update the config asynchronously,
3096 	 * thus we must wait for the async thread to complete any
3097 	 * pending tasks before proceeding.
3098 	 */
3099 	for (;;) {
3100 		boolean_t done;
3101 		mutex_enter(&spa->spa_async_lock);
3102 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3103 		mutex_exit(&spa->spa_async_lock);
3104 		if (done)
3105 			break;
3106 		txg_wait_synced(spa_get_dsl(spa), 0);
3107 		(void) poll(NULL, 0, 100);
3108 	}
3109 
3110 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3111 
3112 	tvd = spa->spa_root_vdev->vdev_child[top];
3113 	new_ms_count = tvd->vdev_ms_count;
3114 	new_class_space = metaslab_class_get_space(mc);
3115 
3116 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3117 		if (ztest_opts.zo_verbose >= 5) {
3118 			(void) printf("Could not verify LUN expansion due to "
3119 			    "intervening vdev offline or remove.\n");
3120 		}
3121 		spa_config_exit(spa, SCL_STATE, spa);
3122 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3123 		return;
3124 	}
3125 
3126 	/*
3127 	 * Make sure we were able to grow the vdev.
3128 	 */
3129 	if (new_ms_count <= old_ms_count)
3130 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3131 		    old_ms_count, new_ms_count);
3132 
3133 	/*
3134 	 * Make sure we were able to grow the pool.
3135 	 */
3136 	if (new_class_space <= old_class_space)
3137 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3138 		    old_class_space, new_class_space);
3139 
3140 	if (ztest_opts.zo_verbose >= 5) {
3141 		char oldnumbuf[6], newnumbuf[6];
3142 
3143 		nicenum(old_class_space, oldnumbuf);
3144 		nicenum(new_class_space, newnumbuf);
3145 		(void) printf("%s grew from %s to %s\n",
3146 		    spa->spa_name, oldnumbuf, newnumbuf);
3147 	}
3148 
3149 	spa_config_exit(spa, SCL_STATE, spa);
3150 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3151 }
3152 
3153 /*
3154  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3155  */
3156 /* ARGSUSED */
3157 static void
3158 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3159 {
3160 	/*
3161 	 * Create the objects common to all ztest datasets.
3162 	 */
3163 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3164 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3165 }
3166 
3167 static int
3168 ztest_dataset_create(char *dsname)
3169 {
3170 	uint64_t zilset = ztest_random(100);
3171 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3172 	    ztest_objset_create_cb, NULL);
3173 
3174 	if (err || zilset < 80)
3175 		return (err);
3176 
3177 	if (ztest_opts.zo_verbose >= 6)
3178 		(void) printf("Setting dataset %s to sync always\n", dsname);
3179 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3180 	    ZFS_SYNC_ALWAYS, B_FALSE));
3181 }
3182 
3183 /* ARGSUSED */
3184 static int
3185 ztest_objset_destroy_cb(const char *name, void *arg)
3186 {
3187 	objset_t *os;
3188 	dmu_object_info_t doi;
3189 	int error;
3190 
3191 	/*
3192 	 * Verify that the dataset contains a directory object.
3193 	 */
3194 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3195 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3196 	if (error != ENOENT) {
3197 		/* We could have crashed in the middle of destroying it */
3198 		ASSERT0(error);
3199 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3200 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3201 	}
3202 	dmu_objset_disown(os, FTAG);
3203 
3204 	/*
3205 	 * Destroy the dataset.
3206 	 */
3207 	if (strchr(name, '@') != NULL) {
3208 		VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3209 	} else {
3210 		VERIFY0(dsl_destroy_head(name));
3211 	}
3212 	return (0);
3213 }
3214 
3215 static boolean_t
3216 ztest_snapshot_create(char *osname, uint64_t id)
3217 {
3218 	char snapname[MAXNAMELEN];
3219 	int error;
3220 
3221 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3222 
3223 	error = dmu_objset_snapshot_one(osname, snapname);
3224 	if (error == ENOSPC) {
3225 		ztest_record_enospc(FTAG);
3226 		return (B_FALSE);
3227 	}
3228 	if (error != 0 && error != EEXIST) {
3229 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3230 		    snapname, error);
3231 	}
3232 	return (B_TRUE);
3233 }
3234 
3235 static boolean_t
3236 ztest_snapshot_destroy(char *osname, uint64_t id)
3237 {
3238 	char snapname[MAXNAMELEN];
3239 	int error;
3240 
3241 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3242 	    (u_longlong_t)id);
3243 
3244 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3245 	if (error != 0 && error != ENOENT)
3246 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3247 	return (B_TRUE);
3248 }
3249 
3250 /* ARGSUSED */
3251 void
3252 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3253 {
3254 	ztest_ds_t zdtmp;
3255 	int iters;
3256 	int error;
3257 	objset_t *os, *os2;
3258 	char name[MAXNAMELEN];
3259 	zilog_t *zilog;
3260 
3261 	(void) rw_rdlock(&ztest_name_lock);
3262 
3263 	(void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3264 	    ztest_opts.zo_pool, (u_longlong_t)id);
3265 
3266 	/*
3267 	 * If this dataset exists from a previous run, process its replay log
3268 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3269 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3270 	 */
3271 	if (ztest_random(2) == 0 &&
3272 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3273 		ztest_zd_init(&zdtmp, NULL, os);
3274 		zil_replay(os, &zdtmp, ztest_replay_vector);
3275 		ztest_zd_fini(&zdtmp);
3276 		dmu_objset_disown(os, FTAG);
3277 	}
3278 
3279 	/*
3280 	 * There may be an old instance of the dataset we're about to
3281 	 * create lying around from a previous run.  If so, destroy it
3282 	 * and all of its snapshots.
3283 	 */
3284 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3285 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3286 
3287 	/*
3288 	 * Verify that the destroyed dataset is no longer in the namespace.
3289 	 */
3290 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3291 	    FTAG, &os));
3292 
3293 	/*
3294 	 * Verify that we can create a new dataset.
3295 	 */
3296 	error = ztest_dataset_create(name);
3297 	if (error) {
3298 		if (error == ENOSPC) {
3299 			ztest_record_enospc(FTAG);
3300 			(void) rw_unlock(&ztest_name_lock);
3301 			return;
3302 		}
3303 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3304 	}
3305 
3306 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3307 
3308 	ztest_zd_init(&zdtmp, NULL, os);
3309 
3310 	/*
3311 	 * Open the intent log for it.
3312 	 */
3313 	zilog = zil_open(os, ztest_get_data);
3314 
3315 	/*
3316 	 * Put some objects in there, do a little I/O to them,
3317 	 * and randomly take a couple of snapshots along the way.
3318 	 */
3319 	iters = ztest_random(5);
3320 	for (int i = 0; i < iters; i++) {
3321 		ztest_dmu_object_alloc_free(&zdtmp, id);
3322 		if (ztest_random(iters) == 0)
3323 			(void) ztest_snapshot_create(name, i);
3324 	}
3325 
3326 	/*
3327 	 * Verify that we cannot create an existing dataset.
3328 	 */
3329 	VERIFY3U(EEXIST, ==,
3330 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3331 
3332 	/*
3333 	 * Verify that we can hold an objset that is also owned.
3334 	 */
3335 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3336 	dmu_objset_rele(os2, FTAG);
3337 
3338 	/*
3339 	 * Verify that we cannot own an objset that is already owned.
3340 	 */
3341 	VERIFY3U(EBUSY, ==,
3342 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3343 
3344 	zil_close(zilog);
3345 	dmu_objset_disown(os, FTAG);
3346 	ztest_zd_fini(&zdtmp);
3347 
3348 	(void) rw_unlock(&ztest_name_lock);
3349 }
3350 
3351 /*
3352  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3353  */
3354 void
3355 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3356 {
3357 	(void) rw_rdlock(&ztest_name_lock);
3358 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3359 	(void) ztest_snapshot_create(zd->zd_name, id);
3360 	(void) rw_unlock(&ztest_name_lock);
3361 }
3362 
3363 /*
3364  * Cleanup non-standard snapshots and clones.
3365  */
3366 void
3367 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3368 {
3369 	char snap1name[MAXNAMELEN];
3370 	char clone1name[MAXNAMELEN];
3371 	char snap2name[MAXNAMELEN];
3372 	char clone2name[MAXNAMELEN];
3373 	char snap3name[MAXNAMELEN];
3374 	int error;
3375 
3376 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3377 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3378 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3379 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3380 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3381 
3382 	error = dsl_destroy_head(clone2name);
3383 	if (error && error != ENOENT)
3384 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3385 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3386 	if (error && error != ENOENT)
3387 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3388 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3389 	if (error && error != ENOENT)
3390 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3391 	error = dsl_destroy_head(clone1name);
3392 	if (error && error != ENOENT)
3393 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3394 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3395 	if (error && error != ENOENT)
3396 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3397 }
3398 
3399 /*
3400  * Verify dsl_dataset_promote handles EBUSY
3401  */
3402 void
3403 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3404 {
3405 	objset_t *os;
3406 	char snap1name[MAXNAMELEN];
3407 	char clone1name[MAXNAMELEN];
3408 	char snap2name[MAXNAMELEN];
3409 	char clone2name[MAXNAMELEN];
3410 	char snap3name[MAXNAMELEN];
3411 	char *osname = zd->zd_name;
3412 	int error;
3413 
3414 	(void) rw_rdlock(&ztest_name_lock);
3415 
3416 	ztest_dsl_dataset_cleanup(osname, id);
3417 
3418 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3419 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3420 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3421 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3422 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3423 
3424 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3425 	if (error && error != EEXIST) {
3426 		if (error == ENOSPC) {
3427 			ztest_record_enospc(FTAG);
3428 			goto out;
3429 		}
3430 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3431 	}
3432 
3433 	error = dmu_objset_clone(clone1name, snap1name);
3434 	if (error) {
3435 		if (error == ENOSPC) {
3436 			ztest_record_enospc(FTAG);
3437 			goto out;
3438 		}
3439 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3440 	}
3441 
3442 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3443 	if (error && error != EEXIST) {
3444 		if (error == ENOSPC) {
3445 			ztest_record_enospc(FTAG);
3446 			goto out;
3447 		}
3448 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3449 	}
3450 
3451 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3452 	if (error && error != EEXIST) {
3453 		if (error == ENOSPC) {
3454 			ztest_record_enospc(FTAG);
3455 			goto out;
3456 		}
3457 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3458 	}
3459 
3460 	error = dmu_objset_clone(clone2name, snap3name);
3461 	if (error) {
3462 		if (error == ENOSPC) {
3463 			ztest_record_enospc(FTAG);
3464 			goto out;
3465 		}
3466 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3467 	}
3468 
3469 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3470 	if (error)
3471 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3472 	error = dsl_dataset_promote(clone2name, NULL);
3473 	if (error != EBUSY)
3474 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3475 		    error);
3476 	dmu_objset_disown(os, FTAG);
3477 
3478 out:
3479 	ztest_dsl_dataset_cleanup(osname, id);
3480 
3481 	(void) rw_unlock(&ztest_name_lock);
3482 }
3483 
3484 /*
3485  * Verify that dmu_object_{alloc,free} work as expected.
3486  */
3487 void
3488 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3489 {
3490 	ztest_od_t od[4];
3491 	int batchsize = sizeof (od) / sizeof (od[0]);
3492 
3493 	for (int b = 0; b < batchsize; b++)
3494 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3495 
3496 	/*
3497 	 * Destroy the previous batch of objects, create a new batch,
3498 	 * and do some I/O on the new objects.
3499 	 */
3500 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3501 		return;
3502 
3503 	while (ztest_random(4 * batchsize) != 0)
3504 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3505 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3506 }
3507 
3508 /*
3509  * Verify that dmu_{read,write} work as expected.
3510  */
3511 void
3512 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3513 {
3514 	objset_t *os = zd->zd_os;
3515 	ztest_od_t od[2];
3516 	dmu_tx_t *tx;
3517 	int i, freeit, error;
3518 	uint64_t n, s, txg;
3519 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3520 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3521 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3522 	uint64_t regions = 997;
3523 	uint64_t stride = 123456789ULL;
3524 	uint64_t width = 40;
3525 	int free_percent = 5;
3526 
3527 	/*
3528 	 * This test uses two objects, packobj and bigobj, that are always
3529 	 * updated together (i.e. in the same tx) so that their contents are
3530 	 * in sync and can be compared.  Their contents relate to each other
3531 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3532 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3533 	 * for any index n, there are three bufwads that should be identical:
3534 	 *
3535 	 *	packobj, at offset n * sizeof (bufwad_t)
3536 	 *	bigobj, at the head of the nth chunk
3537 	 *	bigobj, at the tail of the nth chunk
3538 	 *
3539 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3540 	 * and it doesn't have any relation to the object blocksize.
3541 	 * The only requirement is that it can hold at least two bufwads.
3542 	 *
3543 	 * Normally, we write the bufwad to each of these locations.
3544 	 * However, free_percent of the time we instead write zeroes to
3545 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3546 	 * bigobj to packobj, we can verify that the DMU is correctly
3547 	 * tracking which parts of an object are allocated and free,
3548 	 * and that the contents of the allocated blocks are correct.
3549 	 */
3550 
3551 	/*
3552 	 * Read the directory info.  If it's the first time, set things up.
3553 	 */
3554 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3555 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3556 
3557 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3558 		return;
3559 
3560 	bigobj = od[0].od_object;
3561 	packobj = od[1].od_object;
3562 	chunksize = od[0].od_gen;
3563 	ASSERT(chunksize == od[1].od_gen);
3564 
3565 	/*
3566 	 * Prefetch a random chunk of the big object.
3567 	 * Our aim here is to get some async reads in flight
3568 	 * for blocks that we may free below; the DMU should
3569 	 * handle this race correctly.
3570 	 */
3571 	n = ztest_random(regions) * stride + ztest_random(width);
3572 	s = 1 + ztest_random(2 * width - 1);
3573 	dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3574 
3575 	/*
3576 	 * Pick a random index and compute the offsets into packobj and bigobj.
3577 	 */
3578 	n = ztest_random(regions) * stride + ztest_random(width);
3579 	s = 1 + ztest_random(width - 1);
3580 
3581 	packoff = n * sizeof (bufwad_t);
3582 	packsize = s * sizeof (bufwad_t);
3583 
3584 	bigoff = n * chunksize;
3585 	bigsize = s * chunksize;
3586 
3587 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3588 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3589 
3590 	/*
3591 	 * free_percent of the time, free a range of bigobj rather than
3592 	 * overwriting it.
3593 	 */
3594 	freeit = (ztest_random(100) < free_percent);
3595 
3596 	/*
3597 	 * Read the current contents of our objects.
3598 	 */
3599 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3600 	    DMU_READ_PREFETCH);
3601 	ASSERT0(error);
3602 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3603 	    DMU_READ_PREFETCH);
3604 	ASSERT0(error);
3605 
3606 	/*
3607 	 * Get a tx for the mods to both packobj and bigobj.
3608 	 */
3609 	tx = dmu_tx_create(os);
3610 
3611 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3612 
3613 	if (freeit)
3614 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3615 	else
3616 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3617 
3618 	/* This accounts for setting the checksum/compression. */
3619 	dmu_tx_hold_bonus(tx, bigobj);
3620 
3621 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3622 	if (txg == 0) {
3623 		umem_free(packbuf, packsize);
3624 		umem_free(bigbuf, bigsize);
3625 		return;
3626 	}
3627 
3628 	dmu_object_set_checksum(os, bigobj,
3629 	    (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3630 
3631 	dmu_object_set_compress(os, bigobj,
3632 	    (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3633 
3634 	/*
3635 	 * For each index from n to n + s, verify that the existing bufwad
3636 	 * in packobj matches the bufwads at the head and tail of the
3637 	 * corresponding chunk in bigobj.  Then update all three bufwads
3638 	 * with the new values we want to write out.
3639 	 */
3640 	for (i = 0; i < s; i++) {
3641 		/* LINTED */
3642 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3643 		/* LINTED */
3644 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3645 		/* LINTED */
3646 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3647 
3648 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3649 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3650 
3651 		if (pack->bw_txg > txg)
3652 			fatal(0, "future leak: got %llx, open txg is %llx",
3653 			    pack->bw_txg, txg);
3654 
3655 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3656 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3657 			    pack->bw_index, n, i);
3658 
3659 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3660 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3661 
3662 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3663 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3664 
3665 		if (freeit) {
3666 			bzero(pack, sizeof (bufwad_t));
3667 		} else {
3668 			pack->bw_index = n + i;
3669 			pack->bw_txg = txg;
3670 			pack->bw_data = 1 + ztest_random(-2ULL);
3671 		}
3672 		*bigH = *pack;
3673 		*bigT = *pack;
3674 	}
3675 
3676 	/*
3677 	 * We've verified all the old bufwads, and made new ones.
3678 	 * Now write them out.
3679 	 */
3680 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3681 
3682 	if (freeit) {
3683 		if (ztest_opts.zo_verbose >= 7) {
3684 			(void) printf("freeing offset %llx size %llx"
3685 			    " txg %llx\n",
3686 			    (u_longlong_t)bigoff,
3687 			    (u_longlong_t)bigsize,
3688 			    (u_longlong_t)txg);
3689 		}
3690 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3691 	} else {
3692 		if (ztest_opts.zo_verbose >= 7) {
3693 			(void) printf("writing offset %llx size %llx"
3694 			    " txg %llx\n",
3695 			    (u_longlong_t)bigoff,
3696 			    (u_longlong_t)bigsize,
3697 			    (u_longlong_t)txg);
3698 		}
3699 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3700 	}
3701 
3702 	dmu_tx_commit(tx);
3703 
3704 	/*
3705 	 * Sanity check the stuff we just wrote.
3706 	 */
3707 	{
3708 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3709 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3710 
3711 		VERIFY(0 == dmu_read(os, packobj, packoff,
3712 		    packsize, packcheck, DMU_READ_PREFETCH));
3713 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3714 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3715 
3716 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3717 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3718 
3719 		umem_free(packcheck, packsize);
3720 		umem_free(bigcheck, bigsize);
3721 	}
3722 
3723 	umem_free(packbuf, packsize);
3724 	umem_free(bigbuf, bigsize);
3725 }
3726 
3727 void
3728 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3729     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3730 {
3731 	uint64_t i;
3732 	bufwad_t *pack;
3733 	bufwad_t *bigH;
3734 	bufwad_t *bigT;
3735 
3736 	/*
3737 	 * For each index from n to n + s, verify that the existing bufwad
3738 	 * in packobj matches the bufwads at the head and tail of the
3739 	 * corresponding chunk in bigobj.  Then update all three bufwads
3740 	 * with the new values we want to write out.
3741 	 */
3742 	for (i = 0; i < s; i++) {
3743 		/* LINTED */
3744 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3745 		/* LINTED */
3746 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3747 		/* LINTED */
3748 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3749 
3750 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3751 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3752 
3753 		if (pack->bw_txg > txg)
3754 			fatal(0, "future leak: got %llx, open txg is %llx",
3755 			    pack->bw_txg, txg);
3756 
3757 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3758 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3759 			    pack->bw_index, n, i);
3760 
3761 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3762 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3763 
3764 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3765 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3766 
3767 		pack->bw_index = n + i;
3768 		pack->bw_txg = txg;
3769 		pack->bw_data = 1 + ztest_random(-2ULL);
3770 
3771 		*bigH = *pack;
3772 		*bigT = *pack;
3773 	}
3774 }
3775 
3776 void
3777 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3778 {
3779 	objset_t *os = zd->zd_os;
3780 	ztest_od_t od[2];
3781 	dmu_tx_t *tx;
3782 	uint64_t i;
3783 	int error;
3784 	uint64_t n, s, txg;
3785 	bufwad_t *packbuf, *bigbuf;
3786 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3787 	uint64_t blocksize = ztest_random_blocksize();
3788 	uint64_t chunksize = blocksize;
3789 	uint64_t regions = 997;
3790 	uint64_t stride = 123456789ULL;
3791 	uint64_t width = 9;
3792 	dmu_buf_t *bonus_db;
3793 	arc_buf_t **bigbuf_arcbufs;
3794 	dmu_object_info_t doi;
3795 
3796 	/*
3797 	 * This test uses two objects, packobj and bigobj, that are always
3798 	 * updated together (i.e. in the same tx) so that their contents are
3799 	 * in sync and can be compared.  Their contents relate to each other
3800 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3801 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3802 	 * for any index n, there are three bufwads that should be identical:
3803 	 *
3804 	 *	packobj, at offset n * sizeof (bufwad_t)
3805 	 *	bigobj, at the head of the nth chunk
3806 	 *	bigobj, at the tail of the nth chunk
3807 	 *
3808 	 * The chunk size is set equal to bigobj block size so that
3809 	 * dmu_assign_arcbuf() can be tested for object updates.
3810 	 */
3811 
3812 	/*
3813 	 * Read the directory info.  If it's the first time, set things up.
3814 	 */
3815 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3816 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3817 
3818 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3819 		return;
3820 
3821 	bigobj = od[0].od_object;
3822 	packobj = od[1].od_object;
3823 	blocksize = od[0].od_blocksize;
3824 	chunksize = blocksize;
3825 	ASSERT(chunksize == od[1].od_gen);
3826 
3827 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3828 	VERIFY(ISP2(doi.doi_data_block_size));
3829 	VERIFY(chunksize == doi.doi_data_block_size);
3830 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3831 
3832 	/*
3833 	 * Pick a random index and compute the offsets into packobj and bigobj.
3834 	 */
3835 	n = ztest_random(regions) * stride + ztest_random(width);
3836 	s = 1 + ztest_random(width - 1);
3837 
3838 	packoff = n * sizeof (bufwad_t);
3839 	packsize = s * sizeof (bufwad_t);
3840 
3841 	bigoff = n * chunksize;
3842 	bigsize = s * chunksize;
3843 
3844 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3845 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3846 
3847 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3848 
3849 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3850 
3851 	/*
3852 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3853 	 * Iteration 1 test zcopy to already referenced dbufs.
3854 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3855 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3856 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3857 	 * Iteration 5 test zcopy when it can't be done.
3858 	 * Iteration 6 one more zcopy write.
3859 	 */
3860 	for (i = 0; i < 7; i++) {
3861 		uint64_t j;
3862 		uint64_t off;
3863 
3864 		/*
3865 		 * In iteration 5 (i == 5) use arcbufs
3866 		 * that don't match bigobj blksz to test
3867 		 * dmu_assign_arcbuf() when it can't directly
3868 		 * assign an arcbuf to a dbuf.
3869 		 */
3870 		for (j = 0; j < s; j++) {
3871 			if (i != 5) {
3872 				bigbuf_arcbufs[j] =
3873 				    dmu_request_arcbuf(bonus_db, chunksize);
3874 			} else {
3875 				bigbuf_arcbufs[2 * j] =
3876 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3877 				bigbuf_arcbufs[2 * j + 1] =
3878 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3879 			}
3880 		}
3881 
3882 		/*
3883 		 * Get a tx for the mods to both packobj and bigobj.
3884 		 */
3885 		tx = dmu_tx_create(os);
3886 
3887 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3888 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3889 
3890 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3891 		if (txg == 0) {
3892 			umem_free(packbuf, packsize);
3893 			umem_free(bigbuf, bigsize);
3894 			for (j = 0; j < s; j++) {
3895 				if (i != 5) {
3896 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3897 				} else {
3898 					dmu_return_arcbuf(
3899 					    bigbuf_arcbufs[2 * j]);
3900 					dmu_return_arcbuf(
3901 					    bigbuf_arcbufs[2 * j + 1]);
3902 				}
3903 			}
3904 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3905 			dmu_buf_rele(bonus_db, FTAG);
3906 			return;
3907 		}
3908 
3909 		/*
3910 		 * 50% of the time don't read objects in the 1st iteration to
3911 		 * test dmu_assign_arcbuf() for the case when there're no
3912 		 * existing dbufs for the specified offsets.
3913 		 */
3914 		if (i != 0 || ztest_random(2) != 0) {
3915 			error = dmu_read(os, packobj, packoff,
3916 			    packsize, packbuf, DMU_READ_PREFETCH);
3917 			ASSERT0(error);
3918 			error = dmu_read(os, bigobj, bigoff, bigsize,
3919 			    bigbuf, DMU_READ_PREFETCH);
3920 			ASSERT0(error);
3921 		}
3922 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3923 		    n, chunksize, txg);
3924 
3925 		/*
3926 		 * We've verified all the old bufwads, and made new ones.
3927 		 * Now write them out.
3928 		 */
3929 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3930 		if (ztest_opts.zo_verbose >= 7) {
3931 			(void) printf("writing offset %llx size %llx"
3932 			    " txg %llx\n",
3933 			    (u_longlong_t)bigoff,
3934 			    (u_longlong_t)bigsize,
3935 			    (u_longlong_t)txg);
3936 		}
3937 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3938 			dmu_buf_t *dbt;
3939 			if (i != 5) {
3940 				bcopy((caddr_t)bigbuf + (off - bigoff),
3941 				    bigbuf_arcbufs[j]->b_data, chunksize);
3942 			} else {
3943 				bcopy((caddr_t)bigbuf + (off - bigoff),
3944 				    bigbuf_arcbufs[2 * j]->b_data,
3945 				    chunksize / 2);
3946 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3947 				    chunksize / 2,
3948 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3949 				    chunksize / 2);
3950 			}
3951 
3952 			if (i == 1) {
3953 				VERIFY(dmu_buf_hold(os, bigobj, off,
3954 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3955 			}
3956 			if (i != 5) {
3957 				dmu_assign_arcbuf(bonus_db, off,
3958 				    bigbuf_arcbufs[j], tx);
3959 			} else {
3960 				dmu_assign_arcbuf(bonus_db, off,
3961 				    bigbuf_arcbufs[2 * j], tx);
3962 				dmu_assign_arcbuf(bonus_db,
3963 				    off + chunksize / 2,
3964 				    bigbuf_arcbufs[2 * j + 1], tx);
3965 			}
3966 			if (i == 1) {
3967 				dmu_buf_rele(dbt, FTAG);
3968 			}
3969 		}
3970 		dmu_tx_commit(tx);
3971 
3972 		/*
3973 		 * Sanity check the stuff we just wrote.
3974 		 */
3975 		{
3976 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3977 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3978 
3979 			VERIFY(0 == dmu_read(os, packobj, packoff,
3980 			    packsize, packcheck, DMU_READ_PREFETCH));
3981 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
3982 			    bigsize, bigcheck, DMU_READ_PREFETCH));
3983 
3984 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3985 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3986 
3987 			umem_free(packcheck, packsize);
3988 			umem_free(bigcheck, bigsize);
3989 		}
3990 		if (i == 2) {
3991 			txg_wait_open(dmu_objset_pool(os), 0);
3992 		} else if (i == 3) {
3993 			txg_wait_synced(dmu_objset_pool(os), 0);
3994 		}
3995 	}
3996 
3997 	dmu_buf_rele(bonus_db, FTAG);
3998 	umem_free(packbuf, packsize);
3999 	umem_free(bigbuf, bigsize);
4000 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4001 }
4002 
4003 /* ARGSUSED */
4004 void
4005 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4006 {
4007 	ztest_od_t od[1];
4008 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4009 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4010 
4011 	/*
4012 	 * Have multiple threads write to large offsets in an object
4013 	 * to verify that parallel writes to an object -- even to the
4014 	 * same blocks within the object -- doesn't cause any trouble.
4015 	 */
4016 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4017 
4018 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4019 		return;
4020 
4021 	while (ztest_random(10) != 0)
4022 		ztest_io(zd, od[0].od_object, offset);
4023 }
4024 
4025 void
4026 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4027 {
4028 	ztest_od_t od[1];
4029 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4030 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4031 	uint64_t count = ztest_random(20) + 1;
4032 	uint64_t blocksize = ztest_random_blocksize();
4033 	void *data;
4034 
4035 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4036 
4037 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4038 		return;
4039 
4040 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4041 		return;
4042 
4043 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4044 
4045 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4046 
4047 	while (ztest_random(count) != 0) {
4048 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4049 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4050 		    data) != 0)
4051 			break;
4052 		while (ztest_random(4) != 0)
4053 			ztest_io(zd, od[0].od_object, randoff);
4054 	}
4055 
4056 	umem_free(data, blocksize);
4057 }
4058 
4059 /*
4060  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4061  */
4062 #define	ZTEST_ZAP_MIN_INTS	1
4063 #define	ZTEST_ZAP_MAX_INTS	4
4064 #define	ZTEST_ZAP_MAX_PROPS	1000
4065 
4066 void
4067 ztest_zap(ztest_ds_t *zd, uint64_t id)
4068 {
4069 	objset_t *os = zd->zd_os;
4070 	ztest_od_t od[1];
4071 	uint64_t object;
4072 	uint64_t txg, last_txg;
4073 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4074 	uint64_t zl_ints, zl_intsize, prop;
4075 	int i, ints;
4076 	dmu_tx_t *tx;
4077 	char propname[100], txgname[100];
4078 	int error;
4079 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4080 
4081 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4082 
4083 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4084 		return;
4085 
4086 	object = od[0].od_object;
4087 
4088 	/*
4089 	 * Generate a known hash collision, and verify that
4090 	 * we can lookup and remove both entries.
4091 	 */
4092 	tx = dmu_tx_create(os);
4093 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4094 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4095 	if (txg == 0)
4096 		return;
4097 	for (i = 0; i < 2; i++) {
4098 		value[i] = i;
4099 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4100 		    1, &value[i], tx));
4101 	}
4102 	for (i = 0; i < 2; i++) {
4103 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4104 		    sizeof (uint64_t), 1, &value[i], tx));
4105 		VERIFY3U(0, ==,
4106 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4107 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4108 		ASSERT3U(zl_ints, ==, 1);
4109 	}
4110 	for (i = 0; i < 2; i++) {
4111 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4112 	}
4113 	dmu_tx_commit(tx);
4114 
4115 	/*
4116 	 * Generate a buch of random entries.
4117 	 */
4118 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4119 
4120 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4121 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4122 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4123 	bzero(value, sizeof (value));
4124 	last_txg = 0;
4125 
4126 	/*
4127 	 * If these zap entries already exist, validate their contents.
4128 	 */
4129 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4130 	if (error == 0) {
4131 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4132 		ASSERT3U(zl_ints, ==, 1);
4133 
4134 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4135 		    zl_ints, &last_txg) == 0);
4136 
4137 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4138 		    &zl_ints) == 0);
4139 
4140 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4141 		ASSERT3U(zl_ints, ==, ints);
4142 
4143 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4144 		    zl_ints, value) == 0);
4145 
4146 		for (i = 0; i < ints; i++) {
4147 			ASSERT3U(value[i], ==, last_txg + object + i);
4148 		}
4149 	} else {
4150 		ASSERT3U(error, ==, ENOENT);
4151 	}
4152 
4153 	/*
4154 	 * Atomically update two entries in our zap object.
4155 	 * The first is named txg_%llu, and contains the txg
4156 	 * in which the property was last updated.  The second
4157 	 * is named prop_%llu, and the nth element of its value
4158 	 * should be txg + object + n.
4159 	 */
4160 	tx = dmu_tx_create(os);
4161 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4162 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4163 	if (txg == 0)
4164 		return;
4165 
4166 	if (last_txg > txg)
4167 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4168 
4169 	for (i = 0; i < ints; i++)
4170 		value[i] = txg + object + i;
4171 
4172 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4173 	    1, &txg, tx));
4174 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4175 	    ints, value, tx));
4176 
4177 	dmu_tx_commit(tx);
4178 
4179 	/*
4180 	 * Remove a random pair of entries.
4181 	 */
4182 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4183 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4184 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4185 
4186 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4187 
4188 	if (error == ENOENT)
4189 		return;
4190 
4191 	ASSERT0(error);
4192 
4193 	tx = dmu_tx_create(os);
4194 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4195 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4196 	if (txg == 0)
4197 		return;
4198 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4199 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4200 	dmu_tx_commit(tx);
4201 }
4202 
4203 /*
4204  * Testcase to test the upgrading of a microzap to fatzap.
4205  */
4206 void
4207 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4208 {
4209 	objset_t *os = zd->zd_os;
4210 	ztest_od_t od[1];
4211 	uint64_t object, txg;
4212 
4213 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4214 
4215 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4216 		return;
4217 
4218 	object = od[0].od_object;
4219 
4220 	/*
4221 	 * Add entries to this ZAP and make sure it spills over
4222 	 * and gets upgraded to a fatzap. Also, since we are adding
4223 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4224 	 */
4225 	for (int i = 0; i < 2050; i++) {
4226 		char name[MAXNAMELEN];
4227 		uint64_t value = i;
4228 		dmu_tx_t *tx;
4229 		int error;
4230 
4231 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4232 		    id, value);
4233 
4234 		tx = dmu_tx_create(os);
4235 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4236 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4237 		if (txg == 0)
4238 			return;
4239 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4240 		    &value, tx);
4241 		ASSERT(error == 0 || error == EEXIST);
4242 		dmu_tx_commit(tx);
4243 	}
4244 }
4245 
4246 /* ARGSUSED */
4247 void
4248 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4249 {
4250 	objset_t *os = zd->zd_os;
4251 	ztest_od_t od[1];
4252 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4253 	dmu_tx_t *tx;
4254 	int i, namelen, error;
4255 	int micro = ztest_random(2);
4256 	char name[20], string_value[20];
4257 	void *data;
4258 
4259 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4260 
4261 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4262 		return;
4263 
4264 	object = od[0].od_object;
4265 
4266 	/*
4267 	 * Generate a random name of the form 'xxx.....' where each
4268 	 * x is a random printable character and the dots are dots.
4269 	 * There are 94 such characters, and the name length goes from
4270 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4271 	 */
4272 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4273 
4274 	for (i = 0; i < 3; i++)
4275 		name[i] = '!' + ztest_random('~' - '!' + 1);
4276 	for (; i < namelen - 1; i++)
4277 		name[i] = '.';
4278 	name[i] = '\0';
4279 
4280 	if ((namelen & 1) || micro) {
4281 		wsize = sizeof (txg);
4282 		wc = 1;
4283 		data = &txg;
4284 	} else {
4285 		wsize = 1;
4286 		wc = namelen;
4287 		data = string_value;
4288 	}
4289 
4290 	count = -1ULL;
4291 	VERIFY0(zap_count(os, object, &count));
4292 	ASSERT(count != -1ULL);
4293 
4294 	/*
4295 	 * Select an operation: length, lookup, add, update, remove.
4296 	 */
4297 	i = ztest_random(5);
4298 
4299 	if (i >= 2) {
4300 		tx = dmu_tx_create(os);
4301 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4302 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4303 		if (txg == 0)
4304 			return;
4305 		bcopy(name, string_value, namelen);
4306 	} else {
4307 		tx = NULL;
4308 		txg = 0;
4309 		bzero(string_value, namelen);
4310 	}
4311 
4312 	switch (i) {
4313 
4314 	case 0:
4315 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4316 		if (error == 0) {
4317 			ASSERT3U(wsize, ==, zl_wsize);
4318 			ASSERT3U(wc, ==, zl_wc);
4319 		} else {
4320 			ASSERT3U(error, ==, ENOENT);
4321 		}
4322 		break;
4323 
4324 	case 1:
4325 		error = zap_lookup(os, object, name, wsize, wc, data);
4326 		if (error == 0) {
4327 			if (data == string_value &&
4328 			    bcmp(name, data, namelen) != 0)
4329 				fatal(0, "name '%s' != val '%s' len %d",
4330 				    name, data, namelen);
4331 		} else {
4332 			ASSERT3U(error, ==, ENOENT);
4333 		}
4334 		break;
4335 
4336 	case 2:
4337 		error = zap_add(os, object, name, wsize, wc, data, tx);
4338 		ASSERT(error == 0 || error == EEXIST);
4339 		break;
4340 
4341 	case 3:
4342 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4343 		break;
4344 
4345 	case 4:
4346 		error = zap_remove(os, object, name, tx);
4347 		ASSERT(error == 0 || error == ENOENT);
4348 		break;
4349 	}
4350 
4351 	if (tx != NULL)
4352 		dmu_tx_commit(tx);
4353 }
4354 
4355 /*
4356  * Commit callback data.
4357  */
4358 typedef struct ztest_cb_data {
4359 	list_node_t		zcd_node;
4360 	uint64_t		zcd_txg;
4361 	int			zcd_expected_err;
4362 	boolean_t		zcd_added;
4363 	boolean_t		zcd_called;
4364 	spa_t			*zcd_spa;
4365 } ztest_cb_data_t;
4366 
4367 /* This is the actual commit callback function */
4368 static void
4369 ztest_commit_callback(void *arg, int error)
4370 {
4371 	ztest_cb_data_t *data = arg;
4372 	uint64_t synced_txg;
4373 
4374 	VERIFY(data != NULL);
4375 	VERIFY3S(data->zcd_expected_err, ==, error);
4376 	VERIFY(!data->zcd_called);
4377 
4378 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4379 	if (data->zcd_txg > synced_txg)
4380 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4381 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4382 		    synced_txg);
4383 
4384 	data->zcd_called = B_TRUE;
4385 
4386 	if (error == ECANCELED) {
4387 		ASSERT0(data->zcd_txg);
4388 		ASSERT(!data->zcd_added);
4389 
4390 		/*
4391 		 * The private callback data should be destroyed here, but
4392 		 * since we are going to check the zcd_called field after
4393 		 * dmu_tx_abort(), we will destroy it there.
4394 		 */
4395 		return;
4396 	}
4397 
4398 	/* Was this callback added to the global callback list? */
4399 	if (!data->zcd_added)
4400 		goto out;
4401 
4402 	ASSERT3U(data->zcd_txg, !=, 0);
4403 
4404 	/* Remove our callback from the list */
4405 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4406 	list_remove(&zcl.zcl_callbacks, data);
4407 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4408 
4409 out:
4410 	umem_free(data, sizeof (ztest_cb_data_t));
4411 }
4412 
4413 /* Allocate and initialize callback data structure */
4414 static ztest_cb_data_t *
4415 ztest_create_cb_data(objset_t *os, uint64_t txg)
4416 {
4417 	ztest_cb_data_t *cb_data;
4418 
4419 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4420 
4421 	cb_data->zcd_txg = txg;
4422 	cb_data->zcd_spa = dmu_objset_spa(os);
4423 
4424 	return (cb_data);
4425 }
4426 
4427 /*
4428  * If a number of txgs equal to this threshold have been created after a commit
4429  * callback has been registered but not called, then we assume there is an
4430  * implementation bug.
4431  */
4432 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4433 
4434 /*
4435  * Commit callback test.
4436  */
4437 void
4438 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4439 {
4440 	objset_t *os = zd->zd_os;
4441 	ztest_od_t od[1];
4442 	dmu_tx_t *tx;
4443 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4444 	uint64_t old_txg, txg;
4445 	int i, error;
4446 
4447 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4448 
4449 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4450 		return;
4451 
4452 	tx = dmu_tx_create(os);
4453 
4454 	cb_data[0] = ztest_create_cb_data(os, 0);
4455 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4456 
4457 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4458 
4459 	/* Every once in a while, abort the transaction on purpose */
4460 	if (ztest_random(100) == 0)
4461 		error = -1;
4462 
4463 	if (!error)
4464 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4465 
4466 	txg = error ? 0 : dmu_tx_get_txg(tx);
4467 
4468 	cb_data[0]->zcd_txg = txg;
4469 	cb_data[1] = ztest_create_cb_data(os, txg);
4470 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4471 
4472 	if (error) {
4473 		/*
4474 		 * It's not a strict requirement to call the registered
4475 		 * callbacks from inside dmu_tx_abort(), but that's what
4476 		 * it's supposed to happen in the current implementation
4477 		 * so we will check for that.
4478 		 */
4479 		for (i = 0; i < 2; i++) {
4480 			cb_data[i]->zcd_expected_err = ECANCELED;
4481 			VERIFY(!cb_data[i]->zcd_called);
4482 		}
4483 
4484 		dmu_tx_abort(tx);
4485 
4486 		for (i = 0; i < 2; i++) {
4487 			VERIFY(cb_data[i]->zcd_called);
4488 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4489 		}
4490 
4491 		return;
4492 	}
4493 
4494 	cb_data[2] = ztest_create_cb_data(os, txg);
4495 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4496 
4497 	/*
4498 	 * Read existing data to make sure there isn't a future leak.
4499 	 */
4500 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4501 	    &old_txg, DMU_READ_PREFETCH));
4502 
4503 	if (old_txg > txg)
4504 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4505 		    old_txg, txg);
4506 
4507 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4508 
4509 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4510 
4511 	/*
4512 	 * Since commit callbacks don't have any ordering requirement and since
4513 	 * it is theoretically possible for a commit callback to be called
4514 	 * after an arbitrary amount of time has elapsed since its txg has been
4515 	 * synced, it is difficult to reliably determine whether a commit
4516 	 * callback hasn't been called due to high load or due to a flawed
4517 	 * implementation.
4518 	 *
4519 	 * In practice, we will assume that if after a certain number of txgs a
4520 	 * commit callback hasn't been called, then most likely there's an
4521 	 * implementation bug..
4522 	 */
4523 	tmp_cb = list_head(&zcl.zcl_callbacks);
4524 	if (tmp_cb != NULL &&
4525 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4526 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4527 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4528 	}
4529 
4530 	/*
4531 	 * Let's find the place to insert our callbacks.
4532 	 *
4533 	 * Even though the list is ordered by txg, it is possible for the
4534 	 * insertion point to not be the end because our txg may already be
4535 	 * quiescing at this point and other callbacks in the open txg
4536 	 * (from other objsets) may have sneaked in.
4537 	 */
4538 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4539 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4540 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4541 
4542 	/* Add the 3 callbacks to the list */
4543 	for (i = 0; i < 3; i++) {
4544 		if (tmp_cb == NULL)
4545 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4546 		else
4547 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4548 			    cb_data[i]);
4549 
4550 		cb_data[i]->zcd_added = B_TRUE;
4551 		VERIFY(!cb_data[i]->zcd_called);
4552 
4553 		tmp_cb = cb_data[i];
4554 	}
4555 
4556 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4557 
4558 	dmu_tx_commit(tx);
4559 }
4560 
4561 /* ARGSUSED */
4562 void
4563 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4564 {
4565 	zfs_prop_t proplist[] = {
4566 		ZFS_PROP_CHECKSUM,
4567 		ZFS_PROP_COMPRESSION,
4568 		ZFS_PROP_COPIES,
4569 		ZFS_PROP_DEDUP
4570 	};
4571 
4572 	(void) rw_rdlock(&ztest_name_lock);
4573 
4574 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4575 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4576 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4577 
4578 	(void) rw_unlock(&ztest_name_lock);
4579 }
4580 
4581 /* ARGSUSED */
4582 void
4583 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4584 {
4585 	nvlist_t *props = NULL;
4586 
4587 	(void) rw_rdlock(&ztest_name_lock);
4588 
4589 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4590 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4591 
4592 	VERIFY0(spa_prop_get(ztest_spa, &props));
4593 
4594 	if (ztest_opts.zo_verbose >= 6)
4595 		dump_nvlist(props, 4);
4596 
4597 	nvlist_free(props);
4598 
4599 	(void) rw_unlock(&ztest_name_lock);
4600 }
4601 
4602 static int
4603 user_release_one(const char *snapname, const char *holdname)
4604 {
4605 	nvlist_t *snaps, *holds;
4606 	int error;
4607 
4608 	snaps = fnvlist_alloc();
4609 	holds = fnvlist_alloc();
4610 	fnvlist_add_boolean(holds, holdname);
4611 	fnvlist_add_nvlist(snaps, snapname, holds);
4612 	fnvlist_free(holds);
4613 	error = dsl_dataset_user_release(snaps, NULL);
4614 	fnvlist_free(snaps);
4615 	return (error);
4616 }
4617 
4618 /*
4619  * Test snapshot hold/release and deferred destroy.
4620  */
4621 void
4622 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4623 {
4624 	int error;
4625 	objset_t *os = zd->zd_os;
4626 	objset_t *origin;
4627 	char snapname[100];
4628 	char fullname[100];
4629 	char clonename[100];
4630 	char tag[100];
4631 	char osname[MAXNAMELEN];
4632 	nvlist_t *holds;
4633 
4634 	(void) rw_rdlock(&ztest_name_lock);
4635 
4636 	dmu_objset_name(os, osname);
4637 
4638 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4639 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4640 	(void) snprintf(clonename, sizeof (clonename),
4641 	    "%s/ch1_%llu", osname, id);
4642 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4643 
4644 	/*
4645 	 * Clean up from any previous run.
4646 	 */
4647 	error = dsl_destroy_head(clonename);
4648 	if (error != ENOENT)
4649 		ASSERT0(error);
4650 	error = user_release_one(fullname, tag);
4651 	if (error != ESRCH && error != ENOENT)
4652 		ASSERT0(error);
4653 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4654 	if (error != ENOENT)
4655 		ASSERT0(error);
4656 
4657 	/*
4658 	 * Create snapshot, clone it, mark snap for deferred destroy,
4659 	 * destroy clone, verify snap was also destroyed.
4660 	 */
4661 	error = dmu_objset_snapshot_one(osname, snapname);
4662 	if (error) {
4663 		if (error == ENOSPC) {
4664 			ztest_record_enospc("dmu_objset_snapshot");
4665 			goto out;
4666 		}
4667 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4668 	}
4669 
4670 	error = dmu_objset_clone(clonename, fullname);
4671 	if (error) {
4672 		if (error == ENOSPC) {
4673 			ztest_record_enospc("dmu_objset_clone");
4674 			goto out;
4675 		}
4676 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4677 	}
4678 
4679 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4680 	if (error) {
4681 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4682 		    fullname, error);
4683 	}
4684 
4685 	error = dsl_destroy_head(clonename);
4686 	if (error)
4687 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4688 
4689 	error = dmu_objset_hold(fullname, FTAG, &origin);
4690 	if (error != ENOENT)
4691 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4692 
4693 	/*
4694 	 * Create snapshot, add temporary hold, verify that we can't
4695 	 * destroy a held snapshot, mark for deferred destroy,
4696 	 * release hold, verify snapshot was destroyed.
4697 	 */
4698 	error = dmu_objset_snapshot_one(osname, snapname);
4699 	if (error) {
4700 		if (error == ENOSPC) {
4701 			ztest_record_enospc("dmu_objset_snapshot");
4702 			goto out;
4703 		}
4704 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4705 	}
4706 
4707 	holds = fnvlist_alloc();
4708 	fnvlist_add_string(holds, fullname, tag);
4709 	error = dsl_dataset_user_hold(holds, 0, NULL);
4710 	fnvlist_free(holds);
4711 
4712 	if (error)
4713 		fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4714 
4715 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4716 	if (error != EBUSY) {
4717 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4718 		    fullname, error);
4719 	}
4720 
4721 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4722 	if (error) {
4723 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4724 		    fullname, error);
4725 	}
4726 
4727 	error = user_release_one(fullname, tag);
4728 	if (error)
4729 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4730 
4731 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4732 
4733 out:
4734 	(void) rw_unlock(&ztest_name_lock);
4735 }
4736 
4737 /*
4738  * Inject random faults into the on-disk data.
4739  */
4740 /* ARGSUSED */
4741 void
4742 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4743 {
4744 	ztest_shared_t *zs = ztest_shared;
4745 	spa_t *spa = ztest_spa;
4746 	int fd;
4747 	uint64_t offset;
4748 	uint64_t leaves;
4749 	uint64_t bad = 0x1990c0ffeedecade;
4750 	uint64_t top, leaf;
4751 	char path0[MAXPATHLEN];
4752 	char pathrand[MAXPATHLEN];
4753 	size_t fsize;
4754 	int bshift = SPA_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
4755 	int iters = 1000;
4756 	int maxfaults;
4757 	int mirror_save;
4758 	vdev_t *vd0 = NULL;
4759 	uint64_t guid0 = 0;
4760 	boolean_t islog = B_FALSE;
4761 
4762 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4763 	maxfaults = MAXFAULTS();
4764 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4765 	mirror_save = zs->zs_mirrors;
4766 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4767 
4768 	ASSERT(leaves >= 1);
4769 
4770 	/*
4771 	 * Grab the name lock as reader. There are some operations
4772 	 * which don't like to have their vdevs changed while
4773 	 * they are in progress (i.e. spa_change_guid). Those
4774 	 * operations will have grabbed the name lock as writer.
4775 	 */
4776 	(void) rw_rdlock(&ztest_name_lock);
4777 
4778 	/*
4779 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4780 	 */
4781 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4782 
4783 	if (ztest_random(2) == 0) {
4784 		/*
4785 		 * Inject errors on a normal data device or slog device.
4786 		 */
4787 		top = ztest_random_vdev_top(spa, B_TRUE);
4788 		leaf = ztest_random(leaves) + zs->zs_splits;
4789 
4790 		/*
4791 		 * Generate paths to the first leaf in this top-level vdev,
4792 		 * and to the random leaf we selected.  We'll induce transient
4793 		 * write failures and random online/offline activity on leaf 0,
4794 		 * and we'll write random garbage to the randomly chosen leaf.
4795 		 */
4796 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4797 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4798 		    top * leaves + zs->zs_splits);
4799 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4800 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4801 		    top * leaves + leaf);
4802 
4803 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4804 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4805 			islog = B_TRUE;
4806 
4807 		/*
4808 		 * If the top-level vdev needs to be resilvered
4809 		 * then we only allow faults on the device that is
4810 		 * resilvering.
4811 		 */
4812 		if (vd0 != NULL && maxfaults != 1 &&
4813 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4814 		    vd0->vdev_resilver_txg != 0)) {
4815 			/*
4816 			 * Make vd0 explicitly claim to be unreadable,
4817 			 * or unwriteable, or reach behind its back
4818 			 * and close the underlying fd.  We can do this if
4819 			 * maxfaults == 0 because we'll fail and reexecute,
4820 			 * and we can do it if maxfaults >= 2 because we'll
4821 			 * have enough redundancy.  If maxfaults == 1, the
4822 			 * combination of this with injection of random data
4823 			 * corruption below exceeds the pool's fault tolerance.
4824 			 */
4825 			vdev_file_t *vf = vd0->vdev_tsd;
4826 
4827 			if (vf != NULL && ztest_random(3) == 0) {
4828 				(void) close(vf->vf_vnode->v_fd);
4829 				vf->vf_vnode->v_fd = -1;
4830 			} else if (ztest_random(2) == 0) {
4831 				vd0->vdev_cant_read = B_TRUE;
4832 			} else {
4833 				vd0->vdev_cant_write = B_TRUE;
4834 			}
4835 			guid0 = vd0->vdev_guid;
4836 		}
4837 	} else {
4838 		/*
4839 		 * Inject errors on an l2cache device.
4840 		 */
4841 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4842 
4843 		if (sav->sav_count == 0) {
4844 			spa_config_exit(spa, SCL_STATE, FTAG);
4845 			(void) rw_unlock(&ztest_name_lock);
4846 			return;
4847 		}
4848 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4849 		guid0 = vd0->vdev_guid;
4850 		(void) strcpy(path0, vd0->vdev_path);
4851 		(void) strcpy(pathrand, vd0->vdev_path);
4852 
4853 		leaf = 0;
4854 		leaves = 1;
4855 		maxfaults = INT_MAX;	/* no limit on cache devices */
4856 	}
4857 
4858 	spa_config_exit(spa, SCL_STATE, FTAG);
4859 	(void) rw_unlock(&ztest_name_lock);
4860 
4861 	/*
4862 	 * If we can tolerate two or more faults, or we're dealing
4863 	 * with a slog, randomly online/offline vd0.
4864 	 */
4865 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4866 		if (ztest_random(10) < 6) {
4867 			int flags = (ztest_random(2) == 0 ?
4868 			    ZFS_OFFLINE_TEMPORARY : 0);
4869 
4870 			/*
4871 			 * We have to grab the zs_name_lock as writer to
4872 			 * prevent a race between offlining a slog and
4873 			 * destroying a dataset. Offlining the slog will
4874 			 * grab a reference on the dataset which may cause
4875 			 * dmu_objset_destroy() to fail with EBUSY thus
4876 			 * leaving the dataset in an inconsistent state.
4877 			 */
4878 			if (islog)
4879 				(void) rw_wrlock(&ztest_name_lock);
4880 
4881 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4882 
4883 			if (islog)
4884 				(void) rw_unlock(&ztest_name_lock);
4885 		} else {
4886 			/*
4887 			 * Ideally we would like to be able to randomly
4888 			 * call vdev_[on|off]line without holding locks
4889 			 * to force unpredictable failures but the side
4890 			 * effects of vdev_[on|off]line prevent us from
4891 			 * doing so. We grab the ztest_vdev_lock here to
4892 			 * prevent a race between injection testing and
4893 			 * aux_vdev removal.
4894 			 */
4895 			VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4896 			(void) vdev_online(spa, guid0, 0, NULL);
4897 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4898 		}
4899 	}
4900 
4901 	if (maxfaults == 0)
4902 		return;
4903 
4904 	/*
4905 	 * We have at least single-fault tolerance, so inject data corruption.
4906 	 */
4907 	fd = open(pathrand, O_RDWR);
4908 
4909 	if (fd == -1)	/* we hit a gap in the device namespace */
4910 		return;
4911 
4912 	fsize = lseek(fd, 0, SEEK_END);
4913 
4914 	while (--iters != 0) {
4915 		offset = ztest_random(fsize / (leaves << bshift)) *
4916 		    (leaves << bshift) + (leaf << bshift) +
4917 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4918 
4919 		if (offset >= fsize)
4920 			continue;
4921 
4922 		VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4923 		if (mirror_save != zs->zs_mirrors) {
4924 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4925 			(void) close(fd);
4926 			return;
4927 		}
4928 
4929 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4930 			fatal(1, "can't inject bad word at 0x%llx in %s",
4931 			    offset, pathrand);
4932 
4933 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4934 
4935 		if (ztest_opts.zo_verbose >= 7)
4936 			(void) printf("injected bad word into %s,"
4937 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4938 	}
4939 
4940 	(void) close(fd);
4941 }
4942 
4943 /*
4944  * Verify that DDT repair works as expected.
4945  */
4946 void
4947 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4948 {
4949 	ztest_shared_t *zs = ztest_shared;
4950 	spa_t *spa = ztest_spa;
4951 	objset_t *os = zd->zd_os;
4952 	ztest_od_t od[1];
4953 	uint64_t object, blocksize, txg, pattern, psize;
4954 	enum zio_checksum checksum = spa_dedup_checksum(spa);
4955 	dmu_buf_t *db;
4956 	dmu_tx_t *tx;
4957 	void *buf;
4958 	blkptr_t blk;
4959 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
4960 
4961 	blocksize = ztest_random_blocksize();
4962 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
4963 
4964 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4965 
4966 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4967 		return;
4968 
4969 	/*
4970 	 * Take the name lock as writer to prevent anyone else from changing
4971 	 * the pool and dataset properies we need to maintain during this test.
4972 	 */
4973 	(void) rw_wrlock(&ztest_name_lock);
4974 
4975 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4976 	    B_FALSE) != 0 ||
4977 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4978 	    B_FALSE) != 0) {
4979 		(void) rw_unlock(&ztest_name_lock);
4980 		return;
4981 	}
4982 
4983 	object = od[0].od_object;
4984 	blocksize = od[0].od_blocksize;
4985 	pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4986 
4987 	ASSERT(object != 0);
4988 
4989 	tx = dmu_tx_create(os);
4990 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4991 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4992 	if (txg == 0) {
4993 		(void) rw_unlock(&ztest_name_lock);
4994 		return;
4995 	}
4996 
4997 	/*
4998 	 * Write all the copies of our block.
4999 	 */
5000 	for (int i = 0; i < copies; i++) {
5001 		uint64_t offset = i * blocksize;
5002 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5003 		    DMU_READ_NO_PREFETCH);
5004 		if (error != 0) {
5005 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5006 			    os, (long long)object, (long long) offset, error);
5007 		}
5008 		ASSERT(db->db_offset == offset);
5009 		ASSERT(db->db_size == blocksize);
5010 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5011 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5012 		dmu_buf_will_fill(db, tx);
5013 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5014 		dmu_buf_rele(db, FTAG);
5015 	}
5016 
5017 	dmu_tx_commit(tx);
5018 	txg_wait_synced(spa_get_dsl(spa), txg);
5019 
5020 	/*
5021 	 * Find out what block we got.
5022 	 */
5023 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5024 	    DMU_READ_NO_PREFETCH));
5025 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5026 	dmu_buf_rele(db, FTAG);
5027 
5028 	/*
5029 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5030 	 */
5031 	psize = BP_GET_PSIZE(&blk);
5032 	buf = zio_buf_alloc(psize);
5033 	ztest_pattern_set(buf, psize, ~pattern);
5034 
5035 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5036 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5037 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5038 
5039 	zio_buf_free(buf, psize);
5040 
5041 	(void) rw_unlock(&ztest_name_lock);
5042 }
5043 
5044 /*
5045  * Scrub the pool.
5046  */
5047 /* ARGSUSED */
5048 void
5049 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5050 {
5051 	spa_t *spa = ztest_spa;
5052 
5053 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5054 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5055 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5056 }
5057 
5058 /*
5059  * Change the guid for the pool.
5060  */
5061 /* ARGSUSED */
5062 void
5063 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5064 {
5065 	spa_t *spa = ztest_spa;
5066 	uint64_t orig, load;
5067 	int error;
5068 
5069 	orig = spa_guid(spa);
5070 	load = spa_load_guid(spa);
5071 
5072 	(void) rw_wrlock(&ztest_name_lock);
5073 	error = spa_change_guid(spa);
5074 	(void) rw_unlock(&ztest_name_lock);
5075 
5076 	if (error != 0)
5077 		return;
5078 
5079 	if (ztest_opts.zo_verbose >= 4) {
5080 		(void) printf("Changed guid old %llu -> %llu\n",
5081 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5082 	}
5083 
5084 	VERIFY3U(orig, !=, spa_guid(spa));
5085 	VERIFY3U(load, ==, spa_load_guid(spa));
5086 }
5087 
5088 /*
5089  * Rename the pool to a different name and then rename it back.
5090  */
5091 /* ARGSUSED */
5092 void
5093 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5094 {
5095 	char *oldname, *newname;
5096 	spa_t *spa;
5097 
5098 	(void) rw_wrlock(&ztest_name_lock);
5099 
5100 	oldname = ztest_opts.zo_pool;
5101 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5102 	(void) strcpy(newname, oldname);
5103 	(void) strcat(newname, "_tmp");
5104 
5105 	/*
5106 	 * Do the rename
5107 	 */
5108 	VERIFY3U(0, ==, spa_rename(oldname, newname));
5109 
5110 	/*
5111 	 * Try to open it under the old name, which shouldn't exist
5112 	 */
5113 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5114 
5115 	/*
5116 	 * Open it under the new name and make sure it's still the same spa_t.
5117 	 */
5118 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5119 
5120 	ASSERT(spa == ztest_spa);
5121 	spa_close(spa, FTAG);
5122 
5123 	/*
5124 	 * Rename it back to the original
5125 	 */
5126 	VERIFY3U(0, ==, spa_rename(newname, oldname));
5127 
5128 	/*
5129 	 * Make sure it can still be opened
5130 	 */
5131 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5132 
5133 	ASSERT(spa == ztest_spa);
5134 	spa_close(spa, FTAG);
5135 
5136 	umem_free(newname, strlen(newname) + 1);
5137 
5138 	(void) rw_unlock(&ztest_name_lock);
5139 }
5140 
5141 /*
5142  * Verify pool integrity by running zdb.
5143  */
5144 static void
5145 ztest_run_zdb(char *pool)
5146 {
5147 	int status;
5148 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5149 	char zbuf[1024];
5150 	char *bin;
5151 	char *ztest;
5152 	char *isa;
5153 	int isalen;
5154 	FILE *fp;
5155 
5156 	(void) realpath(getexecname(), zdb);
5157 
5158 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5159 	bin = strstr(zdb, "/usr/bin/");
5160 	ztest = strstr(bin, "/ztest");
5161 	isa = bin + 8;
5162 	isalen = ztest - isa;
5163 	isa = strdup(isa);
5164 	/* LINTED */
5165 	(void) sprintf(bin,
5166 	    "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
5167 	    isalen,
5168 	    isa,
5169 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5170 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5171 	    spa_config_path,
5172 	    pool);
5173 	free(isa);
5174 
5175 	if (ztest_opts.zo_verbose >= 5)
5176 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5177 
5178 	fp = popen(zdb, "r");
5179 
5180 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5181 		if (ztest_opts.zo_verbose >= 3)
5182 			(void) printf("%s", zbuf);
5183 
5184 	status = pclose(fp);
5185 
5186 	if (status == 0)
5187 		return;
5188 
5189 	ztest_dump_core = 0;
5190 	if (WIFEXITED(status))
5191 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5192 	else
5193 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5194 }
5195 
5196 static void
5197 ztest_walk_pool_directory(char *header)
5198 {
5199 	spa_t *spa = NULL;
5200 
5201 	if (ztest_opts.zo_verbose >= 6)
5202 		(void) printf("%s\n", header);
5203 
5204 	mutex_enter(&spa_namespace_lock);
5205 	while ((spa = spa_next(spa)) != NULL)
5206 		if (ztest_opts.zo_verbose >= 6)
5207 			(void) printf("\t%s\n", spa_name(spa));
5208 	mutex_exit(&spa_namespace_lock);
5209 }
5210 
5211 static void
5212 ztest_spa_import_export(char *oldname, char *newname)
5213 {
5214 	nvlist_t *config, *newconfig;
5215 	uint64_t pool_guid;
5216 	spa_t *spa;
5217 	int error;
5218 
5219 	if (ztest_opts.zo_verbose >= 4) {
5220 		(void) printf("import/export: old = %s, new = %s\n",
5221 		    oldname, newname);
5222 	}
5223 
5224 	/*
5225 	 * Clean up from previous runs.
5226 	 */
5227 	(void) spa_destroy(newname);
5228 
5229 	/*
5230 	 * Get the pool's configuration and guid.
5231 	 */
5232 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5233 
5234 	/*
5235 	 * Kick off a scrub to tickle scrub/export races.
5236 	 */
5237 	if (ztest_random(2) == 0)
5238 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5239 
5240 	pool_guid = spa_guid(spa);
5241 	spa_close(spa, FTAG);
5242 
5243 	ztest_walk_pool_directory("pools before export");
5244 
5245 	/*
5246 	 * Export it.
5247 	 */
5248 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5249 
5250 	ztest_walk_pool_directory("pools after export");
5251 
5252 	/*
5253 	 * Try to import it.
5254 	 */
5255 	newconfig = spa_tryimport(config);
5256 	ASSERT(newconfig != NULL);
5257 	nvlist_free(newconfig);
5258 
5259 	/*
5260 	 * Import it under the new name.
5261 	 */
5262 	error = spa_import(newname, config, NULL, 0);
5263 	if (error != 0) {
5264 		dump_nvlist(config, 0);
5265 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5266 		    oldname, newname, error);
5267 	}
5268 
5269 	ztest_walk_pool_directory("pools after import");
5270 
5271 	/*
5272 	 * Try to import it again -- should fail with EEXIST.
5273 	 */
5274 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5275 
5276 	/*
5277 	 * Try to import it under a different name -- should fail with EEXIST.
5278 	 */
5279 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5280 
5281 	/*
5282 	 * Verify that the pool is no longer visible under the old name.
5283 	 */
5284 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5285 
5286 	/*
5287 	 * Verify that we can open and close the pool using the new name.
5288 	 */
5289 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5290 	ASSERT(pool_guid == spa_guid(spa));
5291 	spa_close(spa, FTAG);
5292 
5293 	nvlist_free(config);
5294 }
5295 
5296 static void
5297 ztest_resume(spa_t *spa)
5298 {
5299 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5300 		(void) printf("resuming from suspended state\n");
5301 	spa_vdev_state_enter(spa, SCL_NONE);
5302 	vdev_clear(spa, NULL);
5303 	(void) spa_vdev_state_exit(spa, NULL, 0);
5304 	(void) zio_resume(spa);
5305 }
5306 
5307 static void *
5308 ztest_resume_thread(void *arg)
5309 {
5310 	spa_t *spa = arg;
5311 
5312 	while (!ztest_exiting) {
5313 		if (spa_suspended(spa))
5314 			ztest_resume(spa);
5315 		(void) poll(NULL, 0, 100);
5316 	}
5317 	return (NULL);
5318 }
5319 
5320 static void *
5321 ztest_deadman_thread(void *arg)
5322 {
5323 	ztest_shared_t *zs = arg;
5324 	spa_t *spa = ztest_spa;
5325 	hrtime_t delta, total = 0;
5326 
5327 	for (;;) {
5328 		delta = zs->zs_thread_stop - zs->zs_thread_start +
5329 		    MSEC2NSEC(zfs_deadman_synctime_ms);
5330 
5331 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5332 
5333 		/*
5334 		 * If the pool is suspended then fail immediately. Otherwise,
5335 		 * check to see if the pool is making any progress. If
5336 		 * vdev_deadman() discovers that there hasn't been any recent
5337 		 * I/Os then it will end up aborting the tests.
5338 		 */
5339 		if (spa_suspended(spa)) {
5340 			fatal(0, "aborting test after %llu seconds because "
5341 			    "pool has transitioned to a suspended state.",
5342 			    zfs_deadman_synctime_ms / 1000);
5343 			return (NULL);
5344 		}
5345 		vdev_deadman(spa->spa_root_vdev);
5346 
5347 		total += zfs_deadman_synctime_ms/1000;
5348 		(void) printf("ztest has been running for %lld seconds\n",
5349 		    total);
5350 	}
5351 }
5352 
5353 static void
5354 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5355 {
5356 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5357 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5358 	hrtime_t functime = gethrtime();
5359 
5360 	for (int i = 0; i < zi->zi_iters; i++)
5361 		zi->zi_func(zd, id);
5362 
5363 	functime = gethrtime() - functime;
5364 
5365 	atomic_add_64(&zc->zc_count, 1);
5366 	atomic_add_64(&zc->zc_time, functime);
5367 
5368 	if (ztest_opts.zo_verbose >= 4) {
5369 		Dl_info dli;
5370 		(void) dladdr((void *)zi->zi_func, &dli);
5371 		(void) printf("%6.2f sec in %s\n",
5372 		    (double)functime / NANOSEC, dli.dli_sname);
5373 	}
5374 }
5375 
5376 static void *
5377 ztest_thread(void *arg)
5378 {
5379 	int rand;
5380 	uint64_t id = (uintptr_t)arg;
5381 	ztest_shared_t *zs = ztest_shared;
5382 	uint64_t call_next;
5383 	hrtime_t now;
5384 	ztest_info_t *zi;
5385 	ztest_shared_callstate_t *zc;
5386 
5387 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5388 		/*
5389 		 * See if it's time to force a crash.
5390 		 */
5391 		if (now > zs->zs_thread_kill)
5392 			ztest_kill(zs);
5393 
5394 		/*
5395 		 * If we're getting ENOSPC with some regularity, stop.
5396 		 */
5397 		if (zs->zs_enospc_count > 10)
5398 			break;
5399 
5400 		/*
5401 		 * Pick a random function to execute.
5402 		 */
5403 		rand = ztest_random(ZTEST_FUNCS);
5404 		zi = &ztest_info[rand];
5405 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5406 		call_next = zc->zc_next;
5407 
5408 		if (now >= call_next &&
5409 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5410 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5411 			ztest_execute(rand, zi, id);
5412 		}
5413 	}
5414 
5415 	return (NULL);
5416 }
5417 
5418 static void
5419 ztest_dataset_name(char *dsname, char *pool, int d)
5420 {
5421 	(void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5422 }
5423 
5424 static void
5425 ztest_dataset_destroy(int d)
5426 {
5427 	char name[MAXNAMELEN];
5428 
5429 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5430 
5431 	if (ztest_opts.zo_verbose >= 3)
5432 		(void) printf("Destroying %s to free up space\n", name);
5433 
5434 	/*
5435 	 * Cleanup any non-standard clones and snapshots.  In general,
5436 	 * ztest thread t operates on dataset (t % zopt_datasets),
5437 	 * so there may be more than one thing to clean up.
5438 	 */
5439 	for (int t = d; t < ztest_opts.zo_threads;
5440 	    t += ztest_opts.zo_datasets) {
5441 		ztest_dsl_dataset_cleanup(name, t);
5442 	}
5443 
5444 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5445 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5446 }
5447 
5448 static void
5449 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5450 {
5451 	uint64_t usedobjs, dirobjs, scratch;
5452 
5453 	/*
5454 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5455 	 * Therefore, the number of objects in use should equal the
5456 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5457 	 * If not, we have an object leak.
5458 	 *
5459 	 * Note that we can only check this in ztest_dataset_open(),
5460 	 * when the open-context and syncing-context values agree.
5461 	 * That's because zap_count() returns the open-context value,
5462 	 * while dmu_objset_space() returns the rootbp fill count.
5463 	 */
5464 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5465 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5466 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5467 }
5468 
5469 static int
5470 ztest_dataset_open(int d)
5471 {
5472 	ztest_ds_t *zd = &ztest_ds[d];
5473 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5474 	objset_t *os;
5475 	zilog_t *zilog;
5476 	char name[MAXNAMELEN];
5477 	int error;
5478 
5479 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5480 
5481 	(void) rw_rdlock(&ztest_name_lock);
5482 
5483 	error = ztest_dataset_create(name);
5484 	if (error == ENOSPC) {
5485 		(void) rw_unlock(&ztest_name_lock);
5486 		ztest_record_enospc(FTAG);
5487 		return (error);
5488 	}
5489 	ASSERT(error == 0 || error == EEXIST);
5490 
5491 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5492 	(void) rw_unlock(&ztest_name_lock);
5493 
5494 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5495 
5496 	zilog = zd->zd_zilog;
5497 
5498 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5499 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5500 		fatal(0, "missing log records: claimed %llu < committed %llu",
5501 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5502 
5503 	ztest_dataset_dirobj_verify(zd);
5504 
5505 	zil_replay(os, zd, ztest_replay_vector);
5506 
5507 	ztest_dataset_dirobj_verify(zd);
5508 
5509 	if (ztest_opts.zo_verbose >= 6)
5510 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5511 		    zd->zd_name,
5512 		    (u_longlong_t)zilog->zl_parse_blk_count,
5513 		    (u_longlong_t)zilog->zl_parse_lr_count,
5514 		    (u_longlong_t)zilog->zl_replaying_seq);
5515 
5516 	zilog = zil_open(os, ztest_get_data);
5517 
5518 	if (zilog->zl_replaying_seq != 0 &&
5519 	    zilog->zl_replaying_seq < committed_seq)
5520 		fatal(0, "missing log records: replayed %llu < committed %llu",
5521 		    zilog->zl_replaying_seq, committed_seq);
5522 
5523 	return (0);
5524 }
5525 
5526 static void
5527 ztest_dataset_close(int d)
5528 {
5529 	ztest_ds_t *zd = &ztest_ds[d];
5530 
5531 	zil_close(zd->zd_zilog);
5532 	dmu_objset_disown(zd->zd_os, zd);
5533 
5534 	ztest_zd_fini(zd);
5535 }
5536 
5537 /*
5538  * Kick off threads to run tests on all datasets in parallel.
5539  */
5540 static void
5541 ztest_run(ztest_shared_t *zs)
5542 {
5543 	thread_t *tid;
5544 	spa_t *spa;
5545 	objset_t *os;
5546 	thread_t resume_tid;
5547 	int error;
5548 
5549 	ztest_exiting = B_FALSE;
5550 
5551 	/*
5552 	 * Initialize parent/child shared state.
5553 	 */
5554 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5555 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5556 
5557 	zs->zs_thread_start = gethrtime();
5558 	zs->zs_thread_stop =
5559 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5560 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5561 	zs->zs_thread_kill = zs->zs_thread_stop;
5562 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5563 		zs->zs_thread_kill -=
5564 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5565 	}
5566 
5567 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5568 
5569 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5570 	    offsetof(ztest_cb_data_t, zcd_node));
5571 
5572 	/*
5573 	 * Open our pool.
5574 	 */
5575 	kernel_init(FREAD | FWRITE);
5576 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5577 	spa->spa_debug = B_TRUE;
5578 	ztest_spa = spa;
5579 
5580 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5581 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
5582 	zs->zs_guid = dmu_objset_fsid_guid(os);
5583 	dmu_objset_disown(os, FTAG);
5584 
5585 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5586 
5587 	/*
5588 	 * We don't expect the pool to suspend unless maxfaults == 0,
5589 	 * in which case ztest_fault_inject() temporarily takes away
5590 	 * the only valid replica.
5591 	 */
5592 	if (MAXFAULTS() == 0)
5593 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5594 	else
5595 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5596 
5597 	/*
5598 	 * Create a thread to periodically resume suspended I/O.
5599 	 */
5600 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5601 	    &resume_tid) == 0);
5602 
5603 	/*
5604 	 * Create a deadman thread to abort() if we hang.
5605 	 */
5606 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5607 	    NULL) == 0);
5608 
5609 	/*
5610 	 * Verify that we can safely inquire about about any object,
5611 	 * whether it's allocated or not.  To make it interesting,
5612 	 * we probe a 5-wide window around each power of two.
5613 	 * This hits all edge cases, including zero and the max.
5614 	 */
5615 	for (int t = 0; t < 64; t++) {
5616 		for (int d = -5; d <= 5; d++) {
5617 			error = dmu_object_info(spa->spa_meta_objset,
5618 			    (1ULL << t) + d, NULL);
5619 			ASSERT(error == 0 || error == ENOENT ||
5620 			    error == EINVAL);
5621 		}
5622 	}
5623 
5624 	/*
5625 	 * If we got any ENOSPC errors on the previous run, destroy something.
5626 	 */
5627 	if (zs->zs_enospc_count != 0) {
5628 		int d = ztest_random(ztest_opts.zo_datasets);
5629 		ztest_dataset_destroy(d);
5630 	}
5631 	zs->zs_enospc_count = 0;
5632 
5633 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5634 	    UMEM_NOFAIL);
5635 
5636 	if (ztest_opts.zo_verbose >= 4)
5637 		(void) printf("starting main threads...\n");
5638 
5639 	/*
5640 	 * Kick off all the tests that run in parallel.
5641 	 */
5642 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
5643 		if (t < ztest_opts.zo_datasets &&
5644 		    ztest_dataset_open(t) != 0)
5645 			return;
5646 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5647 		    THR_BOUND, &tid[t]) == 0);
5648 	}
5649 
5650 	/*
5651 	 * Wait for all of the tests to complete.  We go in reverse order
5652 	 * so we don't close datasets while threads are still using them.
5653 	 */
5654 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5655 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5656 		if (t < ztest_opts.zo_datasets)
5657 			ztest_dataset_close(t);
5658 	}
5659 
5660 	txg_wait_synced(spa_get_dsl(spa), 0);
5661 
5662 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5663 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5664 	zfs_dbgmsg_print(FTAG);
5665 
5666 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5667 
5668 	/* Kill the resume thread */
5669 	ztest_exiting = B_TRUE;
5670 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5671 	ztest_resume(spa);
5672 
5673 	/*
5674 	 * Right before closing the pool, kick off a bunch of async I/O;
5675 	 * spa_close() should wait for it to complete.
5676 	 */
5677 	for (uint64_t object = 1; object < 50; object++)
5678 		dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5679 
5680 	spa_close(spa, FTAG);
5681 
5682 	/*
5683 	 * Verify that we can loop over all pools.
5684 	 */
5685 	mutex_enter(&spa_namespace_lock);
5686 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5687 		if (ztest_opts.zo_verbose > 3)
5688 			(void) printf("spa_next: found %s\n", spa_name(spa));
5689 	mutex_exit(&spa_namespace_lock);
5690 
5691 	/*
5692 	 * Verify that we can export the pool and reimport it under a
5693 	 * different name.
5694 	 */
5695 	if (ztest_random(2) == 0) {
5696 		char name[MAXNAMELEN];
5697 		(void) snprintf(name, MAXNAMELEN, "%s_import",
5698 		    ztest_opts.zo_pool);
5699 		ztest_spa_import_export(ztest_opts.zo_pool, name);
5700 		ztest_spa_import_export(name, ztest_opts.zo_pool);
5701 	}
5702 
5703 	kernel_fini();
5704 
5705 	list_destroy(&zcl.zcl_callbacks);
5706 
5707 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5708 
5709 	(void) rwlock_destroy(&ztest_name_lock);
5710 	(void) _mutex_destroy(&ztest_vdev_lock);
5711 }
5712 
5713 static void
5714 ztest_freeze(void)
5715 {
5716 	ztest_ds_t *zd = &ztest_ds[0];
5717 	spa_t *spa;
5718 	int numloops = 0;
5719 
5720 	if (ztest_opts.zo_verbose >= 3)
5721 		(void) printf("testing spa_freeze()...\n");
5722 
5723 	kernel_init(FREAD | FWRITE);
5724 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5725 	VERIFY3U(0, ==, ztest_dataset_open(0));
5726 	spa->spa_debug = B_TRUE;
5727 	ztest_spa = spa;
5728 
5729 	/*
5730 	 * Force the first log block to be transactionally allocated.
5731 	 * We have to do this before we freeze the pool -- otherwise
5732 	 * the log chain won't be anchored.
5733 	 */
5734 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5735 		ztest_dmu_object_alloc_free(zd, 0);
5736 		zil_commit(zd->zd_zilog, 0);
5737 	}
5738 
5739 	txg_wait_synced(spa_get_dsl(spa), 0);
5740 
5741 	/*
5742 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5743 	 * so that the only way to record changes from now on is the ZIL.
5744 	 */
5745 	spa_freeze(spa);
5746 
5747 	/*
5748 	 * Run tests that generate log records but don't alter the pool config
5749 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5750 	 * We do a txg_wait_synced() after each iteration to force the txg
5751 	 * to increase well beyond the last synced value in the uberblock.
5752 	 * The ZIL should be OK with that.
5753 	 */
5754 	while (ztest_random(10) != 0 &&
5755 	    numloops++ < ztest_opts.zo_maxloops) {
5756 		ztest_dmu_write_parallel(zd, 0);
5757 		ztest_dmu_object_alloc_free(zd, 0);
5758 		txg_wait_synced(spa_get_dsl(spa), 0);
5759 	}
5760 
5761 	/*
5762 	 * Commit all of the changes we just generated.
5763 	 */
5764 	zil_commit(zd->zd_zilog, 0);
5765 	txg_wait_synced(spa_get_dsl(spa), 0);
5766 
5767 	/*
5768 	 * Close our dataset and close the pool.
5769 	 */
5770 	ztest_dataset_close(0);
5771 	spa_close(spa, FTAG);
5772 	kernel_fini();
5773 
5774 	/*
5775 	 * Open and close the pool and dataset to induce log replay.
5776 	 */
5777 	kernel_init(FREAD | FWRITE);
5778 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5779 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5780 	VERIFY3U(0, ==, ztest_dataset_open(0));
5781 	ztest_dataset_close(0);
5782 
5783 	spa->spa_debug = B_TRUE;
5784 	ztest_spa = spa;
5785 	txg_wait_synced(spa_get_dsl(spa), 0);
5786 	ztest_reguid(NULL, 0);
5787 
5788 	spa_close(spa, FTAG);
5789 	kernel_fini();
5790 }
5791 
5792 void
5793 print_time(hrtime_t t, char *timebuf)
5794 {
5795 	hrtime_t s = t / NANOSEC;
5796 	hrtime_t m = s / 60;
5797 	hrtime_t h = m / 60;
5798 	hrtime_t d = h / 24;
5799 
5800 	s -= m * 60;
5801 	m -= h * 60;
5802 	h -= d * 24;
5803 
5804 	timebuf[0] = '\0';
5805 
5806 	if (d)
5807 		(void) sprintf(timebuf,
5808 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5809 	else if (h)
5810 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5811 	else if (m)
5812 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5813 	else
5814 		(void) sprintf(timebuf, "%llus", s);
5815 }
5816 
5817 static nvlist_t *
5818 make_random_props()
5819 {
5820 	nvlist_t *props;
5821 
5822 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5823 	if (ztest_random(2) == 0)
5824 		return (props);
5825 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5826 
5827 	return (props);
5828 }
5829 
5830 /*
5831  * Create a storage pool with the given name and initial vdev size.
5832  * Then test spa_freeze() functionality.
5833  */
5834 static void
5835 ztest_init(ztest_shared_t *zs)
5836 {
5837 	spa_t *spa;
5838 	nvlist_t *nvroot, *props;
5839 
5840 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5841 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5842 
5843 	kernel_init(FREAD | FWRITE);
5844 
5845 	/*
5846 	 * Create the storage pool.
5847 	 */
5848 	(void) spa_destroy(ztest_opts.zo_pool);
5849 	ztest_shared->zs_vdev_next_leaf = 0;
5850 	zs->zs_splits = 0;
5851 	zs->zs_mirrors = ztest_opts.zo_mirrors;
5852 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5853 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5854 	props = make_random_props();
5855 	for (int i = 0; i < SPA_FEATURES; i++) {
5856 		char buf[1024];
5857 		(void) snprintf(buf, sizeof (buf), "feature@%s",
5858 		    spa_feature_table[i].fi_uname);
5859 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5860 	}
5861 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5862 	nvlist_free(nvroot);
5863 
5864 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5865 	zs->zs_metaslab_sz =
5866 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5867 
5868 	spa_close(spa, FTAG);
5869 
5870 	kernel_fini();
5871 
5872 	ztest_run_zdb(ztest_opts.zo_pool);
5873 
5874 	ztest_freeze();
5875 
5876 	ztest_run_zdb(ztest_opts.zo_pool);
5877 
5878 	(void) rwlock_destroy(&ztest_name_lock);
5879 	(void) _mutex_destroy(&ztest_vdev_lock);
5880 }
5881 
5882 static void
5883 setup_data_fd(void)
5884 {
5885 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5886 
5887 	ztest_fd_data = mkstemp(ztest_name_data);
5888 	ASSERT3S(ztest_fd_data, >=, 0);
5889 	(void) unlink(ztest_name_data);
5890 }
5891 
5892 
5893 static int
5894 shared_data_size(ztest_shared_hdr_t *hdr)
5895 {
5896 	int size;
5897 
5898 	size = hdr->zh_hdr_size;
5899 	size += hdr->zh_opts_size;
5900 	size += hdr->zh_size;
5901 	size += hdr->zh_stats_size * hdr->zh_stats_count;
5902 	size += hdr->zh_ds_size * hdr->zh_ds_count;
5903 
5904 	return (size);
5905 }
5906 
5907 static void
5908 setup_hdr(void)
5909 {
5910 	int size;
5911 	ztest_shared_hdr_t *hdr;
5912 
5913 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5914 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5915 	ASSERT(hdr != MAP_FAILED);
5916 
5917 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5918 
5919 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5920 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5921 	hdr->zh_size = sizeof (ztest_shared_t);
5922 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5923 	hdr->zh_stats_count = ZTEST_FUNCS;
5924 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5925 	hdr->zh_ds_count = ztest_opts.zo_datasets;
5926 
5927 	size = shared_data_size(hdr);
5928 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5929 
5930 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5931 }
5932 
5933 static void
5934 setup_data(void)
5935 {
5936 	int size, offset;
5937 	ztest_shared_hdr_t *hdr;
5938 	uint8_t *buf;
5939 
5940 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5941 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5942 	ASSERT(hdr != MAP_FAILED);
5943 
5944 	size = shared_data_size(hdr);
5945 
5946 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5947 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5948 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5949 	ASSERT(hdr != MAP_FAILED);
5950 	buf = (uint8_t *)hdr;
5951 
5952 	offset = hdr->zh_hdr_size;
5953 	ztest_shared_opts = (void *)&buf[offset];
5954 	offset += hdr->zh_opts_size;
5955 	ztest_shared = (void *)&buf[offset];
5956 	offset += hdr->zh_size;
5957 	ztest_shared_callstate = (void *)&buf[offset];
5958 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
5959 	ztest_shared_ds = (void *)&buf[offset];
5960 }
5961 
5962 static boolean_t
5963 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5964 {
5965 	pid_t pid;
5966 	int status;
5967 	char *cmdbuf = NULL;
5968 
5969 	pid = fork();
5970 
5971 	if (cmd == NULL) {
5972 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5973 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
5974 		cmd = cmdbuf;
5975 	}
5976 
5977 	if (pid == -1)
5978 		fatal(1, "fork failed");
5979 
5980 	if (pid == 0) {	/* child */
5981 		char *emptyargv[2] = { cmd, NULL };
5982 		char fd_data_str[12];
5983 
5984 		struct rlimit rl = { 1024, 1024 };
5985 		(void) setrlimit(RLIMIT_NOFILE, &rl);
5986 
5987 		(void) close(ztest_fd_rand);
5988 		VERIFY3U(11, >=,
5989 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
5990 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
5991 
5992 		(void) enable_extended_FILE_stdio(-1, -1);
5993 		if (libpath != NULL)
5994 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5995 		(void) execv(cmd, emptyargv);
5996 		ztest_dump_core = B_FALSE;
5997 		fatal(B_TRUE, "exec failed: %s", cmd);
5998 	}
5999 
6000 	if (cmdbuf != NULL) {
6001 		umem_free(cmdbuf, MAXPATHLEN);
6002 		cmd = NULL;
6003 	}
6004 
6005 	while (waitpid(pid, &status, 0) != pid)
6006 		continue;
6007 	if (statusp != NULL)
6008 		*statusp = status;
6009 
6010 	if (WIFEXITED(status)) {
6011 		if (WEXITSTATUS(status) != 0) {
6012 			(void) fprintf(stderr, "child exited with code %d\n",
6013 			    WEXITSTATUS(status));
6014 			exit(2);
6015 		}
6016 		return (B_FALSE);
6017 	} else if (WIFSIGNALED(status)) {
6018 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6019 			(void) fprintf(stderr, "child died with signal %d\n",
6020 			    WTERMSIG(status));
6021 			exit(3);
6022 		}
6023 		return (B_TRUE);
6024 	} else {
6025 		(void) fprintf(stderr, "something strange happened to child\n");
6026 		exit(4);
6027 		/* NOTREACHED */
6028 	}
6029 }
6030 
6031 static void
6032 ztest_run_init(void)
6033 {
6034 	ztest_shared_t *zs = ztest_shared;
6035 
6036 	ASSERT(ztest_opts.zo_init != 0);
6037 
6038 	/*
6039 	 * Blow away any existing copy of zpool.cache
6040 	 */
6041 	(void) remove(spa_config_path);
6042 
6043 	/*
6044 	 * Create and initialize our storage pool.
6045 	 */
6046 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
6047 		bzero(zs, sizeof (ztest_shared_t));
6048 		if (ztest_opts.zo_verbose >= 3 &&
6049 		    ztest_opts.zo_init != 1) {
6050 			(void) printf("ztest_init(), pass %d\n", i);
6051 		}
6052 		ztest_init(zs);
6053 	}
6054 }
6055 
6056 int
6057 main(int argc, char **argv)
6058 {
6059 	int kills = 0;
6060 	int iters = 0;
6061 	int older = 0;
6062 	int newer = 0;
6063 	ztest_shared_t *zs;
6064 	ztest_info_t *zi;
6065 	ztest_shared_callstate_t *zc;
6066 	char timebuf[100];
6067 	char numbuf[6];
6068 	spa_t *spa;
6069 	char *cmd;
6070 	boolean_t hasalt;
6071 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6072 
6073 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6074 
6075 	dprintf_setup(&argc, argv);
6076 	zfs_deadman_synctime_ms = 300000;
6077 
6078 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6079 	ASSERT3S(ztest_fd_rand, >=, 0);
6080 
6081 	if (!fd_data_str) {
6082 		process_options(argc, argv);
6083 
6084 		setup_data_fd();
6085 		setup_hdr();
6086 		setup_data();
6087 		bcopy(&ztest_opts, ztest_shared_opts,
6088 		    sizeof (*ztest_shared_opts));
6089 	} else {
6090 		ztest_fd_data = atoi(fd_data_str);
6091 		setup_data();
6092 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6093 	}
6094 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6095 
6096 	/* Override location of zpool.cache */
6097 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6098 	    ztest_opts.zo_dir), !=, -1);
6099 
6100 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6101 	    UMEM_NOFAIL);
6102 	zs = ztest_shared;
6103 
6104 	if (fd_data_str) {
6105 		metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6106 		metaslab_df_alloc_threshold =
6107 		    zs->zs_metaslab_df_alloc_threshold;
6108 
6109 		if (zs->zs_do_init)
6110 			ztest_run_init();
6111 		else
6112 			ztest_run(zs);
6113 		exit(0);
6114 	}
6115 
6116 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6117 
6118 	if (ztest_opts.zo_verbose >= 1) {
6119 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6120 		    " %llu seconds...\n",
6121 		    (u_longlong_t)ztest_opts.zo_vdevs,
6122 		    ztest_opts.zo_datasets,
6123 		    ztest_opts.zo_threads,
6124 		    (u_longlong_t)ztest_opts.zo_time);
6125 	}
6126 
6127 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6128 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6129 
6130 	zs->zs_do_init = B_TRUE;
6131 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6132 		if (ztest_opts.zo_verbose >= 1) {
6133 			(void) printf("Executing older ztest for "
6134 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6135 		}
6136 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6137 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6138 	} else {
6139 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6140 	}
6141 	zs->zs_do_init = B_FALSE;
6142 
6143 	zs->zs_proc_start = gethrtime();
6144 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6145 
6146 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6147 		zi = &ztest_info[f];
6148 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6149 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6150 			zc->zc_next = UINT64_MAX;
6151 		else
6152 			zc->zc_next = zs->zs_proc_start +
6153 			    ztest_random(2 * zi->zi_interval[0] + 1);
6154 	}
6155 
6156 	/*
6157 	 * Run the tests in a loop.  These tests include fault injection
6158 	 * to verify that self-healing data works, and forced crashes
6159 	 * to verify that we never lose on-disk consistency.
6160 	 */
6161 	while (gethrtime() < zs->zs_proc_stop) {
6162 		int status;
6163 		boolean_t killed;
6164 
6165 		/*
6166 		 * Initialize the workload counters for each function.
6167 		 */
6168 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6169 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6170 			zc->zc_count = 0;
6171 			zc->zc_time = 0;
6172 		}
6173 
6174 		/* Set the allocation switch size */
6175 		zs->zs_metaslab_df_alloc_threshold =
6176 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6177 
6178 		if (!hasalt || ztest_random(2) == 0) {
6179 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6180 				(void) printf("Executing newer ztest: %s\n",
6181 				    cmd);
6182 			}
6183 			newer++;
6184 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6185 		} else {
6186 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6187 				(void) printf("Executing older ztest: %s\n",
6188 				    ztest_opts.zo_alt_ztest);
6189 			}
6190 			older++;
6191 			killed = exec_child(ztest_opts.zo_alt_ztest,
6192 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6193 		}
6194 
6195 		if (killed)
6196 			kills++;
6197 		iters++;
6198 
6199 		if (ztest_opts.zo_verbose >= 1) {
6200 			hrtime_t now = gethrtime();
6201 
6202 			now = MIN(now, zs->zs_proc_stop);
6203 			print_time(zs->zs_proc_stop - now, timebuf);
6204 			nicenum(zs->zs_space, numbuf);
6205 
6206 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6207 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6208 			    iters,
6209 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6210 			    (u_longlong_t)zs->zs_enospc_count,
6211 			    100.0 * zs->zs_alloc / zs->zs_space,
6212 			    numbuf,
6213 			    100.0 * (now - zs->zs_proc_start) /
6214 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6215 		}
6216 
6217 		if (ztest_opts.zo_verbose >= 2) {
6218 			(void) printf("\nWorkload summary:\n\n");
6219 			(void) printf("%7s %9s   %s\n",
6220 			    "Calls", "Time", "Function");
6221 			(void) printf("%7s %9s   %s\n",
6222 			    "-----", "----", "--------");
6223 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6224 				Dl_info dli;
6225 
6226 				zi = &ztest_info[f];
6227 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6228 				print_time(zc->zc_time, timebuf);
6229 				(void) dladdr((void *)zi->zi_func, &dli);
6230 				(void) printf("%7llu %9s   %s\n",
6231 				    (u_longlong_t)zc->zc_count, timebuf,
6232 				    dli.dli_sname);
6233 			}
6234 			(void) printf("\n");
6235 		}
6236 
6237 		/*
6238 		 * It's possible that we killed a child during a rename test,
6239 		 * in which case we'll have a 'ztest_tmp' pool lying around
6240 		 * instead of 'ztest'.  Do a blind rename in case this happened.
6241 		 */
6242 		kernel_init(FREAD);
6243 		if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6244 			spa_close(spa, FTAG);
6245 		} else {
6246 			char tmpname[MAXNAMELEN];
6247 			kernel_fini();
6248 			kernel_init(FREAD | FWRITE);
6249 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6250 			    ztest_opts.zo_pool);
6251 			(void) spa_rename(tmpname, ztest_opts.zo_pool);
6252 		}
6253 		kernel_fini();
6254 
6255 		ztest_run_zdb(ztest_opts.zo_pool);
6256 	}
6257 
6258 	if (ztest_opts.zo_verbose >= 1) {
6259 		if (hasalt) {
6260 			(void) printf("%d runs of older ztest: %s\n", older,
6261 			    ztest_opts.zo_alt_ztest);
6262 			(void) printf("%d runs of newer ztest: %s\n", newer,
6263 			    cmd);
6264 		}
6265 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6266 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6267 	}
6268 
6269 	umem_free(cmd, MAXNAMELEN);
6270 
6271 	return (0);
6272 }
6273