xref: /titanic_50/usr/src/cmd/ztest/ztest.c (revision 25345e466695fbe736faa53b8f3413d8e8f81981)
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) 2012 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  */
26 
27 /*
28  * The objective of this program is to provide a DMU/ZAP/SPA stress test
29  * that runs entirely in userland, is easy to use, and easy to extend.
30  *
31  * The overall design of the ztest program is as follows:
32  *
33  * (1) For each major functional area (e.g. adding vdevs to a pool,
34  *     creating and destroying datasets, reading and writing objects, etc)
35  *     we have a simple routine to test that functionality.  These
36  *     individual routines do not have to do anything "stressful".
37  *
38  * (2) We turn these simple functionality tests into a stress test by
39  *     running them all in parallel, with as many threads as desired,
40  *     and spread across as many datasets, objects, and vdevs as desired.
41  *
42  * (3) While all this is happening, we inject faults into the pool to
43  *     verify that self-healing data really works.
44  *
45  * (4) Every time we open a dataset, we change its checksum and compression
46  *     functions.  Thus even individual objects vary from block to block
47  *     in which checksum they use and whether they're compressed.
48  *
49  * (5) To verify that we never lose on-disk consistency after a crash,
50  *     we run the entire test in a child of the main process.
51  *     At random times, the child self-immolates with a SIGKILL.
52  *     This is the software equivalent of pulling the power cord.
53  *     The parent then runs the test again, using the existing
54  *     storage pool, as many times as desired. If backwards compatability
55  *     testing is enabled ztest will sometimes run the "older" version
56  *     of ztest after a SIGKILL.
57  *
58  * (6) To verify that we don't have future leaks or temporal incursions,
59  *     many of the functional tests record the transaction group number
60  *     as part of their data.  When reading old data, they verify that
61  *     the transaction group number is less than the current, open txg.
62  *     If you add a new test, please do this if applicable.
63  *
64  * When run with no arguments, ztest runs for about five minutes and
65  * produces no output if successful.  To get a little bit of information,
66  * specify -V.  To get more information, specify -VV, and so on.
67  *
68  * To turn this into an overnight stress test, use -T to specify run time.
69  *
70  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
71  * to increase the pool capacity, fanout, and overall stress level.
72  *
73  * Use the -k option to set the desired frequency of kills.
74  *
75  * When ztest invokes itself it passes all relevant information through a
76  * temporary file which is mmap-ed in the child process. This allows shared
77  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
78  * stored at offset 0 of this file and contains information on the size and
79  * number of shared structures in the file. The information stored in this file
80  * must remain backwards compatible with older versions of ztest so that
81  * ztest can invoke them during backwards compatibility testing (-B).
82  */
83 
84 #include <sys/zfs_context.h>
85 #include <sys/spa.h>
86 #include <sys/dmu.h>
87 #include <sys/txg.h>
88 #include <sys/dbuf.h>
89 #include <sys/zap.h>
90 #include <sys/dmu_objset.h>
91 #include <sys/poll.h>
92 #include <sys/stat.h>
93 #include <sys/time.h>
94 #include <sys/wait.h>
95 #include <sys/mman.h>
96 #include <sys/resource.h>
97 #include <sys/zio.h>
98 #include <sys/zil.h>
99 #include <sys/zil_impl.h>
100 #include <sys/vdev_impl.h>
101 #include <sys/vdev_file.h>
102 #include <sys/spa_impl.h>
103 #include <sys/metaslab_impl.h>
104 #include <sys/dsl_prop.h>
105 #include <sys/dsl_dataset.h>
106 #include <sys/dsl_scan.h>
107 #include <sys/zio_checksum.h>
108 #include <sys/refcount.h>
109 #include <sys/zfeature.h>
110 #include <stdio.h>
111 #include <stdio_ext.h>
112 #include <stdlib.h>
113 #include <unistd.h>
114 #include <signal.h>
115 #include <umem.h>
116 #include <dlfcn.h>
117 #include <ctype.h>
118 #include <math.h>
119 #include <sys/fs/zfs.h>
120 #include <libnvpair.h>
121 
122 static int ztest_fd_data = -1;
123 static int ztest_fd_rand = -1;
124 
125 typedef struct ztest_shared_hdr {
126 	uint64_t	zh_hdr_size;
127 	uint64_t	zh_opts_size;
128 	uint64_t	zh_size;
129 	uint64_t	zh_stats_size;
130 	uint64_t	zh_stats_count;
131 	uint64_t	zh_ds_size;
132 	uint64_t	zh_ds_count;
133 } ztest_shared_hdr_t;
134 
135 static ztest_shared_hdr_t *ztest_shared_hdr;
136 
137 typedef struct ztest_shared_opts {
138 	char zo_pool[MAXNAMELEN];
139 	char zo_dir[MAXNAMELEN];
140 	char zo_alt_ztest[MAXNAMELEN];
141 	char zo_alt_libpath[MAXNAMELEN];
142 	uint64_t zo_vdevs;
143 	uint64_t zo_vdevtime;
144 	size_t zo_vdev_size;
145 	int zo_ashift;
146 	int zo_mirrors;
147 	int zo_raidz;
148 	int zo_raidz_parity;
149 	int zo_datasets;
150 	int zo_threads;
151 	uint64_t zo_passtime;
152 	uint64_t zo_killrate;
153 	int zo_verbose;
154 	int zo_init;
155 	uint64_t zo_time;
156 	uint64_t zo_maxloops;
157 	uint64_t zo_metaslab_gang_bang;
158 } ztest_shared_opts_t;
159 
160 static const ztest_shared_opts_t ztest_opts_defaults = {
161 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
162 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
163 	.zo_alt_ztest = { '\0' },
164 	.zo_alt_libpath = { '\0' },
165 	.zo_vdevs = 5,
166 	.zo_ashift = SPA_MINBLOCKSHIFT,
167 	.zo_mirrors = 2,
168 	.zo_raidz = 4,
169 	.zo_raidz_parity = 1,
170 	.zo_vdev_size = SPA_MINDEVSIZE,
171 	.zo_datasets = 7,
172 	.zo_threads = 23,
173 	.zo_passtime = 60,		/* 60 seconds */
174 	.zo_killrate = 70,		/* 70% kill rate */
175 	.zo_verbose = 0,
176 	.zo_init = 1,
177 	.zo_time = 300,			/* 5 minutes */
178 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
179 	.zo_metaslab_gang_bang = 32 << 10
180 };
181 
182 extern uint64_t metaslab_gang_bang;
183 extern uint64_t metaslab_df_alloc_threshold;
184 
185 static ztest_shared_opts_t *ztest_shared_opts;
186 static ztest_shared_opts_t ztest_opts;
187 
188 typedef struct ztest_shared_ds {
189 	uint64_t	zd_seq;
190 } ztest_shared_ds_t;
191 
192 static ztest_shared_ds_t *ztest_shared_ds;
193 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
194 
195 #define	BT_MAGIC	0x123456789abcdefULL
196 #define	MAXFAULTS() \
197 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
198 
199 enum ztest_io_type {
200 	ZTEST_IO_WRITE_TAG,
201 	ZTEST_IO_WRITE_PATTERN,
202 	ZTEST_IO_WRITE_ZEROES,
203 	ZTEST_IO_TRUNCATE,
204 	ZTEST_IO_SETATTR,
205 	ZTEST_IO_TYPES
206 };
207 
208 typedef struct ztest_block_tag {
209 	uint64_t	bt_magic;
210 	uint64_t	bt_objset;
211 	uint64_t	bt_object;
212 	uint64_t	bt_offset;
213 	uint64_t	bt_gen;
214 	uint64_t	bt_txg;
215 	uint64_t	bt_crtxg;
216 } ztest_block_tag_t;
217 
218 typedef struct bufwad {
219 	uint64_t	bw_index;
220 	uint64_t	bw_txg;
221 	uint64_t	bw_data;
222 } bufwad_t;
223 
224 /*
225  * XXX -- fix zfs range locks to be generic so we can use them here.
226  */
227 typedef enum {
228 	RL_READER,
229 	RL_WRITER,
230 	RL_APPEND
231 } rl_type_t;
232 
233 typedef struct rll {
234 	void		*rll_writer;
235 	int		rll_readers;
236 	mutex_t		rll_lock;
237 	cond_t		rll_cv;
238 } rll_t;
239 
240 typedef struct rl {
241 	uint64_t	rl_object;
242 	uint64_t	rl_offset;
243 	uint64_t	rl_size;
244 	rll_t		*rl_lock;
245 } rl_t;
246 
247 #define	ZTEST_RANGE_LOCKS	64
248 #define	ZTEST_OBJECT_LOCKS	64
249 
250 /*
251  * Object descriptor.  Used as a template for object lookup/create/remove.
252  */
253 typedef struct ztest_od {
254 	uint64_t	od_dir;
255 	uint64_t	od_object;
256 	dmu_object_type_t od_type;
257 	dmu_object_type_t od_crtype;
258 	uint64_t	od_blocksize;
259 	uint64_t	od_crblocksize;
260 	uint64_t	od_gen;
261 	uint64_t	od_crgen;
262 	char		od_name[MAXNAMELEN];
263 } ztest_od_t;
264 
265 /*
266  * Per-dataset state.
267  */
268 typedef struct ztest_ds {
269 	ztest_shared_ds_t *zd_shared;
270 	objset_t	*zd_os;
271 	rwlock_t	zd_zilog_lock;
272 	zilog_t		*zd_zilog;
273 	ztest_od_t	*zd_od;		/* debugging aid */
274 	char		zd_name[MAXNAMELEN];
275 	mutex_t		zd_dirobj_lock;
276 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
277 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
278 } ztest_ds_t;
279 
280 /*
281  * Per-iteration state.
282  */
283 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
284 
285 typedef struct ztest_info {
286 	ztest_func_t	*zi_func;	/* test function */
287 	uint64_t	zi_iters;	/* iterations per execution */
288 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
289 } ztest_info_t;
290 
291 typedef struct ztest_shared_callstate {
292 	uint64_t	zc_count;	/* per-pass count */
293 	uint64_t	zc_time;	/* per-pass time */
294 	uint64_t	zc_next;	/* next time to call this function */
295 } ztest_shared_callstate_t;
296 
297 static ztest_shared_callstate_t *ztest_shared_callstate;
298 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
299 
300 /*
301  * Note: these aren't static because we want dladdr() to work.
302  */
303 ztest_func_t ztest_dmu_read_write;
304 ztest_func_t ztest_dmu_write_parallel;
305 ztest_func_t ztest_dmu_object_alloc_free;
306 ztest_func_t ztest_dmu_commit_callbacks;
307 ztest_func_t ztest_zap;
308 ztest_func_t ztest_zap_parallel;
309 ztest_func_t ztest_zil_commit;
310 ztest_func_t ztest_zil_remount;
311 ztest_func_t ztest_dmu_read_write_zcopy;
312 ztest_func_t ztest_dmu_objset_create_destroy;
313 ztest_func_t ztest_dmu_prealloc;
314 ztest_func_t ztest_fzap;
315 ztest_func_t ztest_dmu_snapshot_create_destroy;
316 ztest_func_t ztest_dsl_prop_get_set;
317 ztest_func_t ztest_spa_prop_get_set;
318 ztest_func_t ztest_spa_create_destroy;
319 ztest_func_t ztest_fault_inject;
320 ztest_func_t ztest_ddt_repair;
321 ztest_func_t ztest_dmu_snapshot_hold;
322 ztest_func_t ztest_spa_rename;
323 ztest_func_t ztest_scrub;
324 ztest_func_t ztest_dsl_dataset_promote_busy;
325 ztest_func_t ztest_vdev_attach_detach;
326 ztest_func_t ztest_vdev_LUN_growth;
327 ztest_func_t ztest_vdev_add_remove;
328 ztest_func_t ztest_vdev_aux_add_remove;
329 ztest_func_t ztest_split_pool;
330 ztest_func_t ztest_reguid;
331 ztest_func_t ztest_spa_upgrade;
332 
333 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
334 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
335 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
336 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
337 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
338 
339 ztest_info_t ztest_info[] = {
340 	{ ztest_dmu_read_write,			1,	&zopt_always	},
341 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
342 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
343 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
344 	{ ztest_zap,				30,	&zopt_always	},
345 	{ ztest_zap_parallel,			100,	&zopt_always	},
346 	{ ztest_split_pool,			1,	&zopt_always	},
347 	{ ztest_zil_commit,			1,	&zopt_incessant	},
348 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
349 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
350 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
351 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
352 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
353 #if 0
354 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
355 #endif
356 	{ ztest_fzap,				1,	&zopt_sometimes	},
357 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
358 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
359 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
360 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
361 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
362 	{ ztest_reguid,				1,	&zopt_sometimes },
363 	{ ztest_spa_rename,			1,	&zopt_rarely	},
364 	{ ztest_scrub,				1,	&zopt_rarely	},
365 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
366 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
367 	{ ztest_vdev_attach_detach,		1,	&zopt_rarely	},
368 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
369 	{ ztest_vdev_add_remove,		1,
370 	    &ztest_opts.zo_vdevtime				},
371 	{ ztest_vdev_aux_add_remove,		1,
372 	    &ztest_opts.zo_vdevtime				},
373 };
374 
375 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
376 
377 /*
378  * The following struct is used to hold a list of uncalled commit callbacks.
379  * The callbacks are ordered by txg number.
380  */
381 typedef struct ztest_cb_list {
382 	mutex_t	zcl_callbacks_lock;
383 	list_t	zcl_callbacks;
384 } ztest_cb_list_t;
385 
386 /*
387  * Stuff we need to share writably between parent and child.
388  */
389 typedef struct ztest_shared {
390 	boolean_t	zs_do_init;
391 	hrtime_t	zs_proc_start;
392 	hrtime_t	zs_proc_stop;
393 	hrtime_t	zs_thread_start;
394 	hrtime_t	zs_thread_stop;
395 	hrtime_t	zs_thread_kill;
396 	uint64_t	zs_enospc_count;
397 	uint64_t	zs_vdev_next_leaf;
398 	uint64_t	zs_vdev_aux;
399 	uint64_t	zs_alloc;
400 	uint64_t	zs_space;
401 	uint64_t	zs_splits;
402 	uint64_t	zs_mirrors;
403 	uint64_t	zs_metaslab_sz;
404 	uint64_t	zs_metaslab_df_alloc_threshold;
405 	uint64_t	zs_guid;
406 } ztest_shared_t;
407 
408 #define	ID_PARALLEL	-1ULL
409 
410 static char ztest_dev_template[] = "%s/%s.%llua";
411 static char ztest_aux_template[] = "%s/%s.%s.%llu";
412 ztest_shared_t *ztest_shared;
413 
414 static spa_t *ztest_spa = NULL;
415 static ztest_ds_t *ztest_ds;
416 
417 static mutex_t ztest_vdev_lock;
418 
419 /*
420  * The ztest_name_lock protects the pool and dataset namespace used by
421  * the individual tests. To modify the namespace, consumers must grab
422  * this lock as writer. Grabbing the lock as reader will ensure that the
423  * namespace does not change while the lock is held.
424  */
425 static rwlock_t ztest_name_lock;
426 
427 static boolean_t ztest_dump_core = B_TRUE;
428 static boolean_t ztest_exiting;
429 
430 /* Global commit callback list */
431 static ztest_cb_list_t zcl;
432 
433 enum ztest_object {
434 	ZTEST_META_DNODE = 0,
435 	ZTEST_DIROBJ,
436 	ZTEST_OBJECTS
437 };
438 
439 static void usage(boolean_t) __NORETURN;
440 
441 /*
442  * These libumem hooks provide a reasonable set of defaults for the allocator's
443  * debugging facilities.
444  */
445 const char *
446 _umem_debug_init()
447 {
448 	return ("default,verbose"); /* $UMEM_DEBUG setting */
449 }
450 
451 const char *
452 _umem_logging_init(void)
453 {
454 	return ("fail,contents"); /* $UMEM_LOGGING setting */
455 }
456 
457 #define	FATAL_MSG_SZ	1024
458 
459 char *fatal_msg;
460 
461 static void
462 fatal(int do_perror, char *message, ...)
463 {
464 	va_list args;
465 	int save_errno = errno;
466 	char buf[FATAL_MSG_SZ];
467 
468 	(void) fflush(stdout);
469 
470 	va_start(args, message);
471 	(void) sprintf(buf, "ztest: ");
472 	/* LINTED */
473 	(void) vsprintf(buf + strlen(buf), message, args);
474 	va_end(args);
475 	if (do_perror) {
476 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
477 		    ": %s", strerror(save_errno));
478 	}
479 	(void) fprintf(stderr, "%s\n", buf);
480 	fatal_msg = buf;			/* to ease debugging */
481 	if (ztest_dump_core)
482 		abort();
483 	exit(3);
484 }
485 
486 static int
487 str2shift(const char *buf)
488 {
489 	const char *ends = "BKMGTPEZ";
490 	int i;
491 
492 	if (buf[0] == '\0')
493 		return (0);
494 	for (i = 0; i < strlen(ends); i++) {
495 		if (toupper(buf[0]) == ends[i])
496 			break;
497 	}
498 	if (i == strlen(ends)) {
499 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
500 		    buf);
501 		usage(B_FALSE);
502 	}
503 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
504 		return (10*i);
505 	}
506 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
507 	usage(B_FALSE);
508 	/* NOTREACHED */
509 }
510 
511 static uint64_t
512 nicenumtoull(const char *buf)
513 {
514 	char *end;
515 	uint64_t val;
516 
517 	val = strtoull(buf, &end, 0);
518 	if (end == buf) {
519 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
520 		usage(B_FALSE);
521 	} else if (end[0] == '.') {
522 		double fval = strtod(buf, &end);
523 		fval *= pow(2, str2shift(end));
524 		if (fval > UINT64_MAX) {
525 			(void) fprintf(stderr, "ztest: value too large: %s\n",
526 			    buf);
527 			usage(B_FALSE);
528 		}
529 		val = (uint64_t)fval;
530 	} else {
531 		int shift = str2shift(end);
532 		if (shift >= 64 || (val << shift) >> shift != val) {
533 			(void) fprintf(stderr, "ztest: value too large: %s\n",
534 			    buf);
535 			usage(B_FALSE);
536 		}
537 		val <<= shift;
538 	}
539 	return (val);
540 }
541 
542 static void
543 usage(boolean_t requested)
544 {
545 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
546 
547 	char nice_vdev_size[10];
548 	char nice_gang_bang[10];
549 	FILE *fp = requested ? stdout : stderr;
550 
551 	nicenum(zo->zo_vdev_size, nice_vdev_size);
552 	nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
553 
554 	(void) fprintf(fp, "Usage: %s\n"
555 	    "\t[-v vdevs (default: %llu)]\n"
556 	    "\t[-s size_of_each_vdev (default: %s)]\n"
557 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
558 	    "\t[-m mirror_copies (default: %d)]\n"
559 	    "\t[-r raidz_disks (default: %d)]\n"
560 	    "\t[-R raidz_parity (default: %d)]\n"
561 	    "\t[-d datasets (default: %d)]\n"
562 	    "\t[-t threads (default: %d)]\n"
563 	    "\t[-g gang_block_threshold (default: %s)]\n"
564 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
565 	    "\t[-k kill_percentage (default: %llu%%)]\n"
566 	    "\t[-p pool_name (default: %s)]\n"
567 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
568 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
569 	    "\t[-E] use existing pool instead of creating new one\n"
570 	    "\t[-T time (default: %llu sec)] total run time\n"
571 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
572 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
573 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
574 	    "\t[-h] (print help)\n"
575 	    "",
576 	    zo->zo_pool,
577 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
578 	    nice_vdev_size,				/* -s */
579 	    zo->zo_ashift,				/* -a */
580 	    zo->zo_mirrors,				/* -m */
581 	    zo->zo_raidz,				/* -r */
582 	    zo->zo_raidz_parity,			/* -R */
583 	    zo->zo_datasets,				/* -d */
584 	    zo->zo_threads,				/* -t */
585 	    nice_gang_bang,				/* -g */
586 	    zo->zo_init,				/* -i */
587 	    (u_longlong_t)zo->zo_killrate,		/* -k */
588 	    zo->zo_pool,				/* -p */
589 	    zo->zo_dir,					/* -f */
590 	    (u_longlong_t)zo->zo_time,			/* -T */
591 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
592 	    (u_longlong_t)zo->zo_passtime);
593 	exit(requested ? 0 : 1);
594 }
595 
596 static void
597 process_options(int argc, char **argv)
598 {
599 	char *path;
600 	ztest_shared_opts_t *zo = &ztest_opts;
601 
602 	int opt;
603 	uint64_t value;
604 	char altdir[MAXNAMELEN] = { 0 };
605 
606 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
607 
608 	while ((opt = getopt(argc, argv,
609 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
610 		value = 0;
611 		switch (opt) {
612 		case 'v':
613 		case 's':
614 		case 'a':
615 		case 'm':
616 		case 'r':
617 		case 'R':
618 		case 'd':
619 		case 't':
620 		case 'g':
621 		case 'i':
622 		case 'k':
623 		case 'T':
624 		case 'P':
625 		case 'F':
626 			value = nicenumtoull(optarg);
627 		}
628 		switch (opt) {
629 		case 'v':
630 			zo->zo_vdevs = value;
631 			break;
632 		case 's':
633 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
634 			break;
635 		case 'a':
636 			zo->zo_ashift = value;
637 			break;
638 		case 'm':
639 			zo->zo_mirrors = value;
640 			break;
641 		case 'r':
642 			zo->zo_raidz = MAX(1, value);
643 			break;
644 		case 'R':
645 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
646 			break;
647 		case 'd':
648 			zo->zo_datasets = MAX(1, value);
649 			break;
650 		case 't':
651 			zo->zo_threads = MAX(1, value);
652 			break;
653 		case 'g':
654 			zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
655 			    value);
656 			break;
657 		case 'i':
658 			zo->zo_init = value;
659 			break;
660 		case 'k':
661 			zo->zo_killrate = value;
662 			break;
663 		case 'p':
664 			(void) strlcpy(zo->zo_pool, optarg,
665 			    sizeof (zo->zo_pool));
666 			break;
667 		case 'f':
668 			path = realpath(optarg, NULL);
669 			if (path == NULL) {
670 				(void) fprintf(stderr, "error: %s: %s\n",
671 				    optarg, strerror(errno));
672 				usage(B_FALSE);
673 			} else {
674 				(void) strlcpy(zo->zo_dir, path,
675 				    sizeof (zo->zo_dir));
676 			}
677 			break;
678 		case 'V':
679 			zo->zo_verbose++;
680 			break;
681 		case 'E':
682 			zo->zo_init = 0;
683 			break;
684 		case 'T':
685 			zo->zo_time = value;
686 			break;
687 		case 'P':
688 			zo->zo_passtime = MAX(1, value);
689 			break;
690 		case 'F':
691 			zo->zo_maxloops = MAX(1, value);
692 			break;
693 		case 'B':
694 			(void) strlcpy(altdir, optarg, sizeof (altdir));
695 			break;
696 		case 'h':
697 			usage(B_TRUE);
698 			break;
699 		case '?':
700 		default:
701 			usage(B_FALSE);
702 			break;
703 		}
704 	}
705 
706 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
707 
708 	zo->zo_vdevtime =
709 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
710 	    UINT64_MAX >> 2);
711 
712 	if (strlen(altdir) > 0) {
713 		char *cmd;
714 		char *realaltdir;
715 		char *bin;
716 		char *ztest;
717 		char *isa;
718 		int isalen;
719 
720 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
721 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
722 
723 		VERIFY(NULL != realpath(getexecname(), cmd));
724 		if (0 != access(altdir, F_OK)) {
725 			ztest_dump_core = B_FALSE;
726 			fatal(B_TRUE, "invalid alternate ztest path: %s",
727 			    altdir);
728 		}
729 		VERIFY(NULL != realpath(altdir, realaltdir));
730 
731 		/*
732 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
733 		 * We want to extract <isa> to determine if we should use
734 		 * 32 or 64 bit binaries.
735 		 */
736 		bin = strstr(cmd, "/usr/bin/");
737 		ztest = strstr(bin, "/ztest");
738 		isa = bin + 9;
739 		isalen = ztest - isa;
740 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
741 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
742 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
743 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
744 
745 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
746 			ztest_dump_core = B_FALSE;
747 			fatal(B_TRUE, "invalid alternate ztest: %s",
748 			    zo->zo_alt_ztest);
749 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
750 			ztest_dump_core = B_FALSE;
751 			fatal(B_TRUE, "invalid alternate lib directory %s",
752 			    zo->zo_alt_libpath);
753 		}
754 
755 		umem_free(cmd, MAXPATHLEN);
756 		umem_free(realaltdir, MAXPATHLEN);
757 	}
758 }
759 
760 static void
761 ztest_kill(ztest_shared_t *zs)
762 {
763 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
764 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
765 	(void) kill(getpid(), SIGKILL);
766 }
767 
768 static uint64_t
769 ztest_random(uint64_t range)
770 {
771 	uint64_t r;
772 
773 	ASSERT3S(ztest_fd_rand, >=, 0);
774 
775 	if (range == 0)
776 		return (0);
777 
778 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
779 		fatal(1, "short read from /dev/urandom");
780 
781 	return (r % range);
782 }
783 
784 /* ARGSUSED */
785 static void
786 ztest_record_enospc(const char *s)
787 {
788 	ztest_shared->zs_enospc_count++;
789 }
790 
791 static uint64_t
792 ztest_get_ashift(void)
793 {
794 	if (ztest_opts.zo_ashift == 0)
795 		return (SPA_MINBLOCKSHIFT + ztest_random(3));
796 	return (ztest_opts.zo_ashift);
797 }
798 
799 static nvlist_t *
800 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
801 {
802 	char pathbuf[MAXPATHLEN];
803 	uint64_t vdev;
804 	nvlist_t *file;
805 
806 	if (ashift == 0)
807 		ashift = ztest_get_ashift();
808 
809 	if (path == NULL) {
810 		path = pathbuf;
811 
812 		if (aux != NULL) {
813 			vdev = ztest_shared->zs_vdev_aux;
814 			(void) snprintf(path, sizeof (pathbuf),
815 			    ztest_aux_template, ztest_opts.zo_dir,
816 			    pool == NULL ? ztest_opts.zo_pool : pool,
817 			    aux, vdev);
818 		} else {
819 			vdev = ztest_shared->zs_vdev_next_leaf++;
820 			(void) snprintf(path, sizeof (pathbuf),
821 			    ztest_dev_template, ztest_opts.zo_dir,
822 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
823 		}
824 	}
825 
826 	if (size != 0) {
827 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
828 		if (fd == -1)
829 			fatal(1, "can't open %s", path);
830 		if (ftruncate(fd, size) != 0)
831 			fatal(1, "can't ftruncate %s", path);
832 		(void) close(fd);
833 	}
834 
835 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
836 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
837 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
838 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
839 
840 	return (file);
841 }
842 
843 static nvlist_t *
844 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
845     uint64_t ashift, int r)
846 {
847 	nvlist_t *raidz, **child;
848 	int c;
849 
850 	if (r < 2)
851 		return (make_vdev_file(path, aux, pool, size, ashift));
852 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
853 
854 	for (c = 0; c < r; c++)
855 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
856 
857 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
858 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
859 	    VDEV_TYPE_RAIDZ) == 0);
860 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
861 	    ztest_opts.zo_raidz_parity) == 0);
862 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
863 	    child, r) == 0);
864 
865 	for (c = 0; c < r; c++)
866 		nvlist_free(child[c]);
867 
868 	umem_free(child, r * sizeof (nvlist_t *));
869 
870 	return (raidz);
871 }
872 
873 static nvlist_t *
874 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
875     uint64_t ashift, int r, int m)
876 {
877 	nvlist_t *mirror, **child;
878 	int c;
879 
880 	if (m < 1)
881 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
882 
883 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
884 
885 	for (c = 0; c < m; c++)
886 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
887 
888 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
889 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
890 	    VDEV_TYPE_MIRROR) == 0);
891 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
892 	    child, m) == 0);
893 
894 	for (c = 0; c < m; c++)
895 		nvlist_free(child[c]);
896 
897 	umem_free(child, m * sizeof (nvlist_t *));
898 
899 	return (mirror);
900 }
901 
902 static nvlist_t *
903 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
904     int log, int r, int m, int t)
905 {
906 	nvlist_t *root, **child;
907 	int c;
908 
909 	ASSERT(t > 0);
910 
911 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
912 
913 	for (c = 0; c < t; c++) {
914 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
915 		    r, m);
916 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
917 		    log) == 0);
918 	}
919 
920 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
921 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
922 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
923 	    child, t) == 0);
924 
925 	for (c = 0; c < t; c++)
926 		nvlist_free(child[c]);
927 
928 	umem_free(child, t * sizeof (nvlist_t *));
929 
930 	return (root);
931 }
932 
933 /*
934  * Find a random spa version. Returns back a random spa version in the
935  * range [initial_version, SPA_VERSION_FEATURES].
936  */
937 static uint64_t
938 ztest_random_spa_version(uint64_t initial_version)
939 {
940 	uint64_t version = initial_version;
941 
942 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
943 		version = version +
944 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
945 	}
946 
947 	if (version > SPA_VERSION_BEFORE_FEATURES)
948 		version = SPA_VERSION_FEATURES;
949 
950 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
951 	return (version);
952 }
953 
954 static int
955 ztest_random_blocksize(void)
956 {
957 	return (1 << (SPA_MINBLOCKSHIFT +
958 	    ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
959 }
960 
961 static int
962 ztest_random_ibshift(void)
963 {
964 	return (DN_MIN_INDBLKSHIFT +
965 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
966 }
967 
968 static uint64_t
969 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
970 {
971 	uint64_t top;
972 	vdev_t *rvd = spa->spa_root_vdev;
973 	vdev_t *tvd;
974 
975 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
976 
977 	do {
978 		top = ztest_random(rvd->vdev_children);
979 		tvd = rvd->vdev_child[top];
980 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
981 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
982 
983 	return (top);
984 }
985 
986 static uint64_t
987 ztest_random_dsl_prop(zfs_prop_t prop)
988 {
989 	uint64_t value;
990 
991 	do {
992 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
993 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
994 
995 	return (value);
996 }
997 
998 static int
999 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1000     boolean_t inherit)
1001 {
1002 	const char *propname = zfs_prop_to_name(prop);
1003 	const char *valname;
1004 	char setpoint[MAXPATHLEN];
1005 	uint64_t curval;
1006 	int error;
1007 
1008 	error = dsl_prop_set(osname, propname,
1009 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL),
1010 	    sizeof (value), 1, &value);
1011 
1012 	if (error == ENOSPC) {
1013 		ztest_record_enospc(FTAG);
1014 		return (error);
1015 	}
1016 	ASSERT0(error);
1017 
1018 	VERIFY3U(dsl_prop_get(osname, propname, sizeof (curval),
1019 	    1, &curval, setpoint), ==, 0);
1020 
1021 	if (ztest_opts.zo_verbose >= 6) {
1022 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1023 		(void) printf("%s %s = %s at '%s'\n",
1024 		    osname, propname, valname, setpoint);
1025 	}
1026 
1027 	return (error);
1028 }
1029 
1030 static int
1031 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1032 {
1033 	spa_t *spa = ztest_spa;
1034 	nvlist_t *props = NULL;
1035 	int error;
1036 
1037 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1038 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1039 
1040 	error = spa_prop_set(spa, props);
1041 
1042 	nvlist_free(props);
1043 
1044 	if (error == ENOSPC) {
1045 		ztest_record_enospc(FTAG);
1046 		return (error);
1047 	}
1048 	ASSERT0(error);
1049 
1050 	return (error);
1051 }
1052 
1053 static void
1054 ztest_rll_init(rll_t *rll)
1055 {
1056 	rll->rll_writer = NULL;
1057 	rll->rll_readers = 0;
1058 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1059 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1060 }
1061 
1062 static void
1063 ztest_rll_destroy(rll_t *rll)
1064 {
1065 	ASSERT(rll->rll_writer == NULL);
1066 	ASSERT(rll->rll_readers == 0);
1067 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1068 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
1069 }
1070 
1071 static void
1072 ztest_rll_lock(rll_t *rll, rl_type_t type)
1073 {
1074 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1075 
1076 	if (type == RL_READER) {
1077 		while (rll->rll_writer != NULL)
1078 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1079 		rll->rll_readers++;
1080 	} else {
1081 		while (rll->rll_writer != NULL || rll->rll_readers)
1082 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1083 		rll->rll_writer = curthread;
1084 	}
1085 
1086 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1087 }
1088 
1089 static void
1090 ztest_rll_unlock(rll_t *rll)
1091 {
1092 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1093 
1094 	if (rll->rll_writer) {
1095 		ASSERT(rll->rll_readers == 0);
1096 		rll->rll_writer = NULL;
1097 	} else {
1098 		ASSERT(rll->rll_readers != 0);
1099 		ASSERT(rll->rll_writer == NULL);
1100 		rll->rll_readers--;
1101 	}
1102 
1103 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1104 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1105 
1106 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1107 }
1108 
1109 static void
1110 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1111 {
1112 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1113 
1114 	ztest_rll_lock(rll, type);
1115 }
1116 
1117 static void
1118 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1119 {
1120 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1121 
1122 	ztest_rll_unlock(rll);
1123 }
1124 
1125 static rl_t *
1126 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1127     uint64_t size, rl_type_t type)
1128 {
1129 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1130 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1131 	rl_t *rl;
1132 
1133 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1134 	rl->rl_object = object;
1135 	rl->rl_offset = offset;
1136 	rl->rl_size = size;
1137 	rl->rl_lock = rll;
1138 
1139 	ztest_rll_lock(rll, type);
1140 
1141 	return (rl);
1142 }
1143 
1144 static void
1145 ztest_range_unlock(rl_t *rl)
1146 {
1147 	rll_t *rll = rl->rl_lock;
1148 
1149 	ztest_rll_unlock(rll);
1150 
1151 	umem_free(rl, sizeof (*rl));
1152 }
1153 
1154 static void
1155 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1156 {
1157 	zd->zd_os = os;
1158 	zd->zd_zilog = dmu_objset_zil(os);
1159 	zd->zd_shared = szd;
1160 	dmu_objset_name(os, zd->zd_name);
1161 
1162 	if (zd->zd_shared != NULL)
1163 		zd->zd_shared->zd_seq = 0;
1164 
1165 	VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1166 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1167 
1168 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1169 		ztest_rll_init(&zd->zd_object_lock[l]);
1170 
1171 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1172 		ztest_rll_init(&zd->zd_range_lock[l]);
1173 }
1174 
1175 static void
1176 ztest_zd_fini(ztest_ds_t *zd)
1177 {
1178 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1179 
1180 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1181 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1182 
1183 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1184 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1185 }
1186 
1187 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1188 
1189 static uint64_t
1190 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1191 {
1192 	uint64_t txg;
1193 	int error;
1194 
1195 	/*
1196 	 * Attempt to assign tx to some transaction group.
1197 	 */
1198 	error = dmu_tx_assign(tx, txg_how);
1199 	if (error) {
1200 		if (error == ERESTART) {
1201 			ASSERT(txg_how == TXG_NOWAIT);
1202 			dmu_tx_wait(tx);
1203 		} else {
1204 			ASSERT3U(error, ==, ENOSPC);
1205 			ztest_record_enospc(tag);
1206 		}
1207 		dmu_tx_abort(tx);
1208 		return (0);
1209 	}
1210 	txg = dmu_tx_get_txg(tx);
1211 	ASSERT(txg != 0);
1212 	return (txg);
1213 }
1214 
1215 static void
1216 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1217 {
1218 	uint64_t *ip = buf;
1219 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1220 
1221 	while (ip < ip_end)
1222 		*ip++ = value;
1223 }
1224 
1225 static boolean_t
1226 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1227 {
1228 	uint64_t *ip = buf;
1229 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1230 	uint64_t diff = 0;
1231 
1232 	while (ip < ip_end)
1233 		diff |= (value - *ip++);
1234 
1235 	return (diff == 0);
1236 }
1237 
1238 static void
1239 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1240     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1241 {
1242 	bt->bt_magic = BT_MAGIC;
1243 	bt->bt_objset = dmu_objset_id(os);
1244 	bt->bt_object = object;
1245 	bt->bt_offset = offset;
1246 	bt->bt_gen = gen;
1247 	bt->bt_txg = txg;
1248 	bt->bt_crtxg = crtxg;
1249 }
1250 
1251 static void
1252 ztest_bt_verify(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 	ASSERT(bt->bt_magic == BT_MAGIC);
1256 	ASSERT(bt->bt_objset == dmu_objset_id(os));
1257 	ASSERT(bt->bt_object == object);
1258 	ASSERT(bt->bt_offset == offset);
1259 	ASSERT(bt->bt_gen <= gen);
1260 	ASSERT(bt->bt_txg <= txg);
1261 	ASSERT(bt->bt_crtxg == crtxg);
1262 }
1263 
1264 static ztest_block_tag_t *
1265 ztest_bt_bonus(dmu_buf_t *db)
1266 {
1267 	dmu_object_info_t doi;
1268 	ztest_block_tag_t *bt;
1269 
1270 	dmu_object_info_from_db(db, &doi);
1271 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1272 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1273 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1274 
1275 	return (bt);
1276 }
1277 
1278 /*
1279  * ZIL logging ops
1280  */
1281 
1282 #define	lrz_type	lr_mode
1283 #define	lrz_blocksize	lr_uid
1284 #define	lrz_ibshift	lr_gid
1285 #define	lrz_bonustype	lr_rdev
1286 #define	lrz_bonuslen	lr_crtime[1]
1287 
1288 static void
1289 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1290 {
1291 	char *name = (void *)(lr + 1);		/* name follows lr */
1292 	size_t namesize = strlen(name) + 1;
1293 	itx_t *itx;
1294 
1295 	if (zil_replaying(zd->zd_zilog, tx))
1296 		return;
1297 
1298 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1299 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1300 	    sizeof (*lr) + namesize - sizeof (lr_t));
1301 
1302 	zil_itx_assign(zd->zd_zilog, itx, tx);
1303 }
1304 
1305 static void
1306 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1307 {
1308 	char *name = (void *)(lr + 1);		/* name follows lr */
1309 	size_t namesize = strlen(name) + 1;
1310 	itx_t *itx;
1311 
1312 	if (zil_replaying(zd->zd_zilog, tx))
1313 		return;
1314 
1315 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1316 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1317 	    sizeof (*lr) + namesize - sizeof (lr_t));
1318 
1319 	itx->itx_oid = object;
1320 	zil_itx_assign(zd->zd_zilog, itx, tx);
1321 }
1322 
1323 static void
1324 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1325 {
1326 	itx_t *itx;
1327 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1328 
1329 	if (zil_replaying(zd->zd_zilog, tx))
1330 		return;
1331 
1332 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1333 		write_state = WR_INDIRECT;
1334 
1335 	itx = zil_itx_create(TX_WRITE,
1336 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1337 
1338 	if (write_state == WR_COPIED &&
1339 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1340 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1341 		zil_itx_destroy(itx);
1342 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1343 		write_state = WR_NEED_COPY;
1344 	}
1345 	itx->itx_private = zd;
1346 	itx->itx_wr_state = write_state;
1347 	itx->itx_sync = (ztest_random(8) == 0);
1348 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1349 
1350 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1351 	    sizeof (*lr) - sizeof (lr_t));
1352 
1353 	zil_itx_assign(zd->zd_zilog, itx, tx);
1354 }
1355 
1356 static void
1357 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1358 {
1359 	itx_t *itx;
1360 
1361 	if (zil_replaying(zd->zd_zilog, tx))
1362 		return;
1363 
1364 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1365 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1366 	    sizeof (*lr) - sizeof (lr_t));
1367 
1368 	itx->itx_sync = B_FALSE;
1369 	zil_itx_assign(zd->zd_zilog, itx, tx);
1370 }
1371 
1372 static void
1373 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1374 {
1375 	itx_t *itx;
1376 
1377 	if (zil_replaying(zd->zd_zilog, tx))
1378 		return;
1379 
1380 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1381 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1382 	    sizeof (*lr) - sizeof (lr_t));
1383 
1384 	itx->itx_sync = B_FALSE;
1385 	zil_itx_assign(zd->zd_zilog, itx, tx);
1386 }
1387 
1388 /*
1389  * ZIL replay ops
1390  */
1391 static int
1392 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1393 {
1394 	char *name = (void *)(lr + 1);		/* name follows lr */
1395 	objset_t *os = zd->zd_os;
1396 	ztest_block_tag_t *bbt;
1397 	dmu_buf_t *db;
1398 	dmu_tx_t *tx;
1399 	uint64_t txg;
1400 	int error = 0;
1401 
1402 	if (byteswap)
1403 		byteswap_uint64_array(lr, sizeof (*lr));
1404 
1405 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1406 	ASSERT(name[0] != '\0');
1407 
1408 	tx = dmu_tx_create(os);
1409 
1410 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1411 
1412 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1413 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1414 	} else {
1415 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1416 	}
1417 
1418 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1419 	if (txg == 0)
1420 		return (ENOSPC);
1421 
1422 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1423 
1424 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1425 		if (lr->lr_foid == 0) {
1426 			lr->lr_foid = zap_create(os,
1427 			    lr->lrz_type, lr->lrz_bonustype,
1428 			    lr->lrz_bonuslen, tx);
1429 		} else {
1430 			error = zap_create_claim(os, lr->lr_foid,
1431 			    lr->lrz_type, lr->lrz_bonustype,
1432 			    lr->lrz_bonuslen, tx);
1433 		}
1434 	} else {
1435 		if (lr->lr_foid == 0) {
1436 			lr->lr_foid = dmu_object_alloc(os,
1437 			    lr->lrz_type, 0, lr->lrz_bonustype,
1438 			    lr->lrz_bonuslen, tx);
1439 		} else {
1440 			error = dmu_object_claim(os, lr->lr_foid,
1441 			    lr->lrz_type, 0, lr->lrz_bonustype,
1442 			    lr->lrz_bonuslen, tx);
1443 		}
1444 	}
1445 
1446 	if (error) {
1447 		ASSERT3U(error, ==, EEXIST);
1448 		ASSERT(zd->zd_zilog->zl_replay);
1449 		dmu_tx_commit(tx);
1450 		return (error);
1451 	}
1452 
1453 	ASSERT(lr->lr_foid != 0);
1454 
1455 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1456 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1457 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1458 
1459 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1460 	bbt = ztest_bt_bonus(db);
1461 	dmu_buf_will_dirty(db, tx);
1462 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1463 	dmu_buf_rele(db, FTAG);
1464 
1465 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1466 	    &lr->lr_foid, tx));
1467 
1468 	(void) ztest_log_create(zd, tx, lr);
1469 
1470 	dmu_tx_commit(tx);
1471 
1472 	return (0);
1473 }
1474 
1475 static int
1476 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1477 {
1478 	char *name = (void *)(lr + 1);		/* name follows lr */
1479 	objset_t *os = zd->zd_os;
1480 	dmu_object_info_t doi;
1481 	dmu_tx_t *tx;
1482 	uint64_t object, txg;
1483 
1484 	if (byteswap)
1485 		byteswap_uint64_array(lr, sizeof (*lr));
1486 
1487 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1488 	ASSERT(name[0] != '\0');
1489 
1490 	VERIFY3U(0, ==,
1491 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1492 	ASSERT(object != 0);
1493 
1494 	ztest_object_lock(zd, object, RL_WRITER);
1495 
1496 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1497 
1498 	tx = dmu_tx_create(os);
1499 
1500 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1501 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1502 
1503 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1504 	if (txg == 0) {
1505 		ztest_object_unlock(zd, object);
1506 		return (ENOSPC);
1507 	}
1508 
1509 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1510 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1511 	} else {
1512 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1513 	}
1514 
1515 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1516 
1517 	(void) ztest_log_remove(zd, tx, lr, object);
1518 
1519 	dmu_tx_commit(tx);
1520 
1521 	ztest_object_unlock(zd, object);
1522 
1523 	return (0);
1524 }
1525 
1526 static int
1527 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1528 {
1529 	objset_t *os = zd->zd_os;
1530 	void *data = lr + 1;			/* data follows lr */
1531 	uint64_t offset, length;
1532 	ztest_block_tag_t *bt = data;
1533 	ztest_block_tag_t *bbt;
1534 	uint64_t gen, txg, lrtxg, crtxg;
1535 	dmu_object_info_t doi;
1536 	dmu_tx_t *tx;
1537 	dmu_buf_t *db;
1538 	arc_buf_t *abuf = NULL;
1539 	rl_t *rl;
1540 
1541 	if (byteswap)
1542 		byteswap_uint64_array(lr, sizeof (*lr));
1543 
1544 	offset = lr->lr_offset;
1545 	length = lr->lr_length;
1546 
1547 	/* If it's a dmu_sync() block, write the whole block */
1548 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1549 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1550 		if (length < blocksize) {
1551 			offset -= offset % blocksize;
1552 			length = blocksize;
1553 		}
1554 	}
1555 
1556 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1557 		byteswap_uint64_array(bt, sizeof (*bt));
1558 
1559 	if (bt->bt_magic != BT_MAGIC)
1560 		bt = NULL;
1561 
1562 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1563 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1564 
1565 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1566 
1567 	dmu_object_info_from_db(db, &doi);
1568 
1569 	bbt = ztest_bt_bonus(db);
1570 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1571 	gen = bbt->bt_gen;
1572 	crtxg = bbt->bt_crtxg;
1573 	lrtxg = lr->lr_common.lrc_txg;
1574 
1575 	tx = dmu_tx_create(os);
1576 
1577 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1578 
1579 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1580 	    P2PHASE(offset, length) == 0)
1581 		abuf = dmu_request_arcbuf(db, length);
1582 
1583 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1584 	if (txg == 0) {
1585 		if (abuf != NULL)
1586 			dmu_return_arcbuf(abuf);
1587 		dmu_buf_rele(db, FTAG);
1588 		ztest_range_unlock(rl);
1589 		ztest_object_unlock(zd, lr->lr_foid);
1590 		return (ENOSPC);
1591 	}
1592 
1593 	if (bt != NULL) {
1594 		/*
1595 		 * Usually, verify the old data before writing new data --
1596 		 * but not always, because we also want to verify correct
1597 		 * behavior when the data was not recently read into cache.
1598 		 */
1599 		ASSERT(offset % doi.doi_data_block_size == 0);
1600 		if (ztest_random(4) != 0) {
1601 			int prefetch = ztest_random(2) ?
1602 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1603 			ztest_block_tag_t rbt;
1604 
1605 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1606 			    sizeof (rbt), &rbt, prefetch) == 0);
1607 			if (rbt.bt_magic == BT_MAGIC) {
1608 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1609 				    offset, gen, txg, crtxg);
1610 			}
1611 		}
1612 
1613 		/*
1614 		 * Writes can appear to be newer than the bonus buffer because
1615 		 * the ztest_get_data() callback does a dmu_read() of the
1616 		 * open-context data, which may be different than the data
1617 		 * as it was when the write was generated.
1618 		 */
1619 		if (zd->zd_zilog->zl_replay) {
1620 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1621 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1622 			    bt->bt_crtxg);
1623 		}
1624 
1625 		/*
1626 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1627 		 * so that all of the usual ASSERTs will work.
1628 		 */
1629 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1630 	}
1631 
1632 	if (abuf == NULL) {
1633 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1634 	} else {
1635 		bcopy(data, abuf->b_data, length);
1636 		dmu_assign_arcbuf(db, offset, abuf, tx);
1637 	}
1638 
1639 	(void) ztest_log_write(zd, tx, lr);
1640 
1641 	dmu_buf_rele(db, FTAG);
1642 
1643 	dmu_tx_commit(tx);
1644 
1645 	ztest_range_unlock(rl);
1646 	ztest_object_unlock(zd, lr->lr_foid);
1647 
1648 	return (0);
1649 }
1650 
1651 static int
1652 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1653 {
1654 	objset_t *os = zd->zd_os;
1655 	dmu_tx_t *tx;
1656 	uint64_t txg;
1657 	rl_t *rl;
1658 
1659 	if (byteswap)
1660 		byteswap_uint64_array(lr, sizeof (*lr));
1661 
1662 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1663 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1664 	    RL_WRITER);
1665 
1666 	tx = dmu_tx_create(os);
1667 
1668 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1669 
1670 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1671 	if (txg == 0) {
1672 		ztest_range_unlock(rl);
1673 		ztest_object_unlock(zd, lr->lr_foid);
1674 		return (ENOSPC);
1675 	}
1676 
1677 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1678 	    lr->lr_length, tx) == 0);
1679 
1680 	(void) ztest_log_truncate(zd, tx, lr);
1681 
1682 	dmu_tx_commit(tx);
1683 
1684 	ztest_range_unlock(rl);
1685 	ztest_object_unlock(zd, lr->lr_foid);
1686 
1687 	return (0);
1688 }
1689 
1690 static int
1691 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1692 {
1693 	objset_t *os = zd->zd_os;
1694 	dmu_tx_t *tx;
1695 	dmu_buf_t *db;
1696 	ztest_block_tag_t *bbt;
1697 	uint64_t txg, lrtxg, crtxg;
1698 
1699 	if (byteswap)
1700 		byteswap_uint64_array(lr, sizeof (*lr));
1701 
1702 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1703 
1704 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1705 
1706 	tx = dmu_tx_create(os);
1707 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1708 
1709 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1710 	if (txg == 0) {
1711 		dmu_buf_rele(db, FTAG);
1712 		ztest_object_unlock(zd, lr->lr_foid);
1713 		return (ENOSPC);
1714 	}
1715 
1716 	bbt = ztest_bt_bonus(db);
1717 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1718 	crtxg = bbt->bt_crtxg;
1719 	lrtxg = lr->lr_common.lrc_txg;
1720 
1721 	if (zd->zd_zilog->zl_replay) {
1722 		ASSERT(lr->lr_size != 0);
1723 		ASSERT(lr->lr_mode != 0);
1724 		ASSERT(lrtxg != 0);
1725 	} else {
1726 		/*
1727 		 * Randomly change the size and increment the generation.
1728 		 */
1729 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1730 		    sizeof (*bbt);
1731 		lr->lr_mode = bbt->bt_gen + 1;
1732 		ASSERT(lrtxg == 0);
1733 	}
1734 
1735 	/*
1736 	 * Verify that the current bonus buffer is not newer than our txg.
1737 	 */
1738 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1739 	    MAX(txg, lrtxg), crtxg);
1740 
1741 	dmu_buf_will_dirty(db, tx);
1742 
1743 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1744 	ASSERT3U(lr->lr_size, <=, db->db_size);
1745 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1746 	bbt = ztest_bt_bonus(db);
1747 
1748 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1749 
1750 	dmu_buf_rele(db, FTAG);
1751 
1752 	(void) ztest_log_setattr(zd, tx, lr);
1753 
1754 	dmu_tx_commit(tx);
1755 
1756 	ztest_object_unlock(zd, lr->lr_foid);
1757 
1758 	return (0);
1759 }
1760 
1761 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1762 	NULL,			/* 0 no such transaction type */
1763 	ztest_replay_create,	/* TX_CREATE */
1764 	NULL,			/* TX_MKDIR */
1765 	NULL,			/* TX_MKXATTR */
1766 	NULL,			/* TX_SYMLINK */
1767 	ztest_replay_remove,	/* TX_REMOVE */
1768 	NULL,			/* TX_RMDIR */
1769 	NULL,			/* TX_LINK */
1770 	NULL,			/* TX_RENAME */
1771 	ztest_replay_write,	/* TX_WRITE */
1772 	ztest_replay_truncate,	/* TX_TRUNCATE */
1773 	ztest_replay_setattr,	/* TX_SETATTR */
1774 	NULL,			/* TX_ACL */
1775 	NULL,			/* TX_CREATE_ACL */
1776 	NULL,			/* TX_CREATE_ATTR */
1777 	NULL,			/* TX_CREATE_ACL_ATTR */
1778 	NULL,			/* TX_MKDIR_ACL */
1779 	NULL,			/* TX_MKDIR_ATTR */
1780 	NULL,			/* TX_MKDIR_ACL_ATTR */
1781 	NULL,			/* TX_WRITE2 */
1782 };
1783 
1784 /*
1785  * ZIL get_data callbacks
1786  */
1787 
1788 static void
1789 ztest_get_done(zgd_t *zgd, int error)
1790 {
1791 	ztest_ds_t *zd = zgd->zgd_private;
1792 	uint64_t object = zgd->zgd_rl->rl_object;
1793 
1794 	if (zgd->zgd_db)
1795 		dmu_buf_rele(zgd->zgd_db, zgd);
1796 
1797 	ztest_range_unlock(zgd->zgd_rl);
1798 	ztest_object_unlock(zd, object);
1799 
1800 	if (error == 0 && zgd->zgd_bp)
1801 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1802 
1803 	umem_free(zgd, sizeof (*zgd));
1804 }
1805 
1806 static int
1807 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1808 {
1809 	ztest_ds_t *zd = arg;
1810 	objset_t *os = zd->zd_os;
1811 	uint64_t object = lr->lr_foid;
1812 	uint64_t offset = lr->lr_offset;
1813 	uint64_t size = lr->lr_length;
1814 	blkptr_t *bp = &lr->lr_blkptr;
1815 	uint64_t txg = lr->lr_common.lrc_txg;
1816 	uint64_t crtxg;
1817 	dmu_object_info_t doi;
1818 	dmu_buf_t *db;
1819 	zgd_t *zgd;
1820 	int error;
1821 
1822 	ztest_object_lock(zd, object, RL_READER);
1823 	error = dmu_bonus_hold(os, object, FTAG, &db);
1824 	if (error) {
1825 		ztest_object_unlock(zd, object);
1826 		return (error);
1827 	}
1828 
1829 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1830 
1831 	if (crtxg == 0 || crtxg > txg) {
1832 		dmu_buf_rele(db, FTAG);
1833 		ztest_object_unlock(zd, object);
1834 		return (ENOENT);
1835 	}
1836 
1837 	dmu_object_info_from_db(db, &doi);
1838 	dmu_buf_rele(db, FTAG);
1839 	db = NULL;
1840 
1841 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1842 	zgd->zgd_zilog = zd->zd_zilog;
1843 	zgd->zgd_private = zd;
1844 
1845 	if (buf != NULL) {	/* immediate write */
1846 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1847 		    RL_READER);
1848 
1849 		error = dmu_read(os, object, offset, size, buf,
1850 		    DMU_READ_NO_PREFETCH);
1851 		ASSERT(error == 0);
1852 	} else {
1853 		size = doi.doi_data_block_size;
1854 		if (ISP2(size)) {
1855 			offset = P2ALIGN(offset, size);
1856 		} else {
1857 			ASSERT(offset < size);
1858 			offset = 0;
1859 		}
1860 
1861 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1862 		    RL_READER);
1863 
1864 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1865 		    DMU_READ_NO_PREFETCH);
1866 
1867 		if (error == 0) {
1868 			zgd->zgd_db = db;
1869 			zgd->zgd_bp = bp;
1870 
1871 			ASSERT(db->db_offset == offset);
1872 			ASSERT(db->db_size == size);
1873 
1874 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1875 			    ztest_get_done, zgd);
1876 
1877 			if (error == 0)
1878 				return (0);
1879 		}
1880 	}
1881 
1882 	ztest_get_done(zgd, error);
1883 
1884 	return (error);
1885 }
1886 
1887 static void *
1888 ztest_lr_alloc(size_t lrsize, char *name)
1889 {
1890 	char *lr;
1891 	size_t namesize = name ? strlen(name) + 1 : 0;
1892 
1893 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1894 
1895 	if (name)
1896 		bcopy(name, lr + lrsize, namesize);
1897 
1898 	return (lr);
1899 }
1900 
1901 void
1902 ztest_lr_free(void *lr, size_t lrsize, char *name)
1903 {
1904 	size_t namesize = name ? strlen(name) + 1 : 0;
1905 
1906 	umem_free(lr, lrsize + namesize);
1907 }
1908 
1909 /*
1910  * Lookup a bunch of objects.  Returns the number of objects not found.
1911  */
1912 static int
1913 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1914 {
1915 	int missing = 0;
1916 	int error;
1917 
1918 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1919 
1920 	for (int i = 0; i < count; i++, od++) {
1921 		od->od_object = 0;
1922 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1923 		    sizeof (uint64_t), 1, &od->od_object);
1924 		if (error) {
1925 			ASSERT(error == ENOENT);
1926 			ASSERT(od->od_object == 0);
1927 			missing++;
1928 		} else {
1929 			dmu_buf_t *db;
1930 			ztest_block_tag_t *bbt;
1931 			dmu_object_info_t doi;
1932 
1933 			ASSERT(od->od_object != 0);
1934 			ASSERT(missing == 0);	/* there should be no gaps */
1935 
1936 			ztest_object_lock(zd, od->od_object, RL_READER);
1937 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1938 			    od->od_object, FTAG, &db));
1939 			dmu_object_info_from_db(db, &doi);
1940 			bbt = ztest_bt_bonus(db);
1941 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1942 			od->od_type = doi.doi_type;
1943 			od->od_blocksize = doi.doi_data_block_size;
1944 			od->od_gen = bbt->bt_gen;
1945 			dmu_buf_rele(db, FTAG);
1946 			ztest_object_unlock(zd, od->od_object);
1947 		}
1948 	}
1949 
1950 	return (missing);
1951 }
1952 
1953 static int
1954 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1955 {
1956 	int missing = 0;
1957 
1958 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1959 
1960 	for (int i = 0; i < count; i++, od++) {
1961 		if (missing) {
1962 			od->od_object = 0;
1963 			missing++;
1964 			continue;
1965 		}
1966 
1967 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1968 
1969 		lr->lr_doid = od->od_dir;
1970 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
1971 		lr->lrz_type = od->od_crtype;
1972 		lr->lrz_blocksize = od->od_crblocksize;
1973 		lr->lrz_ibshift = ztest_random_ibshift();
1974 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1975 		lr->lrz_bonuslen = dmu_bonus_max();
1976 		lr->lr_gen = od->od_crgen;
1977 		lr->lr_crtime[0] = time(NULL);
1978 
1979 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1980 			ASSERT(missing == 0);
1981 			od->od_object = 0;
1982 			missing++;
1983 		} else {
1984 			od->od_object = lr->lr_foid;
1985 			od->od_type = od->od_crtype;
1986 			od->od_blocksize = od->od_crblocksize;
1987 			od->od_gen = od->od_crgen;
1988 			ASSERT(od->od_object != 0);
1989 		}
1990 
1991 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
1992 	}
1993 
1994 	return (missing);
1995 }
1996 
1997 static int
1998 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
1999 {
2000 	int missing = 0;
2001 	int error;
2002 
2003 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2004 
2005 	od += count - 1;
2006 
2007 	for (int i = count - 1; i >= 0; i--, od--) {
2008 		if (missing) {
2009 			missing++;
2010 			continue;
2011 		}
2012 
2013 		if (od->od_object == 0)
2014 			continue;
2015 
2016 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2017 
2018 		lr->lr_doid = od->od_dir;
2019 
2020 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2021 			ASSERT3U(error, ==, ENOSPC);
2022 			missing++;
2023 		} else {
2024 			od->od_object = 0;
2025 		}
2026 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2027 	}
2028 
2029 	return (missing);
2030 }
2031 
2032 static int
2033 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2034     void *data)
2035 {
2036 	lr_write_t *lr;
2037 	int error;
2038 
2039 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2040 
2041 	lr->lr_foid = object;
2042 	lr->lr_offset = offset;
2043 	lr->lr_length = size;
2044 	lr->lr_blkoff = 0;
2045 	BP_ZERO(&lr->lr_blkptr);
2046 
2047 	bcopy(data, lr + 1, size);
2048 
2049 	error = ztest_replay_write(zd, lr, B_FALSE);
2050 
2051 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2052 
2053 	return (error);
2054 }
2055 
2056 static int
2057 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2058 {
2059 	lr_truncate_t *lr;
2060 	int error;
2061 
2062 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2063 
2064 	lr->lr_foid = object;
2065 	lr->lr_offset = offset;
2066 	lr->lr_length = size;
2067 
2068 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2069 
2070 	ztest_lr_free(lr, sizeof (*lr), NULL);
2071 
2072 	return (error);
2073 }
2074 
2075 static int
2076 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2077 {
2078 	lr_setattr_t *lr;
2079 	int error;
2080 
2081 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2082 
2083 	lr->lr_foid = object;
2084 	lr->lr_size = 0;
2085 	lr->lr_mode = 0;
2086 
2087 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2088 
2089 	ztest_lr_free(lr, sizeof (*lr), NULL);
2090 
2091 	return (error);
2092 }
2093 
2094 static void
2095 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2096 {
2097 	objset_t *os = zd->zd_os;
2098 	dmu_tx_t *tx;
2099 	uint64_t txg;
2100 	rl_t *rl;
2101 
2102 	txg_wait_synced(dmu_objset_pool(os), 0);
2103 
2104 	ztest_object_lock(zd, object, RL_READER);
2105 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2106 
2107 	tx = dmu_tx_create(os);
2108 
2109 	dmu_tx_hold_write(tx, object, offset, size);
2110 
2111 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2112 
2113 	if (txg != 0) {
2114 		dmu_prealloc(os, object, offset, size, tx);
2115 		dmu_tx_commit(tx);
2116 		txg_wait_synced(dmu_objset_pool(os), txg);
2117 	} else {
2118 		(void) dmu_free_long_range(os, object, offset, size);
2119 	}
2120 
2121 	ztest_range_unlock(rl);
2122 	ztest_object_unlock(zd, object);
2123 }
2124 
2125 static void
2126 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2127 {
2128 	ztest_block_tag_t wbt;
2129 	dmu_object_info_t doi;
2130 	enum ztest_io_type io_type;
2131 	uint64_t blocksize;
2132 	void *data;
2133 
2134 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2135 	blocksize = doi.doi_data_block_size;
2136 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2137 
2138 	/*
2139 	 * Pick an i/o type at random, biased toward writing block tags.
2140 	 */
2141 	io_type = ztest_random(ZTEST_IO_TYPES);
2142 	if (ztest_random(2) == 0)
2143 		io_type = ZTEST_IO_WRITE_TAG;
2144 
2145 	(void) rw_rdlock(&zd->zd_zilog_lock);
2146 
2147 	switch (io_type) {
2148 
2149 	case ZTEST_IO_WRITE_TAG:
2150 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2151 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2152 		break;
2153 
2154 	case ZTEST_IO_WRITE_PATTERN:
2155 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2156 		if (ztest_random(2) == 0) {
2157 			/*
2158 			 * Induce fletcher2 collisions to ensure that
2159 			 * zio_ddt_collision() detects and resolves them
2160 			 * when using fletcher2-verify for deduplication.
2161 			 */
2162 			((uint64_t *)data)[0] ^= 1ULL << 63;
2163 			((uint64_t *)data)[4] ^= 1ULL << 63;
2164 		}
2165 		(void) ztest_write(zd, object, offset, blocksize, data);
2166 		break;
2167 
2168 	case ZTEST_IO_WRITE_ZEROES:
2169 		bzero(data, blocksize);
2170 		(void) ztest_write(zd, object, offset, blocksize, data);
2171 		break;
2172 
2173 	case ZTEST_IO_TRUNCATE:
2174 		(void) ztest_truncate(zd, object, offset, blocksize);
2175 		break;
2176 
2177 	case ZTEST_IO_SETATTR:
2178 		(void) ztest_setattr(zd, object);
2179 		break;
2180 	}
2181 
2182 	(void) rw_unlock(&zd->zd_zilog_lock);
2183 
2184 	umem_free(data, blocksize);
2185 }
2186 
2187 /*
2188  * Initialize an object description template.
2189  */
2190 static void
2191 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2192     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2193 {
2194 	od->od_dir = ZTEST_DIROBJ;
2195 	od->od_object = 0;
2196 
2197 	od->od_crtype = type;
2198 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2199 	od->od_crgen = gen;
2200 
2201 	od->od_type = DMU_OT_NONE;
2202 	od->od_blocksize = 0;
2203 	od->od_gen = 0;
2204 
2205 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2206 	    tag, (int64_t)id, index);
2207 }
2208 
2209 /*
2210  * Lookup or create the objects for a test using the od template.
2211  * If the objects do not all exist, or if 'remove' is specified,
2212  * remove any existing objects and create new ones.  Otherwise,
2213  * use the existing objects.
2214  */
2215 static int
2216 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2217 {
2218 	int count = size / sizeof (*od);
2219 	int rv = 0;
2220 
2221 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2222 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2223 	    (ztest_remove(zd, od, count) != 0 ||
2224 	    ztest_create(zd, od, count) != 0))
2225 		rv = -1;
2226 	zd->zd_od = od;
2227 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2228 
2229 	return (rv);
2230 }
2231 
2232 /* ARGSUSED */
2233 void
2234 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2235 {
2236 	zilog_t *zilog = zd->zd_zilog;
2237 
2238 	(void) rw_rdlock(&zd->zd_zilog_lock);
2239 
2240 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2241 
2242 	/*
2243 	 * Remember the committed values in zd, which is in parent/child
2244 	 * shared memory.  If we die, the next iteration of ztest_run()
2245 	 * will verify that the log really does contain this record.
2246 	 */
2247 	mutex_enter(&zilog->zl_lock);
2248 	ASSERT(zd->zd_shared != NULL);
2249 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2250 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2251 	mutex_exit(&zilog->zl_lock);
2252 
2253 	(void) rw_unlock(&zd->zd_zilog_lock);
2254 }
2255 
2256 /*
2257  * This function is designed to simulate the operations that occur during a
2258  * mount/unmount operation.  We hold the dataset across these operations in an
2259  * attempt to expose any implicit assumptions about ZIL management.
2260  */
2261 /* ARGSUSED */
2262 void
2263 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2264 {
2265 	objset_t *os = zd->zd_os;
2266 
2267 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2268 	(void) rw_wrlock(&zd->zd_zilog_lock);
2269 
2270 	/* zfsvfs_teardown() */
2271 	zil_close(zd->zd_zilog);
2272 
2273 	/* zfsvfs_setup() */
2274 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2275 	zil_replay(os, zd, ztest_replay_vector);
2276 
2277 	(void) rw_unlock(&zd->zd_zilog_lock);
2278 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2279 }
2280 
2281 /*
2282  * Verify that we can't destroy an active pool, create an existing pool,
2283  * or create a pool with a bad vdev spec.
2284  */
2285 /* ARGSUSED */
2286 void
2287 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2288 {
2289 	ztest_shared_opts_t *zo = &ztest_opts;
2290 	spa_t *spa;
2291 	nvlist_t *nvroot;
2292 
2293 	/*
2294 	 * Attempt to create using a bad file.
2295 	 */
2296 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2297 	VERIFY3U(ENOENT, ==,
2298 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2299 	nvlist_free(nvroot);
2300 
2301 	/*
2302 	 * Attempt to create using a bad mirror.
2303 	 */
2304 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2305 	VERIFY3U(ENOENT, ==,
2306 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2307 	nvlist_free(nvroot);
2308 
2309 	/*
2310 	 * Attempt to create an existing pool.  It shouldn't matter
2311 	 * what's in the nvroot; we should fail with EEXIST.
2312 	 */
2313 	(void) rw_rdlock(&ztest_name_lock);
2314 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2315 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2316 	nvlist_free(nvroot);
2317 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2318 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2319 	spa_close(spa, FTAG);
2320 
2321 	(void) rw_unlock(&ztest_name_lock);
2322 }
2323 
2324 /* ARGSUSED */
2325 void
2326 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2327 {
2328 	spa_t *spa;
2329 	uint64_t initial_version = SPA_VERSION_INITIAL;
2330 	uint64_t version, newversion;
2331 	nvlist_t *nvroot, *props;
2332 	char *name;
2333 
2334 	VERIFY0(mutex_lock(&ztest_vdev_lock));
2335 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2336 
2337 	/*
2338 	 * Clean up from previous runs.
2339 	 */
2340 	(void) spa_destroy(name);
2341 
2342 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2343 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2344 
2345 	/*
2346 	 * If we're configuring a RAIDZ device then make sure that the
2347 	 * the initial version is capable of supporting that feature.
2348 	 */
2349 	switch (ztest_opts.zo_raidz_parity) {
2350 	case 0:
2351 	case 1:
2352 		initial_version = SPA_VERSION_INITIAL;
2353 		break;
2354 	case 2:
2355 		initial_version = SPA_VERSION_RAIDZ2;
2356 		break;
2357 	case 3:
2358 		initial_version = SPA_VERSION_RAIDZ3;
2359 		break;
2360 	}
2361 
2362 	/*
2363 	 * Create a pool with a spa version that can be upgraded. Pick
2364 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2365 	 */
2366 	do {
2367 		version = ztest_random_spa_version(initial_version);
2368 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2369 
2370 	props = fnvlist_alloc();
2371 	fnvlist_add_uint64(props,
2372 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2373 	VERIFY0(spa_create(name, nvroot, props, NULL));
2374 	fnvlist_free(nvroot);
2375 	fnvlist_free(props);
2376 
2377 	VERIFY0(spa_open(name, &spa, FTAG));
2378 	VERIFY3U(spa_version(spa), ==, version);
2379 	newversion = ztest_random_spa_version(version + 1);
2380 
2381 	if (ztest_opts.zo_verbose >= 4) {
2382 		(void) printf("upgrading spa version from %llu to %llu\n",
2383 		    (u_longlong_t)version, (u_longlong_t)newversion);
2384 	}
2385 
2386 	spa_upgrade(spa, newversion);
2387 	VERIFY3U(spa_version(spa), >, version);
2388 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2389 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2390 	spa_close(spa, FTAG);
2391 
2392 	strfree(name);
2393 	VERIFY0(mutex_unlock(&ztest_vdev_lock));
2394 }
2395 
2396 static vdev_t *
2397 vdev_lookup_by_path(vdev_t *vd, const char *path)
2398 {
2399 	vdev_t *mvd;
2400 
2401 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2402 		return (vd);
2403 
2404 	for (int c = 0; c < vd->vdev_children; c++)
2405 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2406 		    NULL)
2407 			return (mvd);
2408 
2409 	return (NULL);
2410 }
2411 
2412 /*
2413  * Find the first available hole which can be used as a top-level.
2414  */
2415 int
2416 find_vdev_hole(spa_t *spa)
2417 {
2418 	vdev_t *rvd = spa->spa_root_vdev;
2419 	int c;
2420 
2421 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2422 
2423 	for (c = 0; c < rvd->vdev_children; c++) {
2424 		vdev_t *cvd = rvd->vdev_child[c];
2425 
2426 		if (cvd->vdev_ishole)
2427 			break;
2428 	}
2429 	return (c);
2430 }
2431 
2432 /*
2433  * Verify that vdev_add() works as expected.
2434  */
2435 /* ARGSUSED */
2436 void
2437 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2438 {
2439 	ztest_shared_t *zs = ztest_shared;
2440 	spa_t *spa = ztest_spa;
2441 	uint64_t leaves;
2442 	uint64_t guid;
2443 	nvlist_t *nvroot;
2444 	int error;
2445 
2446 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2447 	leaves =
2448 	    MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2449 
2450 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2451 
2452 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2453 
2454 	/*
2455 	 * If we have slogs then remove them 1/4 of the time.
2456 	 */
2457 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2458 		/*
2459 		 * Grab the guid from the head of the log class rotor.
2460 		 */
2461 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2462 
2463 		spa_config_exit(spa, SCL_VDEV, FTAG);
2464 
2465 		/*
2466 		 * We have to grab the zs_name_lock as writer to
2467 		 * prevent a race between removing a slog (dmu_objset_find)
2468 		 * and destroying a dataset. Removing the slog will
2469 		 * grab a reference on the dataset which may cause
2470 		 * dmu_objset_destroy() to fail with EBUSY thus
2471 		 * leaving the dataset in an inconsistent state.
2472 		 */
2473 		VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2474 		error = spa_vdev_remove(spa, guid, B_FALSE);
2475 		VERIFY(rw_unlock(&ztest_name_lock) == 0);
2476 
2477 		if (error && error != EEXIST)
2478 			fatal(0, "spa_vdev_remove() = %d", error);
2479 	} else {
2480 		spa_config_exit(spa, SCL_VDEV, FTAG);
2481 
2482 		/*
2483 		 * Make 1/4 of the devices be log devices.
2484 		 */
2485 		nvroot = make_vdev_root(NULL, NULL, NULL,
2486 		    ztest_opts.zo_vdev_size, 0,
2487 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2488 		    zs->zs_mirrors, 1);
2489 
2490 		error = spa_vdev_add(spa, nvroot);
2491 		nvlist_free(nvroot);
2492 
2493 		if (error == ENOSPC)
2494 			ztest_record_enospc("spa_vdev_add");
2495 		else if (error != 0)
2496 			fatal(0, "spa_vdev_add() = %d", error);
2497 	}
2498 
2499 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2500 }
2501 
2502 /*
2503  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2504  */
2505 /* ARGSUSED */
2506 void
2507 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2508 {
2509 	ztest_shared_t *zs = ztest_shared;
2510 	spa_t *spa = ztest_spa;
2511 	vdev_t *rvd = spa->spa_root_vdev;
2512 	spa_aux_vdev_t *sav;
2513 	char *aux;
2514 	uint64_t guid = 0;
2515 	int error;
2516 
2517 	if (ztest_random(2) == 0) {
2518 		sav = &spa->spa_spares;
2519 		aux = ZPOOL_CONFIG_SPARES;
2520 	} else {
2521 		sav = &spa->spa_l2cache;
2522 		aux = ZPOOL_CONFIG_L2CACHE;
2523 	}
2524 
2525 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2526 
2527 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2528 
2529 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2530 		/*
2531 		 * Pick a random device to remove.
2532 		 */
2533 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2534 	} else {
2535 		/*
2536 		 * Find an unused device we can add.
2537 		 */
2538 		zs->zs_vdev_aux = 0;
2539 		for (;;) {
2540 			char path[MAXPATHLEN];
2541 			int c;
2542 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2543 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2544 			    zs->zs_vdev_aux);
2545 			for (c = 0; c < sav->sav_count; c++)
2546 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2547 				    path) == 0)
2548 					break;
2549 			if (c == sav->sav_count &&
2550 			    vdev_lookup_by_path(rvd, path) == NULL)
2551 				break;
2552 			zs->zs_vdev_aux++;
2553 		}
2554 	}
2555 
2556 	spa_config_exit(spa, SCL_VDEV, FTAG);
2557 
2558 	if (guid == 0) {
2559 		/*
2560 		 * Add a new device.
2561 		 */
2562 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2563 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2564 		error = spa_vdev_add(spa, nvroot);
2565 		if (error != 0)
2566 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2567 		nvlist_free(nvroot);
2568 	} else {
2569 		/*
2570 		 * Remove an existing device.  Sometimes, dirty its
2571 		 * vdev state first to make sure we handle removal
2572 		 * of devices that have pending state changes.
2573 		 */
2574 		if (ztest_random(2) == 0)
2575 			(void) vdev_online(spa, guid, 0, NULL);
2576 
2577 		error = spa_vdev_remove(spa, guid, B_FALSE);
2578 		if (error != 0 && error != EBUSY)
2579 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2580 	}
2581 
2582 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2583 }
2584 
2585 /*
2586  * split a pool if it has mirror tlvdevs
2587  */
2588 /* ARGSUSED */
2589 void
2590 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2591 {
2592 	ztest_shared_t *zs = ztest_shared;
2593 	spa_t *spa = ztest_spa;
2594 	vdev_t *rvd = spa->spa_root_vdev;
2595 	nvlist_t *tree, **child, *config, *split, **schild;
2596 	uint_t c, children, schildren = 0, lastlogid = 0;
2597 	int error = 0;
2598 
2599 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2600 
2601 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2602 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2603 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2604 		return;
2605 	}
2606 
2607 	/* clean up the old pool, if any */
2608 	(void) spa_destroy("splitp");
2609 
2610 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2611 
2612 	/* generate a config from the existing config */
2613 	mutex_enter(&spa->spa_props_lock);
2614 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2615 	    &tree) == 0);
2616 	mutex_exit(&spa->spa_props_lock);
2617 
2618 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2619 	    &children) == 0);
2620 
2621 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2622 	for (c = 0; c < children; c++) {
2623 		vdev_t *tvd = rvd->vdev_child[c];
2624 		nvlist_t **mchild;
2625 		uint_t mchildren;
2626 
2627 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2628 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2629 			    0) == 0);
2630 			VERIFY(nvlist_add_string(schild[schildren],
2631 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2632 			VERIFY(nvlist_add_uint64(schild[schildren],
2633 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2634 			if (lastlogid == 0)
2635 				lastlogid = schildren;
2636 			++schildren;
2637 			continue;
2638 		}
2639 		lastlogid = 0;
2640 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2641 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2642 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2643 	}
2644 
2645 	/* OK, create a config that can be used to split */
2646 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2647 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2648 	    VDEV_TYPE_ROOT) == 0);
2649 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2650 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2651 
2652 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2653 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2654 
2655 	for (c = 0; c < schildren; c++)
2656 		nvlist_free(schild[c]);
2657 	free(schild);
2658 	nvlist_free(split);
2659 
2660 	spa_config_exit(spa, SCL_VDEV, FTAG);
2661 
2662 	(void) rw_wrlock(&ztest_name_lock);
2663 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2664 	(void) rw_unlock(&ztest_name_lock);
2665 
2666 	nvlist_free(config);
2667 
2668 	if (error == 0) {
2669 		(void) printf("successful split - results:\n");
2670 		mutex_enter(&spa_namespace_lock);
2671 		show_pool_stats(spa);
2672 		show_pool_stats(spa_lookup("splitp"));
2673 		mutex_exit(&spa_namespace_lock);
2674 		++zs->zs_splits;
2675 		--zs->zs_mirrors;
2676 	}
2677 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2678 
2679 }
2680 
2681 /*
2682  * Verify that we can attach and detach devices.
2683  */
2684 /* ARGSUSED */
2685 void
2686 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2687 {
2688 	ztest_shared_t *zs = ztest_shared;
2689 	spa_t *spa = ztest_spa;
2690 	spa_aux_vdev_t *sav = &spa->spa_spares;
2691 	vdev_t *rvd = spa->spa_root_vdev;
2692 	vdev_t *oldvd, *newvd, *pvd;
2693 	nvlist_t *root;
2694 	uint64_t leaves;
2695 	uint64_t leaf, top;
2696 	uint64_t ashift = ztest_get_ashift();
2697 	uint64_t oldguid, pguid;
2698 	size_t oldsize, newsize;
2699 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2700 	int replacing;
2701 	int oldvd_has_siblings = B_FALSE;
2702 	int newvd_is_spare = B_FALSE;
2703 	int oldvd_is_log;
2704 	int error, expected_error;
2705 
2706 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2707 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2708 
2709 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2710 
2711 	/*
2712 	 * Decide whether to do an attach or a replace.
2713 	 */
2714 	replacing = ztest_random(2);
2715 
2716 	/*
2717 	 * Pick a random top-level vdev.
2718 	 */
2719 	top = ztest_random_vdev_top(spa, B_TRUE);
2720 
2721 	/*
2722 	 * Pick a random leaf within it.
2723 	 */
2724 	leaf = ztest_random(leaves);
2725 
2726 	/*
2727 	 * Locate this vdev.
2728 	 */
2729 	oldvd = rvd->vdev_child[top];
2730 	if (zs->zs_mirrors >= 1) {
2731 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2732 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2733 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2734 	}
2735 	if (ztest_opts.zo_raidz > 1) {
2736 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2737 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2738 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2739 	}
2740 
2741 	/*
2742 	 * If we're already doing an attach or replace, oldvd may be a
2743 	 * mirror vdev -- in which case, pick a random child.
2744 	 */
2745 	while (oldvd->vdev_children != 0) {
2746 		oldvd_has_siblings = B_TRUE;
2747 		ASSERT(oldvd->vdev_children >= 2);
2748 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2749 	}
2750 
2751 	oldguid = oldvd->vdev_guid;
2752 	oldsize = vdev_get_min_asize(oldvd);
2753 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2754 	(void) strcpy(oldpath, oldvd->vdev_path);
2755 	pvd = oldvd->vdev_parent;
2756 	pguid = pvd->vdev_guid;
2757 
2758 	/*
2759 	 * If oldvd has siblings, then half of the time, detach it.
2760 	 */
2761 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2762 		spa_config_exit(spa, SCL_VDEV, FTAG);
2763 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2764 		if (error != 0 && error != ENODEV && error != EBUSY &&
2765 		    error != ENOTSUP)
2766 			fatal(0, "detach (%s) returned %d", oldpath, error);
2767 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2768 		return;
2769 	}
2770 
2771 	/*
2772 	 * For the new vdev, choose with equal probability between the two
2773 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2774 	 */
2775 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2776 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2777 		newvd_is_spare = B_TRUE;
2778 		(void) strcpy(newpath, newvd->vdev_path);
2779 	} else {
2780 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2781 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2782 		    top * leaves + leaf);
2783 		if (ztest_random(2) == 0)
2784 			newpath[strlen(newpath) - 1] = 'b';
2785 		newvd = vdev_lookup_by_path(rvd, newpath);
2786 	}
2787 
2788 	if (newvd) {
2789 		newsize = vdev_get_min_asize(newvd);
2790 	} else {
2791 		/*
2792 		 * Make newsize a little bigger or smaller than oldsize.
2793 		 * If it's smaller, the attach should fail.
2794 		 * If it's larger, and we're doing a replace,
2795 		 * we should get dynamic LUN growth when we're done.
2796 		 */
2797 		newsize = 10 * oldsize / (9 + ztest_random(3));
2798 	}
2799 
2800 	/*
2801 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2802 	 * unless it's a replace; in that case any non-replacing parent is OK.
2803 	 *
2804 	 * If newvd is already part of the pool, it should fail with EBUSY.
2805 	 *
2806 	 * If newvd is too small, it should fail with EOVERFLOW.
2807 	 */
2808 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2809 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2810 	    pvd->vdev_ops == &vdev_replacing_ops ||
2811 	    pvd->vdev_ops == &vdev_spare_ops))
2812 		expected_error = ENOTSUP;
2813 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2814 		expected_error = ENOTSUP;
2815 	else if (newvd == oldvd)
2816 		expected_error = replacing ? 0 : EBUSY;
2817 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2818 		expected_error = EBUSY;
2819 	else if (newsize < oldsize)
2820 		expected_error = EOVERFLOW;
2821 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2822 		expected_error = EDOM;
2823 	else
2824 		expected_error = 0;
2825 
2826 	spa_config_exit(spa, SCL_VDEV, FTAG);
2827 
2828 	/*
2829 	 * Build the nvlist describing newpath.
2830 	 */
2831 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2832 	    ashift, 0, 0, 0, 1);
2833 
2834 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2835 
2836 	nvlist_free(root);
2837 
2838 	/*
2839 	 * If our parent was the replacing vdev, but the replace completed,
2840 	 * then instead of failing with ENOTSUP we may either succeed,
2841 	 * fail with ENODEV, or fail with EOVERFLOW.
2842 	 */
2843 	if (expected_error == ENOTSUP &&
2844 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2845 		expected_error = error;
2846 
2847 	/*
2848 	 * If someone grew the LUN, the replacement may be too small.
2849 	 */
2850 	if (error == EOVERFLOW || error == EBUSY)
2851 		expected_error = error;
2852 
2853 	/* XXX workaround 6690467 */
2854 	if (error != expected_error && expected_error != EBUSY) {
2855 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2856 		    "returned %d, expected %d",
2857 		    oldpath, (longlong_t)oldsize, newpath,
2858 		    (longlong_t)newsize, replacing, error, expected_error);
2859 	}
2860 
2861 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2862 }
2863 
2864 /*
2865  * Callback function which expands the physical size of the vdev.
2866  */
2867 vdev_t *
2868 grow_vdev(vdev_t *vd, void *arg)
2869 {
2870 	spa_t *spa = vd->vdev_spa;
2871 	size_t *newsize = arg;
2872 	size_t fsize;
2873 	int fd;
2874 
2875 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2876 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2877 
2878 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2879 		return (vd);
2880 
2881 	fsize = lseek(fd, 0, SEEK_END);
2882 	(void) ftruncate(fd, *newsize);
2883 
2884 	if (ztest_opts.zo_verbose >= 6) {
2885 		(void) printf("%s grew from %lu to %lu bytes\n",
2886 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2887 	}
2888 	(void) close(fd);
2889 	return (NULL);
2890 }
2891 
2892 /*
2893  * Callback function which expands a given vdev by calling vdev_online().
2894  */
2895 /* ARGSUSED */
2896 vdev_t *
2897 online_vdev(vdev_t *vd, void *arg)
2898 {
2899 	spa_t *spa = vd->vdev_spa;
2900 	vdev_t *tvd = vd->vdev_top;
2901 	uint64_t guid = vd->vdev_guid;
2902 	uint64_t generation = spa->spa_config_generation + 1;
2903 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2904 	int error;
2905 
2906 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2907 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2908 
2909 	/* Calling vdev_online will initialize the new metaslabs */
2910 	spa_config_exit(spa, SCL_STATE, spa);
2911 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2912 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2913 
2914 	/*
2915 	 * If vdev_online returned an error or the underlying vdev_open
2916 	 * failed then we abort the expand. The only way to know that
2917 	 * vdev_open fails is by checking the returned newstate.
2918 	 */
2919 	if (error || newstate != VDEV_STATE_HEALTHY) {
2920 		if (ztest_opts.zo_verbose >= 5) {
2921 			(void) printf("Unable to expand vdev, state %llu, "
2922 			    "error %d\n", (u_longlong_t)newstate, error);
2923 		}
2924 		return (vd);
2925 	}
2926 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2927 
2928 	/*
2929 	 * Since we dropped the lock we need to ensure that we're
2930 	 * still talking to the original vdev. It's possible this
2931 	 * vdev may have been detached/replaced while we were
2932 	 * trying to online it.
2933 	 */
2934 	if (generation != spa->spa_config_generation) {
2935 		if (ztest_opts.zo_verbose >= 5) {
2936 			(void) printf("vdev configuration has changed, "
2937 			    "guid %llu, state %llu, expected gen %llu, "
2938 			    "got gen %llu\n",
2939 			    (u_longlong_t)guid,
2940 			    (u_longlong_t)tvd->vdev_state,
2941 			    (u_longlong_t)generation,
2942 			    (u_longlong_t)spa->spa_config_generation);
2943 		}
2944 		return (vd);
2945 	}
2946 	return (NULL);
2947 }
2948 
2949 /*
2950  * Traverse the vdev tree calling the supplied function.
2951  * We continue to walk the tree until we either have walked all
2952  * children or we receive a non-NULL return from the callback.
2953  * If a NULL callback is passed, then we just return back the first
2954  * leaf vdev we encounter.
2955  */
2956 vdev_t *
2957 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2958 {
2959 	if (vd->vdev_ops->vdev_op_leaf) {
2960 		if (func == NULL)
2961 			return (vd);
2962 		else
2963 			return (func(vd, arg));
2964 	}
2965 
2966 	for (uint_t c = 0; c < vd->vdev_children; c++) {
2967 		vdev_t *cvd = vd->vdev_child[c];
2968 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
2969 			return (cvd);
2970 	}
2971 	return (NULL);
2972 }
2973 
2974 /*
2975  * Verify that dynamic LUN growth works as expected.
2976  */
2977 /* ARGSUSED */
2978 void
2979 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
2980 {
2981 	spa_t *spa = ztest_spa;
2982 	vdev_t *vd, *tvd;
2983 	metaslab_class_t *mc;
2984 	metaslab_group_t *mg;
2985 	size_t psize, newsize;
2986 	uint64_t top;
2987 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
2988 
2989 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2990 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2991 
2992 	top = ztest_random_vdev_top(spa, B_TRUE);
2993 
2994 	tvd = spa->spa_root_vdev->vdev_child[top];
2995 	mg = tvd->vdev_mg;
2996 	mc = mg->mg_class;
2997 	old_ms_count = tvd->vdev_ms_count;
2998 	old_class_space = metaslab_class_get_space(mc);
2999 
3000 	/*
3001 	 * Determine the size of the first leaf vdev associated with
3002 	 * our top-level device.
3003 	 */
3004 	vd = vdev_walk_tree(tvd, NULL, NULL);
3005 	ASSERT3P(vd, !=, NULL);
3006 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3007 
3008 	psize = vd->vdev_psize;
3009 
3010 	/*
3011 	 * We only try to expand the vdev if it's healthy, less than 4x its
3012 	 * original size, and it has a valid psize.
3013 	 */
3014 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3015 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3016 		spa_config_exit(spa, SCL_STATE, spa);
3017 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3018 		return;
3019 	}
3020 	ASSERT(psize > 0);
3021 	newsize = psize + psize / 8;
3022 	ASSERT3U(newsize, >, psize);
3023 
3024 	if (ztest_opts.zo_verbose >= 6) {
3025 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3026 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3027 	}
3028 
3029 	/*
3030 	 * Growing the vdev is a two step process:
3031 	 *	1). expand the physical size (i.e. relabel)
3032 	 *	2). online the vdev to create the new metaslabs
3033 	 */
3034 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3035 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3036 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3037 		if (ztest_opts.zo_verbose >= 5) {
3038 			(void) printf("Could not expand LUN because "
3039 			    "the vdev configuration changed.\n");
3040 		}
3041 		spa_config_exit(spa, SCL_STATE, spa);
3042 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3043 		return;
3044 	}
3045 
3046 	spa_config_exit(spa, SCL_STATE, spa);
3047 
3048 	/*
3049 	 * Expanding the LUN will update the config asynchronously,
3050 	 * thus we must wait for the async thread to complete any
3051 	 * pending tasks before proceeding.
3052 	 */
3053 	for (;;) {
3054 		boolean_t done;
3055 		mutex_enter(&spa->spa_async_lock);
3056 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3057 		mutex_exit(&spa->spa_async_lock);
3058 		if (done)
3059 			break;
3060 		txg_wait_synced(spa_get_dsl(spa), 0);
3061 		(void) poll(NULL, 0, 100);
3062 	}
3063 
3064 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3065 
3066 	tvd = spa->spa_root_vdev->vdev_child[top];
3067 	new_ms_count = tvd->vdev_ms_count;
3068 	new_class_space = metaslab_class_get_space(mc);
3069 
3070 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3071 		if (ztest_opts.zo_verbose >= 5) {
3072 			(void) printf("Could not verify LUN expansion due to "
3073 			    "intervening vdev offline or remove.\n");
3074 		}
3075 		spa_config_exit(spa, SCL_STATE, spa);
3076 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3077 		return;
3078 	}
3079 
3080 	/*
3081 	 * Make sure we were able to grow the vdev.
3082 	 */
3083 	if (new_ms_count <= old_ms_count)
3084 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3085 		    old_ms_count, new_ms_count);
3086 
3087 	/*
3088 	 * Make sure we were able to grow the pool.
3089 	 */
3090 	if (new_class_space <= old_class_space)
3091 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3092 		    old_class_space, new_class_space);
3093 
3094 	if (ztest_opts.zo_verbose >= 5) {
3095 		char oldnumbuf[6], newnumbuf[6];
3096 
3097 		nicenum(old_class_space, oldnumbuf);
3098 		nicenum(new_class_space, newnumbuf);
3099 		(void) printf("%s grew from %s to %s\n",
3100 		    spa->spa_name, oldnumbuf, newnumbuf);
3101 	}
3102 
3103 	spa_config_exit(spa, SCL_STATE, spa);
3104 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3105 }
3106 
3107 /*
3108  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3109  */
3110 /* ARGSUSED */
3111 static void
3112 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3113 {
3114 	/*
3115 	 * Create the objects common to all ztest datasets.
3116 	 */
3117 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3118 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3119 }
3120 
3121 static int
3122 ztest_dataset_create(char *dsname)
3123 {
3124 	uint64_t zilset = ztest_random(100);
3125 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3126 	    ztest_objset_create_cb, NULL);
3127 
3128 	if (err || zilset < 80)
3129 		return (err);
3130 
3131 	if (ztest_opts.zo_verbose >= 6)
3132 		(void) printf("Setting dataset %s to sync always\n", dsname);
3133 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3134 	    ZFS_SYNC_ALWAYS, B_FALSE));
3135 }
3136 
3137 /* ARGSUSED */
3138 static int
3139 ztest_objset_destroy_cb(const char *name, void *arg)
3140 {
3141 	objset_t *os;
3142 	dmu_object_info_t doi;
3143 	int error;
3144 
3145 	/*
3146 	 * Verify that the dataset contains a directory object.
3147 	 */
3148 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os));
3149 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3150 	if (error != ENOENT) {
3151 		/* We could have crashed in the middle of destroying it */
3152 		ASSERT0(error);
3153 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3154 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3155 	}
3156 	dmu_objset_rele(os, FTAG);
3157 
3158 	/*
3159 	 * Destroy the dataset.
3160 	 */
3161 	VERIFY3U(0, ==, dmu_objset_destroy(name, B_FALSE));
3162 	return (0);
3163 }
3164 
3165 static boolean_t
3166 ztest_snapshot_create(char *osname, uint64_t id)
3167 {
3168 	char snapname[MAXNAMELEN];
3169 	int error;
3170 
3171 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3172 	    (u_longlong_t)id);
3173 
3174 	error = dmu_objset_snapshot_one(osname, strchr(snapname, '@') + 1);
3175 	if (error == ENOSPC) {
3176 		ztest_record_enospc(FTAG);
3177 		return (B_FALSE);
3178 	}
3179 	if (error != 0 && error != EEXIST)
3180 		fatal(0, "ztest_snapshot_create(%s) = %d", snapname, error);
3181 	return (B_TRUE);
3182 }
3183 
3184 static boolean_t
3185 ztest_snapshot_destroy(char *osname, uint64_t id)
3186 {
3187 	char snapname[MAXNAMELEN];
3188 	int error;
3189 
3190 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3191 	    (u_longlong_t)id);
3192 
3193 	error = dmu_objset_destroy(snapname, B_FALSE);
3194 	if (error != 0 && error != ENOENT)
3195 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3196 	return (B_TRUE);
3197 }
3198 
3199 /* ARGSUSED */
3200 void
3201 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3202 {
3203 	ztest_ds_t zdtmp;
3204 	int iters;
3205 	int error;
3206 	objset_t *os, *os2;
3207 	char name[MAXNAMELEN];
3208 	zilog_t *zilog;
3209 
3210 	(void) rw_rdlock(&ztest_name_lock);
3211 
3212 	(void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3213 	    ztest_opts.zo_pool, (u_longlong_t)id);
3214 
3215 	/*
3216 	 * If this dataset exists from a previous run, process its replay log
3217 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3218 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3219 	 */
3220 	if (ztest_random(2) == 0 &&
3221 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3222 		ztest_zd_init(&zdtmp, NULL, os);
3223 		zil_replay(os, &zdtmp, ztest_replay_vector);
3224 		ztest_zd_fini(&zdtmp);
3225 		dmu_objset_disown(os, FTAG);
3226 	}
3227 
3228 	/*
3229 	 * There may be an old instance of the dataset we're about to
3230 	 * create lying around from a previous run.  If so, destroy it
3231 	 * and all of its snapshots.
3232 	 */
3233 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3234 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3235 
3236 	/*
3237 	 * Verify that the destroyed dataset is no longer in the namespace.
3238 	 */
3239 	VERIFY3U(ENOENT, ==, dmu_objset_hold(name, FTAG, &os));
3240 
3241 	/*
3242 	 * Verify that we can create a new dataset.
3243 	 */
3244 	error = ztest_dataset_create(name);
3245 	if (error) {
3246 		if (error == ENOSPC) {
3247 			ztest_record_enospc(FTAG);
3248 			(void) rw_unlock(&ztest_name_lock);
3249 			return;
3250 		}
3251 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3252 	}
3253 
3254 	VERIFY3U(0, ==,
3255 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3256 
3257 	ztest_zd_init(&zdtmp, NULL, os);
3258 
3259 	/*
3260 	 * Open the intent log for it.
3261 	 */
3262 	zilog = zil_open(os, ztest_get_data);
3263 
3264 	/*
3265 	 * Put some objects in there, do a little I/O to them,
3266 	 * and randomly take a couple of snapshots along the way.
3267 	 */
3268 	iters = ztest_random(5);
3269 	for (int i = 0; i < iters; i++) {
3270 		ztest_dmu_object_alloc_free(&zdtmp, id);
3271 		if (ztest_random(iters) == 0)
3272 			(void) ztest_snapshot_create(name, i);
3273 	}
3274 
3275 	/*
3276 	 * Verify that we cannot create an existing dataset.
3277 	 */
3278 	VERIFY3U(EEXIST, ==,
3279 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3280 
3281 	/*
3282 	 * Verify that we can hold an objset that is also owned.
3283 	 */
3284 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3285 	dmu_objset_rele(os2, FTAG);
3286 
3287 	/*
3288 	 * Verify that we cannot own an objset that is already owned.
3289 	 */
3290 	VERIFY3U(EBUSY, ==,
3291 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3292 
3293 	zil_close(zilog);
3294 	dmu_objset_disown(os, FTAG);
3295 	ztest_zd_fini(&zdtmp);
3296 
3297 	(void) rw_unlock(&ztest_name_lock);
3298 }
3299 
3300 /*
3301  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3302  */
3303 void
3304 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3305 {
3306 	(void) rw_rdlock(&ztest_name_lock);
3307 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3308 	(void) ztest_snapshot_create(zd->zd_name, id);
3309 	(void) rw_unlock(&ztest_name_lock);
3310 }
3311 
3312 /*
3313  * Cleanup non-standard snapshots and clones.
3314  */
3315 void
3316 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3317 {
3318 	char snap1name[MAXNAMELEN];
3319 	char clone1name[MAXNAMELEN];
3320 	char snap2name[MAXNAMELEN];
3321 	char clone2name[MAXNAMELEN];
3322 	char snap3name[MAXNAMELEN];
3323 	int error;
3324 
3325 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3326 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3327 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3328 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3329 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3330 
3331 	error = dmu_objset_destroy(clone2name, B_FALSE);
3332 	if (error && error != ENOENT)
3333 		fatal(0, "dmu_objset_destroy(%s) = %d", clone2name, error);
3334 	error = dmu_objset_destroy(snap3name, B_FALSE);
3335 	if (error && error != ENOENT)
3336 		fatal(0, "dmu_objset_destroy(%s) = %d", snap3name, error);
3337 	error = dmu_objset_destroy(snap2name, B_FALSE);
3338 	if (error && error != ENOENT)
3339 		fatal(0, "dmu_objset_destroy(%s) = %d", snap2name, error);
3340 	error = dmu_objset_destroy(clone1name, B_FALSE);
3341 	if (error && error != ENOENT)
3342 		fatal(0, "dmu_objset_destroy(%s) = %d", clone1name, error);
3343 	error = dmu_objset_destroy(snap1name, B_FALSE);
3344 	if (error && error != ENOENT)
3345 		fatal(0, "dmu_objset_destroy(%s) = %d", snap1name, error);
3346 }
3347 
3348 /*
3349  * Verify dsl_dataset_promote handles EBUSY
3350  */
3351 void
3352 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3353 {
3354 	objset_t *clone;
3355 	dsl_dataset_t *ds;
3356 	char snap1name[MAXNAMELEN];
3357 	char clone1name[MAXNAMELEN];
3358 	char snap2name[MAXNAMELEN];
3359 	char clone2name[MAXNAMELEN];
3360 	char snap3name[MAXNAMELEN];
3361 	char *osname = zd->zd_name;
3362 	int error;
3363 
3364 	(void) rw_rdlock(&ztest_name_lock);
3365 
3366 	ztest_dsl_dataset_cleanup(osname, id);
3367 
3368 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3369 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3370 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3371 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3372 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3373 
3374 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3375 	if (error && error != EEXIST) {
3376 		if (error == ENOSPC) {
3377 			ztest_record_enospc(FTAG);
3378 			goto out;
3379 		}
3380 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3381 	}
3382 
3383 	error = dmu_objset_hold(snap1name, FTAG, &clone);
3384 	if (error)
3385 		fatal(0, "dmu_open_snapshot(%s) = %d", snap1name, error);
3386 
3387 	error = dmu_objset_clone(clone1name, dmu_objset_ds(clone), 0);
3388 	dmu_objset_rele(clone, FTAG);
3389 	if (error) {
3390 		if (error == ENOSPC) {
3391 			ztest_record_enospc(FTAG);
3392 			goto out;
3393 		}
3394 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3395 	}
3396 
3397 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3398 	if (error && error != EEXIST) {
3399 		if (error == ENOSPC) {
3400 			ztest_record_enospc(FTAG);
3401 			goto out;
3402 		}
3403 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3404 	}
3405 
3406 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3407 	if (error && error != EEXIST) {
3408 		if (error == ENOSPC) {
3409 			ztest_record_enospc(FTAG);
3410 			goto out;
3411 		}
3412 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3413 	}
3414 
3415 	error = dmu_objset_hold(snap3name, FTAG, &clone);
3416 	if (error)
3417 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3418 
3419 	error = dmu_objset_clone(clone2name, dmu_objset_ds(clone), 0);
3420 	dmu_objset_rele(clone, FTAG);
3421 	if (error) {
3422 		if (error == ENOSPC) {
3423 			ztest_record_enospc(FTAG);
3424 			goto out;
3425 		}
3426 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3427 	}
3428 
3429 	error = dsl_dataset_own(snap2name, B_FALSE, FTAG, &ds);
3430 	if (error)
3431 		fatal(0, "dsl_dataset_own(%s) = %d", snap2name, error);
3432 	error = dsl_dataset_promote(clone2name, NULL);
3433 	if (error != EBUSY)
3434 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3435 		    error);
3436 	dsl_dataset_disown(ds, FTAG);
3437 
3438 out:
3439 	ztest_dsl_dataset_cleanup(osname, id);
3440 
3441 	(void) rw_unlock(&ztest_name_lock);
3442 }
3443 
3444 /*
3445  * Verify that dmu_object_{alloc,free} work as expected.
3446  */
3447 void
3448 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3449 {
3450 	ztest_od_t od[4];
3451 	int batchsize = sizeof (od) / sizeof (od[0]);
3452 
3453 	for (int b = 0; b < batchsize; b++)
3454 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3455 
3456 	/*
3457 	 * Destroy the previous batch of objects, create a new batch,
3458 	 * and do some I/O on the new objects.
3459 	 */
3460 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3461 		return;
3462 
3463 	while (ztest_random(4 * batchsize) != 0)
3464 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3465 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3466 }
3467 
3468 /*
3469  * Verify that dmu_{read,write} work as expected.
3470  */
3471 void
3472 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3473 {
3474 	objset_t *os = zd->zd_os;
3475 	ztest_od_t od[2];
3476 	dmu_tx_t *tx;
3477 	int i, freeit, error;
3478 	uint64_t n, s, txg;
3479 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3480 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3481 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3482 	uint64_t regions = 997;
3483 	uint64_t stride = 123456789ULL;
3484 	uint64_t width = 40;
3485 	int free_percent = 5;
3486 
3487 	/*
3488 	 * This test uses two objects, packobj and bigobj, that are always
3489 	 * updated together (i.e. in the same tx) so that their contents are
3490 	 * in sync and can be compared.  Their contents relate to each other
3491 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3492 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3493 	 * for any index n, there are three bufwads that should be identical:
3494 	 *
3495 	 *	packobj, at offset n * sizeof (bufwad_t)
3496 	 *	bigobj, at the head of the nth chunk
3497 	 *	bigobj, at the tail of the nth chunk
3498 	 *
3499 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3500 	 * and it doesn't have any relation to the object blocksize.
3501 	 * The only requirement is that it can hold at least two bufwads.
3502 	 *
3503 	 * Normally, we write the bufwad to each of these locations.
3504 	 * However, free_percent of the time we instead write zeroes to
3505 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3506 	 * bigobj to packobj, we can verify that the DMU is correctly
3507 	 * tracking which parts of an object are allocated and free,
3508 	 * and that the contents of the allocated blocks are correct.
3509 	 */
3510 
3511 	/*
3512 	 * Read the directory info.  If it's the first time, set things up.
3513 	 */
3514 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3515 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3516 
3517 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3518 		return;
3519 
3520 	bigobj = od[0].od_object;
3521 	packobj = od[1].od_object;
3522 	chunksize = od[0].od_gen;
3523 	ASSERT(chunksize == od[1].od_gen);
3524 
3525 	/*
3526 	 * Prefetch a random chunk of the big object.
3527 	 * Our aim here is to get some async reads in flight
3528 	 * for blocks that we may free below; the DMU should
3529 	 * handle this race correctly.
3530 	 */
3531 	n = ztest_random(regions) * stride + ztest_random(width);
3532 	s = 1 + ztest_random(2 * width - 1);
3533 	dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3534 
3535 	/*
3536 	 * Pick a random index and compute the offsets into packobj and bigobj.
3537 	 */
3538 	n = ztest_random(regions) * stride + ztest_random(width);
3539 	s = 1 + ztest_random(width - 1);
3540 
3541 	packoff = n * sizeof (bufwad_t);
3542 	packsize = s * sizeof (bufwad_t);
3543 
3544 	bigoff = n * chunksize;
3545 	bigsize = s * chunksize;
3546 
3547 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3548 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3549 
3550 	/*
3551 	 * free_percent of the time, free a range of bigobj rather than
3552 	 * overwriting it.
3553 	 */
3554 	freeit = (ztest_random(100) < free_percent);
3555 
3556 	/*
3557 	 * Read the current contents of our objects.
3558 	 */
3559 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3560 	    DMU_READ_PREFETCH);
3561 	ASSERT0(error);
3562 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3563 	    DMU_READ_PREFETCH);
3564 	ASSERT0(error);
3565 
3566 	/*
3567 	 * Get a tx for the mods to both packobj and bigobj.
3568 	 */
3569 	tx = dmu_tx_create(os);
3570 
3571 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3572 
3573 	if (freeit)
3574 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3575 	else
3576 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3577 
3578 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3579 	if (txg == 0) {
3580 		umem_free(packbuf, packsize);
3581 		umem_free(bigbuf, bigsize);
3582 		return;
3583 	}
3584 
3585 	dmu_object_set_checksum(os, bigobj,
3586 	    (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3587 
3588 	dmu_object_set_compress(os, bigobj,
3589 	    (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3590 
3591 	/*
3592 	 * For each index from n to n + s, verify that the existing bufwad
3593 	 * in packobj matches the bufwads at the head and tail of the
3594 	 * corresponding chunk in bigobj.  Then update all three bufwads
3595 	 * with the new values we want to write out.
3596 	 */
3597 	for (i = 0; i < s; i++) {
3598 		/* LINTED */
3599 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3600 		/* LINTED */
3601 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3602 		/* LINTED */
3603 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3604 
3605 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3606 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3607 
3608 		if (pack->bw_txg > txg)
3609 			fatal(0, "future leak: got %llx, open txg is %llx",
3610 			    pack->bw_txg, txg);
3611 
3612 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3613 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3614 			    pack->bw_index, n, i);
3615 
3616 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3617 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3618 
3619 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3620 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3621 
3622 		if (freeit) {
3623 			bzero(pack, sizeof (bufwad_t));
3624 		} else {
3625 			pack->bw_index = n + i;
3626 			pack->bw_txg = txg;
3627 			pack->bw_data = 1 + ztest_random(-2ULL);
3628 		}
3629 		*bigH = *pack;
3630 		*bigT = *pack;
3631 	}
3632 
3633 	/*
3634 	 * We've verified all the old bufwads, and made new ones.
3635 	 * Now write them out.
3636 	 */
3637 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3638 
3639 	if (freeit) {
3640 		if (ztest_opts.zo_verbose >= 7) {
3641 			(void) printf("freeing offset %llx size %llx"
3642 			    " txg %llx\n",
3643 			    (u_longlong_t)bigoff,
3644 			    (u_longlong_t)bigsize,
3645 			    (u_longlong_t)txg);
3646 		}
3647 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3648 	} else {
3649 		if (ztest_opts.zo_verbose >= 7) {
3650 			(void) printf("writing offset %llx size %llx"
3651 			    " txg %llx\n",
3652 			    (u_longlong_t)bigoff,
3653 			    (u_longlong_t)bigsize,
3654 			    (u_longlong_t)txg);
3655 		}
3656 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3657 	}
3658 
3659 	dmu_tx_commit(tx);
3660 
3661 	/*
3662 	 * Sanity check the stuff we just wrote.
3663 	 */
3664 	{
3665 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3666 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3667 
3668 		VERIFY(0 == dmu_read(os, packobj, packoff,
3669 		    packsize, packcheck, DMU_READ_PREFETCH));
3670 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3671 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3672 
3673 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3674 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3675 
3676 		umem_free(packcheck, packsize);
3677 		umem_free(bigcheck, bigsize);
3678 	}
3679 
3680 	umem_free(packbuf, packsize);
3681 	umem_free(bigbuf, bigsize);
3682 }
3683 
3684 void
3685 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3686     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3687 {
3688 	uint64_t i;
3689 	bufwad_t *pack;
3690 	bufwad_t *bigH;
3691 	bufwad_t *bigT;
3692 
3693 	/*
3694 	 * For each index from n to n + s, verify that the existing bufwad
3695 	 * in packobj matches the bufwads at the head and tail of the
3696 	 * corresponding chunk in bigobj.  Then update all three bufwads
3697 	 * with the new values we want to write out.
3698 	 */
3699 	for (i = 0; i < s; i++) {
3700 		/* LINTED */
3701 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3702 		/* LINTED */
3703 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3704 		/* LINTED */
3705 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3706 
3707 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3708 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3709 
3710 		if (pack->bw_txg > txg)
3711 			fatal(0, "future leak: got %llx, open txg is %llx",
3712 			    pack->bw_txg, txg);
3713 
3714 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3715 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3716 			    pack->bw_index, n, i);
3717 
3718 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3719 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3720 
3721 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3722 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3723 
3724 		pack->bw_index = n + i;
3725 		pack->bw_txg = txg;
3726 		pack->bw_data = 1 + ztest_random(-2ULL);
3727 
3728 		*bigH = *pack;
3729 		*bigT = *pack;
3730 	}
3731 }
3732 
3733 void
3734 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3735 {
3736 	objset_t *os = zd->zd_os;
3737 	ztest_od_t od[2];
3738 	dmu_tx_t *tx;
3739 	uint64_t i;
3740 	int error;
3741 	uint64_t n, s, txg;
3742 	bufwad_t *packbuf, *bigbuf;
3743 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3744 	uint64_t blocksize = ztest_random_blocksize();
3745 	uint64_t chunksize = blocksize;
3746 	uint64_t regions = 997;
3747 	uint64_t stride = 123456789ULL;
3748 	uint64_t width = 9;
3749 	dmu_buf_t *bonus_db;
3750 	arc_buf_t **bigbuf_arcbufs;
3751 	dmu_object_info_t doi;
3752 
3753 	/*
3754 	 * This test uses two objects, packobj and bigobj, that are always
3755 	 * updated together (i.e. in the same tx) so that their contents are
3756 	 * in sync and can be compared.  Their contents relate to each other
3757 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3758 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3759 	 * for any index n, there are three bufwads that should be identical:
3760 	 *
3761 	 *	packobj, at offset n * sizeof (bufwad_t)
3762 	 *	bigobj, at the head of the nth chunk
3763 	 *	bigobj, at the tail of the nth chunk
3764 	 *
3765 	 * The chunk size is set equal to bigobj block size so that
3766 	 * dmu_assign_arcbuf() can be tested for object updates.
3767 	 */
3768 
3769 	/*
3770 	 * Read the directory info.  If it's the first time, set things up.
3771 	 */
3772 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3773 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3774 
3775 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3776 		return;
3777 
3778 	bigobj = od[0].od_object;
3779 	packobj = od[1].od_object;
3780 	blocksize = od[0].od_blocksize;
3781 	chunksize = blocksize;
3782 	ASSERT(chunksize == od[1].od_gen);
3783 
3784 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3785 	VERIFY(ISP2(doi.doi_data_block_size));
3786 	VERIFY(chunksize == doi.doi_data_block_size);
3787 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3788 
3789 	/*
3790 	 * Pick a random index and compute the offsets into packobj and bigobj.
3791 	 */
3792 	n = ztest_random(regions) * stride + ztest_random(width);
3793 	s = 1 + ztest_random(width - 1);
3794 
3795 	packoff = n * sizeof (bufwad_t);
3796 	packsize = s * sizeof (bufwad_t);
3797 
3798 	bigoff = n * chunksize;
3799 	bigsize = s * chunksize;
3800 
3801 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3802 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3803 
3804 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3805 
3806 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3807 
3808 	/*
3809 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3810 	 * Iteration 1 test zcopy to already referenced dbufs.
3811 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3812 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3813 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3814 	 * Iteration 5 test zcopy when it can't be done.
3815 	 * Iteration 6 one more zcopy write.
3816 	 */
3817 	for (i = 0; i < 7; i++) {
3818 		uint64_t j;
3819 		uint64_t off;
3820 
3821 		/*
3822 		 * In iteration 5 (i == 5) use arcbufs
3823 		 * that don't match bigobj blksz to test
3824 		 * dmu_assign_arcbuf() when it can't directly
3825 		 * assign an arcbuf to a dbuf.
3826 		 */
3827 		for (j = 0; j < s; j++) {
3828 			if (i != 5) {
3829 				bigbuf_arcbufs[j] =
3830 				    dmu_request_arcbuf(bonus_db, chunksize);
3831 			} else {
3832 				bigbuf_arcbufs[2 * j] =
3833 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3834 				bigbuf_arcbufs[2 * j + 1] =
3835 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3836 			}
3837 		}
3838 
3839 		/*
3840 		 * Get a tx for the mods to both packobj and bigobj.
3841 		 */
3842 		tx = dmu_tx_create(os);
3843 
3844 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3845 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3846 
3847 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3848 		if (txg == 0) {
3849 			umem_free(packbuf, packsize);
3850 			umem_free(bigbuf, bigsize);
3851 			for (j = 0; j < s; j++) {
3852 				if (i != 5) {
3853 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3854 				} else {
3855 					dmu_return_arcbuf(
3856 					    bigbuf_arcbufs[2 * j]);
3857 					dmu_return_arcbuf(
3858 					    bigbuf_arcbufs[2 * j + 1]);
3859 				}
3860 			}
3861 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3862 			dmu_buf_rele(bonus_db, FTAG);
3863 			return;
3864 		}
3865 
3866 		/*
3867 		 * 50% of the time don't read objects in the 1st iteration to
3868 		 * test dmu_assign_arcbuf() for the case when there're no
3869 		 * existing dbufs for the specified offsets.
3870 		 */
3871 		if (i != 0 || ztest_random(2) != 0) {
3872 			error = dmu_read(os, packobj, packoff,
3873 			    packsize, packbuf, DMU_READ_PREFETCH);
3874 			ASSERT0(error);
3875 			error = dmu_read(os, bigobj, bigoff, bigsize,
3876 			    bigbuf, DMU_READ_PREFETCH);
3877 			ASSERT0(error);
3878 		}
3879 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3880 		    n, chunksize, txg);
3881 
3882 		/*
3883 		 * We've verified all the old bufwads, and made new ones.
3884 		 * Now write them out.
3885 		 */
3886 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3887 		if (ztest_opts.zo_verbose >= 7) {
3888 			(void) printf("writing offset %llx size %llx"
3889 			    " txg %llx\n",
3890 			    (u_longlong_t)bigoff,
3891 			    (u_longlong_t)bigsize,
3892 			    (u_longlong_t)txg);
3893 		}
3894 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3895 			dmu_buf_t *dbt;
3896 			if (i != 5) {
3897 				bcopy((caddr_t)bigbuf + (off - bigoff),
3898 				    bigbuf_arcbufs[j]->b_data, chunksize);
3899 			} else {
3900 				bcopy((caddr_t)bigbuf + (off - bigoff),
3901 				    bigbuf_arcbufs[2 * j]->b_data,
3902 				    chunksize / 2);
3903 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3904 				    chunksize / 2,
3905 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3906 				    chunksize / 2);
3907 			}
3908 
3909 			if (i == 1) {
3910 				VERIFY(dmu_buf_hold(os, bigobj, off,
3911 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3912 			}
3913 			if (i != 5) {
3914 				dmu_assign_arcbuf(bonus_db, off,
3915 				    bigbuf_arcbufs[j], tx);
3916 			} else {
3917 				dmu_assign_arcbuf(bonus_db, off,
3918 				    bigbuf_arcbufs[2 * j], tx);
3919 				dmu_assign_arcbuf(bonus_db,
3920 				    off + chunksize / 2,
3921 				    bigbuf_arcbufs[2 * j + 1], tx);
3922 			}
3923 			if (i == 1) {
3924 				dmu_buf_rele(dbt, FTAG);
3925 			}
3926 		}
3927 		dmu_tx_commit(tx);
3928 
3929 		/*
3930 		 * Sanity check the stuff we just wrote.
3931 		 */
3932 		{
3933 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3934 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3935 
3936 			VERIFY(0 == dmu_read(os, packobj, packoff,
3937 			    packsize, packcheck, DMU_READ_PREFETCH));
3938 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
3939 			    bigsize, bigcheck, DMU_READ_PREFETCH));
3940 
3941 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3942 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3943 
3944 			umem_free(packcheck, packsize);
3945 			umem_free(bigcheck, bigsize);
3946 		}
3947 		if (i == 2) {
3948 			txg_wait_open(dmu_objset_pool(os), 0);
3949 		} else if (i == 3) {
3950 			txg_wait_synced(dmu_objset_pool(os), 0);
3951 		}
3952 	}
3953 
3954 	dmu_buf_rele(bonus_db, FTAG);
3955 	umem_free(packbuf, packsize);
3956 	umem_free(bigbuf, bigsize);
3957 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3958 }
3959 
3960 /* ARGSUSED */
3961 void
3962 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3963 {
3964 	ztest_od_t od[1];
3965 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3966 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3967 
3968 	/*
3969 	 * Have multiple threads write to large offsets in an object
3970 	 * to verify that parallel writes to an object -- even to the
3971 	 * same blocks within the object -- doesn't cause any trouble.
3972 	 */
3973 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
3974 
3975 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3976 		return;
3977 
3978 	while (ztest_random(10) != 0)
3979 		ztest_io(zd, od[0].od_object, offset);
3980 }
3981 
3982 void
3983 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
3984 {
3985 	ztest_od_t od[1];
3986 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
3987 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3988 	uint64_t count = ztest_random(20) + 1;
3989 	uint64_t blocksize = ztest_random_blocksize();
3990 	void *data;
3991 
3992 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3993 
3994 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
3995 		return;
3996 
3997 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
3998 		return;
3999 
4000 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4001 
4002 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4003 
4004 	while (ztest_random(count) != 0) {
4005 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4006 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4007 		    data) != 0)
4008 			break;
4009 		while (ztest_random(4) != 0)
4010 			ztest_io(zd, od[0].od_object, randoff);
4011 	}
4012 
4013 	umem_free(data, blocksize);
4014 }
4015 
4016 /*
4017  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4018  */
4019 #define	ZTEST_ZAP_MIN_INTS	1
4020 #define	ZTEST_ZAP_MAX_INTS	4
4021 #define	ZTEST_ZAP_MAX_PROPS	1000
4022 
4023 void
4024 ztest_zap(ztest_ds_t *zd, uint64_t id)
4025 {
4026 	objset_t *os = zd->zd_os;
4027 	ztest_od_t od[1];
4028 	uint64_t object;
4029 	uint64_t txg, last_txg;
4030 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4031 	uint64_t zl_ints, zl_intsize, prop;
4032 	int i, ints;
4033 	dmu_tx_t *tx;
4034 	char propname[100], txgname[100];
4035 	int error;
4036 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4037 
4038 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4039 
4040 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4041 		return;
4042 
4043 	object = od[0].od_object;
4044 
4045 	/*
4046 	 * Generate a known hash collision, and verify that
4047 	 * we can lookup and remove both entries.
4048 	 */
4049 	tx = dmu_tx_create(os);
4050 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4051 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4052 	if (txg == 0)
4053 		return;
4054 	for (i = 0; i < 2; i++) {
4055 		value[i] = i;
4056 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4057 		    1, &value[i], tx));
4058 	}
4059 	for (i = 0; i < 2; i++) {
4060 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4061 		    sizeof (uint64_t), 1, &value[i], tx));
4062 		VERIFY3U(0, ==,
4063 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4064 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4065 		ASSERT3U(zl_ints, ==, 1);
4066 	}
4067 	for (i = 0; i < 2; i++) {
4068 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4069 	}
4070 	dmu_tx_commit(tx);
4071 
4072 	/*
4073 	 * Generate a buch of random entries.
4074 	 */
4075 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4076 
4077 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4078 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4079 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4080 	bzero(value, sizeof (value));
4081 	last_txg = 0;
4082 
4083 	/*
4084 	 * If these zap entries already exist, validate their contents.
4085 	 */
4086 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4087 	if (error == 0) {
4088 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4089 		ASSERT3U(zl_ints, ==, 1);
4090 
4091 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4092 		    zl_ints, &last_txg) == 0);
4093 
4094 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4095 		    &zl_ints) == 0);
4096 
4097 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4098 		ASSERT3U(zl_ints, ==, ints);
4099 
4100 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4101 		    zl_ints, value) == 0);
4102 
4103 		for (i = 0; i < ints; i++) {
4104 			ASSERT3U(value[i], ==, last_txg + object + i);
4105 		}
4106 	} else {
4107 		ASSERT3U(error, ==, ENOENT);
4108 	}
4109 
4110 	/*
4111 	 * Atomically update two entries in our zap object.
4112 	 * The first is named txg_%llu, and contains the txg
4113 	 * in which the property was last updated.  The second
4114 	 * is named prop_%llu, and the nth element of its value
4115 	 * should be txg + object + n.
4116 	 */
4117 	tx = dmu_tx_create(os);
4118 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4119 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4120 	if (txg == 0)
4121 		return;
4122 
4123 	if (last_txg > txg)
4124 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4125 
4126 	for (i = 0; i < ints; i++)
4127 		value[i] = txg + object + i;
4128 
4129 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4130 	    1, &txg, tx));
4131 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4132 	    ints, value, tx));
4133 
4134 	dmu_tx_commit(tx);
4135 
4136 	/*
4137 	 * Remove a random pair of entries.
4138 	 */
4139 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4140 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4141 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4142 
4143 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4144 
4145 	if (error == ENOENT)
4146 		return;
4147 
4148 	ASSERT0(error);
4149 
4150 	tx = dmu_tx_create(os);
4151 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4152 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4153 	if (txg == 0)
4154 		return;
4155 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4156 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4157 	dmu_tx_commit(tx);
4158 }
4159 
4160 /*
4161  * Testcase to test the upgrading of a microzap to fatzap.
4162  */
4163 void
4164 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4165 {
4166 	objset_t *os = zd->zd_os;
4167 	ztest_od_t od[1];
4168 	uint64_t object, txg;
4169 
4170 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4171 
4172 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4173 		return;
4174 
4175 	object = od[0].od_object;
4176 
4177 	/*
4178 	 * Add entries to this ZAP and make sure it spills over
4179 	 * and gets upgraded to a fatzap. Also, since we are adding
4180 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4181 	 */
4182 	for (int i = 0; i < 2050; i++) {
4183 		char name[MAXNAMELEN];
4184 		uint64_t value = i;
4185 		dmu_tx_t *tx;
4186 		int error;
4187 
4188 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4189 		    id, value);
4190 
4191 		tx = dmu_tx_create(os);
4192 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4193 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4194 		if (txg == 0)
4195 			return;
4196 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4197 		    &value, tx);
4198 		ASSERT(error == 0 || error == EEXIST);
4199 		dmu_tx_commit(tx);
4200 	}
4201 }
4202 
4203 /* ARGSUSED */
4204 void
4205 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4206 {
4207 	objset_t *os = zd->zd_os;
4208 	ztest_od_t od[1];
4209 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4210 	dmu_tx_t *tx;
4211 	int i, namelen, error;
4212 	int micro = ztest_random(2);
4213 	char name[20], string_value[20];
4214 	void *data;
4215 
4216 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4217 
4218 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4219 		return;
4220 
4221 	object = od[0].od_object;
4222 
4223 	/*
4224 	 * Generate a random name of the form 'xxx.....' where each
4225 	 * x is a random printable character and the dots are dots.
4226 	 * There are 94 such characters, and the name length goes from
4227 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4228 	 */
4229 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4230 
4231 	for (i = 0; i < 3; i++)
4232 		name[i] = '!' + ztest_random('~' - '!' + 1);
4233 	for (; i < namelen - 1; i++)
4234 		name[i] = '.';
4235 	name[i] = '\0';
4236 
4237 	if ((namelen & 1) || micro) {
4238 		wsize = sizeof (txg);
4239 		wc = 1;
4240 		data = &txg;
4241 	} else {
4242 		wsize = 1;
4243 		wc = namelen;
4244 		data = string_value;
4245 	}
4246 
4247 	count = -1ULL;
4248 	VERIFY(zap_count(os, object, &count) == 0);
4249 	ASSERT(count != -1ULL);
4250 
4251 	/*
4252 	 * Select an operation: length, lookup, add, update, remove.
4253 	 */
4254 	i = ztest_random(5);
4255 
4256 	if (i >= 2) {
4257 		tx = dmu_tx_create(os);
4258 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4259 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4260 		if (txg == 0)
4261 			return;
4262 		bcopy(name, string_value, namelen);
4263 	} else {
4264 		tx = NULL;
4265 		txg = 0;
4266 		bzero(string_value, namelen);
4267 	}
4268 
4269 	switch (i) {
4270 
4271 	case 0:
4272 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4273 		if (error == 0) {
4274 			ASSERT3U(wsize, ==, zl_wsize);
4275 			ASSERT3U(wc, ==, zl_wc);
4276 		} else {
4277 			ASSERT3U(error, ==, ENOENT);
4278 		}
4279 		break;
4280 
4281 	case 1:
4282 		error = zap_lookup(os, object, name, wsize, wc, data);
4283 		if (error == 0) {
4284 			if (data == string_value &&
4285 			    bcmp(name, data, namelen) != 0)
4286 				fatal(0, "name '%s' != val '%s' len %d",
4287 				    name, data, namelen);
4288 		} else {
4289 			ASSERT3U(error, ==, ENOENT);
4290 		}
4291 		break;
4292 
4293 	case 2:
4294 		error = zap_add(os, object, name, wsize, wc, data, tx);
4295 		ASSERT(error == 0 || error == EEXIST);
4296 		break;
4297 
4298 	case 3:
4299 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4300 		break;
4301 
4302 	case 4:
4303 		error = zap_remove(os, object, name, tx);
4304 		ASSERT(error == 0 || error == ENOENT);
4305 		break;
4306 	}
4307 
4308 	if (tx != NULL)
4309 		dmu_tx_commit(tx);
4310 }
4311 
4312 /*
4313  * Commit callback data.
4314  */
4315 typedef struct ztest_cb_data {
4316 	list_node_t		zcd_node;
4317 	uint64_t		zcd_txg;
4318 	int			zcd_expected_err;
4319 	boolean_t		zcd_added;
4320 	boolean_t		zcd_called;
4321 	spa_t			*zcd_spa;
4322 } ztest_cb_data_t;
4323 
4324 /* This is the actual commit callback function */
4325 static void
4326 ztest_commit_callback(void *arg, int error)
4327 {
4328 	ztest_cb_data_t *data = arg;
4329 	uint64_t synced_txg;
4330 
4331 	VERIFY(data != NULL);
4332 	VERIFY3S(data->zcd_expected_err, ==, error);
4333 	VERIFY(!data->zcd_called);
4334 
4335 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4336 	if (data->zcd_txg > synced_txg)
4337 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4338 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4339 		    synced_txg);
4340 
4341 	data->zcd_called = B_TRUE;
4342 
4343 	if (error == ECANCELED) {
4344 		ASSERT0(data->zcd_txg);
4345 		ASSERT(!data->zcd_added);
4346 
4347 		/*
4348 		 * The private callback data should be destroyed here, but
4349 		 * since we are going to check the zcd_called field after
4350 		 * dmu_tx_abort(), we will destroy it there.
4351 		 */
4352 		return;
4353 	}
4354 
4355 	/* Was this callback added to the global callback list? */
4356 	if (!data->zcd_added)
4357 		goto out;
4358 
4359 	ASSERT3U(data->zcd_txg, !=, 0);
4360 
4361 	/* Remove our callback from the list */
4362 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4363 	list_remove(&zcl.zcl_callbacks, data);
4364 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4365 
4366 out:
4367 	umem_free(data, sizeof (ztest_cb_data_t));
4368 }
4369 
4370 /* Allocate and initialize callback data structure */
4371 static ztest_cb_data_t *
4372 ztest_create_cb_data(objset_t *os, uint64_t txg)
4373 {
4374 	ztest_cb_data_t *cb_data;
4375 
4376 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4377 
4378 	cb_data->zcd_txg = txg;
4379 	cb_data->zcd_spa = dmu_objset_spa(os);
4380 
4381 	return (cb_data);
4382 }
4383 
4384 /*
4385  * If a number of txgs equal to this threshold have been created after a commit
4386  * callback has been registered but not called, then we assume there is an
4387  * implementation bug.
4388  */
4389 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4390 
4391 /*
4392  * Commit callback test.
4393  */
4394 void
4395 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4396 {
4397 	objset_t *os = zd->zd_os;
4398 	ztest_od_t od[1];
4399 	dmu_tx_t *tx;
4400 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4401 	uint64_t old_txg, txg;
4402 	int i, error;
4403 
4404 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4405 
4406 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4407 		return;
4408 
4409 	tx = dmu_tx_create(os);
4410 
4411 	cb_data[0] = ztest_create_cb_data(os, 0);
4412 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4413 
4414 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4415 
4416 	/* Every once in a while, abort the transaction on purpose */
4417 	if (ztest_random(100) == 0)
4418 		error = -1;
4419 
4420 	if (!error)
4421 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4422 
4423 	txg = error ? 0 : dmu_tx_get_txg(tx);
4424 
4425 	cb_data[0]->zcd_txg = txg;
4426 	cb_data[1] = ztest_create_cb_data(os, txg);
4427 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4428 
4429 	if (error) {
4430 		/*
4431 		 * It's not a strict requirement to call the registered
4432 		 * callbacks from inside dmu_tx_abort(), but that's what
4433 		 * it's supposed to happen in the current implementation
4434 		 * so we will check for that.
4435 		 */
4436 		for (i = 0; i < 2; i++) {
4437 			cb_data[i]->zcd_expected_err = ECANCELED;
4438 			VERIFY(!cb_data[i]->zcd_called);
4439 		}
4440 
4441 		dmu_tx_abort(tx);
4442 
4443 		for (i = 0; i < 2; i++) {
4444 			VERIFY(cb_data[i]->zcd_called);
4445 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4446 		}
4447 
4448 		return;
4449 	}
4450 
4451 	cb_data[2] = ztest_create_cb_data(os, txg);
4452 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4453 
4454 	/*
4455 	 * Read existing data to make sure there isn't a future leak.
4456 	 */
4457 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4458 	    &old_txg, DMU_READ_PREFETCH));
4459 
4460 	if (old_txg > txg)
4461 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4462 		    old_txg, txg);
4463 
4464 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4465 
4466 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4467 
4468 	/*
4469 	 * Since commit callbacks don't have any ordering requirement and since
4470 	 * it is theoretically possible for a commit callback to be called
4471 	 * after an arbitrary amount of time has elapsed since its txg has been
4472 	 * synced, it is difficult to reliably determine whether a commit
4473 	 * callback hasn't been called due to high load or due to a flawed
4474 	 * implementation.
4475 	 *
4476 	 * In practice, we will assume that if after a certain number of txgs a
4477 	 * commit callback hasn't been called, then most likely there's an
4478 	 * implementation bug..
4479 	 */
4480 	tmp_cb = list_head(&zcl.zcl_callbacks);
4481 	if (tmp_cb != NULL &&
4482 	    tmp_cb->zcd_txg > txg - ZTEST_COMMIT_CALLBACK_THRESH) {
4483 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4484 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4485 	}
4486 
4487 	/*
4488 	 * Let's find the place to insert our callbacks.
4489 	 *
4490 	 * Even though the list is ordered by txg, it is possible for the
4491 	 * insertion point to not be the end because our txg may already be
4492 	 * quiescing at this point and other callbacks in the open txg
4493 	 * (from other objsets) may have sneaked in.
4494 	 */
4495 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4496 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4497 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4498 
4499 	/* Add the 3 callbacks to the list */
4500 	for (i = 0; i < 3; i++) {
4501 		if (tmp_cb == NULL)
4502 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4503 		else
4504 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4505 			    cb_data[i]);
4506 
4507 		cb_data[i]->zcd_added = B_TRUE;
4508 		VERIFY(!cb_data[i]->zcd_called);
4509 
4510 		tmp_cb = cb_data[i];
4511 	}
4512 
4513 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4514 
4515 	dmu_tx_commit(tx);
4516 }
4517 
4518 /* ARGSUSED */
4519 void
4520 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4521 {
4522 	zfs_prop_t proplist[] = {
4523 		ZFS_PROP_CHECKSUM,
4524 		ZFS_PROP_COMPRESSION,
4525 		ZFS_PROP_COPIES,
4526 		ZFS_PROP_DEDUP
4527 	};
4528 
4529 	(void) rw_rdlock(&ztest_name_lock);
4530 
4531 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4532 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4533 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4534 
4535 	(void) rw_unlock(&ztest_name_lock);
4536 }
4537 
4538 /* ARGSUSED */
4539 void
4540 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4541 {
4542 	nvlist_t *props = NULL;
4543 
4544 	(void) rw_rdlock(&ztest_name_lock);
4545 
4546 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4547 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4548 
4549 	VERIFY0(spa_prop_get(ztest_spa, &props));
4550 
4551 	if (ztest_opts.zo_verbose >= 6)
4552 		dump_nvlist(props, 4);
4553 
4554 	nvlist_free(props);
4555 
4556 	(void) rw_unlock(&ztest_name_lock);
4557 }
4558 
4559 /*
4560  * Test snapshot hold/release and deferred destroy.
4561  */
4562 void
4563 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4564 {
4565 	int error;
4566 	objset_t *os = zd->zd_os;
4567 	objset_t *origin;
4568 	char snapname[100];
4569 	char fullname[100];
4570 	char clonename[100];
4571 	char tag[100];
4572 	char osname[MAXNAMELEN];
4573 
4574 	(void) rw_rdlock(&ztest_name_lock);
4575 
4576 	dmu_objset_name(os, osname);
4577 
4578 	(void) snprintf(snapname, 100, "sh1_%llu", id);
4579 	(void) snprintf(fullname, 100, "%s@%s", osname, snapname);
4580 	(void) snprintf(clonename, 100, "%s/ch1_%llu", osname, id);
4581 	(void) snprintf(tag, 100, "%tag_%llu", id);
4582 
4583 	/*
4584 	 * Clean up from any previous run.
4585 	 */
4586 	(void) dmu_objset_destroy(clonename, B_FALSE);
4587 	(void) dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4588 	(void) dmu_objset_destroy(fullname, B_FALSE);
4589 
4590 	/*
4591 	 * Create snapshot, clone it, mark snap for deferred destroy,
4592 	 * destroy clone, verify snap was also destroyed.
4593 	 */
4594 	error = dmu_objset_snapshot_one(osname, snapname);
4595 	if (error) {
4596 		if (error == ENOSPC) {
4597 			ztest_record_enospc("dmu_objset_snapshot");
4598 			goto out;
4599 		}
4600 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4601 	}
4602 
4603 	error = dmu_objset_hold(fullname, FTAG, &origin);
4604 	if (error)
4605 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4606 
4607 	error = dmu_objset_clone(clonename, dmu_objset_ds(origin), 0);
4608 	dmu_objset_rele(origin, FTAG);
4609 	if (error) {
4610 		if (error == ENOSPC) {
4611 			ztest_record_enospc("dmu_objset_clone");
4612 			goto out;
4613 		}
4614 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4615 	}
4616 
4617 	error = dmu_objset_destroy(fullname, B_TRUE);
4618 	if (error) {
4619 		fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4620 		    fullname, error);
4621 	}
4622 
4623 	error = dmu_objset_destroy(clonename, B_FALSE);
4624 	if (error)
4625 		fatal(0, "dmu_objset_destroy(%s) = %d", clonename, error);
4626 
4627 	error = dmu_objset_hold(fullname, FTAG, &origin);
4628 	if (error != ENOENT)
4629 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4630 
4631 	/*
4632 	 * Create snapshot, add temporary hold, verify that we can't
4633 	 * destroy a held snapshot, mark for deferred destroy,
4634 	 * release hold, verify snapshot was destroyed.
4635 	 */
4636 	error = dmu_objset_snapshot_one(osname, snapname);
4637 	if (error) {
4638 		if (error == ENOSPC) {
4639 			ztest_record_enospc("dmu_objset_snapshot");
4640 			goto out;
4641 		}
4642 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4643 	}
4644 
4645 	error = dsl_dataset_user_hold(osname, snapname, tag, B_FALSE,
4646 	    B_TRUE, -1);
4647 	if (error)
4648 		fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4649 
4650 	error = dmu_objset_destroy(fullname, B_FALSE);
4651 	if (error != EBUSY) {
4652 		fatal(0, "dmu_objset_destroy(%s, B_FALSE) = %d",
4653 		    fullname, error);
4654 	}
4655 
4656 	error = dmu_objset_destroy(fullname, B_TRUE);
4657 	if (error) {
4658 		fatal(0, "dmu_objset_destroy(%s, B_TRUE) = %d",
4659 		    fullname, error);
4660 	}
4661 
4662 	error = dsl_dataset_user_release(osname, snapname, tag, B_FALSE);
4663 	if (error)
4664 		fatal(0, "dsl_dataset_user_release(%s)", fullname, tag);
4665 
4666 	VERIFY(dmu_objset_hold(fullname, FTAG, &origin) == ENOENT);
4667 
4668 out:
4669 	(void) rw_unlock(&ztest_name_lock);
4670 }
4671 
4672 /*
4673  * Inject random faults into the on-disk data.
4674  */
4675 /* ARGSUSED */
4676 void
4677 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4678 {
4679 	ztest_shared_t *zs = ztest_shared;
4680 	spa_t *spa = ztest_spa;
4681 	int fd;
4682 	uint64_t offset;
4683 	uint64_t leaves;
4684 	uint64_t bad = 0x1990c0ffeedecade;
4685 	uint64_t top, leaf;
4686 	char path0[MAXPATHLEN];
4687 	char pathrand[MAXPATHLEN];
4688 	size_t fsize;
4689 	int bshift = SPA_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
4690 	int iters = 1000;
4691 	int maxfaults;
4692 	int mirror_save;
4693 	vdev_t *vd0 = NULL;
4694 	uint64_t guid0 = 0;
4695 	boolean_t islog = B_FALSE;
4696 
4697 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4698 	maxfaults = MAXFAULTS();
4699 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4700 	mirror_save = zs->zs_mirrors;
4701 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4702 
4703 	ASSERT(leaves >= 1);
4704 
4705 	/*
4706 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4707 	 */
4708 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4709 
4710 	if (ztest_random(2) == 0) {
4711 		/*
4712 		 * Inject errors on a normal data device or slog device.
4713 		 */
4714 		top = ztest_random_vdev_top(spa, B_TRUE);
4715 		leaf = ztest_random(leaves) + zs->zs_splits;
4716 
4717 		/*
4718 		 * Generate paths to the first leaf in this top-level vdev,
4719 		 * and to the random leaf we selected.  We'll induce transient
4720 		 * write failures and random online/offline activity on leaf 0,
4721 		 * and we'll write random garbage to the randomly chosen leaf.
4722 		 */
4723 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4724 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4725 		    top * leaves + zs->zs_splits);
4726 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4727 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4728 		    top * leaves + leaf);
4729 
4730 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4731 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4732 			islog = B_TRUE;
4733 
4734 		if (vd0 != NULL && maxfaults != 1) {
4735 			/*
4736 			 * Make vd0 explicitly claim to be unreadable,
4737 			 * or unwriteable, or reach behind its back
4738 			 * and close the underlying fd.  We can do this if
4739 			 * maxfaults == 0 because we'll fail and reexecute,
4740 			 * and we can do it if maxfaults >= 2 because we'll
4741 			 * have enough redundancy.  If maxfaults == 1, the
4742 			 * combination of this with injection of random data
4743 			 * corruption below exceeds the pool's fault tolerance.
4744 			 */
4745 			vdev_file_t *vf = vd0->vdev_tsd;
4746 
4747 			if (vf != NULL && ztest_random(3) == 0) {
4748 				(void) close(vf->vf_vnode->v_fd);
4749 				vf->vf_vnode->v_fd = -1;
4750 			} else if (ztest_random(2) == 0) {
4751 				vd0->vdev_cant_read = B_TRUE;
4752 			} else {
4753 				vd0->vdev_cant_write = B_TRUE;
4754 			}
4755 			guid0 = vd0->vdev_guid;
4756 		}
4757 	} else {
4758 		/*
4759 		 * Inject errors on an l2cache device.
4760 		 */
4761 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4762 
4763 		if (sav->sav_count == 0) {
4764 			spa_config_exit(spa, SCL_STATE, FTAG);
4765 			return;
4766 		}
4767 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4768 		guid0 = vd0->vdev_guid;
4769 		(void) strcpy(path0, vd0->vdev_path);
4770 		(void) strcpy(pathrand, vd0->vdev_path);
4771 
4772 		leaf = 0;
4773 		leaves = 1;
4774 		maxfaults = INT_MAX;	/* no limit on cache devices */
4775 	}
4776 
4777 	spa_config_exit(spa, SCL_STATE, FTAG);
4778 
4779 	/*
4780 	 * If we can tolerate two or more faults, or we're dealing
4781 	 * with a slog, randomly online/offline vd0.
4782 	 */
4783 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4784 		if (ztest_random(10) < 6) {
4785 			int flags = (ztest_random(2) == 0 ?
4786 			    ZFS_OFFLINE_TEMPORARY : 0);
4787 
4788 			/*
4789 			 * We have to grab the zs_name_lock as writer to
4790 			 * prevent a race between offlining a slog and
4791 			 * destroying a dataset. Offlining the slog will
4792 			 * grab a reference on the dataset which may cause
4793 			 * dmu_objset_destroy() to fail with EBUSY thus
4794 			 * leaving the dataset in an inconsistent state.
4795 			 */
4796 			if (islog)
4797 				(void) rw_wrlock(&ztest_name_lock);
4798 
4799 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4800 
4801 			if (islog)
4802 				(void) rw_unlock(&ztest_name_lock);
4803 		} else {
4804 			/*
4805 			 * Ideally we would like to be able to randomly
4806 			 * call vdev_[on|off]line without holding locks
4807 			 * to force unpredictable failures but the side
4808 			 * effects of vdev_[on|off]line prevent us from
4809 			 * doing so. We grab the ztest_vdev_lock here to
4810 			 * prevent a race between injection testing and
4811 			 * aux_vdev removal.
4812 			 */
4813 			VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4814 			(void) vdev_online(spa, guid0, 0, NULL);
4815 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4816 		}
4817 	}
4818 
4819 	if (maxfaults == 0)
4820 		return;
4821 
4822 	/*
4823 	 * We have at least single-fault tolerance, so inject data corruption.
4824 	 */
4825 	fd = open(pathrand, O_RDWR);
4826 
4827 	if (fd == -1)	/* we hit a gap in the device namespace */
4828 		return;
4829 
4830 	fsize = lseek(fd, 0, SEEK_END);
4831 
4832 	while (--iters != 0) {
4833 		offset = ztest_random(fsize / (leaves << bshift)) *
4834 		    (leaves << bshift) + (leaf << bshift) +
4835 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4836 
4837 		if (offset >= fsize)
4838 			continue;
4839 
4840 		VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4841 		if (mirror_save != zs->zs_mirrors) {
4842 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4843 			(void) close(fd);
4844 			return;
4845 		}
4846 
4847 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4848 			fatal(1, "can't inject bad word at 0x%llx in %s",
4849 			    offset, pathrand);
4850 
4851 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4852 
4853 		if (ztest_opts.zo_verbose >= 7)
4854 			(void) printf("injected bad word into %s,"
4855 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4856 	}
4857 
4858 	(void) close(fd);
4859 }
4860 
4861 /*
4862  * Verify that DDT repair works as expected.
4863  */
4864 void
4865 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4866 {
4867 	ztest_shared_t *zs = ztest_shared;
4868 	spa_t *spa = ztest_spa;
4869 	objset_t *os = zd->zd_os;
4870 	ztest_od_t od[1];
4871 	uint64_t object, blocksize, txg, pattern, psize;
4872 	enum zio_checksum checksum = spa_dedup_checksum(spa);
4873 	dmu_buf_t *db;
4874 	dmu_tx_t *tx;
4875 	void *buf;
4876 	blkptr_t blk;
4877 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
4878 
4879 	blocksize = ztest_random_blocksize();
4880 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
4881 
4882 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4883 
4884 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4885 		return;
4886 
4887 	/*
4888 	 * Take the name lock as writer to prevent anyone else from changing
4889 	 * the pool and dataset properies we need to maintain during this test.
4890 	 */
4891 	(void) rw_wrlock(&ztest_name_lock);
4892 
4893 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4894 	    B_FALSE) != 0 ||
4895 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4896 	    B_FALSE) != 0) {
4897 		(void) rw_unlock(&ztest_name_lock);
4898 		return;
4899 	}
4900 
4901 	object = od[0].od_object;
4902 	blocksize = od[0].od_blocksize;
4903 	pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4904 
4905 	ASSERT(object != 0);
4906 
4907 	tx = dmu_tx_create(os);
4908 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4909 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4910 	if (txg == 0) {
4911 		(void) rw_unlock(&ztest_name_lock);
4912 		return;
4913 	}
4914 
4915 	/*
4916 	 * Write all the copies of our block.
4917 	 */
4918 	for (int i = 0; i < copies; i++) {
4919 		uint64_t offset = i * blocksize;
4920 		VERIFY(dmu_buf_hold(os, object, offset, FTAG, &db,
4921 		    DMU_READ_NO_PREFETCH) == 0);
4922 		ASSERT(db->db_offset == offset);
4923 		ASSERT(db->db_size == blocksize);
4924 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4925 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4926 		dmu_buf_will_fill(db, tx);
4927 		ztest_pattern_set(db->db_data, db->db_size, pattern);
4928 		dmu_buf_rele(db, FTAG);
4929 	}
4930 
4931 	dmu_tx_commit(tx);
4932 	txg_wait_synced(spa_get_dsl(spa), txg);
4933 
4934 	/*
4935 	 * Find out what block we got.
4936 	 */
4937 	VERIFY(dmu_buf_hold(os, object, 0, FTAG, &db,
4938 	    DMU_READ_NO_PREFETCH) == 0);
4939 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4940 	dmu_buf_rele(db, FTAG);
4941 
4942 	/*
4943 	 * Damage the block.  Dedup-ditto will save us when we read it later.
4944 	 */
4945 	psize = BP_GET_PSIZE(&blk);
4946 	buf = zio_buf_alloc(psize);
4947 	ztest_pattern_set(buf, psize, ~pattern);
4948 
4949 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
4950 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
4951 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
4952 
4953 	zio_buf_free(buf, psize);
4954 
4955 	(void) rw_unlock(&ztest_name_lock);
4956 }
4957 
4958 /*
4959  * Scrub the pool.
4960  */
4961 /* ARGSUSED */
4962 void
4963 ztest_scrub(ztest_ds_t *zd, uint64_t id)
4964 {
4965 	spa_t *spa = ztest_spa;
4966 
4967 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
4968 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
4969 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
4970 }
4971 
4972 /*
4973  * Change the guid for the pool.
4974  */
4975 /* ARGSUSED */
4976 void
4977 ztest_reguid(ztest_ds_t *zd, uint64_t id)
4978 {
4979 	spa_t *spa = ztest_spa;
4980 	uint64_t orig, load;
4981 	int error;
4982 
4983 	orig = spa_guid(spa);
4984 	load = spa_load_guid(spa);
4985 
4986 	(void) rw_wrlock(&ztest_name_lock);
4987 	error = spa_change_guid(spa);
4988 	(void) rw_unlock(&ztest_name_lock);
4989 
4990 	if (error != 0)
4991 		return;
4992 
4993 	if (ztest_opts.zo_verbose >= 4) {
4994 		(void) printf("Changed guid old %llu -> %llu\n",
4995 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
4996 	}
4997 
4998 	VERIFY3U(orig, !=, spa_guid(spa));
4999 	VERIFY3U(load, ==, spa_load_guid(spa));
5000 }
5001 
5002 /*
5003  * Rename the pool to a different name and then rename it back.
5004  */
5005 /* ARGSUSED */
5006 void
5007 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5008 {
5009 	char *oldname, *newname;
5010 	spa_t *spa;
5011 
5012 	(void) rw_wrlock(&ztest_name_lock);
5013 
5014 	oldname = ztest_opts.zo_pool;
5015 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5016 	(void) strcpy(newname, oldname);
5017 	(void) strcat(newname, "_tmp");
5018 
5019 	/*
5020 	 * Do the rename
5021 	 */
5022 	VERIFY3U(0, ==, spa_rename(oldname, newname));
5023 
5024 	/*
5025 	 * Try to open it under the old name, which shouldn't exist
5026 	 */
5027 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5028 
5029 	/*
5030 	 * Open it under the new name and make sure it's still the same spa_t.
5031 	 */
5032 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5033 
5034 	ASSERT(spa == ztest_spa);
5035 	spa_close(spa, FTAG);
5036 
5037 	/*
5038 	 * Rename it back to the original
5039 	 */
5040 	VERIFY3U(0, ==, spa_rename(newname, oldname));
5041 
5042 	/*
5043 	 * Make sure it can still be opened
5044 	 */
5045 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5046 
5047 	ASSERT(spa == ztest_spa);
5048 	spa_close(spa, FTAG);
5049 
5050 	umem_free(newname, strlen(newname) + 1);
5051 
5052 	(void) rw_unlock(&ztest_name_lock);
5053 }
5054 
5055 /*
5056  * Verify pool integrity by running zdb.
5057  */
5058 static void
5059 ztest_run_zdb(char *pool)
5060 {
5061 	int status;
5062 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5063 	char zbuf[1024];
5064 	char *bin;
5065 	char *ztest;
5066 	char *isa;
5067 	int isalen;
5068 	FILE *fp;
5069 
5070 	(void) realpath(getexecname(), zdb);
5071 
5072 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5073 	bin = strstr(zdb, "/usr/bin/");
5074 	ztest = strstr(bin, "/ztest");
5075 	isa = bin + 8;
5076 	isalen = ztest - isa;
5077 	isa = strdup(isa);
5078 	/* LINTED */
5079 	(void) sprintf(bin,
5080 	    "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
5081 	    isalen,
5082 	    isa,
5083 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5084 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5085 	    spa_config_path,
5086 	    pool);
5087 	free(isa);
5088 
5089 	if (ztest_opts.zo_verbose >= 5)
5090 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5091 
5092 	fp = popen(zdb, "r");
5093 
5094 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5095 		if (ztest_opts.zo_verbose >= 3)
5096 			(void) printf("%s", zbuf);
5097 
5098 	status = pclose(fp);
5099 
5100 	if (status == 0)
5101 		return;
5102 
5103 	ztest_dump_core = 0;
5104 	if (WIFEXITED(status))
5105 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5106 	else
5107 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5108 }
5109 
5110 static void
5111 ztest_walk_pool_directory(char *header)
5112 {
5113 	spa_t *spa = NULL;
5114 
5115 	if (ztest_opts.zo_verbose >= 6)
5116 		(void) printf("%s\n", header);
5117 
5118 	mutex_enter(&spa_namespace_lock);
5119 	while ((spa = spa_next(spa)) != NULL)
5120 		if (ztest_opts.zo_verbose >= 6)
5121 			(void) printf("\t%s\n", spa_name(spa));
5122 	mutex_exit(&spa_namespace_lock);
5123 }
5124 
5125 static void
5126 ztest_spa_import_export(char *oldname, char *newname)
5127 {
5128 	nvlist_t *config, *newconfig;
5129 	uint64_t pool_guid;
5130 	spa_t *spa;
5131 
5132 	if (ztest_opts.zo_verbose >= 4) {
5133 		(void) printf("import/export: old = %s, new = %s\n",
5134 		    oldname, newname);
5135 	}
5136 
5137 	/*
5138 	 * Clean up from previous runs.
5139 	 */
5140 	(void) spa_destroy(newname);
5141 
5142 	/*
5143 	 * Get the pool's configuration and guid.
5144 	 */
5145 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5146 
5147 	/*
5148 	 * Kick off a scrub to tickle scrub/export races.
5149 	 */
5150 	if (ztest_random(2) == 0)
5151 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5152 
5153 	pool_guid = spa_guid(spa);
5154 	spa_close(spa, FTAG);
5155 
5156 	ztest_walk_pool_directory("pools before export");
5157 
5158 	/*
5159 	 * Export it.
5160 	 */
5161 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5162 
5163 	ztest_walk_pool_directory("pools after export");
5164 
5165 	/*
5166 	 * Try to import it.
5167 	 */
5168 	newconfig = spa_tryimport(config);
5169 	ASSERT(newconfig != NULL);
5170 	nvlist_free(newconfig);
5171 
5172 	/*
5173 	 * Import it under the new name.
5174 	 */
5175 	VERIFY3U(0, ==, spa_import(newname, config, NULL, 0));
5176 
5177 	ztest_walk_pool_directory("pools after import");
5178 
5179 	/*
5180 	 * Try to import it again -- should fail with EEXIST.
5181 	 */
5182 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5183 
5184 	/*
5185 	 * Try to import it under a different name -- should fail with EEXIST.
5186 	 */
5187 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5188 
5189 	/*
5190 	 * Verify that the pool is no longer visible under the old name.
5191 	 */
5192 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5193 
5194 	/*
5195 	 * Verify that we can open and close the pool using the new name.
5196 	 */
5197 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5198 	ASSERT(pool_guid == spa_guid(spa));
5199 	spa_close(spa, FTAG);
5200 
5201 	nvlist_free(config);
5202 }
5203 
5204 static void
5205 ztest_resume(spa_t *spa)
5206 {
5207 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5208 		(void) printf("resuming from suspended state\n");
5209 	spa_vdev_state_enter(spa, SCL_NONE);
5210 	vdev_clear(spa, NULL);
5211 	(void) spa_vdev_state_exit(spa, NULL, 0);
5212 	(void) zio_resume(spa);
5213 }
5214 
5215 static void *
5216 ztest_resume_thread(void *arg)
5217 {
5218 	spa_t *spa = arg;
5219 
5220 	while (!ztest_exiting) {
5221 		if (spa_suspended(spa))
5222 			ztest_resume(spa);
5223 		(void) poll(NULL, 0, 100);
5224 	}
5225 	return (NULL);
5226 }
5227 
5228 static void *
5229 ztest_deadman_thread(void *arg)
5230 {
5231 	ztest_shared_t *zs = arg;
5232 	int grace = 300;
5233 	hrtime_t delta;
5234 
5235 	delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
5236 
5237 	(void) poll(NULL, 0, (int)(1000 * delta));
5238 
5239 	fatal(0, "failed to complete within %d seconds of deadline", grace);
5240 
5241 	return (NULL);
5242 }
5243 
5244 static void
5245 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5246 {
5247 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5248 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5249 	hrtime_t functime = gethrtime();
5250 
5251 	for (int i = 0; i < zi->zi_iters; i++)
5252 		zi->zi_func(zd, id);
5253 
5254 	functime = gethrtime() - functime;
5255 
5256 	atomic_add_64(&zc->zc_count, 1);
5257 	atomic_add_64(&zc->zc_time, functime);
5258 
5259 	if (ztest_opts.zo_verbose >= 4) {
5260 		Dl_info dli;
5261 		(void) dladdr((void *)zi->zi_func, &dli);
5262 		(void) printf("%6.2f sec in %s\n",
5263 		    (double)functime / NANOSEC, dli.dli_sname);
5264 	}
5265 }
5266 
5267 static void *
5268 ztest_thread(void *arg)
5269 {
5270 	int rand;
5271 	uint64_t id = (uintptr_t)arg;
5272 	ztest_shared_t *zs = ztest_shared;
5273 	uint64_t call_next;
5274 	hrtime_t now;
5275 	ztest_info_t *zi;
5276 	ztest_shared_callstate_t *zc;
5277 
5278 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5279 		/*
5280 		 * See if it's time to force a crash.
5281 		 */
5282 		if (now > zs->zs_thread_kill)
5283 			ztest_kill(zs);
5284 
5285 		/*
5286 		 * If we're getting ENOSPC with some regularity, stop.
5287 		 */
5288 		if (zs->zs_enospc_count > 10)
5289 			break;
5290 
5291 		/*
5292 		 * Pick a random function to execute.
5293 		 */
5294 		rand = ztest_random(ZTEST_FUNCS);
5295 		zi = &ztest_info[rand];
5296 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5297 		call_next = zc->zc_next;
5298 
5299 		if (now >= call_next &&
5300 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5301 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5302 			ztest_execute(rand, zi, id);
5303 		}
5304 	}
5305 
5306 	return (NULL);
5307 }
5308 
5309 static void
5310 ztest_dataset_name(char *dsname, char *pool, int d)
5311 {
5312 	(void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5313 }
5314 
5315 static void
5316 ztest_dataset_destroy(int d)
5317 {
5318 	char name[MAXNAMELEN];
5319 
5320 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5321 
5322 	if (ztest_opts.zo_verbose >= 3)
5323 		(void) printf("Destroying %s to free up space\n", name);
5324 
5325 	/*
5326 	 * Cleanup any non-standard clones and snapshots.  In general,
5327 	 * ztest thread t operates on dataset (t % zopt_datasets),
5328 	 * so there may be more than one thing to clean up.
5329 	 */
5330 	for (int t = d; t < ztest_opts.zo_threads;
5331 	    t += ztest_opts.zo_datasets) {
5332 		ztest_dsl_dataset_cleanup(name, t);
5333 	}
5334 
5335 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5336 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5337 }
5338 
5339 static void
5340 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5341 {
5342 	uint64_t usedobjs, dirobjs, scratch;
5343 
5344 	/*
5345 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5346 	 * Therefore, the number of objects in use should equal the
5347 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5348 	 * If not, we have an object leak.
5349 	 *
5350 	 * Note that we can only check this in ztest_dataset_open(),
5351 	 * when the open-context and syncing-context values agree.
5352 	 * That's because zap_count() returns the open-context value,
5353 	 * while dmu_objset_space() returns the rootbp fill count.
5354 	 */
5355 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5356 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5357 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5358 }
5359 
5360 static int
5361 ztest_dataset_open(int d)
5362 {
5363 	ztest_ds_t *zd = &ztest_ds[d];
5364 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5365 	objset_t *os;
5366 	zilog_t *zilog;
5367 	char name[MAXNAMELEN];
5368 	int error;
5369 
5370 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5371 
5372 	(void) rw_rdlock(&ztest_name_lock);
5373 
5374 	error = ztest_dataset_create(name);
5375 	if (error == ENOSPC) {
5376 		(void) rw_unlock(&ztest_name_lock);
5377 		ztest_record_enospc(FTAG);
5378 		return (error);
5379 	}
5380 	ASSERT(error == 0 || error == EEXIST);
5381 
5382 	VERIFY0(dmu_objset_hold(name, zd, &os));
5383 	(void) rw_unlock(&ztest_name_lock);
5384 
5385 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5386 
5387 	zilog = zd->zd_zilog;
5388 
5389 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5390 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5391 		fatal(0, "missing log records: claimed %llu < committed %llu",
5392 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5393 
5394 	ztest_dataset_dirobj_verify(zd);
5395 
5396 	zil_replay(os, zd, ztest_replay_vector);
5397 
5398 	ztest_dataset_dirobj_verify(zd);
5399 
5400 	if (ztest_opts.zo_verbose >= 6)
5401 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5402 		    zd->zd_name,
5403 		    (u_longlong_t)zilog->zl_parse_blk_count,
5404 		    (u_longlong_t)zilog->zl_parse_lr_count,
5405 		    (u_longlong_t)zilog->zl_replaying_seq);
5406 
5407 	zilog = zil_open(os, ztest_get_data);
5408 
5409 	if (zilog->zl_replaying_seq != 0 &&
5410 	    zilog->zl_replaying_seq < committed_seq)
5411 		fatal(0, "missing log records: replayed %llu < committed %llu",
5412 		    zilog->zl_replaying_seq, committed_seq);
5413 
5414 	return (0);
5415 }
5416 
5417 static void
5418 ztest_dataset_close(int d)
5419 {
5420 	ztest_ds_t *zd = &ztest_ds[d];
5421 
5422 	zil_close(zd->zd_zilog);
5423 	dmu_objset_rele(zd->zd_os, zd);
5424 
5425 	ztest_zd_fini(zd);
5426 }
5427 
5428 /*
5429  * Kick off threads to run tests on all datasets in parallel.
5430  */
5431 static void
5432 ztest_run(ztest_shared_t *zs)
5433 {
5434 	thread_t *tid;
5435 	spa_t *spa;
5436 	objset_t *os;
5437 	thread_t resume_tid;
5438 	int error;
5439 
5440 	ztest_exiting = B_FALSE;
5441 
5442 	/*
5443 	 * Initialize parent/child shared state.
5444 	 */
5445 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5446 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5447 
5448 	zs->zs_thread_start = gethrtime();
5449 	zs->zs_thread_stop =
5450 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5451 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5452 	zs->zs_thread_kill = zs->zs_thread_stop;
5453 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5454 		zs->zs_thread_kill -=
5455 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5456 	}
5457 
5458 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5459 
5460 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5461 	    offsetof(ztest_cb_data_t, zcd_node));
5462 
5463 	/*
5464 	 * Open our pool.
5465 	 */
5466 	kernel_init(FREAD | FWRITE);
5467 	VERIFY(spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0);
5468 	spa->spa_debug = B_TRUE;
5469 	ztest_spa = spa;
5470 
5471 	VERIFY3U(0, ==, dmu_objset_hold(ztest_opts.zo_pool, FTAG, &os));
5472 	zs->zs_guid = dmu_objset_fsid_guid(os);
5473 	dmu_objset_rele(os, FTAG);
5474 
5475 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5476 
5477 	/*
5478 	 * We don't expect the pool to suspend unless maxfaults == 0,
5479 	 * in which case ztest_fault_inject() temporarily takes away
5480 	 * the only valid replica.
5481 	 */
5482 	if (MAXFAULTS() == 0)
5483 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5484 	else
5485 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5486 
5487 	/*
5488 	 * Create a thread to periodically resume suspended I/O.
5489 	 */
5490 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5491 	    &resume_tid) == 0);
5492 
5493 	/*
5494 	 * Create a deadman thread to abort() if we hang.
5495 	 */
5496 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5497 	    NULL) == 0);
5498 
5499 	/*
5500 	 * Verify that we can safely inquire about about any object,
5501 	 * whether it's allocated or not.  To make it interesting,
5502 	 * we probe a 5-wide window around each power of two.
5503 	 * This hits all edge cases, including zero and the max.
5504 	 */
5505 	for (int t = 0; t < 64; t++) {
5506 		for (int d = -5; d <= 5; d++) {
5507 			error = dmu_object_info(spa->spa_meta_objset,
5508 			    (1ULL << t) + d, NULL);
5509 			ASSERT(error == 0 || error == ENOENT ||
5510 			    error == EINVAL);
5511 		}
5512 	}
5513 
5514 	/*
5515 	 * If we got any ENOSPC errors on the previous run, destroy something.
5516 	 */
5517 	if (zs->zs_enospc_count != 0) {
5518 		int d = ztest_random(ztest_opts.zo_datasets);
5519 		ztest_dataset_destroy(d);
5520 	}
5521 	zs->zs_enospc_count = 0;
5522 
5523 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5524 	    UMEM_NOFAIL);
5525 
5526 	if (ztest_opts.zo_verbose >= 4)
5527 		(void) printf("starting main threads...\n");
5528 
5529 	/*
5530 	 * Kick off all the tests that run in parallel.
5531 	 */
5532 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
5533 		if (t < ztest_opts.zo_datasets &&
5534 		    ztest_dataset_open(t) != 0)
5535 			return;
5536 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5537 		    THR_BOUND, &tid[t]) == 0);
5538 	}
5539 
5540 	/*
5541 	 * Wait for all of the tests to complete.  We go in reverse order
5542 	 * so we don't close datasets while threads are still using them.
5543 	 */
5544 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5545 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5546 		if (t < ztest_opts.zo_datasets)
5547 			ztest_dataset_close(t);
5548 	}
5549 
5550 	txg_wait_synced(spa_get_dsl(spa), 0);
5551 
5552 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5553 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5554 
5555 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5556 
5557 	/* Kill the resume thread */
5558 	ztest_exiting = B_TRUE;
5559 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5560 	ztest_resume(spa);
5561 
5562 	/*
5563 	 * Right before closing the pool, kick off a bunch of async I/O;
5564 	 * spa_close() should wait for it to complete.
5565 	 */
5566 	for (uint64_t object = 1; object < 50; object++)
5567 		dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5568 
5569 	spa_close(spa, FTAG);
5570 
5571 	/*
5572 	 * Verify that we can loop over all pools.
5573 	 */
5574 	mutex_enter(&spa_namespace_lock);
5575 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5576 		if (ztest_opts.zo_verbose > 3)
5577 			(void) printf("spa_next: found %s\n", spa_name(spa));
5578 	mutex_exit(&spa_namespace_lock);
5579 
5580 	/*
5581 	 * Verify that we can export the pool and reimport it under a
5582 	 * different name.
5583 	 */
5584 	if (ztest_random(2) == 0) {
5585 		char name[MAXNAMELEN];
5586 		(void) snprintf(name, MAXNAMELEN, "%s_import",
5587 		    ztest_opts.zo_pool);
5588 		ztest_spa_import_export(ztest_opts.zo_pool, name);
5589 		ztest_spa_import_export(name, ztest_opts.zo_pool);
5590 	}
5591 
5592 	kernel_fini();
5593 
5594 	list_destroy(&zcl.zcl_callbacks);
5595 
5596 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5597 
5598 	(void) rwlock_destroy(&ztest_name_lock);
5599 	(void) _mutex_destroy(&ztest_vdev_lock);
5600 }
5601 
5602 static void
5603 ztest_freeze(void)
5604 {
5605 	ztest_ds_t *zd = &ztest_ds[0];
5606 	spa_t *spa;
5607 	int numloops = 0;
5608 
5609 	if (ztest_opts.zo_verbose >= 3)
5610 		(void) printf("testing spa_freeze()...\n");
5611 
5612 	kernel_init(FREAD | FWRITE);
5613 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5614 	VERIFY3U(0, ==, ztest_dataset_open(0));
5615 
5616 	/*
5617 	 * Force the first log block to be transactionally allocated.
5618 	 * We have to do this before we freeze the pool -- otherwise
5619 	 * the log chain won't be anchored.
5620 	 */
5621 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5622 		ztest_dmu_object_alloc_free(zd, 0);
5623 		zil_commit(zd->zd_zilog, 0);
5624 	}
5625 
5626 	txg_wait_synced(spa_get_dsl(spa), 0);
5627 
5628 	/*
5629 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5630 	 * so that the only way to record changes from now on is the ZIL.
5631 	 */
5632 	spa_freeze(spa);
5633 
5634 	/*
5635 	 * Run tests that generate log records but don't alter the pool config
5636 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5637 	 * We do a txg_wait_synced() after each iteration to force the txg
5638 	 * to increase well beyond the last synced value in the uberblock.
5639 	 * The ZIL should be OK with that.
5640 	 */
5641 	while (ztest_random(10) != 0 &&
5642 	    numloops++ < ztest_opts.zo_maxloops) {
5643 		ztest_dmu_write_parallel(zd, 0);
5644 		ztest_dmu_object_alloc_free(zd, 0);
5645 		txg_wait_synced(spa_get_dsl(spa), 0);
5646 	}
5647 
5648 	/*
5649 	 * Commit all of the changes we just generated.
5650 	 */
5651 	zil_commit(zd->zd_zilog, 0);
5652 	txg_wait_synced(spa_get_dsl(spa), 0);
5653 
5654 	/*
5655 	 * Close our dataset and close the pool.
5656 	 */
5657 	ztest_dataset_close(0);
5658 	spa_close(spa, FTAG);
5659 	kernel_fini();
5660 
5661 	/*
5662 	 * Open and close the pool and dataset to induce log replay.
5663 	 */
5664 	kernel_init(FREAD | FWRITE);
5665 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5666 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5667 	VERIFY3U(0, ==, ztest_dataset_open(0));
5668 	ztest_dataset_close(0);
5669 
5670 	spa->spa_debug = B_TRUE;
5671 	ztest_spa = spa;
5672 	txg_wait_synced(spa_get_dsl(spa), 0);
5673 	ztest_reguid(NULL, 0);
5674 
5675 	spa_close(spa, FTAG);
5676 	kernel_fini();
5677 }
5678 
5679 void
5680 print_time(hrtime_t t, char *timebuf)
5681 {
5682 	hrtime_t s = t / NANOSEC;
5683 	hrtime_t m = s / 60;
5684 	hrtime_t h = m / 60;
5685 	hrtime_t d = h / 24;
5686 
5687 	s -= m * 60;
5688 	m -= h * 60;
5689 	h -= d * 24;
5690 
5691 	timebuf[0] = '\0';
5692 
5693 	if (d)
5694 		(void) sprintf(timebuf,
5695 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5696 	else if (h)
5697 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5698 	else if (m)
5699 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5700 	else
5701 		(void) sprintf(timebuf, "%llus", s);
5702 }
5703 
5704 static nvlist_t *
5705 make_random_props()
5706 {
5707 	nvlist_t *props;
5708 
5709 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5710 	if (ztest_random(2) == 0)
5711 		return (props);
5712 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5713 
5714 	return (props);
5715 }
5716 
5717 /*
5718  * Create a storage pool with the given name and initial vdev size.
5719  * Then test spa_freeze() functionality.
5720  */
5721 static void
5722 ztest_init(ztest_shared_t *zs)
5723 {
5724 	spa_t *spa;
5725 	nvlist_t *nvroot, *props;
5726 
5727 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5728 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5729 
5730 	kernel_init(FREAD | FWRITE);
5731 
5732 	/*
5733 	 * Create the storage pool.
5734 	 */
5735 	(void) spa_destroy(ztest_opts.zo_pool);
5736 	ztest_shared->zs_vdev_next_leaf = 0;
5737 	zs->zs_splits = 0;
5738 	zs->zs_mirrors = ztest_opts.zo_mirrors;
5739 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5740 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5741 	props = make_random_props();
5742 	for (int i = 0; i < SPA_FEATURES; i++) {
5743 		char buf[1024];
5744 		(void) snprintf(buf, sizeof (buf), "feature@%s",
5745 		    spa_feature_table[i].fi_uname);
5746 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5747 	}
5748 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5749 	nvlist_free(nvroot);
5750 
5751 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5752 	zs->zs_metaslab_sz =
5753 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5754 
5755 	spa_close(spa, FTAG);
5756 
5757 	kernel_fini();
5758 
5759 	ztest_run_zdb(ztest_opts.zo_pool);
5760 
5761 	ztest_freeze();
5762 
5763 	ztest_run_zdb(ztest_opts.zo_pool);
5764 
5765 	(void) rwlock_destroy(&ztest_name_lock);
5766 	(void) _mutex_destroy(&ztest_vdev_lock);
5767 }
5768 
5769 static void
5770 setup_data_fd(void)
5771 {
5772 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5773 
5774 	ztest_fd_data = mkstemp(ztest_name_data);
5775 	ASSERT3S(ztest_fd_data, >=, 0);
5776 	(void) unlink(ztest_name_data);
5777 }
5778 
5779 
5780 static int
5781 shared_data_size(ztest_shared_hdr_t *hdr)
5782 {
5783 	int size;
5784 
5785 	size = hdr->zh_hdr_size;
5786 	size += hdr->zh_opts_size;
5787 	size += hdr->zh_size;
5788 	size += hdr->zh_stats_size * hdr->zh_stats_count;
5789 	size += hdr->zh_ds_size * hdr->zh_ds_count;
5790 
5791 	return (size);
5792 }
5793 
5794 static void
5795 setup_hdr(void)
5796 {
5797 	int size;
5798 	ztest_shared_hdr_t *hdr;
5799 
5800 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5801 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5802 	ASSERT(hdr != MAP_FAILED);
5803 
5804 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5805 
5806 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5807 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5808 	hdr->zh_size = sizeof (ztest_shared_t);
5809 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5810 	hdr->zh_stats_count = ZTEST_FUNCS;
5811 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5812 	hdr->zh_ds_count = ztest_opts.zo_datasets;
5813 
5814 	size = shared_data_size(hdr);
5815 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5816 
5817 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5818 }
5819 
5820 static void
5821 setup_data(void)
5822 {
5823 	int size, offset;
5824 	ztest_shared_hdr_t *hdr;
5825 	uint8_t *buf;
5826 
5827 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5828 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5829 	ASSERT(hdr != MAP_FAILED);
5830 
5831 	size = shared_data_size(hdr);
5832 
5833 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5834 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5835 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5836 	ASSERT(hdr != MAP_FAILED);
5837 	buf = (uint8_t *)hdr;
5838 
5839 	offset = hdr->zh_hdr_size;
5840 	ztest_shared_opts = (void *)&buf[offset];
5841 	offset += hdr->zh_opts_size;
5842 	ztest_shared = (void *)&buf[offset];
5843 	offset += hdr->zh_size;
5844 	ztest_shared_callstate = (void *)&buf[offset];
5845 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
5846 	ztest_shared_ds = (void *)&buf[offset];
5847 }
5848 
5849 static boolean_t
5850 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5851 {
5852 	pid_t pid;
5853 	int status;
5854 	char *cmdbuf = NULL;
5855 
5856 	pid = fork();
5857 
5858 	if (cmd == NULL) {
5859 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5860 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
5861 		cmd = cmdbuf;
5862 	}
5863 
5864 	if (pid == -1)
5865 		fatal(1, "fork failed");
5866 
5867 	if (pid == 0) {	/* child */
5868 		char *emptyargv[2] = { cmd, NULL };
5869 		char fd_data_str[12];
5870 
5871 		struct rlimit rl = { 1024, 1024 };
5872 		(void) setrlimit(RLIMIT_NOFILE, &rl);
5873 
5874 		(void) close(ztest_fd_rand);
5875 		VERIFY3U(11, >=,
5876 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
5877 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
5878 
5879 		(void) enable_extended_FILE_stdio(-1, -1);
5880 		if (libpath != NULL)
5881 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5882 		(void) execv(cmd, emptyargv);
5883 		ztest_dump_core = B_FALSE;
5884 		fatal(B_TRUE, "exec failed: %s", cmd);
5885 	}
5886 
5887 	if (cmdbuf != NULL) {
5888 		umem_free(cmdbuf, MAXPATHLEN);
5889 		cmd = NULL;
5890 	}
5891 
5892 	while (waitpid(pid, &status, 0) != pid)
5893 		continue;
5894 	if (statusp != NULL)
5895 		*statusp = status;
5896 
5897 	if (WIFEXITED(status)) {
5898 		if (WEXITSTATUS(status) != 0) {
5899 			(void) fprintf(stderr, "child exited with code %d\n",
5900 			    WEXITSTATUS(status));
5901 			exit(2);
5902 		}
5903 		return (B_FALSE);
5904 	} else if (WIFSIGNALED(status)) {
5905 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
5906 			(void) fprintf(stderr, "child died with signal %d\n",
5907 			    WTERMSIG(status));
5908 			exit(3);
5909 		}
5910 		return (B_TRUE);
5911 	} else {
5912 		(void) fprintf(stderr, "something strange happened to child\n");
5913 		exit(4);
5914 		/* NOTREACHED */
5915 	}
5916 }
5917 
5918 static void
5919 ztest_run_init(void)
5920 {
5921 	ztest_shared_t *zs = ztest_shared;
5922 
5923 	ASSERT(ztest_opts.zo_init != 0);
5924 
5925 	/*
5926 	 * Blow away any existing copy of zpool.cache
5927 	 */
5928 	(void) remove(spa_config_path);
5929 
5930 	/*
5931 	 * Create and initialize our storage pool.
5932 	 */
5933 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
5934 		bzero(zs, sizeof (ztest_shared_t));
5935 		if (ztest_opts.zo_verbose >= 3 &&
5936 		    ztest_opts.zo_init != 1) {
5937 			(void) printf("ztest_init(), pass %d\n", i);
5938 		}
5939 		ztest_init(zs);
5940 	}
5941 }
5942 
5943 int
5944 main(int argc, char **argv)
5945 {
5946 	int kills = 0;
5947 	int iters = 0;
5948 	int older = 0;
5949 	int newer = 0;
5950 	ztest_shared_t *zs;
5951 	ztest_info_t *zi;
5952 	ztest_shared_callstate_t *zc;
5953 	char timebuf[100];
5954 	char numbuf[6];
5955 	spa_t *spa;
5956 	char *cmd;
5957 	boolean_t hasalt;
5958 	char *fd_data_str = getenv("ZTEST_FD_DATA");
5959 
5960 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
5961 
5962 	dprintf_setup(&argc, argv);
5963 
5964 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
5965 	ASSERT3S(ztest_fd_rand, >=, 0);
5966 
5967 	if (!fd_data_str) {
5968 		process_options(argc, argv);
5969 
5970 		setup_data_fd();
5971 		setup_hdr();
5972 		setup_data();
5973 		bcopy(&ztest_opts, ztest_shared_opts,
5974 		    sizeof (*ztest_shared_opts));
5975 	} else {
5976 		ztest_fd_data = atoi(fd_data_str);
5977 		setup_data();
5978 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
5979 	}
5980 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
5981 
5982 	/* Override location of zpool.cache */
5983 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
5984 	    ztest_opts.zo_dir), !=, -1);
5985 
5986 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
5987 	    UMEM_NOFAIL);
5988 	zs = ztest_shared;
5989 
5990 	if (fd_data_str) {
5991 		metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
5992 		metaslab_df_alloc_threshold =
5993 		    zs->zs_metaslab_df_alloc_threshold;
5994 
5995 		if (zs->zs_do_init)
5996 			ztest_run_init();
5997 		else
5998 			ztest_run(zs);
5999 		exit(0);
6000 	}
6001 
6002 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6003 
6004 	if (ztest_opts.zo_verbose >= 1) {
6005 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6006 		    " %llu seconds...\n",
6007 		    (u_longlong_t)ztest_opts.zo_vdevs,
6008 		    ztest_opts.zo_datasets,
6009 		    ztest_opts.zo_threads,
6010 		    (u_longlong_t)ztest_opts.zo_time);
6011 	}
6012 
6013 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6014 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6015 
6016 	zs->zs_do_init = B_TRUE;
6017 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6018 		if (ztest_opts.zo_verbose >= 1) {
6019 			(void) printf("Executing older ztest for "
6020 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6021 		}
6022 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6023 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6024 	} else {
6025 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6026 	}
6027 	zs->zs_do_init = B_FALSE;
6028 
6029 	zs->zs_proc_start = gethrtime();
6030 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6031 
6032 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6033 		zi = &ztest_info[f];
6034 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6035 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6036 			zc->zc_next = UINT64_MAX;
6037 		else
6038 			zc->zc_next = zs->zs_proc_start +
6039 			    ztest_random(2 * zi->zi_interval[0] + 1);
6040 	}
6041 
6042 	/*
6043 	 * Run the tests in a loop.  These tests include fault injection
6044 	 * to verify that self-healing data works, and forced crashes
6045 	 * to verify that we never lose on-disk consistency.
6046 	 */
6047 	while (gethrtime() < zs->zs_proc_stop) {
6048 		int status;
6049 		boolean_t killed;
6050 
6051 		/*
6052 		 * Initialize the workload counters for each function.
6053 		 */
6054 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6055 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6056 			zc->zc_count = 0;
6057 			zc->zc_time = 0;
6058 		}
6059 
6060 		/* Set the allocation switch size */
6061 		zs->zs_metaslab_df_alloc_threshold =
6062 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6063 
6064 		if (!hasalt || ztest_random(2) == 0) {
6065 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6066 				(void) printf("Executing newer ztest: %s\n",
6067 				    cmd);
6068 			}
6069 			newer++;
6070 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6071 		} else {
6072 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6073 				(void) printf("Executing older ztest: %s\n",
6074 				    ztest_opts.zo_alt_ztest);
6075 			}
6076 			older++;
6077 			killed = exec_child(ztest_opts.zo_alt_ztest,
6078 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6079 		}
6080 
6081 		if (killed)
6082 			kills++;
6083 		iters++;
6084 
6085 		if (ztest_opts.zo_verbose >= 1) {
6086 			hrtime_t now = gethrtime();
6087 
6088 			now = MIN(now, zs->zs_proc_stop);
6089 			print_time(zs->zs_proc_stop - now, timebuf);
6090 			nicenum(zs->zs_space, numbuf);
6091 
6092 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6093 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6094 			    iters,
6095 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6096 			    (u_longlong_t)zs->zs_enospc_count,
6097 			    100.0 * zs->zs_alloc / zs->zs_space,
6098 			    numbuf,
6099 			    100.0 * (now - zs->zs_proc_start) /
6100 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6101 		}
6102 
6103 		if (ztest_opts.zo_verbose >= 2) {
6104 			(void) printf("\nWorkload summary:\n\n");
6105 			(void) printf("%7s %9s   %s\n",
6106 			    "Calls", "Time", "Function");
6107 			(void) printf("%7s %9s   %s\n",
6108 			    "-----", "----", "--------");
6109 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6110 				Dl_info dli;
6111 
6112 				zi = &ztest_info[f];
6113 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6114 				print_time(zc->zc_time, timebuf);
6115 				(void) dladdr((void *)zi->zi_func, &dli);
6116 				(void) printf("%7llu %9s   %s\n",
6117 				    (u_longlong_t)zc->zc_count, timebuf,
6118 				    dli.dli_sname);
6119 			}
6120 			(void) printf("\n");
6121 		}
6122 
6123 		/*
6124 		 * It's possible that we killed a child during a rename test,
6125 		 * in which case we'll have a 'ztest_tmp' pool lying around
6126 		 * instead of 'ztest'.  Do a blind rename in case this happened.
6127 		 */
6128 		kernel_init(FREAD);
6129 		if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6130 			spa_close(spa, FTAG);
6131 		} else {
6132 			char tmpname[MAXNAMELEN];
6133 			kernel_fini();
6134 			kernel_init(FREAD | FWRITE);
6135 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6136 			    ztest_opts.zo_pool);
6137 			(void) spa_rename(tmpname, ztest_opts.zo_pool);
6138 		}
6139 		kernel_fini();
6140 
6141 		ztest_run_zdb(ztest_opts.zo_pool);
6142 	}
6143 
6144 	if (ztest_opts.zo_verbose >= 1) {
6145 		if (hasalt) {
6146 			(void) printf("%d runs of older ztest: %s\n", older,
6147 			    ztest_opts.zo_alt_ztest);
6148 			(void) printf("%d runs of newer ztest: %s\n", newer,
6149 			    cmd);
6150 		}
6151 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6152 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6153 	}
6154 
6155 	umem_free(cmd, MAXNAMELEN);
6156 
6157 	return (0);
6158 }
6159