xref: /titanic_41/usr/src/cmd/ztest/ztest.c (revision 6774dc07ebd5a9d69956f033d5ac5a68803bd66b)
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  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  */
27 
28 /*
29  * The objective of this program is to provide a DMU/ZAP/SPA stress test
30  * that runs entirely in userland, is easy to use, and easy to extend.
31  *
32  * The overall design of the ztest program is as follows:
33  *
34  * (1) For each major functional area (e.g. adding vdevs to a pool,
35  *     creating and destroying datasets, reading and writing objects, etc)
36  *     we have a simple routine to test that functionality.  These
37  *     individual routines do not have to do anything "stressful".
38  *
39  * (2) We turn these simple functionality tests into a stress test by
40  *     running them all in parallel, with as many threads as desired,
41  *     and spread across as many datasets, objects, and vdevs as desired.
42  *
43  * (3) While all this is happening, we inject faults into the pool to
44  *     verify that self-healing data really works.
45  *
46  * (4) Every time we open a dataset, we change its checksum and compression
47  *     functions.  Thus even individual objects vary from block to block
48  *     in which checksum they use and whether they're compressed.
49  *
50  * (5) To verify that we never lose on-disk consistency after a crash,
51  *     we run the entire test in a child of the main process.
52  *     At random times, the child self-immolates with a SIGKILL.
53  *     This is the software equivalent of pulling the power cord.
54  *     The parent then runs the test again, using the existing
55  *     storage pool, as many times as desired. If backwards compatability
56  *     testing is enabled ztest will sometimes run the "older" version
57  *     of ztest after a SIGKILL.
58  *
59  * (6) To verify that we don't have future leaks or temporal incursions,
60  *     many of the functional tests record the transaction group number
61  *     as part of their data.  When reading old data, they verify that
62  *     the transaction group number is less than the current, open txg.
63  *     If you add a new test, please do this if applicable.
64  *
65  * When run with no arguments, ztest runs for about five minutes and
66  * produces no output if successful.  To get a little bit of information,
67  * specify -V.  To get more information, specify -VV, and so on.
68  *
69  * To turn this into an overnight stress test, use -T to specify run time.
70  *
71  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
72  * to increase the pool capacity, fanout, and overall stress level.
73  *
74  * Use the -k option to set the desired frequency of kills.
75  *
76  * When ztest invokes itself it passes all relevant information through a
77  * temporary file which is mmap-ed in the child process. This allows shared
78  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
79  * stored at offset 0 of this file and contains information on the size and
80  * number of shared structures in the file. The information stored in this file
81  * must remain backwards compatible with older versions of ztest so that
82  * ztest can invoke them during backwards compatibility testing (-B).
83  */
84 
85 #include <sys/zfs_context.h>
86 #include <sys/spa.h>
87 #include <sys/dmu.h>
88 #include <sys/txg.h>
89 #include <sys/dbuf.h>
90 #include <sys/zap.h>
91 #include <sys/dmu_objset.h>
92 #include <sys/poll.h>
93 #include <sys/stat.h>
94 #include <sys/time.h>
95 #include <sys/wait.h>
96 #include <sys/mman.h>
97 #include <sys/resource.h>
98 #include <sys/zio.h>
99 #include <sys/zil.h>
100 #include <sys/zil_impl.h>
101 #include <sys/vdev_impl.h>
102 #include <sys/vdev_file.h>
103 #include <sys/spa_impl.h>
104 #include <sys/metaslab_impl.h>
105 #include <sys/dsl_prop.h>
106 #include <sys/dsl_dataset.h>
107 #include <sys/dsl_destroy.h>
108 #include <sys/dsl_scan.h>
109 #include <sys/zio_checksum.h>
110 #include <sys/refcount.h>
111 #include <sys/zfeature.h>
112 #include <sys/dsl_userhold.h>
113 #include <stdio.h>
114 #include <stdio_ext.h>
115 #include <stdlib.h>
116 #include <unistd.h>
117 #include <signal.h>
118 #include <umem.h>
119 #include <dlfcn.h>
120 #include <ctype.h>
121 #include <math.h>
122 #include <sys/fs/zfs.h>
123 #include <libnvpair.h>
124 
125 static int ztest_fd_data = -1;
126 static int ztest_fd_rand = -1;
127 
128 typedef struct ztest_shared_hdr {
129 	uint64_t	zh_hdr_size;
130 	uint64_t	zh_opts_size;
131 	uint64_t	zh_size;
132 	uint64_t	zh_stats_size;
133 	uint64_t	zh_stats_count;
134 	uint64_t	zh_ds_size;
135 	uint64_t	zh_ds_count;
136 } ztest_shared_hdr_t;
137 
138 static ztest_shared_hdr_t *ztest_shared_hdr;
139 
140 typedef struct ztest_shared_opts {
141 	char zo_pool[MAXNAMELEN];
142 	char zo_dir[MAXNAMELEN];
143 	char zo_alt_ztest[MAXNAMELEN];
144 	char zo_alt_libpath[MAXNAMELEN];
145 	uint64_t zo_vdevs;
146 	uint64_t zo_vdevtime;
147 	size_t zo_vdev_size;
148 	int zo_ashift;
149 	int zo_mirrors;
150 	int zo_raidz;
151 	int zo_raidz_parity;
152 	int zo_datasets;
153 	int zo_threads;
154 	uint64_t zo_passtime;
155 	uint64_t zo_killrate;
156 	int zo_verbose;
157 	int zo_init;
158 	uint64_t zo_time;
159 	uint64_t zo_maxloops;
160 	uint64_t zo_metaslab_gang_bang;
161 } ztest_shared_opts_t;
162 
163 static const ztest_shared_opts_t ztest_opts_defaults = {
164 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
165 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
166 	.zo_alt_ztest = { '\0' },
167 	.zo_alt_libpath = { '\0' },
168 	.zo_vdevs = 5,
169 	.zo_ashift = SPA_MINBLOCKSHIFT,
170 	.zo_mirrors = 2,
171 	.zo_raidz = 4,
172 	.zo_raidz_parity = 1,
173 	.zo_vdev_size = SPA_MINDEVSIZE,
174 	.zo_datasets = 7,
175 	.zo_threads = 23,
176 	.zo_passtime = 60,		/* 60 seconds */
177 	.zo_killrate = 70,		/* 70% kill rate */
178 	.zo_verbose = 0,
179 	.zo_init = 1,
180 	.zo_time = 300,			/* 5 minutes */
181 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
182 	.zo_metaslab_gang_bang = 32 << 10
183 };
184 
185 extern uint64_t metaslab_gang_bang;
186 extern uint64_t metaslab_df_alloc_threshold;
187 
188 static ztest_shared_opts_t *ztest_shared_opts;
189 static ztest_shared_opts_t ztest_opts;
190 
191 typedef struct ztest_shared_ds {
192 	uint64_t	zd_seq;
193 } ztest_shared_ds_t;
194 
195 static ztest_shared_ds_t *ztest_shared_ds;
196 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
197 
198 #define	BT_MAGIC	0x123456789abcdefULL
199 #define	MAXFAULTS() \
200 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
201 
202 enum ztest_io_type {
203 	ZTEST_IO_WRITE_TAG,
204 	ZTEST_IO_WRITE_PATTERN,
205 	ZTEST_IO_WRITE_ZEROES,
206 	ZTEST_IO_TRUNCATE,
207 	ZTEST_IO_SETATTR,
208 	ZTEST_IO_REWRITE,
209 	ZTEST_IO_TYPES
210 };
211 
212 typedef struct ztest_block_tag {
213 	uint64_t	bt_magic;
214 	uint64_t	bt_objset;
215 	uint64_t	bt_object;
216 	uint64_t	bt_offset;
217 	uint64_t	bt_gen;
218 	uint64_t	bt_txg;
219 	uint64_t	bt_crtxg;
220 } ztest_block_tag_t;
221 
222 typedef struct bufwad {
223 	uint64_t	bw_index;
224 	uint64_t	bw_txg;
225 	uint64_t	bw_data;
226 } bufwad_t;
227 
228 /*
229  * XXX -- fix zfs range locks to be generic so we can use them here.
230  */
231 typedef enum {
232 	RL_READER,
233 	RL_WRITER,
234 	RL_APPEND
235 } rl_type_t;
236 
237 typedef struct rll {
238 	void		*rll_writer;
239 	int		rll_readers;
240 	mutex_t		rll_lock;
241 	cond_t		rll_cv;
242 } rll_t;
243 
244 typedef struct rl {
245 	uint64_t	rl_object;
246 	uint64_t	rl_offset;
247 	uint64_t	rl_size;
248 	rll_t		*rl_lock;
249 } rl_t;
250 
251 #define	ZTEST_RANGE_LOCKS	64
252 #define	ZTEST_OBJECT_LOCKS	64
253 
254 /*
255  * Object descriptor.  Used as a template for object lookup/create/remove.
256  */
257 typedef struct ztest_od {
258 	uint64_t	od_dir;
259 	uint64_t	od_object;
260 	dmu_object_type_t od_type;
261 	dmu_object_type_t od_crtype;
262 	uint64_t	od_blocksize;
263 	uint64_t	od_crblocksize;
264 	uint64_t	od_gen;
265 	uint64_t	od_crgen;
266 	char		od_name[MAXNAMELEN];
267 } ztest_od_t;
268 
269 /*
270  * Per-dataset state.
271  */
272 typedef struct ztest_ds {
273 	ztest_shared_ds_t *zd_shared;
274 	objset_t	*zd_os;
275 	rwlock_t	zd_zilog_lock;
276 	zilog_t		*zd_zilog;
277 	ztest_od_t	*zd_od;		/* debugging aid */
278 	char		zd_name[MAXNAMELEN];
279 	mutex_t		zd_dirobj_lock;
280 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
281 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
282 } ztest_ds_t;
283 
284 /*
285  * Per-iteration state.
286  */
287 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
288 
289 typedef struct ztest_info {
290 	ztest_func_t	*zi_func;	/* test function */
291 	uint64_t	zi_iters;	/* iterations per execution */
292 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
293 } ztest_info_t;
294 
295 typedef struct ztest_shared_callstate {
296 	uint64_t	zc_count;	/* per-pass count */
297 	uint64_t	zc_time;	/* per-pass time */
298 	uint64_t	zc_next;	/* next time to call this function */
299 } ztest_shared_callstate_t;
300 
301 static ztest_shared_callstate_t *ztest_shared_callstate;
302 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
303 
304 /*
305  * Note: these aren't static because we want dladdr() to work.
306  */
307 ztest_func_t ztest_dmu_read_write;
308 ztest_func_t ztest_dmu_write_parallel;
309 ztest_func_t ztest_dmu_object_alloc_free;
310 ztest_func_t ztest_dmu_commit_callbacks;
311 ztest_func_t ztest_zap;
312 ztest_func_t ztest_zap_parallel;
313 ztest_func_t ztest_zil_commit;
314 ztest_func_t ztest_zil_remount;
315 ztest_func_t ztest_dmu_read_write_zcopy;
316 ztest_func_t ztest_dmu_objset_create_destroy;
317 ztest_func_t ztest_dmu_prealloc;
318 ztest_func_t ztest_fzap;
319 ztest_func_t ztest_dmu_snapshot_create_destroy;
320 ztest_func_t ztest_dsl_prop_get_set;
321 ztest_func_t ztest_spa_prop_get_set;
322 ztest_func_t ztest_spa_create_destroy;
323 ztest_func_t ztest_fault_inject;
324 ztest_func_t ztest_ddt_repair;
325 ztest_func_t ztest_dmu_snapshot_hold;
326 ztest_func_t ztest_spa_rename;
327 ztest_func_t ztest_scrub;
328 ztest_func_t ztest_dsl_dataset_promote_busy;
329 ztest_func_t ztest_vdev_attach_detach;
330 ztest_func_t ztest_vdev_LUN_growth;
331 ztest_func_t ztest_vdev_add_remove;
332 ztest_func_t ztest_vdev_aux_add_remove;
333 ztest_func_t ztest_split_pool;
334 ztest_func_t ztest_reguid;
335 ztest_func_t ztest_spa_upgrade;
336 
337 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
338 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
339 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
340 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
341 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
342 
343 ztest_info_t ztest_info[] = {
344 	{ ztest_dmu_read_write,			1,	&zopt_always	},
345 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
346 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
347 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
348 	{ ztest_zap,				30,	&zopt_always	},
349 	{ ztest_zap_parallel,			100,	&zopt_always	},
350 	{ ztest_split_pool,			1,	&zopt_always	},
351 	{ ztest_zil_commit,			1,	&zopt_incessant	},
352 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
353 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
354 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
355 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
356 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
357 #if 0
358 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
359 #endif
360 	{ ztest_fzap,				1,	&zopt_sometimes	},
361 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
362 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
363 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
364 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
365 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
366 	{ ztest_reguid,				1,	&zopt_sometimes },
367 	{ ztest_spa_rename,			1,	&zopt_rarely	},
368 	{ ztest_scrub,				1,	&zopt_rarely	},
369 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
370 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
371 	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
372 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
373 	{ ztest_vdev_add_remove,		1,
374 	    &ztest_opts.zo_vdevtime				},
375 	{ ztest_vdev_aux_add_remove,		1,
376 	    &ztest_opts.zo_vdevtime				},
377 };
378 
379 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
380 
381 /*
382  * The following struct is used to hold a list of uncalled commit callbacks.
383  * The callbacks are ordered by txg number.
384  */
385 typedef struct ztest_cb_list {
386 	mutex_t	zcl_callbacks_lock;
387 	list_t	zcl_callbacks;
388 } ztest_cb_list_t;
389 
390 /*
391  * Stuff we need to share writably between parent and child.
392  */
393 typedef struct ztest_shared {
394 	boolean_t	zs_do_init;
395 	hrtime_t	zs_proc_start;
396 	hrtime_t	zs_proc_stop;
397 	hrtime_t	zs_thread_start;
398 	hrtime_t	zs_thread_stop;
399 	hrtime_t	zs_thread_kill;
400 	uint64_t	zs_enospc_count;
401 	uint64_t	zs_vdev_next_leaf;
402 	uint64_t	zs_vdev_aux;
403 	uint64_t	zs_alloc;
404 	uint64_t	zs_space;
405 	uint64_t	zs_splits;
406 	uint64_t	zs_mirrors;
407 	uint64_t	zs_metaslab_sz;
408 	uint64_t	zs_metaslab_df_alloc_threshold;
409 	uint64_t	zs_guid;
410 } ztest_shared_t;
411 
412 #define	ID_PARALLEL	-1ULL
413 
414 static char ztest_dev_template[] = "%s/%s.%llua";
415 static char ztest_aux_template[] = "%s/%s.%s.%llu";
416 ztest_shared_t *ztest_shared;
417 
418 static spa_t *ztest_spa = NULL;
419 static ztest_ds_t *ztest_ds;
420 
421 static mutex_t ztest_vdev_lock;
422 
423 /*
424  * The ztest_name_lock protects the pool and dataset namespace used by
425  * the individual tests. To modify the namespace, consumers must grab
426  * this lock as writer. Grabbing the lock as reader will ensure that the
427  * namespace does not change while the lock is held.
428  */
429 static rwlock_t ztest_name_lock;
430 
431 static boolean_t ztest_dump_core = B_TRUE;
432 static boolean_t ztest_exiting;
433 
434 /* Global commit callback list */
435 static ztest_cb_list_t zcl;
436 
437 enum ztest_object {
438 	ZTEST_META_DNODE = 0,
439 	ZTEST_DIROBJ,
440 	ZTEST_OBJECTS
441 };
442 
443 static void usage(boolean_t) __NORETURN;
444 
445 /*
446  * These libumem hooks provide a reasonable set of defaults for the allocator's
447  * debugging facilities.
448  */
449 const char *
450 _umem_debug_init()
451 {
452 	return ("default,verbose"); /* $UMEM_DEBUG setting */
453 }
454 
455 const char *
456 _umem_logging_init(void)
457 {
458 	return ("fail,contents"); /* $UMEM_LOGGING setting */
459 }
460 
461 #define	FATAL_MSG_SZ	1024
462 
463 char *fatal_msg;
464 
465 static void
466 fatal(int do_perror, char *message, ...)
467 {
468 	va_list args;
469 	int save_errno = errno;
470 	char buf[FATAL_MSG_SZ];
471 
472 	(void) fflush(stdout);
473 
474 	va_start(args, message);
475 	(void) sprintf(buf, "ztest: ");
476 	/* LINTED */
477 	(void) vsprintf(buf + strlen(buf), message, args);
478 	va_end(args);
479 	if (do_perror) {
480 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
481 		    ": %s", strerror(save_errno));
482 	}
483 	(void) fprintf(stderr, "%s\n", buf);
484 	fatal_msg = buf;			/* to ease debugging */
485 	if (ztest_dump_core)
486 		abort();
487 	exit(3);
488 }
489 
490 static int
491 str2shift(const char *buf)
492 {
493 	const char *ends = "BKMGTPEZ";
494 	int i;
495 
496 	if (buf[0] == '\0')
497 		return (0);
498 	for (i = 0; i < strlen(ends); i++) {
499 		if (toupper(buf[0]) == ends[i])
500 			break;
501 	}
502 	if (i == strlen(ends)) {
503 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
504 		    buf);
505 		usage(B_FALSE);
506 	}
507 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
508 		return (10*i);
509 	}
510 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
511 	usage(B_FALSE);
512 	/* NOTREACHED */
513 }
514 
515 static uint64_t
516 nicenumtoull(const char *buf)
517 {
518 	char *end;
519 	uint64_t val;
520 
521 	val = strtoull(buf, &end, 0);
522 	if (end == buf) {
523 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
524 		usage(B_FALSE);
525 	} else if (end[0] == '.') {
526 		double fval = strtod(buf, &end);
527 		fval *= pow(2, str2shift(end));
528 		if (fval > UINT64_MAX) {
529 			(void) fprintf(stderr, "ztest: value too large: %s\n",
530 			    buf);
531 			usage(B_FALSE);
532 		}
533 		val = (uint64_t)fval;
534 	} else {
535 		int shift = str2shift(end);
536 		if (shift >= 64 || (val << shift) >> shift != val) {
537 			(void) fprintf(stderr, "ztest: value too large: %s\n",
538 			    buf);
539 			usage(B_FALSE);
540 		}
541 		val <<= shift;
542 	}
543 	return (val);
544 }
545 
546 static void
547 usage(boolean_t requested)
548 {
549 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
550 
551 	char nice_vdev_size[10];
552 	char nice_gang_bang[10];
553 	FILE *fp = requested ? stdout : stderr;
554 
555 	nicenum(zo->zo_vdev_size, nice_vdev_size);
556 	nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
557 
558 	(void) fprintf(fp, "Usage: %s\n"
559 	    "\t[-v vdevs (default: %llu)]\n"
560 	    "\t[-s size_of_each_vdev (default: %s)]\n"
561 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
562 	    "\t[-m mirror_copies (default: %d)]\n"
563 	    "\t[-r raidz_disks (default: %d)]\n"
564 	    "\t[-R raidz_parity (default: %d)]\n"
565 	    "\t[-d datasets (default: %d)]\n"
566 	    "\t[-t threads (default: %d)]\n"
567 	    "\t[-g gang_block_threshold (default: %s)]\n"
568 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
569 	    "\t[-k kill_percentage (default: %llu%%)]\n"
570 	    "\t[-p pool_name (default: %s)]\n"
571 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
572 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
573 	    "\t[-E] use existing pool instead of creating new one\n"
574 	    "\t[-T time (default: %llu sec)] total run time\n"
575 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
576 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
577 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
578 	    "\t[-h] (print help)\n"
579 	    "",
580 	    zo->zo_pool,
581 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
582 	    nice_vdev_size,				/* -s */
583 	    zo->zo_ashift,				/* -a */
584 	    zo->zo_mirrors,				/* -m */
585 	    zo->zo_raidz,				/* -r */
586 	    zo->zo_raidz_parity,			/* -R */
587 	    zo->zo_datasets,				/* -d */
588 	    zo->zo_threads,				/* -t */
589 	    nice_gang_bang,				/* -g */
590 	    zo->zo_init,				/* -i */
591 	    (u_longlong_t)zo->zo_killrate,		/* -k */
592 	    zo->zo_pool,				/* -p */
593 	    zo->zo_dir,					/* -f */
594 	    (u_longlong_t)zo->zo_time,			/* -T */
595 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
596 	    (u_longlong_t)zo->zo_passtime);
597 	exit(requested ? 0 : 1);
598 }
599 
600 static void
601 process_options(int argc, char **argv)
602 {
603 	char *path;
604 	ztest_shared_opts_t *zo = &ztest_opts;
605 
606 	int opt;
607 	uint64_t value;
608 	char altdir[MAXNAMELEN] = { 0 };
609 
610 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
611 
612 	while ((opt = getopt(argc, argv,
613 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
614 		value = 0;
615 		switch (opt) {
616 		case 'v':
617 		case 's':
618 		case 'a':
619 		case 'm':
620 		case 'r':
621 		case 'R':
622 		case 'd':
623 		case 't':
624 		case 'g':
625 		case 'i':
626 		case 'k':
627 		case 'T':
628 		case 'P':
629 		case 'F':
630 			value = nicenumtoull(optarg);
631 		}
632 		switch (opt) {
633 		case 'v':
634 			zo->zo_vdevs = value;
635 			break;
636 		case 's':
637 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
638 			break;
639 		case 'a':
640 			zo->zo_ashift = value;
641 			break;
642 		case 'm':
643 			zo->zo_mirrors = value;
644 			break;
645 		case 'r':
646 			zo->zo_raidz = MAX(1, value);
647 			break;
648 		case 'R':
649 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
650 			break;
651 		case 'd':
652 			zo->zo_datasets = MAX(1, value);
653 			break;
654 		case 't':
655 			zo->zo_threads = MAX(1, value);
656 			break;
657 		case 'g':
658 			zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
659 			    value);
660 			break;
661 		case 'i':
662 			zo->zo_init = value;
663 			break;
664 		case 'k':
665 			zo->zo_killrate = value;
666 			break;
667 		case 'p':
668 			(void) strlcpy(zo->zo_pool, optarg,
669 			    sizeof (zo->zo_pool));
670 			break;
671 		case 'f':
672 			path = realpath(optarg, NULL);
673 			if (path == NULL) {
674 				(void) fprintf(stderr, "error: %s: %s\n",
675 				    optarg, strerror(errno));
676 				usage(B_FALSE);
677 			} else {
678 				(void) strlcpy(zo->zo_dir, path,
679 				    sizeof (zo->zo_dir));
680 			}
681 			break;
682 		case 'V':
683 			zo->zo_verbose++;
684 			break;
685 		case 'E':
686 			zo->zo_init = 0;
687 			break;
688 		case 'T':
689 			zo->zo_time = value;
690 			break;
691 		case 'P':
692 			zo->zo_passtime = MAX(1, value);
693 			break;
694 		case 'F':
695 			zo->zo_maxloops = MAX(1, value);
696 			break;
697 		case 'B':
698 			(void) strlcpy(altdir, optarg, sizeof (altdir));
699 			break;
700 		case 'h':
701 			usage(B_TRUE);
702 			break;
703 		case '?':
704 		default:
705 			usage(B_FALSE);
706 			break;
707 		}
708 	}
709 
710 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
711 
712 	zo->zo_vdevtime =
713 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
714 	    UINT64_MAX >> 2);
715 
716 	if (strlen(altdir) > 0) {
717 		char *cmd;
718 		char *realaltdir;
719 		char *bin;
720 		char *ztest;
721 		char *isa;
722 		int isalen;
723 
724 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
725 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
726 
727 		VERIFY(NULL != realpath(getexecname(), cmd));
728 		if (0 != access(altdir, F_OK)) {
729 			ztest_dump_core = B_FALSE;
730 			fatal(B_TRUE, "invalid alternate ztest path: %s",
731 			    altdir);
732 		}
733 		VERIFY(NULL != realpath(altdir, realaltdir));
734 
735 		/*
736 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
737 		 * We want to extract <isa> to determine if we should use
738 		 * 32 or 64 bit binaries.
739 		 */
740 		bin = strstr(cmd, "/usr/bin/");
741 		ztest = strstr(bin, "/ztest");
742 		isa = bin + 9;
743 		isalen = ztest - isa;
744 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
745 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
746 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
747 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
748 
749 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
750 			ztest_dump_core = B_FALSE;
751 			fatal(B_TRUE, "invalid alternate ztest: %s",
752 			    zo->zo_alt_ztest);
753 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
754 			ztest_dump_core = B_FALSE;
755 			fatal(B_TRUE, "invalid alternate lib directory %s",
756 			    zo->zo_alt_libpath);
757 		}
758 
759 		umem_free(cmd, MAXPATHLEN);
760 		umem_free(realaltdir, MAXPATHLEN);
761 	}
762 }
763 
764 static void
765 ztest_kill(ztest_shared_t *zs)
766 {
767 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
768 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
769 	(void) kill(getpid(), SIGKILL);
770 }
771 
772 static uint64_t
773 ztest_random(uint64_t range)
774 {
775 	uint64_t r;
776 
777 	ASSERT3S(ztest_fd_rand, >=, 0);
778 
779 	if (range == 0)
780 		return (0);
781 
782 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
783 		fatal(1, "short read from /dev/urandom");
784 
785 	return (r % range);
786 }
787 
788 /* ARGSUSED */
789 static void
790 ztest_record_enospc(const char *s)
791 {
792 	ztest_shared->zs_enospc_count++;
793 }
794 
795 static uint64_t
796 ztest_get_ashift(void)
797 {
798 	if (ztest_opts.zo_ashift == 0)
799 		return (SPA_MINBLOCKSHIFT + ztest_random(3));
800 	return (ztest_opts.zo_ashift);
801 }
802 
803 static nvlist_t *
804 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
805 {
806 	char pathbuf[MAXPATHLEN];
807 	uint64_t vdev;
808 	nvlist_t *file;
809 
810 	if (ashift == 0)
811 		ashift = ztest_get_ashift();
812 
813 	if (path == NULL) {
814 		path = pathbuf;
815 
816 		if (aux != NULL) {
817 			vdev = ztest_shared->zs_vdev_aux;
818 			(void) snprintf(path, sizeof (pathbuf),
819 			    ztest_aux_template, ztest_opts.zo_dir,
820 			    pool == NULL ? ztest_opts.zo_pool : pool,
821 			    aux, vdev);
822 		} else {
823 			vdev = ztest_shared->zs_vdev_next_leaf++;
824 			(void) snprintf(path, sizeof (pathbuf),
825 			    ztest_dev_template, ztest_opts.zo_dir,
826 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
827 		}
828 	}
829 
830 	if (size != 0) {
831 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
832 		if (fd == -1)
833 			fatal(1, "can't open %s", path);
834 		if (ftruncate(fd, size) != 0)
835 			fatal(1, "can't ftruncate %s", path);
836 		(void) close(fd);
837 	}
838 
839 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
840 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
841 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
842 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
843 
844 	return (file);
845 }
846 
847 static nvlist_t *
848 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
849     uint64_t ashift, int r)
850 {
851 	nvlist_t *raidz, **child;
852 	int c;
853 
854 	if (r < 2)
855 		return (make_vdev_file(path, aux, pool, size, ashift));
856 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
857 
858 	for (c = 0; c < r; c++)
859 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
860 
861 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
862 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
863 	    VDEV_TYPE_RAIDZ) == 0);
864 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
865 	    ztest_opts.zo_raidz_parity) == 0);
866 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
867 	    child, r) == 0);
868 
869 	for (c = 0; c < r; c++)
870 		nvlist_free(child[c]);
871 
872 	umem_free(child, r * sizeof (nvlist_t *));
873 
874 	return (raidz);
875 }
876 
877 static nvlist_t *
878 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
879     uint64_t ashift, int r, int m)
880 {
881 	nvlist_t *mirror, **child;
882 	int c;
883 
884 	if (m < 1)
885 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
886 
887 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
888 
889 	for (c = 0; c < m; c++)
890 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
891 
892 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
893 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
894 	    VDEV_TYPE_MIRROR) == 0);
895 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
896 	    child, m) == 0);
897 
898 	for (c = 0; c < m; c++)
899 		nvlist_free(child[c]);
900 
901 	umem_free(child, m * sizeof (nvlist_t *));
902 
903 	return (mirror);
904 }
905 
906 static nvlist_t *
907 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
908     int log, int r, int m, int t)
909 {
910 	nvlist_t *root, **child;
911 	int c;
912 
913 	ASSERT(t > 0);
914 
915 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
916 
917 	for (c = 0; c < t; c++) {
918 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
919 		    r, m);
920 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
921 		    log) == 0);
922 	}
923 
924 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
925 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
926 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
927 	    child, t) == 0);
928 
929 	for (c = 0; c < t; c++)
930 		nvlist_free(child[c]);
931 
932 	umem_free(child, t * sizeof (nvlist_t *));
933 
934 	return (root);
935 }
936 
937 /*
938  * Find a random spa version. Returns back a random spa version in the
939  * range [initial_version, SPA_VERSION_FEATURES].
940  */
941 static uint64_t
942 ztest_random_spa_version(uint64_t initial_version)
943 {
944 	uint64_t version = initial_version;
945 
946 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
947 		version = version +
948 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
949 	}
950 
951 	if (version > SPA_VERSION_BEFORE_FEATURES)
952 		version = SPA_VERSION_FEATURES;
953 
954 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
955 	return (version);
956 }
957 
958 static int
959 ztest_random_blocksize(void)
960 {
961 	return (1 << (SPA_MINBLOCKSHIFT +
962 	    ztest_random(SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)));
963 }
964 
965 static int
966 ztest_random_ibshift(void)
967 {
968 	return (DN_MIN_INDBLKSHIFT +
969 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
970 }
971 
972 static uint64_t
973 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
974 {
975 	uint64_t top;
976 	vdev_t *rvd = spa->spa_root_vdev;
977 	vdev_t *tvd;
978 
979 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
980 
981 	do {
982 		top = ztest_random(rvd->vdev_children);
983 		tvd = rvd->vdev_child[top];
984 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
985 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
986 
987 	return (top);
988 }
989 
990 static uint64_t
991 ztest_random_dsl_prop(zfs_prop_t prop)
992 {
993 	uint64_t value;
994 
995 	do {
996 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
997 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
998 
999 	return (value);
1000 }
1001 
1002 static int
1003 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1004     boolean_t inherit)
1005 {
1006 	const char *propname = zfs_prop_to_name(prop);
1007 	const char *valname;
1008 	char setpoint[MAXPATHLEN];
1009 	uint64_t curval;
1010 	int error;
1011 
1012 	error = dsl_prop_set_int(osname, propname,
1013 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1014 
1015 	if (error == ENOSPC) {
1016 		ztest_record_enospc(FTAG);
1017 		return (error);
1018 	}
1019 	ASSERT0(error);
1020 
1021 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1022 
1023 	if (ztest_opts.zo_verbose >= 6) {
1024 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1025 		(void) printf("%s %s = %s at '%s'\n",
1026 		    osname, propname, valname, setpoint);
1027 	}
1028 
1029 	return (error);
1030 }
1031 
1032 static int
1033 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1034 {
1035 	spa_t *spa = ztest_spa;
1036 	nvlist_t *props = NULL;
1037 	int error;
1038 
1039 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1040 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1041 
1042 	error = spa_prop_set(spa, props);
1043 
1044 	nvlist_free(props);
1045 
1046 	if (error == ENOSPC) {
1047 		ztest_record_enospc(FTAG);
1048 		return (error);
1049 	}
1050 	ASSERT0(error);
1051 
1052 	return (error);
1053 }
1054 
1055 static void
1056 ztest_rll_init(rll_t *rll)
1057 {
1058 	rll->rll_writer = NULL;
1059 	rll->rll_readers = 0;
1060 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1061 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1062 }
1063 
1064 static void
1065 ztest_rll_destroy(rll_t *rll)
1066 {
1067 	ASSERT(rll->rll_writer == NULL);
1068 	ASSERT(rll->rll_readers == 0);
1069 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1070 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
1071 }
1072 
1073 static void
1074 ztest_rll_lock(rll_t *rll, rl_type_t type)
1075 {
1076 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1077 
1078 	if (type == RL_READER) {
1079 		while (rll->rll_writer != NULL)
1080 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1081 		rll->rll_readers++;
1082 	} else {
1083 		while (rll->rll_writer != NULL || rll->rll_readers)
1084 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1085 		rll->rll_writer = curthread;
1086 	}
1087 
1088 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1089 }
1090 
1091 static void
1092 ztest_rll_unlock(rll_t *rll)
1093 {
1094 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1095 
1096 	if (rll->rll_writer) {
1097 		ASSERT(rll->rll_readers == 0);
1098 		rll->rll_writer = NULL;
1099 	} else {
1100 		ASSERT(rll->rll_readers != 0);
1101 		ASSERT(rll->rll_writer == NULL);
1102 		rll->rll_readers--;
1103 	}
1104 
1105 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1106 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1107 
1108 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1109 }
1110 
1111 static void
1112 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1113 {
1114 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1115 
1116 	ztest_rll_lock(rll, type);
1117 }
1118 
1119 static void
1120 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1121 {
1122 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1123 
1124 	ztest_rll_unlock(rll);
1125 }
1126 
1127 static rl_t *
1128 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1129     uint64_t size, rl_type_t type)
1130 {
1131 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1132 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1133 	rl_t *rl;
1134 
1135 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1136 	rl->rl_object = object;
1137 	rl->rl_offset = offset;
1138 	rl->rl_size = size;
1139 	rl->rl_lock = rll;
1140 
1141 	ztest_rll_lock(rll, type);
1142 
1143 	return (rl);
1144 }
1145 
1146 static void
1147 ztest_range_unlock(rl_t *rl)
1148 {
1149 	rll_t *rll = rl->rl_lock;
1150 
1151 	ztest_rll_unlock(rll);
1152 
1153 	umem_free(rl, sizeof (*rl));
1154 }
1155 
1156 static void
1157 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1158 {
1159 	zd->zd_os = os;
1160 	zd->zd_zilog = dmu_objset_zil(os);
1161 	zd->zd_shared = szd;
1162 	dmu_objset_name(os, zd->zd_name);
1163 
1164 	if (zd->zd_shared != NULL)
1165 		zd->zd_shared->zd_seq = 0;
1166 
1167 	VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1168 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1169 
1170 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1171 		ztest_rll_init(&zd->zd_object_lock[l]);
1172 
1173 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1174 		ztest_rll_init(&zd->zd_range_lock[l]);
1175 }
1176 
1177 static void
1178 ztest_zd_fini(ztest_ds_t *zd)
1179 {
1180 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1181 
1182 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1183 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1184 
1185 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1186 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1187 }
1188 
1189 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1190 
1191 static uint64_t
1192 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1193 {
1194 	uint64_t txg;
1195 	int error;
1196 
1197 	/*
1198 	 * Attempt to assign tx to some transaction group.
1199 	 */
1200 	error = dmu_tx_assign(tx, txg_how);
1201 	if (error) {
1202 		if (error == ERESTART) {
1203 			ASSERT(txg_how == TXG_NOWAIT);
1204 			dmu_tx_wait(tx);
1205 		} else {
1206 			ASSERT3U(error, ==, ENOSPC);
1207 			ztest_record_enospc(tag);
1208 		}
1209 		dmu_tx_abort(tx);
1210 		return (0);
1211 	}
1212 	txg = dmu_tx_get_txg(tx);
1213 	ASSERT(txg != 0);
1214 	return (txg);
1215 }
1216 
1217 static void
1218 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1219 {
1220 	uint64_t *ip = buf;
1221 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1222 
1223 	while (ip < ip_end)
1224 		*ip++ = value;
1225 }
1226 
1227 static boolean_t
1228 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1229 {
1230 	uint64_t *ip = buf;
1231 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1232 	uint64_t diff = 0;
1233 
1234 	while (ip < ip_end)
1235 		diff |= (value - *ip++);
1236 
1237 	return (diff == 0);
1238 }
1239 
1240 static void
1241 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1242     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1243 {
1244 	bt->bt_magic = BT_MAGIC;
1245 	bt->bt_objset = dmu_objset_id(os);
1246 	bt->bt_object = object;
1247 	bt->bt_offset = offset;
1248 	bt->bt_gen = gen;
1249 	bt->bt_txg = txg;
1250 	bt->bt_crtxg = crtxg;
1251 }
1252 
1253 static void
1254 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1255     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1256 {
1257 	ASSERT(bt->bt_magic == BT_MAGIC);
1258 	ASSERT(bt->bt_objset == dmu_objset_id(os));
1259 	ASSERT(bt->bt_object == object);
1260 	ASSERT(bt->bt_offset == offset);
1261 	ASSERT(bt->bt_gen <= gen);
1262 	ASSERT(bt->bt_txg <= txg);
1263 	ASSERT(bt->bt_crtxg == crtxg);
1264 }
1265 
1266 static ztest_block_tag_t *
1267 ztest_bt_bonus(dmu_buf_t *db)
1268 {
1269 	dmu_object_info_t doi;
1270 	ztest_block_tag_t *bt;
1271 
1272 	dmu_object_info_from_db(db, &doi);
1273 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1274 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1275 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1276 
1277 	return (bt);
1278 }
1279 
1280 /*
1281  * ZIL logging ops
1282  */
1283 
1284 #define	lrz_type	lr_mode
1285 #define	lrz_blocksize	lr_uid
1286 #define	lrz_ibshift	lr_gid
1287 #define	lrz_bonustype	lr_rdev
1288 #define	lrz_bonuslen	lr_crtime[1]
1289 
1290 static void
1291 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1292 {
1293 	char *name = (void *)(lr + 1);		/* name follows lr */
1294 	size_t namesize = strlen(name) + 1;
1295 	itx_t *itx;
1296 
1297 	if (zil_replaying(zd->zd_zilog, tx))
1298 		return;
1299 
1300 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1301 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1302 	    sizeof (*lr) + namesize - sizeof (lr_t));
1303 
1304 	zil_itx_assign(zd->zd_zilog, itx, tx);
1305 }
1306 
1307 static void
1308 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1309 {
1310 	char *name = (void *)(lr + 1);		/* name follows lr */
1311 	size_t namesize = strlen(name) + 1;
1312 	itx_t *itx;
1313 
1314 	if (zil_replaying(zd->zd_zilog, tx))
1315 		return;
1316 
1317 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1318 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1319 	    sizeof (*lr) + namesize - sizeof (lr_t));
1320 
1321 	itx->itx_oid = object;
1322 	zil_itx_assign(zd->zd_zilog, itx, tx);
1323 }
1324 
1325 static void
1326 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1327 {
1328 	itx_t *itx;
1329 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1330 
1331 	if (zil_replaying(zd->zd_zilog, tx))
1332 		return;
1333 
1334 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1335 		write_state = WR_INDIRECT;
1336 
1337 	itx = zil_itx_create(TX_WRITE,
1338 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1339 
1340 	if (write_state == WR_COPIED &&
1341 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1342 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1343 		zil_itx_destroy(itx);
1344 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1345 		write_state = WR_NEED_COPY;
1346 	}
1347 	itx->itx_private = zd;
1348 	itx->itx_wr_state = write_state;
1349 	itx->itx_sync = (ztest_random(8) == 0);
1350 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1351 
1352 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1353 	    sizeof (*lr) - sizeof (lr_t));
1354 
1355 	zil_itx_assign(zd->zd_zilog, itx, tx);
1356 }
1357 
1358 static void
1359 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1360 {
1361 	itx_t *itx;
1362 
1363 	if (zil_replaying(zd->zd_zilog, tx))
1364 		return;
1365 
1366 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1367 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1368 	    sizeof (*lr) - sizeof (lr_t));
1369 
1370 	itx->itx_sync = B_FALSE;
1371 	zil_itx_assign(zd->zd_zilog, itx, tx);
1372 }
1373 
1374 static void
1375 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1376 {
1377 	itx_t *itx;
1378 
1379 	if (zil_replaying(zd->zd_zilog, tx))
1380 		return;
1381 
1382 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1383 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1384 	    sizeof (*lr) - sizeof (lr_t));
1385 
1386 	itx->itx_sync = B_FALSE;
1387 	zil_itx_assign(zd->zd_zilog, itx, tx);
1388 }
1389 
1390 /*
1391  * ZIL replay ops
1392  */
1393 static int
1394 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1395 {
1396 	char *name = (void *)(lr + 1);		/* name follows lr */
1397 	objset_t *os = zd->zd_os;
1398 	ztest_block_tag_t *bbt;
1399 	dmu_buf_t *db;
1400 	dmu_tx_t *tx;
1401 	uint64_t txg;
1402 	int error = 0;
1403 
1404 	if (byteswap)
1405 		byteswap_uint64_array(lr, sizeof (*lr));
1406 
1407 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1408 	ASSERT(name[0] != '\0');
1409 
1410 	tx = dmu_tx_create(os);
1411 
1412 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1413 
1414 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1415 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1416 	} else {
1417 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1418 	}
1419 
1420 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1421 	if (txg == 0)
1422 		return (ENOSPC);
1423 
1424 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1425 
1426 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1427 		if (lr->lr_foid == 0) {
1428 			lr->lr_foid = zap_create(os,
1429 			    lr->lrz_type, lr->lrz_bonustype,
1430 			    lr->lrz_bonuslen, tx);
1431 		} else {
1432 			error = zap_create_claim(os, lr->lr_foid,
1433 			    lr->lrz_type, lr->lrz_bonustype,
1434 			    lr->lrz_bonuslen, tx);
1435 		}
1436 	} else {
1437 		if (lr->lr_foid == 0) {
1438 			lr->lr_foid = dmu_object_alloc(os,
1439 			    lr->lrz_type, 0, lr->lrz_bonustype,
1440 			    lr->lrz_bonuslen, tx);
1441 		} else {
1442 			error = dmu_object_claim(os, lr->lr_foid,
1443 			    lr->lrz_type, 0, lr->lrz_bonustype,
1444 			    lr->lrz_bonuslen, tx);
1445 		}
1446 	}
1447 
1448 	if (error) {
1449 		ASSERT3U(error, ==, EEXIST);
1450 		ASSERT(zd->zd_zilog->zl_replay);
1451 		dmu_tx_commit(tx);
1452 		return (error);
1453 	}
1454 
1455 	ASSERT(lr->lr_foid != 0);
1456 
1457 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1458 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1459 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1460 
1461 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1462 	bbt = ztest_bt_bonus(db);
1463 	dmu_buf_will_dirty(db, tx);
1464 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1465 	dmu_buf_rele(db, FTAG);
1466 
1467 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1468 	    &lr->lr_foid, tx));
1469 
1470 	(void) ztest_log_create(zd, tx, lr);
1471 
1472 	dmu_tx_commit(tx);
1473 
1474 	return (0);
1475 }
1476 
1477 static int
1478 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1479 {
1480 	char *name = (void *)(lr + 1);		/* name follows lr */
1481 	objset_t *os = zd->zd_os;
1482 	dmu_object_info_t doi;
1483 	dmu_tx_t *tx;
1484 	uint64_t object, txg;
1485 
1486 	if (byteswap)
1487 		byteswap_uint64_array(lr, sizeof (*lr));
1488 
1489 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1490 	ASSERT(name[0] != '\0');
1491 
1492 	VERIFY3U(0, ==,
1493 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1494 	ASSERT(object != 0);
1495 
1496 	ztest_object_lock(zd, object, RL_WRITER);
1497 
1498 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1499 
1500 	tx = dmu_tx_create(os);
1501 
1502 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1503 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1504 
1505 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1506 	if (txg == 0) {
1507 		ztest_object_unlock(zd, object);
1508 		return (ENOSPC);
1509 	}
1510 
1511 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1512 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1513 	} else {
1514 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1515 	}
1516 
1517 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1518 
1519 	(void) ztest_log_remove(zd, tx, lr, object);
1520 
1521 	dmu_tx_commit(tx);
1522 
1523 	ztest_object_unlock(zd, object);
1524 
1525 	return (0);
1526 }
1527 
1528 static int
1529 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1530 {
1531 	objset_t *os = zd->zd_os;
1532 	void *data = lr + 1;			/* data follows lr */
1533 	uint64_t offset, length;
1534 	ztest_block_tag_t *bt = data;
1535 	ztest_block_tag_t *bbt;
1536 	uint64_t gen, txg, lrtxg, crtxg;
1537 	dmu_object_info_t doi;
1538 	dmu_tx_t *tx;
1539 	dmu_buf_t *db;
1540 	arc_buf_t *abuf = NULL;
1541 	rl_t *rl;
1542 
1543 	if (byteswap)
1544 		byteswap_uint64_array(lr, sizeof (*lr));
1545 
1546 	offset = lr->lr_offset;
1547 	length = lr->lr_length;
1548 
1549 	/* If it's a dmu_sync() block, write the whole block */
1550 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1551 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1552 		if (length < blocksize) {
1553 			offset -= offset % blocksize;
1554 			length = blocksize;
1555 		}
1556 	}
1557 
1558 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1559 		byteswap_uint64_array(bt, sizeof (*bt));
1560 
1561 	if (bt->bt_magic != BT_MAGIC)
1562 		bt = NULL;
1563 
1564 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1565 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1566 
1567 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1568 
1569 	dmu_object_info_from_db(db, &doi);
1570 
1571 	bbt = ztest_bt_bonus(db);
1572 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1573 	gen = bbt->bt_gen;
1574 	crtxg = bbt->bt_crtxg;
1575 	lrtxg = lr->lr_common.lrc_txg;
1576 
1577 	tx = dmu_tx_create(os);
1578 
1579 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1580 
1581 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1582 	    P2PHASE(offset, length) == 0)
1583 		abuf = dmu_request_arcbuf(db, length);
1584 
1585 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1586 	if (txg == 0) {
1587 		if (abuf != NULL)
1588 			dmu_return_arcbuf(abuf);
1589 		dmu_buf_rele(db, FTAG);
1590 		ztest_range_unlock(rl);
1591 		ztest_object_unlock(zd, lr->lr_foid);
1592 		return (ENOSPC);
1593 	}
1594 
1595 	if (bt != NULL) {
1596 		/*
1597 		 * Usually, verify the old data before writing new data --
1598 		 * but not always, because we also want to verify correct
1599 		 * behavior when the data was not recently read into cache.
1600 		 */
1601 		ASSERT(offset % doi.doi_data_block_size == 0);
1602 		if (ztest_random(4) != 0) {
1603 			int prefetch = ztest_random(2) ?
1604 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1605 			ztest_block_tag_t rbt;
1606 
1607 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1608 			    sizeof (rbt), &rbt, prefetch) == 0);
1609 			if (rbt.bt_magic == BT_MAGIC) {
1610 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1611 				    offset, gen, txg, crtxg);
1612 			}
1613 		}
1614 
1615 		/*
1616 		 * Writes can appear to be newer than the bonus buffer because
1617 		 * the ztest_get_data() callback does a dmu_read() of the
1618 		 * open-context data, which may be different than the data
1619 		 * as it was when the write was generated.
1620 		 */
1621 		if (zd->zd_zilog->zl_replay) {
1622 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1623 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1624 			    bt->bt_crtxg);
1625 		}
1626 
1627 		/*
1628 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1629 		 * so that all of the usual ASSERTs will work.
1630 		 */
1631 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1632 	}
1633 
1634 	if (abuf == NULL) {
1635 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1636 	} else {
1637 		bcopy(data, abuf->b_data, length);
1638 		dmu_assign_arcbuf(db, offset, abuf, tx);
1639 	}
1640 
1641 	(void) ztest_log_write(zd, tx, lr);
1642 
1643 	dmu_buf_rele(db, FTAG);
1644 
1645 	dmu_tx_commit(tx);
1646 
1647 	ztest_range_unlock(rl);
1648 	ztest_object_unlock(zd, lr->lr_foid);
1649 
1650 	return (0);
1651 }
1652 
1653 static int
1654 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1655 {
1656 	objset_t *os = zd->zd_os;
1657 	dmu_tx_t *tx;
1658 	uint64_t txg;
1659 	rl_t *rl;
1660 
1661 	if (byteswap)
1662 		byteswap_uint64_array(lr, sizeof (*lr));
1663 
1664 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1665 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1666 	    RL_WRITER);
1667 
1668 	tx = dmu_tx_create(os);
1669 
1670 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1671 
1672 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1673 	if (txg == 0) {
1674 		ztest_range_unlock(rl);
1675 		ztest_object_unlock(zd, lr->lr_foid);
1676 		return (ENOSPC);
1677 	}
1678 
1679 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1680 	    lr->lr_length, tx) == 0);
1681 
1682 	(void) ztest_log_truncate(zd, tx, lr);
1683 
1684 	dmu_tx_commit(tx);
1685 
1686 	ztest_range_unlock(rl);
1687 	ztest_object_unlock(zd, lr->lr_foid);
1688 
1689 	return (0);
1690 }
1691 
1692 static int
1693 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1694 {
1695 	objset_t *os = zd->zd_os;
1696 	dmu_tx_t *tx;
1697 	dmu_buf_t *db;
1698 	ztest_block_tag_t *bbt;
1699 	uint64_t txg, lrtxg, crtxg;
1700 
1701 	if (byteswap)
1702 		byteswap_uint64_array(lr, sizeof (*lr));
1703 
1704 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1705 
1706 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1707 
1708 	tx = dmu_tx_create(os);
1709 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1710 
1711 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1712 	if (txg == 0) {
1713 		dmu_buf_rele(db, FTAG);
1714 		ztest_object_unlock(zd, lr->lr_foid);
1715 		return (ENOSPC);
1716 	}
1717 
1718 	bbt = ztest_bt_bonus(db);
1719 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1720 	crtxg = bbt->bt_crtxg;
1721 	lrtxg = lr->lr_common.lrc_txg;
1722 
1723 	if (zd->zd_zilog->zl_replay) {
1724 		ASSERT(lr->lr_size != 0);
1725 		ASSERT(lr->lr_mode != 0);
1726 		ASSERT(lrtxg != 0);
1727 	} else {
1728 		/*
1729 		 * Randomly change the size and increment the generation.
1730 		 */
1731 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1732 		    sizeof (*bbt);
1733 		lr->lr_mode = bbt->bt_gen + 1;
1734 		ASSERT(lrtxg == 0);
1735 	}
1736 
1737 	/*
1738 	 * Verify that the current bonus buffer is not newer than our txg.
1739 	 */
1740 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1741 	    MAX(txg, lrtxg), crtxg);
1742 
1743 	dmu_buf_will_dirty(db, tx);
1744 
1745 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1746 	ASSERT3U(lr->lr_size, <=, db->db_size);
1747 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1748 	bbt = ztest_bt_bonus(db);
1749 
1750 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1751 
1752 	dmu_buf_rele(db, FTAG);
1753 
1754 	(void) ztest_log_setattr(zd, tx, lr);
1755 
1756 	dmu_tx_commit(tx);
1757 
1758 	ztest_object_unlock(zd, lr->lr_foid);
1759 
1760 	return (0);
1761 }
1762 
1763 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1764 	NULL,			/* 0 no such transaction type */
1765 	ztest_replay_create,	/* TX_CREATE */
1766 	NULL,			/* TX_MKDIR */
1767 	NULL,			/* TX_MKXATTR */
1768 	NULL,			/* TX_SYMLINK */
1769 	ztest_replay_remove,	/* TX_REMOVE */
1770 	NULL,			/* TX_RMDIR */
1771 	NULL,			/* TX_LINK */
1772 	NULL,			/* TX_RENAME */
1773 	ztest_replay_write,	/* TX_WRITE */
1774 	ztest_replay_truncate,	/* TX_TRUNCATE */
1775 	ztest_replay_setattr,	/* TX_SETATTR */
1776 	NULL,			/* TX_ACL */
1777 	NULL,			/* TX_CREATE_ACL */
1778 	NULL,			/* TX_CREATE_ATTR */
1779 	NULL,			/* TX_CREATE_ACL_ATTR */
1780 	NULL,			/* TX_MKDIR_ACL */
1781 	NULL,			/* TX_MKDIR_ATTR */
1782 	NULL,			/* TX_MKDIR_ACL_ATTR */
1783 	NULL,			/* TX_WRITE2 */
1784 };
1785 
1786 /*
1787  * ZIL get_data callbacks
1788  */
1789 
1790 static void
1791 ztest_get_done(zgd_t *zgd, int error)
1792 {
1793 	ztest_ds_t *zd = zgd->zgd_private;
1794 	uint64_t object = zgd->zgd_rl->rl_object;
1795 
1796 	if (zgd->zgd_db)
1797 		dmu_buf_rele(zgd->zgd_db, zgd);
1798 
1799 	ztest_range_unlock(zgd->zgd_rl);
1800 	ztest_object_unlock(zd, object);
1801 
1802 	if (error == 0 && zgd->zgd_bp)
1803 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1804 
1805 	umem_free(zgd, sizeof (*zgd));
1806 }
1807 
1808 static int
1809 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1810 {
1811 	ztest_ds_t *zd = arg;
1812 	objset_t *os = zd->zd_os;
1813 	uint64_t object = lr->lr_foid;
1814 	uint64_t offset = lr->lr_offset;
1815 	uint64_t size = lr->lr_length;
1816 	blkptr_t *bp = &lr->lr_blkptr;
1817 	uint64_t txg = lr->lr_common.lrc_txg;
1818 	uint64_t crtxg;
1819 	dmu_object_info_t doi;
1820 	dmu_buf_t *db;
1821 	zgd_t *zgd;
1822 	int error;
1823 
1824 	ztest_object_lock(zd, object, RL_READER);
1825 	error = dmu_bonus_hold(os, object, FTAG, &db);
1826 	if (error) {
1827 		ztest_object_unlock(zd, object);
1828 		return (error);
1829 	}
1830 
1831 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1832 
1833 	if (crtxg == 0 || crtxg > txg) {
1834 		dmu_buf_rele(db, FTAG);
1835 		ztest_object_unlock(zd, object);
1836 		return (ENOENT);
1837 	}
1838 
1839 	dmu_object_info_from_db(db, &doi);
1840 	dmu_buf_rele(db, FTAG);
1841 	db = NULL;
1842 
1843 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1844 	zgd->zgd_zilog = zd->zd_zilog;
1845 	zgd->zgd_private = zd;
1846 
1847 	if (buf != NULL) {	/* immediate write */
1848 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1849 		    RL_READER);
1850 
1851 		error = dmu_read(os, object, offset, size, buf,
1852 		    DMU_READ_NO_PREFETCH);
1853 		ASSERT(error == 0);
1854 	} else {
1855 		size = doi.doi_data_block_size;
1856 		if (ISP2(size)) {
1857 			offset = P2ALIGN(offset, size);
1858 		} else {
1859 			ASSERT(offset < size);
1860 			offset = 0;
1861 		}
1862 
1863 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1864 		    RL_READER);
1865 
1866 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1867 		    DMU_READ_NO_PREFETCH);
1868 
1869 		if (error == 0) {
1870 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1871 			if (obp) {
1872 				ASSERT(BP_IS_HOLE(bp));
1873 				*bp = *obp;
1874 			}
1875 
1876 			zgd->zgd_db = db;
1877 			zgd->zgd_bp = bp;
1878 
1879 			ASSERT(db->db_offset == offset);
1880 			ASSERT(db->db_size == size);
1881 
1882 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1883 			    ztest_get_done, zgd);
1884 
1885 			if (error == 0)
1886 				return (0);
1887 		}
1888 	}
1889 
1890 	ztest_get_done(zgd, error);
1891 
1892 	return (error);
1893 }
1894 
1895 static void *
1896 ztest_lr_alloc(size_t lrsize, char *name)
1897 {
1898 	char *lr;
1899 	size_t namesize = name ? strlen(name) + 1 : 0;
1900 
1901 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1902 
1903 	if (name)
1904 		bcopy(name, lr + lrsize, namesize);
1905 
1906 	return (lr);
1907 }
1908 
1909 void
1910 ztest_lr_free(void *lr, size_t lrsize, char *name)
1911 {
1912 	size_t namesize = name ? strlen(name) + 1 : 0;
1913 
1914 	umem_free(lr, lrsize + namesize);
1915 }
1916 
1917 /*
1918  * Lookup a bunch of objects.  Returns the number of objects not found.
1919  */
1920 static int
1921 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1922 {
1923 	int missing = 0;
1924 	int error;
1925 
1926 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1927 
1928 	for (int i = 0; i < count; i++, od++) {
1929 		od->od_object = 0;
1930 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1931 		    sizeof (uint64_t), 1, &od->od_object);
1932 		if (error) {
1933 			ASSERT(error == ENOENT);
1934 			ASSERT(od->od_object == 0);
1935 			missing++;
1936 		} else {
1937 			dmu_buf_t *db;
1938 			ztest_block_tag_t *bbt;
1939 			dmu_object_info_t doi;
1940 
1941 			ASSERT(od->od_object != 0);
1942 			ASSERT(missing == 0);	/* there should be no gaps */
1943 
1944 			ztest_object_lock(zd, od->od_object, RL_READER);
1945 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1946 			    od->od_object, FTAG, &db));
1947 			dmu_object_info_from_db(db, &doi);
1948 			bbt = ztest_bt_bonus(db);
1949 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1950 			od->od_type = doi.doi_type;
1951 			od->od_blocksize = doi.doi_data_block_size;
1952 			od->od_gen = bbt->bt_gen;
1953 			dmu_buf_rele(db, FTAG);
1954 			ztest_object_unlock(zd, od->od_object);
1955 		}
1956 	}
1957 
1958 	return (missing);
1959 }
1960 
1961 static int
1962 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1963 {
1964 	int missing = 0;
1965 
1966 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1967 
1968 	for (int i = 0; i < count; i++, od++) {
1969 		if (missing) {
1970 			od->od_object = 0;
1971 			missing++;
1972 			continue;
1973 		}
1974 
1975 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1976 
1977 		lr->lr_doid = od->od_dir;
1978 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
1979 		lr->lrz_type = od->od_crtype;
1980 		lr->lrz_blocksize = od->od_crblocksize;
1981 		lr->lrz_ibshift = ztest_random_ibshift();
1982 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
1983 		lr->lrz_bonuslen = dmu_bonus_max();
1984 		lr->lr_gen = od->od_crgen;
1985 		lr->lr_crtime[0] = time(NULL);
1986 
1987 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
1988 			ASSERT(missing == 0);
1989 			od->od_object = 0;
1990 			missing++;
1991 		} else {
1992 			od->od_object = lr->lr_foid;
1993 			od->od_type = od->od_crtype;
1994 			od->od_blocksize = od->od_crblocksize;
1995 			od->od_gen = od->od_crgen;
1996 			ASSERT(od->od_object != 0);
1997 		}
1998 
1999 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2000 	}
2001 
2002 	return (missing);
2003 }
2004 
2005 static int
2006 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2007 {
2008 	int missing = 0;
2009 	int error;
2010 
2011 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2012 
2013 	od += count - 1;
2014 
2015 	for (int i = count - 1; i >= 0; i--, od--) {
2016 		if (missing) {
2017 			missing++;
2018 			continue;
2019 		}
2020 
2021 		/*
2022 		 * No object was found.
2023 		 */
2024 		if (od->od_object == 0)
2025 			continue;
2026 
2027 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2028 
2029 		lr->lr_doid = od->od_dir;
2030 
2031 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2032 			ASSERT3U(error, ==, ENOSPC);
2033 			missing++;
2034 		} else {
2035 			od->od_object = 0;
2036 		}
2037 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2038 	}
2039 
2040 	return (missing);
2041 }
2042 
2043 static int
2044 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2045     void *data)
2046 {
2047 	lr_write_t *lr;
2048 	int error;
2049 
2050 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2051 
2052 	lr->lr_foid = object;
2053 	lr->lr_offset = offset;
2054 	lr->lr_length = size;
2055 	lr->lr_blkoff = 0;
2056 	BP_ZERO(&lr->lr_blkptr);
2057 
2058 	bcopy(data, lr + 1, size);
2059 
2060 	error = ztest_replay_write(zd, lr, B_FALSE);
2061 
2062 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2063 
2064 	return (error);
2065 }
2066 
2067 static int
2068 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2069 {
2070 	lr_truncate_t *lr;
2071 	int error;
2072 
2073 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2074 
2075 	lr->lr_foid = object;
2076 	lr->lr_offset = offset;
2077 	lr->lr_length = size;
2078 
2079 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2080 
2081 	ztest_lr_free(lr, sizeof (*lr), NULL);
2082 
2083 	return (error);
2084 }
2085 
2086 static int
2087 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2088 {
2089 	lr_setattr_t *lr;
2090 	int error;
2091 
2092 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2093 
2094 	lr->lr_foid = object;
2095 	lr->lr_size = 0;
2096 	lr->lr_mode = 0;
2097 
2098 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2099 
2100 	ztest_lr_free(lr, sizeof (*lr), NULL);
2101 
2102 	return (error);
2103 }
2104 
2105 static void
2106 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2107 {
2108 	objset_t *os = zd->zd_os;
2109 	dmu_tx_t *tx;
2110 	uint64_t txg;
2111 	rl_t *rl;
2112 
2113 	txg_wait_synced(dmu_objset_pool(os), 0);
2114 
2115 	ztest_object_lock(zd, object, RL_READER);
2116 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2117 
2118 	tx = dmu_tx_create(os);
2119 
2120 	dmu_tx_hold_write(tx, object, offset, size);
2121 
2122 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2123 
2124 	if (txg != 0) {
2125 		dmu_prealloc(os, object, offset, size, tx);
2126 		dmu_tx_commit(tx);
2127 		txg_wait_synced(dmu_objset_pool(os), txg);
2128 	} else {
2129 		(void) dmu_free_long_range(os, object, offset, size);
2130 	}
2131 
2132 	ztest_range_unlock(rl);
2133 	ztest_object_unlock(zd, object);
2134 }
2135 
2136 static void
2137 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2138 {
2139 	int err;
2140 	ztest_block_tag_t wbt;
2141 	dmu_object_info_t doi;
2142 	enum ztest_io_type io_type;
2143 	uint64_t blocksize;
2144 	void *data;
2145 
2146 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2147 	blocksize = doi.doi_data_block_size;
2148 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2149 
2150 	/*
2151 	 * Pick an i/o type at random, biased toward writing block tags.
2152 	 */
2153 	io_type = ztest_random(ZTEST_IO_TYPES);
2154 	if (ztest_random(2) == 0)
2155 		io_type = ZTEST_IO_WRITE_TAG;
2156 
2157 	(void) rw_rdlock(&zd->zd_zilog_lock);
2158 
2159 	switch (io_type) {
2160 
2161 	case ZTEST_IO_WRITE_TAG:
2162 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2163 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2164 		break;
2165 
2166 	case ZTEST_IO_WRITE_PATTERN:
2167 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2168 		if (ztest_random(2) == 0) {
2169 			/*
2170 			 * Induce fletcher2 collisions to ensure that
2171 			 * zio_ddt_collision() detects and resolves them
2172 			 * when using fletcher2-verify for deduplication.
2173 			 */
2174 			((uint64_t *)data)[0] ^= 1ULL << 63;
2175 			((uint64_t *)data)[4] ^= 1ULL << 63;
2176 		}
2177 		(void) ztest_write(zd, object, offset, blocksize, data);
2178 		break;
2179 
2180 	case ZTEST_IO_WRITE_ZEROES:
2181 		bzero(data, blocksize);
2182 		(void) ztest_write(zd, object, offset, blocksize, data);
2183 		break;
2184 
2185 	case ZTEST_IO_TRUNCATE:
2186 		(void) ztest_truncate(zd, object, offset, blocksize);
2187 		break;
2188 
2189 	case ZTEST_IO_SETATTR:
2190 		(void) ztest_setattr(zd, object);
2191 		break;
2192 
2193 	case ZTEST_IO_REWRITE:
2194 		(void) rw_rdlock(&ztest_name_lock);
2195 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2196 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2197 		    B_FALSE);
2198 		VERIFY(err == 0 || err == ENOSPC);
2199 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2200 		    ZFS_PROP_COMPRESSION,
2201 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2202 		    B_FALSE);
2203 		VERIFY(err == 0 || err == ENOSPC);
2204 		(void) rw_unlock(&ztest_name_lock);
2205 
2206 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2207 		    DMU_READ_NO_PREFETCH));
2208 
2209 		(void) ztest_write(zd, object, offset, blocksize, data);
2210 		break;
2211 	}
2212 
2213 	(void) rw_unlock(&zd->zd_zilog_lock);
2214 
2215 	umem_free(data, blocksize);
2216 }
2217 
2218 /*
2219  * Initialize an object description template.
2220  */
2221 static void
2222 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2223     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2224 {
2225 	od->od_dir = ZTEST_DIROBJ;
2226 	od->od_object = 0;
2227 
2228 	od->od_crtype = type;
2229 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2230 	od->od_crgen = gen;
2231 
2232 	od->od_type = DMU_OT_NONE;
2233 	od->od_blocksize = 0;
2234 	od->od_gen = 0;
2235 
2236 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2237 	    tag, (int64_t)id, index);
2238 }
2239 
2240 /*
2241  * Lookup or create the objects for a test using the od template.
2242  * If the objects do not all exist, or if 'remove' is specified,
2243  * remove any existing objects and create new ones.  Otherwise,
2244  * use the existing objects.
2245  */
2246 static int
2247 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2248 {
2249 	int count = size / sizeof (*od);
2250 	int rv = 0;
2251 
2252 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2253 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2254 	    (ztest_remove(zd, od, count) != 0 ||
2255 	    ztest_create(zd, od, count) != 0))
2256 		rv = -1;
2257 	zd->zd_od = od;
2258 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2259 
2260 	return (rv);
2261 }
2262 
2263 /* ARGSUSED */
2264 void
2265 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2266 {
2267 	zilog_t *zilog = zd->zd_zilog;
2268 
2269 	(void) rw_rdlock(&zd->zd_zilog_lock);
2270 
2271 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2272 
2273 	/*
2274 	 * Remember the committed values in zd, which is in parent/child
2275 	 * shared memory.  If we die, the next iteration of ztest_run()
2276 	 * will verify that the log really does contain this record.
2277 	 */
2278 	mutex_enter(&zilog->zl_lock);
2279 	ASSERT(zd->zd_shared != NULL);
2280 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2281 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2282 	mutex_exit(&zilog->zl_lock);
2283 
2284 	(void) rw_unlock(&zd->zd_zilog_lock);
2285 }
2286 
2287 /*
2288  * This function is designed to simulate the operations that occur during a
2289  * mount/unmount operation.  We hold the dataset across these operations in an
2290  * attempt to expose any implicit assumptions about ZIL management.
2291  */
2292 /* ARGSUSED */
2293 void
2294 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2295 {
2296 	objset_t *os = zd->zd_os;
2297 
2298 	/*
2299 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2300 	 * updating the zil (i.e. adding in-memory log records) and the
2301 	 * zd_zilog_lock to block any I/O.
2302 	 */
2303 	VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2304 	(void) rw_wrlock(&zd->zd_zilog_lock);
2305 
2306 	/* zfsvfs_teardown() */
2307 	zil_close(zd->zd_zilog);
2308 
2309 	/* zfsvfs_setup() */
2310 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2311 	zil_replay(os, zd, ztest_replay_vector);
2312 
2313 	(void) rw_unlock(&zd->zd_zilog_lock);
2314 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2315 }
2316 
2317 /*
2318  * Verify that we can't destroy an active pool, create an existing pool,
2319  * or create a pool with a bad vdev spec.
2320  */
2321 /* ARGSUSED */
2322 void
2323 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2324 {
2325 	ztest_shared_opts_t *zo = &ztest_opts;
2326 	spa_t *spa;
2327 	nvlist_t *nvroot;
2328 
2329 	/*
2330 	 * Attempt to create using a bad file.
2331 	 */
2332 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2333 	VERIFY3U(ENOENT, ==,
2334 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2335 	nvlist_free(nvroot);
2336 
2337 	/*
2338 	 * Attempt to create using a bad mirror.
2339 	 */
2340 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2341 	VERIFY3U(ENOENT, ==,
2342 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2343 	nvlist_free(nvroot);
2344 
2345 	/*
2346 	 * Attempt to create an existing pool.  It shouldn't matter
2347 	 * what's in the nvroot; we should fail with EEXIST.
2348 	 */
2349 	(void) rw_rdlock(&ztest_name_lock);
2350 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2351 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2352 	nvlist_free(nvroot);
2353 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2354 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2355 	spa_close(spa, FTAG);
2356 
2357 	(void) rw_unlock(&ztest_name_lock);
2358 }
2359 
2360 /* ARGSUSED */
2361 void
2362 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2363 {
2364 	spa_t *spa;
2365 	uint64_t initial_version = SPA_VERSION_INITIAL;
2366 	uint64_t version, newversion;
2367 	nvlist_t *nvroot, *props;
2368 	char *name;
2369 
2370 	VERIFY0(mutex_lock(&ztest_vdev_lock));
2371 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2372 
2373 	/*
2374 	 * Clean up from previous runs.
2375 	 */
2376 	(void) spa_destroy(name);
2377 
2378 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2379 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2380 
2381 	/*
2382 	 * If we're configuring a RAIDZ device then make sure that the
2383 	 * the initial version is capable of supporting that feature.
2384 	 */
2385 	switch (ztest_opts.zo_raidz_parity) {
2386 	case 0:
2387 	case 1:
2388 		initial_version = SPA_VERSION_INITIAL;
2389 		break;
2390 	case 2:
2391 		initial_version = SPA_VERSION_RAIDZ2;
2392 		break;
2393 	case 3:
2394 		initial_version = SPA_VERSION_RAIDZ3;
2395 		break;
2396 	}
2397 
2398 	/*
2399 	 * Create a pool with a spa version that can be upgraded. Pick
2400 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2401 	 */
2402 	do {
2403 		version = ztest_random_spa_version(initial_version);
2404 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2405 
2406 	props = fnvlist_alloc();
2407 	fnvlist_add_uint64(props,
2408 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2409 	VERIFY0(spa_create(name, nvroot, props, NULL));
2410 	fnvlist_free(nvroot);
2411 	fnvlist_free(props);
2412 
2413 	VERIFY0(spa_open(name, &spa, FTAG));
2414 	VERIFY3U(spa_version(spa), ==, version);
2415 	newversion = ztest_random_spa_version(version + 1);
2416 
2417 	if (ztest_opts.zo_verbose >= 4) {
2418 		(void) printf("upgrading spa version from %llu to %llu\n",
2419 		    (u_longlong_t)version, (u_longlong_t)newversion);
2420 	}
2421 
2422 	spa_upgrade(spa, newversion);
2423 	VERIFY3U(spa_version(spa), >, version);
2424 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2425 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2426 	spa_close(spa, FTAG);
2427 
2428 	strfree(name);
2429 	VERIFY0(mutex_unlock(&ztest_vdev_lock));
2430 }
2431 
2432 static vdev_t *
2433 vdev_lookup_by_path(vdev_t *vd, const char *path)
2434 {
2435 	vdev_t *mvd;
2436 
2437 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2438 		return (vd);
2439 
2440 	for (int c = 0; c < vd->vdev_children; c++)
2441 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2442 		    NULL)
2443 			return (mvd);
2444 
2445 	return (NULL);
2446 }
2447 
2448 /*
2449  * Find the first available hole which can be used as a top-level.
2450  */
2451 int
2452 find_vdev_hole(spa_t *spa)
2453 {
2454 	vdev_t *rvd = spa->spa_root_vdev;
2455 	int c;
2456 
2457 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2458 
2459 	for (c = 0; c < rvd->vdev_children; c++) {
2460 		vdev_t *cvd = rvd->vdev_child[c];
2461 
2462 		if (cvd->vdev_ishole)
2463 			break;
2464 	}
2465 	return (c);
2466 }
2467 
2468 /*
2469  * Verify that vdev_add() works as expected.
2470  */
2471 /* ARGSUSED */
2472 void
2473 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2474 {
2475 	ztest_shared_t *zs = ztest_shared;
2476 	spa_t *spa = ztest_spa;
2477 	uint64_t leaves;
2478 	uint64_t guid;
2479 	nvlist_t *nvroot;
2480 	int error;
2481 
2482 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2483 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2484 
2485 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2486 
2487 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2488 
2489 	/*
2490 	 * If we have slogs then remove them 1/4 of the time.
2491 	 */
2492 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2493 		/*
2494 		 * Grab the guid from the head of the log class rotor.
2495 		 */
2496 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2497 
2498 		spa_config_exit(spa, SCL_VDEV, FTAG);
2499 
2500 		/*
2501 		 * We have to grab the zs_name_lock as writer to
2502 		 * prevent a race between removing a slog (dmu_objset_find)
2503 		 * and destroying a dataset. Removing the slog will
2504 		 * grab a reference on the dataset which may cause
2505 		 * dmu_objset_destroy() to fail with EBUSY thus
2506 		 * leaving the dataset in an inconsistent state.
2507 		 */
2508 		VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2509 		error = spa_vdev_remove(spa, guid, B_FALSE);
2510 		VERIFY(rw_unlock(&ztest_name_lock) == 0);
2511 
2512 		if (error && error != EEXIST)
2513 			fatal(0, "spa_vdev_remove() = %d", error);
2514 	} else {
2515 		spa_config_exit(spa, SCL_VDEV, FTAG);
2516 
2517 		/*
2518 		 * Make 1/4 of the devices be log devices.
2519 		 */
2520 		nvroot = make_vdev_root(NULL, NULL, NULL,
2521 		    ztest_opts.zo_vdev_size, 0,
2522 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2523 		    zs->zs_mirrors, 1);
2524 
2525 		error = spa_vdev_add(spa, nvroot);
2526 		nvlist_free(nvroot);
2527 
2528 		if (error == ENOSPC)
2529 			ztest_record_enospc("spa_vdev_add");
2530 		else if (error != 0)
2531 			fatal(0, "spa_vdev_add() = %d", error);
2532 	}
2533 
2534 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2535 }
2536 
2537 /*
2538  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2539  */
2540 /* ARGSUSED */
2541 void
2542 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2543 {
2544 	ztest_shared_t *zs = ztest_shared;
2545 	spa_t *spa = ztest_spa;
2546 	vdev_t *rvd = spa->spa_root_vdev;
2547 	spa_aux_vdev_t *sav;
2548 	char *aux;
2549 	uint64_t guid = 0;
2550 	int error;
2551 
2552 	if (ztest_random(2) == 0) {
2553 		sav = &spa->spa_spares;
2554 		aux = ZPOOL_CONFIG_SPARES;
2555 	} else {
2556 		sav = &spa->spa_l2cache;
2557 		aux = ZPOOL_CONFIG_L2CACHE;
2558 	}
2559 
2560 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2561 
2562 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2563 
2564 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2565 		/*
2566 		 * Pick a random device to remove.
2567 		 */
2568 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2569 	} else {
2570 		/*
2571 		 * Find an unused device we can add.
2572 		 */
2573 		zs->zs_vdev_aux = 0;
2574 		for (;;) {
2575 			char path[MAXPATHLEN];
2576 			int c;
2577 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2578 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2579 			    zs->zs_vdev_aux);
2580 			for (c = 0; c < sav->sav_count; c++)
2581 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2582 				    path) == 0)
2583 					break;
2584 			if (c == sav->sav_count &&
2585 			    vdev_lookup_by_path(rvd, path) == NULL)
2586 				break;
2587 			zs->zs_vdev_aux++;
2588 		}
2589 	}
2590 
2591 	spa_config_exit(spa, SCL_VDEV, FTAG);
2592 
2593 	if (guid == 0) {
2594 		/*
2595 		 * Add a new device.
2596 		 */
2597 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2598 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2599 		error = spa_vdev_add(spa, nvroot);
2600 		if (error != 0)
2601 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2602 		nvlist_free(nvroot);
2603 	} else {
2604 		/*
2605 		 * Remove an existing device.  Sometimes, dirty its
2606 		 * vdev state first to make sure we handle removal
2607 		 * of devices that have pending state changes.
2608 		 */
2609 		if (ztest_random(2) == 0)
2610 			(void) vdev_online(spa, guid, 0, NULL);
2611 
2612 		error = spa_vdev_remove(spa, guid, B_FALSE);
2613 		if (error != 0 && error != EBUSY)
2614 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2615 	}
2616 
2617 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2618 }
2619 
2620 /*
2621  * split a pool if it has mirror tlvdevs
2622  */
2623 /* ARGSUSED */
2624 void
2625 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2626 {
2627 	ztest_shared_t *zs = ztest_shared;
2628 	spa_t *spa = ztest_spa;
2629 	vdev_t *rvd = spa->spa_root_vdev;
2630 	nvlist_t *tree, **child, *config, *split, **schild;
2631 	uint_t c, children, schildren = 0, lastlogid = 0;
2632 	int error = 0;
2633 
2634 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2635 
2636 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2637 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2638 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2639 		return;
2640 	}
2641 
2642 	/* clean up the old pool, if any */
2643 	(void) spa_destroy("splitp");
2644 
2645 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2646 
2647 	/* generate a config from the existing config */
2648 	mutex_enter(&spa->spa_props_lock);
2649 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2650 	    &tree) == 0);
2651 	mutex_exit(&spa->spa_props_lock);
2652 
2653 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2654 	    &children) == 0);
2655 
2656 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2657 	for (c = 0; c < children; c++) {
2658 		vdev_t *tvd = rvd->vdev_child[c];
2659 		nvlist_t **mchild;
2660 		uint_t mchildren;
2661 
2662 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2663 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2664 			    0) == 0);
2665 			VERIFY(nvlist_add_string(schild[schildren],
2666 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2667 			VERIFY(nvlist_add_uint64(schild[schildren],
2668 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2669 			if (lastlogid == 0)
2670 				lastlogid = schildren;
2671 			++schildren;
2672 			continue;
2673 		}
2674 		lastlogid = 0;
2675 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2676 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2677 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2678 	}
2679 
2680 	/* OK, create a config that can be used to split */
2681 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2682 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2683 	    VDEV_TYPE_ROOT) == 0);
2684 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2685 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2686 
2687 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2688 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2689 
2690 	for (c = 0; c < schildren; c++)
2691 		nvlist_free(schild[c]);
2692 	free(schild);
2693 	nvlist_free(split);
2694 
2695 	spa_config_exit(spa, SCL_VDEV, FTAG);
2696 
2697 	(void) rw_wrlock(&ztest_name_lock);
2698 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2699 	(void) rw_unlock(&ztest_name_lock);
2700 
2701 	nvlist_free(config);
2702 
2703 	if (error == 0) {
2704 		(void) printf("successful split - results:\n");
2705 		mutex_enter(&spa_namespace_lock);
2706 		show_pool_stats(spa);
2707 		show_pool_stats(spa_lookup("splitp"));
2708 		mutex_exit(&spa_namespace_lock);
2709 		++zs->zs_splits;
2710 		--zs->zs_mirrors;
2711 	}
2712 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2713 
2714 }
2715 
2716 /*
2717  * Verify that we can attach and detach devices.
2718  */
2719 /* ARGSUSED */
2720 void
2721 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2722 {
2723 	ztest_shared_t *zs = ztest_shared;
2724 	spa_t *spa = ztest_spa;
2725 	spa_aux_vdev_t *sav = &spa->spa_spares;
2726 	vdev_t *rvd = spa->spa_root_vdev;
2727 	vdev_t *oldvd, *newvd, *pvd;
2728 	nvlist_t *root;
2729 	uint64_t leaves;
2730 	uint64_t leaf, top;
2731 	uint64_t ashift = ztest_get_ashift();
2732 	uint64_t oldguid, pguid;
2733 	size_t oldsize, newsize;
2734 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2735 	int replacing;
2736 	int oldvd_has_siblings = B_FALSE;
2737 	int newvd_is_spare = B_FALSE;
2738 	int oldvd_is_log;
2739 	int error, expected_error;
2740 
2741 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2742 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2743 
2744 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2745 
2746 	/*
2747 	 * Decide whether to do an attach or a replace.
2748 	 */
2749 	replacing = ztest_random(2);
2750 
2751 	/*
2752 	 * Pick a random top-level vdev.
2753 	 */
2754 	top = ztest_random_vdev_top(spa, B_TRUE);
2755 
2756 	/*
2757 	 * Pick a random leaf within it.
2758 	 */
2759 	leaf = ztest_random(leaves);
2760 
2761 	/*
2762 	 * Locate this vdev.
2763 	 */
2764 	oldvd = rvd->vdev_child[top];
2765 	if (zs->zs_mirrors >= 1) {
2766 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2767 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2768 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2769 	}
2770 	if (ztest_opts.zo_raidz > 1) {
2771 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2772 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2773 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2774 	}
2775 
2776 	/*
2777 	 * If we're already doing an attach or replace, oldvd may be a
2778 	 * mirror vdev -- in which case, pick a random child.
2779 	 */
2780 	while (oldvd->vdev_children != 0) {
2781 		oldvd_has_siblings = B_TRUE;
2782 		ASSERT(oldvd->vdev_children >= 2);
2783 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2784 	}
2785 
2786 	oldguid = oldvd->vdev_guid;
2787 	oldsize = vdev_get_min_asize(oldvd);
2788 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2789 	(void) strcpy(oldpath, oldvd->vdev_path);
2790 	pvd = oldvd->vdev_parent;
2791 	pguid = pvd->vdev_guid;
2792 
2793 	/*
2794 	 * If oldvd has siblings, then half of the time, detach it.
2795 	 */
2796 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2797 		spa_config_exit(spa, SCL_VDEV, FTAG);
2798 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2799 		if (error != 0 && error != ENODEV && error != EBUSY &&
2800 		    error != ENOTSUP)
2801 			fatal(0, "detach (%s) returned %d", oldpath, error);
2802 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2803 		return;
2804 	}
2805 
2806 	/*
2807 	 * For the new vdev, choose with equal probability between the two
2808 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2809 	 */
2810 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2811 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2812 		newvd_is_spare = B_TRUE;
2813 		(void) strcpy(newpath, newvd->vdev_path);
2814 	} else {
2815 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2816 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2817 		    top * leaves + leaf);
2818 		if (ztest_random(2) == 0)
2819 			newpath[strlen(newpath) - 1] = 'b';
2820 		newvd = vdev_lookup_by_path(rvd, newpath);
2821 	}
2822 
2823 	if (newvd) {
2824 		newsize = vdev_get_min_asize(newvd);
2825 	} else {
2826 		/*
2827 		 * Make newsize a little bigger or smaller than oldsize.
2828 		 * If it's smaller, the attach should fail.
2829 		 * If it's larger, and we're doing a replace,
2830 		 * we should get dynamic LUN growth when we're done.
2831 		 */
2832 		newsize = 10 * oldsize / (9 + ztest_random(3));
2833 	}
2834 
2835 	/*
2836 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2837 	 * unless it's a replace; in that case any non-replacing parent is OK.
2838 	 *
2839 	 * If newvd is already part of the pool, it should fail with EBUSY.
2840 	 *
2841 	 * If newvd is too small, it should fail with EOVERFLOW.
2842 	 */
2843 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2844 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2845 	    pvd->vdev_ops == &vdev_replacing_ops ||
2846 	    pvd->vdev_ops == &vdev_spare_ops))
2847 		expected_error = ENOTSUP;
2848 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2849 		expected_error = ENOTSUP;
2850 	else if (newvd == oldvd)
2851 		expected_error = replacing ? 0 : EBUSY;
2852 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2853 		expected_error = EBUSY;
2854 	else if (newsize < oldsize)
2855 		expected_error = EOVERFLOW;
2856 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2857 		expected_error = EDOM;
2858 	else
2859 		expected_error = 0;
2860 
2861 	spa_config_exit(spa, SCL_VDEV, FTAG);
2862 
2863 	/*
2864 	 * Build the nvlist describing newpath.
2865 	 */
2866 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2867 	    ashift, 0, 0, 0, 1);
2868 
2869 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2870 
2871 	nvlist_free(root);
2872 
2873 	/*
2874 	 * If our parent was the replacing vdev, but the replace completed,
2875 	 * then instead of failing with ENOTSUP we may either succeed,
2876 	 * fail with ENODEV, or fail with EOVERFLOW.
2877 	 */
2878 	if (expected_error == ENOTSUP &&
2879 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2880 		expected_error = error;
2881 
2882 	/*
2883 	 * If someone grew the LUN, the replacement may be too small.
2884 	 */
2885 	if (error == EOVERFLOW || error == EBUSY)
2886 		expected_error = error;
2887 
2888 	/* XXX workaround 6690467 */
2889 	if (error != expected_error && expected_error != EBUSY) {
2890 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2891 		    "returned %d, expected %d",
2892 		    oldpath, (longlong_t)oldsize, newpath,
2893 		    (longlong_t)newsize, replacing, error, expected_error);
2894 	}
2895 
2896 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2897 }
2898 
2899 /*
2900  * Callback function which expands the physical size of the vdev.
2901  */
2902 vdev_t *
2903 grow_vdev(vdev_t *vd, void *arg)
2904 {
2905 	spa_t *spa = vd->vdev_spa;
2906 	size_t *newsize = arg;
2907 	size_t fsize;
2908 	int fd;
2909 
2910 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2911 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2912 
2913 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2914 		return (vd);
2915 
2916 	fsize = lseek(fd, 0, SEEK_END);
2917 	(void) ftruncate(fd, *newsize);
2918 
2919 	if (ztest_opts.zo_verbose >= 6) {
2920 		(void) printf("%s grew from %lu to %lu bytes\n",
2921 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2922 	}
2923 	(void) close(fd);
2924 	return (NULL);
2925 }
2926 
2927 /*
2928  * Callback function which expands a given vdev by calling vdev_online().
2929  */
2930 /* ARGSUSED */
2931 vdev_t *
2932 online_vdev(vdev_t *vd, void *arg)
2933 {
2934 	spa_t *spa = vd->vdev_spa;
2935 	vdev_t *tvd = vd->vdev_top;
2936 	uint64_t guid = vd->vdev_guid;
2937 	uint64_t generation = spa->spa_config_generation + 1;
2938 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2939 	int error;
2940 
2941 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2942 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2943 
2944 	/* Calling vdev_online will initialize the new metaslabs */
2945 	spa_config_exit(spa, SCL_STATE, spa);
2946 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2947 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2948 
2949 	/*
2950 	 * If vdev_online returned an error or the underlying vdev_open
2951 	 * failed then we abort the expand. The only way to know that
2952 	 * vdev_open fails is by checking the returned newstate.
2953 	 */
2954 	if (error || newstate != VDEV_STATE_HEALTHY) {
2955 		if (ztest_opts.zo_verbose >= 5) {
2956 			(void) printf("Unable to expand vdev, state %llu, "
2957 			    "error %d\n", (u_longlong_t)newstate, error);
2958 		}
2959 		return (vd);
2960 	}
2961 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2962 
2963 	/*
2964 	 * Since we dropped the lock we need to ensure that we're
2965 	 * still talking to the original vdev. It's possible this
2966 	 * vdev may have been detached/replaced while we were
2967 	 * trying to online it.
2968 	 */
2969 	if (generation != spa->spa_config_generation) {
2970 		if (ztest_opts.zo_verbose >= 5) {
2971 			(void) printf("vdev configuration has changed, "
2972 			    "guid %llu, state %llu, expected gen %llu, "
2973 			    "got gen %llu\n",
2974 			    (u_longlong_t)guid,
2975 			    (u_longlong_t)tvd->vdev_state,
2976 			    (u_longlong_t)generation,
2977 			    (u_longlong_t)spa->spa_config_generation);
2978 		}
2979 		return (vd);
2980 	}
2981 	return (NULL);
2982 }
2983 
2984 /*
2985  * Traverse the vdev tree calling the supplied function.
2986  * We continue to walk the tree until we either have walked all
2987  * children or we receive a non-NULL return from the callback.
2988  * If a NULL callback is passed, then we just return back the first
2989  * leaf vdev we encounter.
2990  */
2991 vdev_t *
2992 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
2993 {
2994 	if (vd->vdev_ops->vdev_op_leaf) {
2995 		if (func == NULL)
2996 			return (vd);
2997 		else
2998 			return (func(vd, arg));
2999 	}
3000 
3001 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3002 		vdev_t *cvd = vd->vdev_child[c];
3003 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3004 			return (cvd);
3005 	}
3006 	return (NULL);
3007 }
3008 
3009 /*
3010  * Verify that dynamic LUN growth works as expected.
3011  */
3012 /* ARGSUSED */
3013 void
3014 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3015 {
3016 	spa_t *spa = ztest_spa;
3017 	vdev_t *vd, *tvd;
3018 	metaslab_class_t *mc;
3019 	metaslab_group_t *mg;
3020 	size_t psize, newsize;
3021 	uint64_t top;
3022 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3023 
3024 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3025 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3026 
3027 	top = ztest_random_vdev_top(spa, B_TRUE);
3028 
3029 	tvd = spa->spa_root_vdev->vdev_child[top];
3030 	mg = tvd->vdev_mg;
3031 	mc = mg->mg_class;
3032 	old_ms_count = tvd->vdev_ms_count;
3033 	old_class_space = metaslab_class_get_space(mc);
3034 
3035 	/*
3036 	 * Determine the size of the first leaf vdev associated with
3037 	 * our top-level device.
3038 	 */
3039 	vd = vdev_walk_tree(tvd, NULL, NULL);
3040 	ASSERT3P(vd, !=, NULL);
3041 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3042 
3043 	psize = vd->vdev_psize;
3044 
3045 	/*
3046 	 * We only try to expand the vdev if it's healthy, less than 4x its
3047 	 * original size, and it has a valid psize.
3048 	 */
3049 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3050 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3051 		spa_config_exit(spa, SCL_STATE, spa);
3052 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3053 		return;
3054 	}
3055 	ASSERT(psize > 0);
3056 	newsize = psize + psize / 8;
3057 	ASSERT3U(newsize, >, psize);
3058 
3059 	if (ztest_opts.zo_verbose >= 6) {
3060 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3061 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3062 	}
3063 
3064 	/*
3065 	 * Growing the vdev is a two step process:
3066 	 *	1). expand the physical size (i.e. relabel)
3067 	 *	2). online the vdev to create the new metaslabs
3068 	 */
3069 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3070 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3071 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3072 		if (ztest_opts.zo_verbose >= 5) {
3073 			(void) printf("Could not expand LUN because "
3074 			    "the vdev configuration changed.\n");
3075 		}
3076 		spa_config_exit(spa, SCL_STATE, spa);
3077 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3078 		return;
3079 	}
3080 
3081 	spa_config_exit(spa, SCL_STATE, spa);
3082 
3083 	/*
3084 	 * Expanding the LUN will update the config asynchronously,
3085 	 * thus we must wait for the async thread to complete any
3086 	 * pending tasks before proceeding.
3087 	 */
3088 	for (;;) {
3089 		boolean_t done;
3090 		mutex_enter(&spa->spa_async_lock);
3091 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3092 		mutex_exit(&spa->spa_async_lock);
3093 		if (done)
3094 			break;
3095 		txg_wait_synced(spa_get_dsl(spa), 0);
3096 		(void) poll(NULL, 0, 100);
3097 	}
3098 
3099 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3100 
3101 	tvd = spa->spa_root_vdev->vdev_child[top];
3102 	new_ms_count = tvd->vdev_ms_count;
3103 	new_class_space = metaslab_class_get_space(mc);
3104 
3105 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3106 		if (ztest_opts.zo_verbose >= 5) {
3107 			(void) printf("Could not verify LUN expansion due to "
3108 			    "intervening vdev offline or remove.\n");
3109 		}
3110 		spa_config_exit(spa, SCL_STATE, spa);
3111 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3112 		return;
3113 	}
3114 
3115 	/*
3116 	 * Make sure we were able to grow the vdev.
3117 	 */
3118 	if (new_ms_count <= old_ms_count)
3119 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3120 		    old_ms_count, new_ms_count);
3121 
3122 	/*
3123 	 * Make sure we were able to grow the pool.
3124 	 */
3125 	if (new_class_space <= old_class_space)
3126 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3127 		    old_class_space, new_class_space);
3128 
3129 	if (ztest_opts.zo_verbose >= 5) {
3130 		char oldnumbuf[6], newnumbuf[6];
3131 
3132 		nicenum(old_class_space, oldnumbuf);
3133 		nicenum(new_class_space, newnumbuf);
3134 		(void) printf("%s grew from %s to %s\n",
3135 		    spa->spa_name, oldnumbuf, newnumbuf);
3136 	}
3137 
3138 	spa_config_exit(spa, SCL_STATE, spa);
3139 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3140 }
3141 
3142 /*
3143  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3144  */
3145 /* ARGSUSED */
3146 static void
3147 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3148 {
3149 	/*
3150 	 * Create the objects common to all ztest datasets.
3151 	 */
3152 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3153 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3154 }
3155 
3156 static int
3157 ztest_dataset_create(char *dsname)
3158 {
3159 	uint64_t zilset = ztest_random(100);
3160 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3161 	    ztest_objset_create_cb, NULL);
3162 
3163 	if (err || zilset < 80)
3164 		return (err);
3165 
3166 	if (ztest_opts.zo_verbose >= 6)
3167 		(void) printf("Setting dataset %s to sync always\n", dsname);
3168 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3169 	    ZFS_SYNC_ALWAYS, B_FALSE));
3170 }
3171 
3172 /* ARGSUSED */
3173 static int
3174 ztest_objset_destroy_cb(const char *name, void *arg)
3175 {
3176 	objset_t *os;
3177 	dmu_object_info_t doi;
3178 	int error;
3179 
3180 	/*
3181 	 * Verify that the dataset contains a directory object.
3182 	 */
3183 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3184 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3185 	if (error != ENOENT) {
3186 		/* We could have crashed in the middle of destroying it */
3187 		ASSERT0(error);
3188 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3189 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3190 	}
3191 	dmu_objset_disown(os, FTAG);
3192 
3193 	/*
3194 	 * Destroy the dataset.
3195 	 */
3196 	if (strchr(name, '@') != NULL) {
3197 		VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3198 	} else {
3199 		VERIFY0(dsl_destroy_head(name));
3200 	}
3201 	return (0);
3202 }
3203 
3204 static boolean_t
3205 ztest_snapshot_create(char *osname, uint64_t id)
3206 {
3207 	char snapname[MAXNAMELEN];
3208 	int error;
3209 
3210 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3211 
3212 	error = dmu_objset_snapshot_one(osname, snapname);
3213 	if (error == ENOSPC) {
3214 		ztest_record_enospc(FTAG);
3215 		return (B_FALSE);
3216 	}
3217 	if (error != 0 && error != EEXIST) {
3218 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3219 		    snapname, error);
3220 	}
3221 	return (B_TRUE);
3222 }
3223 
3224 static boolean_t
3225 ztest_snapshot_destroy(char *osname, uint64_t id)
3226 {
3227 	char snapname[MAXNAMELEN];
3228 	int error;
3229 
3230 	(void) snprintf(snapname, MAXNAMELEN, "%s@%llu", osname,
3231 	    (u_longlong_t)id);
3232 
3233 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3234 	if (error != 0 && error != ENOENT)
3235 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3236 	return (B_TRUE);
3237 }
3238 
3239 /* ARGSUSED */
3240 void
3241 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3242 {
3243 	ztest_ds_t zdtmp;
3244 	int iters;
3245 	int error;
3246 	objset_t *os, *os2;
3247 	char name[MAXNAMELEN];
3248 	zilog_t *zilog;
3249 
3250 	(void) rw_rdlock(&ztest_name_lock);
3251 
3252 	(void) snprintf(name, MAXNAMELEN, "%s/temp_%llu",
3253 	    ztest_opts.zo_pool, (u_longlong_t)id);
3254 
3255 	/*
3256 	 * If this dataset exists from a previous run, process its replay log
3257 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3258 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3259 	 */
3260 	if (ztest_random(2) == 0 &&
3261 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3262 		ztest_zd_init(&zdtmp, NULL, os);
3263 		zil_replay(os, &zdtmp, ztest_replay_vector);
3264 		ztest_zd_fini(&zdtmp);
3265 		dmu_objset_disown(os, FTAG);
3266 	}
3267 
3268 	/*
3269 	 * There may be an old instance of the dataset we're about to
3270 	 * create lying around from a previous run.  If so, destroy it
3271 	 * and all of its snapshots.
3272 	 */
3273 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3274 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3275 
3276 	/*
3277 	 * Verify that the destroyed dataset is no longer in the namespace.
3278 	 */
3279 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3280 	    FTAG, &os));
3281 
3282 	/*
3283 	 * Verify that we can create a new dataset.
3284 	 */
3285 	error = ztest_dataset_create(name);
3286 	if (error) {
3287 		if (error == ENOSPC) {
3288 			ztest_record_enospc(FTAG);
3289 			(void) rw_unlock(&ztest_name_lock);
3290 			return;
3291 		}
3292 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3293 	}
3294 
3295 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3296 
3297 	ztest_zd_init(&zdtmp, NULL, os);
3298 
3299 	/*
3300 	 * Open the intent log for it.
3301 	 */
3302 	zilog = zil_open(os, ztest_get_data);
3303 
3304 	/*
3305 	 * Put some objects in there, do a little I/O to them,
3306 	 * and randomly take a couple of snapshots along the way.
3307 	 */
3308 	iters = ztest_random(5);
3309 	for (int i = 0; i < iters; i++) {
3310 		ztest_dmu_object_alloc_free(&zdtmp, id);
3311 		if (ztest_random(iters) == 0)
3312 			(void) ztest_snapshot_create(name, i);
3313 	}
3314 
3315 	/*
3316 	 * Verify that we cannot create an existing dataset.
3317 	 */
3318 	VERIFY3U(EEXIST, ==,
3319 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3320 
3321 	/*
3322 	 * Verify that we can hold an objset that is also owned.
3323 	 */
3324 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3325 	dmu_objset_rele(os2, FTAG);
3326 
3327 	/*
3328 	 * Verify that we cannot own an objset that is already owned.
3329 	 */
3330 	VERIFY3U(EBUSY, ==,
3331 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3332 
3333 	zil_close(zilog);
3334 	dmu_objset_disown(os, FTAG);
3335 	ztest_zd_fini(&zdtmp);
3336 
3337 	(void) rw_unlock(&ztest_name_lock);
3338 }
3339 
3340 /*
3341  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3342  */
3343 void
3344 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3345 {
3346 	(void) rw_rdlock(&ztest_name_lock);
3347 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3348 	(void) ztest_snapshot_create(zd->zd_name, id);
3349 	(void) rw_unlock(&ztest_name_lock);
3350 }
3351 
3352 /*
3353  * Cleanup non-standard snapshots and clones.
3354  */
3355 void
3356 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3357 {
3358 	char snap1name[MAXNAMELEN];
3359 	char clone1name[MAXNAMELEN];
3360 	char snap2name[MAXNAMELEN];
3361 	char clone2name[MAXNAMELEN];
3362 	char snap3name[MAXNAMELEN];
3363 	int error;
3364 
3365 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3366 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3367 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3368 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3369 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3370 
3371 	error = dsl_destroy_head(clone2name);
3372 	if (error && error != ENOENT)
3373 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3374 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3375 	if (error && error != ENOENT)
3376 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3377 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3378 	if (error && error != ENOENT)
3379 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3380 	error = dsl_destroy_head(clone1name);
3381 	if (error && error != ENOENT)
3382 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3383 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3384 	if (error && error != ENOENT)
3385 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3386 }
3387 
3388 /*
3389  * Verify dsl_dataset_promote handles EBUSY
3390  */
3391 void
3392 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3393 {
3394 	objset_t *os;
3395 	char snap1name[MAXNAMELEN];
3396 	char clone1name[MAXNAMELEN];
3397 	char snap2name[MAXNAMELEN];
3398 	char clone2name[MAXNAMELEN];
3399 	char snap3name[MAXNAMELEN];
3400 	char *osname = zd->zd_name;
3401 	int error;
3402 
3403 	(void) rw_rdlock(&ztest_name_lock);
3404 
3405 	ztest_dsl_dataset_cleanup(osname, id);
3406 
3407 	(void) snprintf(snap1name, MAXNAMELEN, "%s@s1_%llu", osname, id);
3408 	(void) snprintf(clone1name, MAXNAMELEN, "%s/c1_%llu", osname, id);
3409 	(void) snprintf(snap2name, MAXNAMELEN, "%s@s2_%llu", clone1name, id);
3410 	(void) snprintf(clone2name, MAXNAMELEN, "%s/c2_%llu", osname, id);
3411 	(void) snprintf(snap3name, MAXNAMELEN, "%s@s3_%llu", clone1name, id);
3412 
3413 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3414 	if (error && error != EEXIST) {
3415 		if (error == ENOSPC) {
3416 			ztest_record_enospc(FTAG);
3417 			goto out;
3418 		}
3419 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3420 	}
3421 
3422 	error = dmu_objset_clone(clone1name, snap1name);
3423 	if (error) {
3424 		if (error == ENOSPC) {
3425 			ztest_record_enospc(FTAG);
3426 			goto out;
3427 		}
3428 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3429 	}
3430 
3431 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3432 	if (error && error != EEXIST) {
3433 		if (error == ENOSPC) {
3434 			ztest_record_enospc(FTAG);
3435 			goto out;
3436 		}
3437 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3438 	}
3439 
3440 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3441 	if (error && error != EEXIST) {
3442 		if (error == ENOSPC) {
3443 			ztest_record_enospc(FTAG);
3444 			goto out;
3445 		}
3446 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3447 	}
3448 
3449 	error = dmu_objset_clone(clone2name, snap3name);
3450 	if (error) {
3451 		if (error == ENOSPC) {
3452 			ztest_record_enospc(FTAG);
3453 			goto out;
3454 		}
3455 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3456 	}
3457 
3458 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3459 	if (error)
3460 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3461 	error = dsl_dataset_promote(clone2name, NULL);
3462 	if (error != EBUSY)
3463 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3464 		    error);
3465 	dmu_objset_disown(os, FTAG);
3466 
3467 out:
3468 	ztest_dsl_dataset_cleanup(osname, id);
3469 
3470 	(void) rw_unlock(&ztest_name_lock);
3471 }
3472 
3473 /*
3474  * Verify that dmu_object_{alloc,free} work as expected.
3475  */
3476 void
3477 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3478 {
3479 	ztest_od_t od[4];
3480 	int batchsize = sizeof (od) / sizeof (od[0]);
3481 
3482 	for (int b = 0; b < batchsize; b++)
3483 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3484 
3485 	/*
3486 	 * Destroy the previous batch of objects, create a new batch,
3487 	 * and do some I/O on the new objects.
3488 	 */
3489 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3490 		return;
3491 
3492 	while (ztest_random(4 * batchsize) != 0)
3493 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3494 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3495 }
3496 
3497 /*
3498  * Verify that dmu_{read,write} work as expected.
3499  */
3500 void
3501 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3502 {
3503 	objset_t *os = zd->zd_os;
3504 	ztest_od_t od[2];
3505 	dmu_tx_t *tx;
3506 	int i, freeit, error;
3507 	uint64_t n, s, txg;
3508 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3509 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3510 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3511 	uint64_t regions = 997;
3512 	uint64_t stride = 123456789ULL;
3513 	uint64_t width = 40;
3514 	int free_percent = 5;
3515 
3516 	/*
3517 	 * This test uses two objects, packobj and bigobj, that are always
3518 	 * updated together (i.e. in the same tx) so that their contents are
3519 	 * in sync and can be compared.  Their contents relate to each other
3520 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3521 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3522 	 * for any index n, there are three bufwads that should be identical:
3523 	 *
3524 	 *	packobj, at offset n * sizeof (bufwad_t)
3525 	 *	bigobj, at the head of the nth chunk
3526 	 *	bigobj, at the tail of the nth chunk
3527 	 *
3528 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3529 	 * and it doesn't have any relation to the object blocksize.
3530 	 * The only requirement is that it can hold at least two bufwads.
3531 	 *
3532 	 * Normally, we write the bufwad to each of these locations.
3533 	 * However, free_percent of the time we instead write zeroes to
3534 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3535 	 * bigobj to packobj, we can verify that the DMU is correctly
3536 	 * tracking which parts of an object are allocated and free,
3537 	 * and that the contents of the allocated blocks are correct.
3538 	 */
3539 
3540 	/*
3541 	 * Read the directory info.  If it's the first time, set things up.
3542 	 */
3543 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3544 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3545 
3546 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3547 		return;
3548 
3549 	bigobj = od[0].od_object;
3550 	packobj = od[1].od_object;
3551 	chunksize = od[0].od_gen;
3552 	ASSERT(chunksize == od[1].od_gen);
3553 
3554 	/*
3555 	 * Prefetch a random chunk of the big object.
3556 	 * Our aim here is to get some async reads in flight
3557 	 * for blocks that we may free below; the DMU should
3558 	 * handle this race correctly.
3559 	 */
3560 	n = ztest_random(regions) * stride + ztest_random(width);
3561 	s = 1 + ztest_random(2 * width - 1);
3562 	dmu_prefetch(os, bigobj, n * chunksize, s * chunksize);
3563 
3564 	/*
3565 	 * Pick a random index and compute the offsets into packobj and bigobj.
3566 	 */
3567 	n = ztest_random(regions) * stride + ztest_random(width);
3568 	s = 1 + ztest_random(width - 1);
3569 
3570 	packoff = n * sizeof (bufwad_t);
3571 	packsize = s * sizeof (bufwad_t);
3572 
3573 	bigoff = n * chunksize;
3574 	bigsize = s * chunksize;
3575 
3576 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3577 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3578 
3579 	/*
3580 	 * free_percent of the time, free a range of bigobj rather than
3581 	 * overwriting it.
3582 	 */
3583 	freeit = (ztest_random(100) < free_percent);
3584 
3585 	/*
3586 	 * Read the current contents of our objects.
3587 	 */
3588 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3589 	    DMU_READ_PREFETCH);
3590 	ASSERT0(error);
3591 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3592 	    DMU_READ_PREFETCH);
3593 	ASSERT0(error);
3594 
3595 	/*
3596 	 * Get a tx for the mods to both packobj and bigobj.
3597 	 */
3598 	tx = dmu_tx_create(os);
3599 
3600 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3601 
3602 	if (freeit)
3603 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3604 	else
3605 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3606 
3607 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3608 	if (txg == 0) {
3609 		umem_free(packbuf, packsize);
3610 		umem_free(bigbuf, bigsize);
3611 		return;
3612 	}
3613 
3614 	dmu_object_set_checksum(os, bigobj,
3615 	    (enum zio_checksum)ztest_random_dsl_prop(ZFS_PROP_CHECKSUM), tx);
3616 
3617 	dmu_object_set_compress(os, bigobj,
3618 	    (enum zio_compress)ztest_random_dsl_prop(ZFS_PROP_COMPRESSION), tx);
3619 
3620 	/*
3621 	 * For each index from n to n + s, verify that the existing bufwad
3622 	 * in packobj matches the bufwads at the head and tail of the
3623 	 * corresponding chunk in bigobj.  Then update all three bufwads
3624 	 * with the new values we want to write out.
3625 	 */
3626 	for (i = 0; i < s; i++) {
3627 		/* LINTED */
3628 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3629 		/* LINTED */
3630 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3631 		/* LINTED */
3632 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3633 
3634 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3635 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3636 
3637 		if (pack->bw_txg > txg)
3638 			fatal(0, "future leak: got %llx, open txg is %llx",
3639 			    pack->bw_txg, txg);
3640 
3641 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3642 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3643 			    pack->bw_index, n, i);
3644 
3645 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3646 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3647 
3648 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3649 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3650 
3651 		if (freeit) {
3652 			bzero(pack, sizeof (bufwad_t));
3653 		} else {
3654 			pack->bw_index = n + i;
3655 			pack->bw_txg = txg;
3656 			pack->bw_data = 1 + ztest_random(-2ULL);
3657 		}
3658 		*bigH = *pack;
3659 		*bigT = *pack;
3660 	}
3661 
3662 	/*
3663 	 * We've verified all the old bufwads, and made new ones.
3664 	 * Now write them out.
3665 	 */
3666 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3667 
3668 	if (freeit) {
3669 		if (ztest_opts.zo_verbose >= 7) {
3670 			(void) printf("freeing offset %llx size %llx"
3671 			    " txg %llx\n",
3672 			    (u_longlong_t)bigoff,
3673 			    (u_longlong_t)bigsize,
3674 			    (u_longlong_t)txg);
3675 		}
3676 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3677 	} else {
3678 		if (ztest_opts.zo_verbose >= 7) {
3679 			(void) printf("writing offset %llx size %llx"
3680 			    " txg %llx\n",
3681 			    (u_longlong_t)bigoff,
3682 			    (u_longlong_t)bigsize,
3683 			    (u_longlong_t)txg);
3684 		}
3685 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3686 	}
3687 
3688 	dmu_tx_commit(tx);
3689 
3690 	/*
3691 	 * Sanity check the stuff we just wrote.
3692 	 */
3693 	{
3694 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3695 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3696 
3697 		VERIFY(0 == dmu_read(os, packobj, packoff,
3698 		    packsize, packcheck, DMU_READ_PREFETCH));
3699 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3700 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3701 
3702 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3703 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3704 
3705 		umem_free(packcheck, packsize);
3706 		umem_free(bigcheck, bigsize);
3707 	}
3708 
3709 	umem_free(packbuf, packsize);
3710 	umem_free(bigbuf, bigsize);
3711 }
3712 
3713 void
3714 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3715     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3716 {
3717 	uint64_t i;
3718 	bufwad_t *pack;
3719 	bufwad_t *bigH;
3720 	bufwad_t *bigT;
3721 
3722 	/*
3723 	 * For each index from n to n + s, verify that the existing bufwad
3724 	 * in packobj matches the bufwads at the head and tail of the
3725 	 * corresponding chunk in bigobj.  Then update all three bufwads
3726 	 * with the new values we want to write out.
3727 	 */
3728 	for (i = 0; i < s; i++) {
3729 		/* LINTED */
3730 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3731 		/* LINTED */
3732 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3733 		/* LINTED */
3734 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3735 
3736 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3737 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3738 
3739 		if (pack->bw_txg > txg)
3740 			fatal(0, "future leak: got %llx, open txg is %llx",
3741 			    pack->bw_txg, txg);
3742 
3743 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3744 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3745 			    pack->bw_index, n, i);
3746 
3747 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3748 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3749 
3750 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3751 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3752 
3753 		pack->bw_index = n + i;
3754 		pack->bw_txg = txg;
3755 		pack->bw_data = 1 + ztest_random(-2ULL);
3756 
3757 		*bigH = *pack;
3758 		*bigT = *pack;
3759 	}
3760 }
3761 
3762 void
3763 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3764 {
3765 	objset_t *os = zd->zd_os;
3766 	ztest_od_t od[2];
3767 	dmu_tx_t *tx;
3768 	uint64_t i;
3769 	int error;
3770 	uint64_t n, s, txg;
3771 	bufwad_t *packbuf, *bigbuf;
3772 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3773 	uint64_t blocksize = ztest_random_blocksize();
3774 	uint64_t chunksize = blocksize;
3775 	uint64_t regions = 997;
3776 	uint64_t stride = 123456789ULL;
3777 	uint64_t width = 9;
3778 	dmu_buf_t *bonus_db;
3779 	arc_buf_t **bigbuf_arcbufs;
3780 	dmu_object_info_t doi;
3781 
3782 	/*
3783 	 * This test uses two objects, packobj and bigobj, that are always
3784 	 * updated together (i.e. in the same tx) so that their contents are
3785 	 * in sync and can be compared.  Their contents relate to each other
3786 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3787 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3788 	 * for any index n, there are three bufwads that should be identical:
3789 	 *
3790 	 *	packobj, at offset n * sizeof (bufwad_t)
3791 	 *	bigobj, at the head of the nth chunk
3792 	 *	bigobj, at the tail of the nth chunk
3793 	 *
3794 	 * The chunk size is set equal to bigobj block size so that
3795 	 * dmu_assign_arcbuf() can be tested for object updates.
3796 	 */
3797 
3798 	/*
3799 	 * Read the directory info.  If it's the first time, set things up.
3800 	 */
3801 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3802 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3803 
3804 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3805 		return;
3806 
3807 	bigobj = od[0].od_object;
3808 	packobj = od[1].od_object;
3809 	blocksize = od[0].od_blocksize;
3810 	chunksize = blocksize;
3811 	ASSERT(chunksize == od[1].od_gen);
3812 
3813 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3814 	VERIFY(ISP2(doi.doi_data_block_size));
3815 	VERIFY(chunksize == doi.doi_data_block_size);
3816 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3817 
3818 	/*
3819 	 * Pick a random index and compute the offsets into packobj and bigobj.
3820 	 */
3821 	n = ztest_random(regions) * stride + ztest_random(width);
3822 	s = 1 + ztest_random(width - 1);
3823 
3824 	packoff = n * sizeof (bufwad_t);
3825 	packsize = s * sizeof (bufwad_t);
3826 
3827 	bigoff = n * chunksize;
3828 	bigsize = s * chunksize;
3829 
3830 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3831 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3832 
3833 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3834 
3835 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3836 
3837 	/*
3838 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3839 	 * Iteration 1 test zcopy to already referenced dbufs.
3840 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3841 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3842 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3843 	 * Iteration 5 test zcopy when it can't be done.
3844 	 * Iteration 6 one more zcopy write.
3845 	 */
3846 	for (i = 0; i < 7; i++) {
3847 		uint64_t j;
3848 		uint64_t off;
3849 
3850 		/*
3851 		 * In iteration 5 (i == 5) use arcbufs
3852 		 * that don't match bigobj blksz to test
3853 		 * dmu_assign_arcbuf() when it can't directly
3854 		 * assign an arcbuf to a dbuf.
3855 		 */
3856 		for (j = 0; j < s; j++) {
3857 			if (i != 5) {
3858 				bigbuf_arcbufs[j] =
3859 				    dmu_request_arcbuf(bonus_db, chunksize);
3860 			} else {
3861 				bigbuf_arcbufs[2 * j] =
3862 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3863 				bigbuf_arcbufs[2 * j + 1] =
3864 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3865 			}
3866 		}
3867 
3868 		/*
3869 		 * Get a tx for the mods to both packobj and bigobj.
3870 		 */
3871 		tx = dmu_tx_create(os);
3872 
3873 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3874 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3875 
3876 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3877 		if (txg == 0) {
3878 			umem_free(packbuf, packsize);
3879 			umem_free(bigbuf, bigsize);
3880 			for (j = 0; j < s; j++) {
3881 				if (i != 5) {
3882 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3883 				} else {
3884 					dmu_return_arcbuf(
3885 					    bigbuf_arcbufs[2 * j]);
3886 					dmu_return_arcbuf(
3887 					    bigbuf_arcbufs[2 * j + 1]);
3888 				}
3889 			}
3890 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3891 			dmu_buf_rele(bonus_db, FTAG);
3892 			return;
3893 		}
3894 
3895 		/*
3896 		 * 50% of the time don't read objects in the 1st iteration to
3897 		 * test dmu_assign_arcbuf() for the case when there're no
3898 		 * existing dbufs for the specified offsets.
3899 		 */
3900 		if (i != 0 || ztest_random(2) != 0) {
3901 			error = dmu_read(os, packobj, packoff,
3902 			    packsize, packbuf, DMU_READ_PREFETCH);
3903 			ASSERT0(error);
3904 			error = dmu_read(os, bigobj, bigoff, bigsize,
3905 			    bigbuf, DMU_READ_PREFETCH);
3906 			ASSERT0(error);
3907 		}
3908 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3909 		    n, chunksize, txg);
3910 
3911 		/*
3912 		 * We've verified all the old bufwads, and made new ones.
3913 		 * Now write them out.
3914 		 */
3915 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3916 		if (ztest_opts.zo_verbose >= 7) {
3917 			(void) printf("writing offset %llx size %llx"
3918 			    " txg %llx\n",
3919 			    (u_longlong_t)bigoff,
3920 			    (u_longlong_t)bigsize,
3921 			    (u_longlong_t)txg);
3922 		}
3923 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3924 			dmu_buf_t *dbt;
3925 			if (i != 5) {
3926 				bcopy((caddr_t)bigbuf + (off - bigoff),
3927 				    bigbuf_arcbufs[j]->b_data, chunksize);
3928 			} else {
3929 				bcopy((caddr_t)bigbuf + (off - bigoff),
3930 				    bigbuf_arcbufs[2 * j]->b_data,
3931 				    chunksize / 2);
3932 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3933 				    chunksize / 2,
3934 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3935 				    chunksize / 2);
3936 			}
3937 
3938 			if (i == 1) {
3939 				VERIFY(dmu_buf_hold(os, bigobj, off,
3940 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3941 			}
3942 			if (i != 5) {
3943 				dmu_assign_arcbuf(bonus_db, off,
3944 				    bigbuf_arcbufs[j], tx);
3945 			} else {
3946 				dmu_assign_arcbuf(bonus_db, off,
3947 				    bigbuf_arcbufs[2 * j], tx);
3948 				dmu_assign_arcbuf(bonus_db,
3949 				    off + chunksize / 2,
3950 				    bigbuf_arcbufs[2 * j + 1], tx);
3951 			}
3952 			if (i == 1) {
3953 				dmu_buf_rele(dbt, FTAG);
3954 			}
3955 		}
3956 		dmu_tx_commit(tx);
3957 
3958 		/*
3959 		 * Sanity check the stuff we just wrote.
3960 		 */
3961 		{
3962 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3963 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3964 
3965 			VERIFY(0 == dmu_read(os, packobj, packoff,
3966 			    packsize, packcheck, DMU_READ_PREFETCH));
3967 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
3968 			    bigsize, bigcheck, DMU_READ_PREFETCH));
3969 
3970 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3971 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3972 
3973 			umem_free(packcheck, packsize);
3974 			umem_free(bigcheck, bigsize);
3975 		}
3976 		if (i == 2) {
3977 			txg_wait_open(dmu_objset_pool(os), 0);
3978 		} else if (i == 3) {
3979 			txg_wait_synced(dmu_objset_pool(os), 0);
3980 		}
3981 	}
3982 
3983 	dmu_buf_rele(bonus_db, FTAG);
3984 	umem_free(packbuf, packsize);
3985 	umem_free(bigbuf, bigsize);
3986 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3987 }
3988 
3989 /* ARGSUSED */
3990 void
3991 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
3992 {
3993 	ztest_od_t od[1];
3994 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
3995 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3996 
3997 	/*
3998 	 * Have multiple threads write to large offsets in an object
3999 	 * to verify that parallel writes to an object -- even to the
4000 	 * same blocks within the object -- doesn't cause any trouble.
4001 	 */
4002 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4003 
4004 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4005 		return;
4006 
4007 	while (ztest_random(10) != 0)
4008 		ztest_io(zd, od[0].od_object, offset);
4009 }
4010 
4011 void
4012 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4013 {
4014 	ztest_od_t od[1];
4015 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4016 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4017 	uint64_t count = ztest_random(20) + 1;
4018 	uint64_t blocksize = ztest_random_blocksize();
4019 	void *data;
4020 
4021 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4022 
4023 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4024 		return;
4025 
4026 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4027 		return;
4028 
4029 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4030 
4031 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4032 
4033 	while (ztest_random(count) != 0) {
4034 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4035 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4036 		    data) != 0)
4037 			break;
4038 		while (ztest_random(4) != 0)
4039 			ztest_io(zd, od[0].od_object, randoff);
4040 	}
4041 
4042 	umem_free(data, blocksize);
4043 }
4044 
4045 /*
4046  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4047  */
4048 #define	ZTEST_ZAP_MIN_INTS	1
4049 #define	ZTEST_ZAP_MAX_INTS	4
4050 #define	ZTEST_ZAP_MAX_PROPS	1000
4051 
4052 void
4053 ztest_zap(ztest_ds_t *zd, uint64_t id)
4054 {
4055 	objset_t *os = zd->zd_os;
4056 	ztest_od_t od[1];
4057 	uint64_t object;
4058 	uint64_t txg, last_txg;
4059 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4060 	uint64_t zl_ints, zl_intsize, prop;
4061 	int i, ints;
4062 	dmu_tx_t *tx;
4063 	char propname[100], txgname[100];
4064 	int error;
4065 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4066 
4067 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4068 
4069 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4070 		return;
4071 
4072 	object = od[0].od_object;
4073 
4074 	/*
4075 	 * Generate a known hash collision, and verify that
4076 	 * we can lookup and remove both entries.
4077 	 */
4078 	tx = dmu_tx_create(os);
4079 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4080 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4081 	if (txg == 0)
4082 		return;
4083 	for (i = 0; i < 2; i++) {
4084 		value[i] = i;
4085 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4086 		    1, &value[i], tx));
4087 	}
4088 	for (i = 0; i < 2; i++) {
4089 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4090 		    sizeof (uint64_t), 1, &value[i], tx));
4091 		VERIFY3U(0, ==,
4092 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4093 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4094 		ASSERT3U(zl_ints, ==, 1);
4095 	}
4096 	for (i = 0; i < 2; i++) {
4097 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4098 	}
4099 	dmu_tx_commit(tx);
4100 
4101 	/*
4102 	 * Generate a buch of random entries.
4103 	 */
4104 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4105 
4106 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4107 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4108 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4109 	bzero(value, sizeof (value));
4110 	last_txg = 0;
4111 
4112 	/*
4113 	 * If these zap entries already exist, validate their contents.
4114 	 */
4115 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4116 	if (error == 0) {
4117 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4118 		ASSERT3U(zl_ints, ==, 1);
4119 
4120 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4121 		    zl_ints, &last_txg) == 0);
4122 
4123 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4124 		    &zl_ints) == 0);
4125 
4126 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4127 		ASSERT3U(zl_ints, ==, ints);
4128 
4129 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4130 		    zl_ints, value) == 0);
4131 
4132 		for (i = 0; i < ints; i++) {
4133 			ASSERT3U(value[i], ==, last_txg + object + i);
4134 		}
4135 	} else {
4136 		ASSERT3U(error, ==, ENOENT);
4137 	}
4138 
4139 	/*
4140 	 * Atomically update two entries in our zap object.
4141 	 * The first is named txg_%llu, and contains the txg
4142 	 * in which the property was last updated.  The second
4143 	 * is named prop_%llu, and the nth element of its value
4144 	 * should be txg + object + n.
4145 	 */
4146 	tx = dmu_tx_create(os);
4147 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4148 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4149 	if (txg == 0)
4150 		return;
4151 
4152 	if (last_txg > txg)
4153 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4154 
4155 	for (i = 0; i < ints; i++)
4156 		value[i] = txg + object + i;
4157 
4158 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4159 	    1, &txg, tx));
4160 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4161 	    ints, value, tx));
4162 
4163 	dmu_tx_commit(tx);
4164 
4165 	/*
4166 	 * Remove a random pair of entries.
4167 	 */
4168 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4169 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4170 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4171 
4172 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4173 
4174 	if (error == ENOENT)
4175 		return;
4176 
4177 	ASSERT0(error);
4178 
4179 	tx = dmu_tx_create(os);
4180 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4181 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4182 	if (txg == 0)
4183 		return;
4184 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4185 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4186 	dmu_tx_commit(tx);
4187 }
4188 
4189 /*
4190  * Testcase to test the upgrading of a microzap to fatzap.
4191  */
4192 void
4193 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4194 {
4195 	objset_t *os = zd->zd_os;
4196 	ztest_od_t od[1];
4197 	uint64_t object, txg;
4198 
4199 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4200 
4201 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4202 		return;
4203 
4204 	object = od[0].od_object;
4205 
4206 	/*
4207 	 * Add entries to this ZAP and make sure it spills over
4208 	 * and gets upgraded to a fatzap. Also, since we are adding
4209 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4210 	 */
4211 	for (int i = 0; i < 2050; i++) {
4212 		char name[MAXNAMELEN];
4213 		uint64_t value = i;
4214 		dmu_tx_t *tx;
4215 		int error;
4216 
4217 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4218 		    id, value);
4219 
4220 		tx = dmu_tx_create(os);
4221 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4222 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4223 		if (txg == 0)
4224 			return;
4225 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4226 		    &value, tx);
4227 		ASSERT(error == 0 || error == EEXIST);
4228 		dmu_tx_commit(tx);
4229 	}
4230 }
4231 
4232 /* ARGSUSED */
4233 void
4234 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4235 {
4236 	objset_t *os = zd->zd_os;
4237 	ztest_od_t od[1];
4238 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4239 	dmu_tx_t *tx;
4240 	int i, namelen, error;
4241 	int micro = ztest_random(2);
4242 	char name[20], string_value[20];
4243 	void *data;
4244 
4245 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4246 
4247 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4248 		return;
4249 
4250 	object = od[0].od_object;
4251 
4252 	/*
4253 	 * Generate a random name of the form 'xxx.....' where each
4254 	 * x is a random printable character and the dots are dots.
4255 	 * There are 94 such characters, and the name length goes from
4256 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4257 	 */
4258 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4259 
4260 	for (i = 0; i < 3; i++)
4261 		name[i] = '!' + ztest_random('~' - '!' + 1);
4262 	for (; i < namelen - 1; i++)
4263 		name[i] = '.';
4264 	name[i] = '\0';
4265 
4266 	if ((namelen & 1) || micro) {
4267 		wsize = sizeof (txg);
4268 		wc = 1;
4269 		data = &txg;
4270 	} else {
4271 		wsize = 1;
4272 		wc = namelen;
4273 		data = string_value;
4274 	}
4275 
4276 	count = -1ULL;
4277 	VERIFY0(zap_count(os, object, &count));
4278 	ASSERT(count != -1ULL);
4279 
4280 	/*
4281 	 * Select an operation: length, lookup, add, update, remove.
4282 	 */
4283 	i = ztest_random(5);
4284 
4285 	if (i >= 2) {
4286 		tx = dmu_tx_create(os);
4287 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4288 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4289 		if (txg == 0)
4290 			return;
4291 		bcopy(name, string_value, namelen);
4292 	} else {
4293 		tx = NULL;
4294 		txg = 0;
4295 		bzero(string_value, namelen);
4296 	}
4297 
4298 	switch (i) {
4299 
4300 	case 0:
4301 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4302 		if (error == 0) {
4303 			ASSERT3U(wsize, ==, zl_wsize);
4304 			ASSERT3U(wc, ==, zl_wc);
4305 		} else {
4306 			ASSERT3U(error, ==, ENOENT);
4307 		}
4308 		break;
4309 
4310 	case 1:
4311 		error = zap_lookup(os, object, name, wsize, wc, data);
4312 		if (error == 0) {
4313 			if (data == string_value &&
4314 			    bcmp(name, data, namelen) != 0)
4315 				fatal(0, "name '%s' != val '%s' len %d",
4316 				    name, data, namelen);
4317 		} else {
4318 			ASSERT3U(error, ==, ENOENT);
4319 		}
4320 		break;
4321 
4322 	case 2:
4323 		error = zap_add(os, object, name, wsize, wc, data, tx);
4324 		ASSERT(error == 0 || error == EEXIST);
4325 		break;
4326 
4327 	case 3:
4328 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4329 		break;
4330 
4331 	case 4:
4332 		error = zap_remove(os, object, name, tx);
4333 		ASSERT(error == 0 || error == ENOENT);
4334 		break;
4335 	}
4336 
4337 	if (tx != NULL)
4338 		dmu_tx_commit(tx);
4339 }
4340 
4341 /*
4342  * Commit callback data.
4343  */
4344 typedef struct ztest_cb_data {
4345 	list_node_t		zcd_node;
4346 	uint64_t		zcd_txg;
4347 	int			zcd_expected_err;
4348 	boolean_t		zcd_added;
4349 	boolean_t		zcd_called;
4350 	spa_t			*zcd_spa;
4351 } ztest_cb_data_t;
4352 
4353 /* This is the actual commit callback function */
4354 static void
4355 ztest_commit_callback(void *arg, int error)
4356 {
4357 	ztest_cb_data_t *data = arg;
4358 	uint64_t synced_txg;
4359 
4360 	VERIFY(data != NULL);
4361 	VERIFY3S(data->zcd_expected_err, ==, error);
4362 	VERIFY(!data->zcd_called);
4363 
4364 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4365 	if (data->zcd_txg > synced_txg)
4366 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4367 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4368 		    synced_txg);
4369 
4370 	data->zcd_called = B_TRUE;
4371 
4372 	if (error == ECANCELED) {
4373 		ASSERT0(data->zcd_txg);
4374 		ASSERT(!data->zcd_added);
4375 
4376 		/*
4377 		 * The private callback data should be destroyed here, but
4378 		 * since we are going to check the zcd_called field after
4379 		 * dmu_tx_abort(), we will destroy it there.
4380 		 */
4381 		return;
4382 	}
4383 
4384 	/* Was this callback added to the global callback list? */
4385 	if (!data->zcd_added)
4386 		goto out;
4387 
4388 	ASSERT3U(data->zcd_txg, !=, 0);
4389 
4390 	/* Remove our callback from the list */
4391 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4392 	list_remove(&zcl.zcl_callbacks, data);
4393 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4394 
4395 out:
4396 	umem_free(data, sizeof (ztest_cb_data_t));
4397 }
4398 
4399 /* Allocate and initialize callback data structure */
4400 static ztest_cb_data_t *
4401 ztest_create_cb_data(objset_t *os, uint64_t txg)
4402 {
4403 	ztest_cb_data_t *cb_data;
4404 
4405 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4406 
4407 	cb_data->zcd_txg = txg;
4408 	cb_data->zcd_spa = dmu_objset_spa(os);
4409 
4410 	return (cb_data);
4411 }
4412 
4413 /*
4414  * If a number of txgs equal to this threshold have been created after a commit
4415  * callback has been registered but not called, then we assume there is an
4416  * implementation bug.
4417  */
4418 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4419 
4420 /*
4421  * Commit callback test.
4422  */
4423 void
4424 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4425 {
4426 	objset_t *os = zd->zd_os;
4427 	ztest_od_t od[1];
4428 	dmu_tx_t *tx;
4429 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4430 	uint64_t old_txg, txg;
4431 	int i, error;
4432 
4433 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4434 
4435 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4436 		return;
4437 
4438 	tx = dmu_tx_create(os);
4439 
4440 	cb_data[0] = ztest_create_cb_data(os, 0);
4441 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4442 
4443 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4444 
4445 	/* Every once in a while, abort the transaction on purpose */
4446 	if (ztest_random(100) == 0)
4447 		error = -1;
4448 
4449 	if (!error)
4450 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4451 
4452 	txg = error ? 0 : dmu_tx_get_txg(tx);
4453 
4454 	cb_data[0]->zcd_txg = txg;
4455 	cb_data[1] = ztest_create_cb_data(os, txg);
4456 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4457 
4458 	if (error) {
4459 		/*
4460 		 * It's not a strict requirement to call the registered
4461 		 * callbacks from inside dmu_tx_abort(), but that's what
4462 		 * it's supposed to happen in the current implementation
4463 		 * so we will check for that.
4464 		 */
4465 		for (i = 0; i < 2; i++) {
4466 			cb_data[i]->zcd_expected_err = ECANCELED;
4467 			VERIFY(!cb_data[i]->zcd_called);
4468 		}
4469 
4470 		dmu_tx_abort(tx);
4471 
4472 		for (i = 0; i < 2; i++) {
4473 			VERIFY(cb_data[i]->zcd_called);
4474 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4475 		}
4476 
4477 		return;
4478 	}
4479 
4480 	cb_data[2] = ztest_create_cb_data(os, txg);
4481 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4482 
4483 	/*
4484 	 * Read existing data to make sure there isn't a future leak.
4485 	 */
4486 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4487 	    &old_txg, DMU_READ_PREFETCH));
4488 
4489 	if (old_txg > txg)
4490 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4491 		    old_txg, txg);
4492 
4493 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4494 
4495 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4496 
4497 	/*
4498 	 * Since commit callbacks don't have any ordering requirement and since
4499 	 * it is theoretically possible for a commit callback to be called
4500 	 * after an arbitrary amount of time has elapsed since its txg has been
4501 	 * synced, it is difficult to reliably determine whether a commit
4502 	 * callback hasn't been called due to high load or due to a flawed
4503 	 * implementation.
4504 	 *
4505 	 * In practice, we will assume that if after a certain number of txgs a
4506 	 * commit callback hasn't been called, then most likely there's an
4507 	 * implementation bug..
4508 	 */
4509 	tmp_cb = list_head(&zcl.zcl_callbacks);
4510 	if (tmp_cb != NULL &&
4511 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4512 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4513 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4514 	}
4515 
4516 	/*
4517 	 * Let's find the place to insert our callbacks.
4518 	 *
4519 	 * Even though the list is ordered by txg, it is possible for the
4520 	 * insertion point to not be the end because our txg may already be
4521 	 * quiescing at this point and other callbacks in the open txg
4522 	 * (from other objsets) may have sneaked in.
4523 	 */
4524 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4525 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4526 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4527 
4528 	/* Add the 3 callbacks to the list */
4529 	for (i = 0; i < 3; i++) {
4530 		if (tmp_cb == NULL)
4531 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4532 		else
4533 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4534 			    cb_data[i]);
4535 
4536 		cb_data[i]->zcd_added = B_TRUE;
4537 		VERIFY(!cb_data[i]->zcd_called);
4538 
4539 		tmp_cb = cb_data[i];
4540 	}
4541 
4542 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4543 
4544 	dmu_tx_commit(tx);
4545 }
4546 
4547 /* ARGSUSED */
4548 void
4549 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4550 {
4551 	zfs_prop_t proplist[] = {
4552 		ZFS_PROP_CHECKSUM,
4553 		ZFS_PROP_COMPRESSION,
4554 		ZFS_PROP_COPIES,
4555 		ZFS_PROP_DEDUP
4556 	};
4557 
4558 	(void) rw_rdlock(&ztest_name_lock);
4559 
4560 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4561 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4562 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4563 
4564 	(void) rw_unlock(&ztest_name_lock);
4565 }
4566 
4567 /* ARGSUSED */
4568 void
4569 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4570 {
4571 	nvlist_t *props = NULL;
4572 
4573 	(void) rw_rdlock(&ztest_name_lock);
4574 
4575 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4576 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4577 
4578 	VERIFY0(spa_prop_get(ztest_spa, &props));
4579 
4580 	if (ztest_opts.zo_verbose >= 6)
4581 		dump_nvlist(props, 4);
4582 
4583 	nvlist_free(props);
4584 
4585 	(void) rw_unlock(&ztest_name_lock);
4586 }
4587 
4588 static int
4589 user_release_one(const char *snapname, const char *holdname)
4590 {
4591 	nvlist_t *snaps, *holds;
4592 	int error;
4593 
4594 	snaps = fnvlist_alloc();
4595 	holds = fnvlist_alloc();
4596 	fnvlist_add_boolean(holds, holdname);
4597 	fnvlist_add_nvlist(snaps, snapname, holds);
4598 	fnvlist_free(holds);
4599 	error = dsl_dataset_user_release(snaps, NULL);
4600 	fnvlist_free(snaps);
4601 	return (error);
4602 }
4603 
4604 /*
4605  * Test snapshot hold/release and deferred destroy.
4606  */
4607 void
4608 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4609 {
4610 	int error;
4611 	objset_t *os = zd->zd_os;
4612 	objset_t *origin;
4613 	char snapname[100];
4614 	char fullname[100];
4615 	char clonename[100];
4616 	char tag[100];
4617 	char osname[MAXNAMELEN];
4618 	nvlist_t *holds;
4619 
4620 	(void) rw_rdlock(&ztest_name_lock);
4621 
4622 	dmu_objset_name(os, osname);
4623 
4624 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4625 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4626 	(void) snprintf(clonename, sizeof (clonename),
4627 	    "%s/ch1_%llu", osname, id);
4628 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4629 
4630 	/*
4631 	 * Clean up from any previous run.
4632 	 */
4633 	error = dsl_destroy_head(clonename);
4634 	if (error != ENOENT)
4635 		ASSERT0(error);
4636 	error = user_release_one(fullname, tag);
4637 	if (error != ESRCH && error != ENOENT)
4638 		ASSERT0(error);
4639 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4640 	if (error != ENOENT)
4641 		ASSERT0(error);
4642 
4643 	/*
4644 	 * Create snapshot, clone it, mark snap for deferred destroy,
4645 	 * destroy clone, verify snap was also destroyed.
4646 	 */
4647 	error = dmu_objset_snapshot_one(osname, snapname);
4648 	if (error) {
4649 		if (error == ENOSPC) {
4650 			ztest_record_enospc("dmu_objset_snapshot");
4651 			goto out;
4652 		}
4653 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4654 	}
4655 
4656 	error = dmu_objset_clone(clonename, fullname);
4657 	if (error) {
4658 		if (error == ENOSPC) {
4659 			ztest_record_enospc("dmu_objset_clone");
4660 			goto out;
4661 		}
4662 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4663 	}
4664 
4665 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4666 	if (error) {
4667 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4668 		    fullname, error);
4669 	}
4670 
4671 	error = dsl_destroy_head(clonename);
4672 	if (error)
4673 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4674 
4675 	error = dmu_objset_hold(fullname, FTAG, &origin);
4676 	if (error != ENOENT)
4677 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4678 
4679 	/*
4680 	 * Create snapshot, add temporary hold, verify that we can't
4681 	 * destroy a held snapshot, mark for deferred destroy,
4682 	 * release hold, verify snapshot was destroyed.
4683 	 */
4684 	error = dmu_objset_snapshot_one(osname, snapname);
4685 	if (error) {
4686 		if (error == ENOSPC) {
4687 			ztest_record_enospc("dmu_objset_snapshot");
4688 			goto out;
4689 		}
4690 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4691 	}
4692 
4693 	holds = fnvlist_alloc();
4694 	fnvlist_add_string(holds, fullname, tag);
4695 	error = dsl_dataset_user_hold(holds, 0, NULL);
4696 	fnvlist_free(holds);
4697 
4698 	if (error)
4699 		fatal(0, "dsl_dataset_user_hold(%s)", fullname, tag);
4700 
4701 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4702 	if (error != EBUSY) {
4703 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4704 		    fullname, error);
4705 	}
4706 
4707 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4708 	if (error) {
4709 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4710 		    fullname, error);
4711 	}
4712 
4713 	error = user_release_one(fullname, tag);
4714 	if (error)
4715 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4716 
4717 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4718 
4719 out:
4720 	(void) rw_unlock(&ztest_name_lock);
4721 }
4722 
4723 /*
4724  * Inject random faults into the on-disk data.
4725  */
4726 /* ARGSUSED */
4727 void
4728 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4729 {
4730 	ztest_shared_t *zs = ztest_shared;
4731 	spa_t *spa = ztest_spa;
4732 	int fd;
4733 	uint64_t offset;
4734 	uint64_t leaves;
4735 	uint64_t bad = 0x1990c0ffeedecade;
4736 	uint64_t top, leaf;
4737 	char path0[MAXPATHLEN];
4738 	char pathrand[MAXPATHLEN];
4739 	size_t fsize;
4740 	int bshift = SPA_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
4741 	int iters = 1000;
4742 	int maxfaults;
4743 	int mirror_save;
4744 	vdev_t *vd0 = NULL;
4745 	uint64_t guid0 = 0;
4746 	boolean_t islog = B_FALSE;
4747 
4748 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4749 	maxfaults = MAXFAULTS();
4750 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4751 	mirror_save = zs->zs_mirrors;
4752 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4753 
4754 	ASSERT(leaves >= 1);
4755 
4756 	/*
4757 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4758 	 */
4759 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4760 
4761 	if (ztest_random(2) == 0) {
4762 		/*
4763 		 * Inject errors on a normal data device or slog device.
4764 		 */
4765 		top = ztest_random_vdev_top(spa, B_TRUE);
4766 		leaf = ztest_random(leaves) + zs->zs_splits;
4767 
4768 		/*
4769 		 * Generate paths to the first leaf in this top-level vdev,
4770 		 * and to the random leaf we selected.  We'll induce transient
4771 		 * write failures and random online/offline activity on leaf 0,
4772 		 * and we'll write random garbage to the randomly chosen leaf.
4773 		 */
4774 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4775 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4776 		    top * leaves + zs->zs_splits);
4777 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4778 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4779 		    top * leaves + leaf);
4780 
4781 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4782 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4783 			islog = B_TRUE;
4784 
4785 		if (vd0 != NULL && maxfaults != 1) {
4786 			/*
4787 			 * Make vd0 explicitly claim to be unreadable,
4788 			 * or unwriteable, or reach behind its back
4789 			 * and close the underlying fd.  We can do this if
4790 			 * maxfaults == 0 because we'll fail and reexecute,
4791 			 * and we can do it if maxfaults >= 2 because we'll
4792 			 * have enough redundancy.  If maxfaults == 1, the
4793 			 * combination of this with injection of random data
4794 			 * corruption below exceeds the pool's fault tolerance.
4795 			 */
4796 			vdev_file_t *vf = vd0->vdev_tsd;
4797 
4798 			if (vf != NULL && ztest_random(3) == 0) {
4799 				(void) close(vf->vf_vnode->v_fd);
4800 				vf->vf_vnode->v_fd = -1;
4801 			} else if (ztest_random(2) == 0) {
4802 				vd0->vdev_cant_read = B_TRUE;
4803 			} else {
4804 				vd0->vdev_cant_write = B_TRUE;
4805 			}
4806 			guid0 = vd0->vdev_guid;
4807 		}
4808 	} else {
4809 		/*
4810 		 * Inject errors on an l2cache device.
4811 		 */
4812 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4813 
4814 		if (sav->sav_count == 0) {
4815 			spa_config_exit(spa, SCL_STATE, FTAG);
4816 			return;
4817 		}
4818 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4819 		guid0 = vd0->vdev_guid;
4820 		(void) strcpy(path0, vd0->vdev_path);
4821 		(void) strcpy(pathrand, vd0->vdev_path);
4822 
4823 		leaf = 0;
4824 		leaves = 1;
4825 		maxfaults = INT_MAX;	/* no limit on cache devices */
4826 	}
4827 
4828 	spa_config_exit(spa, SCL_STATE, FTAG);
4829 
4830 	/*
4831 	 * If we can tolerate two or more faults, or we're dealing
4832 	 * with a slog, randomly online/offline vd0.
4833 	 */
4834 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4835 		if (ztest_random(10) < 6) {
4836 			int flags = (ztest_random(2) == 0 ?
4837 			    ZFS_OFFLINE_TEMPORARY : 0);
4838 
4839 			/*
4840 			 * We have to grab the zs_name_lock as writer to
4841 			 * prevent a race between offlining a slog and
4842 			 * destroying a dataset. Offlining the slog will
4843 			 * grab a reference on the dataset which may cause
4844 			 * dmu_objset_destroy() to fail with EBUSY thus
4845 			 * leaving the dataset in an inconsistent state.
4846 			 */
4847 			if (islog)
4848 				(void) rw_wrlock(&ztest_name_lock);
4849 
4850 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4851 
4852 			if (islog)
4853 				(void) rw_unlock(&ztest_name_lock);
4854 		} else {
4855 			/*
4856 			 * Ideally we would like to be able to randomly
4857 			 * call vdev_[on|off]line without holding locks
4858 			 * to force unpredictable failures but the side
4859 			 * effects of vdev_[on|off]line prevent us from
4860 			 * doing so. We grab the ztest_vdev_lock here to
4861 			 * prevent a race between injection testing and
4862 			 * aux_vdev removal.
4863 			 */
4864 			VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4865 			(void) vdev_online(spa, guid0, 0, NULL);
4866 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4867 		}
4868 	}
4869 
4870 	if (maxfaults == 0)
4871 		return;
4872 
4873 	/*
4874 	 * We have at least single-fault tolerance, so inject data corruption.
4875 	 */
4876 	fd = open(pathrand, O_RDWR);
4877 
4878 	if (fd == -1)	/* we hit a gap in the device namespace */
4879 		return;
4880 
4881 	fsize = lseek(fd, 0, SEEK_END);
4882 
4883 	while (--iters != 0) {
4884 		offset = ztest_random(fsize / (leaves << bshift)) *
4885 		    (leaves << bshift) + (leaf << bshift) +
4886 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4887 
4888 		if (offset >= fsize)
4889 			continue;
4890 
4891 		VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4892 		if (mirror_save != zs->zs_mirrors) {
4893 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4894 			(void) close(fd);
4895 			return;
4896 		}
4897 
4898 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4899 			fatal(1, "can't inject bad word at 0x%llx in %s",
4900 			    offset, pathrand);
4901 
4902 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4903 
4904 		if (ztest_opts.zo_verbose >= 7)
4905 			(void) printf("injected bad word into %s,"
4906 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4907 	}
4908 
4909 	(void) close(fd);
4910 }
4911 
4912 /*
4913  * Verify that DDT repair works as expected.
4914  */
4915 void
4916 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4917 {
4918 	ztest_shared_t *zs = ztest_shared;
4919 	spa_t *spa = ztest_spa;
4920 	objset_t *os = zd->zd_os;
4921 	ztest_od_t od[1];
4922 	uint64_t object, blocksize, txg, pattern, psize;
4923 	enum zio_checksum checksum = spa_dedup_checksum(spa);
4924 	dmu_buf_t *db;
4925 	dmu_tx_t *tx;
4926 	void *buf;
4927 	blkptr_t blk;
4928 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
4929 
4930 	blocksize = ztest_random_blocksize();
4931 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
4932 
4933 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4934 
4935 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4936 		return;
4937 
4938 	/*
4939 	 * Take the name lock as writer to prevent anyone else from changing
4940 	 * the pool and dataset properies we need to maintain during this test.
4941 	 */
4942 	(void) rw_wrlock(&ztest_name_lock);
4943 
4944 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
4945 	    B_FALSE) != 0 ||
4946 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
4947 	    B_FALSE) != 0) {
4948 		(void) rw_unlock(&ztest_name_lock);
4949 		return;
4950 	}
4951 
4952 	object = od[0].od_object;
4953 	blocksize = od[0].od_blocksize;
4954 	pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
4955 
4956 	ASSERT(object != 0);
4957 
4958 	tx = dmu_tx_create(os);
4959 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
4960 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
4961 	if (txg == 0) {
4962 		(void) rw_unlock(&ztest_name_lock);
4963 		return;
4964 	}
4965 
4966 	/*
4967 	 * Write all the copies of our block.
4968 	 */
4969 	for (int i = 0; i < copies; i++) {
4970 		uint64_t offset = i * blocksize;
4971 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
4972 		    DMU_READ_NO_PREFETCH);
4973 		if (error != 0) {
4974 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
4975 			    os, (long long)object, (long long) offset, error);
4976 		}
4977 		ASSERT(db->db_offset == offset);
4978 		ASSERT(db->db_size == blocksize);
4979 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
4980 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
4981 		dmu_buf_will_fill(db, tx);
4982 		ztest_pattern_set(db->db_data, db->db_size, pattern);
4983 		dmu_buf_rele(db, FTAG);
4984 	}
4985 
4986 	dmu_tx_commit(tx);
4987 	txg_wait_synced(spa_get_dsl(spa), txg);
4988 
4989 	/*
4990 	 * Find out what block we got.
4991 	 */
4992 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
4993 	    DMU_READ_NO_PREFETCH));
4994 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
4995 	dmu_buf_rele(db, FTAG);
4996 
4997 	/*
4998 	 * Damage the block.  Dedup-ditto will save us when we read it later.
4999 	 */
5000 	psize = BP_GET_PSIZE(&blk);
5001 	buf = zio_buf_alloc(psize);
5002 	ztest_pattern_set(buf, psize, ~pattern);
5003 
5004 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5005 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5006 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5007 
5008 	zio_buf_free(buf, psize);
5009 
5010 	(void) rw_unlock(&ztest_name_lock);
5011 }
5012 
5013 /*
5014  * Scrub the pool.
5015  */
5016 /* ARGSUSED */
5017 void
5018 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5019 {
5020 	spa_t *spa = ztest_spa;
5021 
5022 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5023 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5024 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5025 }
5026 
5027 /*
5028  * Change the guid for the pool.
5029  */
5030 /* ARGSUSED */
5031 void
5032 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5033 {
5034 	spa_t *spa = ztest_spa;
5035 	uint64_t orig, load;
5036 	int error;
5037 
5038 	orig = spa_guid(spa);
5039 	load = spa_load_guid(spa);
5040 
5041 	(void) rw_wrlock(&ztest_name_lock);
5042 	error = spa_change_guid(spa);
5043 	(void) rw_unlock(&ztest_name_lock);
5044 
5045 	if (error != 0)
5046 		return;
5047 
5048 	if (ztest_opts.zo_verbose >= 4) {
5049 		(void) printf("Changed guid old %llu -> %llu\n",
5050 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5051 	}
5052 
5053 	VERIFY3U(orig, !=, spa_guid(spa));
5054 	VERIFY3U(load, ==, spa_load_guid(spa));
5055 }
5056 
5057 /*
5058  * Rename the pool to a different name and then rename it back.
5059  */
5060 /* ARGSUSED */
5061 void
5062 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5063 {
5064 	char *oldname, *newname;
5065 	spa_t *spa;
5066 
5067 	(void) rw_wrlock(&ztest_name_lock);
5068 
5069 	oldname = ztest_opts.zo_pool;
5070 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5071 	(void) strcpy(newname, oldname);
5072 	(void) strcat(newname, "_tmp");
5073 
5074 	/*
5075 	 * Do the rename
5076 	 */
5077 	VERIFY3U(0, ==, spa_rename(oldname, newname));
5078 
5079 	/*
5080 	 * Try to open it under the old name, which shouldn't exist
5081 	 */
5082 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5083 
5084 	/*
5085 	 * Open it under the new name and make sure it's still the same spa_t.
5086 	 */
5087 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5088 
5089 	ASSERT(spa == ztest_spa);
5090 	spa_close(spa, FTAG);
5091 
5092 	/*
5093 	 * Rename it back to the original
5094 	 */
5095 	VERIFY3U(0, ==, spa_rename(newname, oldname));
5096 
5097 	/*
5098 	 * Make sure it can still be opened
5099 	 */
5100 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5101 
5102 	ASSERT(spa == ztest_spa);
5103 	spa_close(spa, FTAG);
5104 
5105 	umem_free(newname, strlen(newname) + 1);
5106 
5107 	(void) rw_unlock(&ztest_name_lock);
5108 }
5109 
5110 /*
5111  * Verify pool integrity by running zdb.
5112  */
5113 static void
5114 ztest_run_zdb(char *pool)
5115 {
5116 	int status;
5117 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5118 	char zbuf[1024];
5119 	char *bin;
5120 	char *ztest;
5121 	char *isa;
5122 	int isalen;
5123 	FILE *fp;
5124 
5125 	(void) realpath(getexecname(), zdb);
5126 
5127 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5128 	bin = strstr(zdb, "/usr/bin/");
5129 	ztest = strstr(bin, "/ztest");
5130 	isa = bin + 8;
5131 	isalen = ztest - isa;
5132 	isa = strdup(isa);
5133 	/* LINTED */
5134 	(void) sprintf(bin,
5135 	    "/usr/sbin%.*s/zdb -bcc%s%s -U %s %s",
5136 	    isalen,
5137 	    isa,
5138 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5139 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5140 	    spa_config_path,
5141 	    pool);
5142 	free(isa);
5143 
5144 	if (ztest_opts.zo_verbose >= 5)
5145 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5146 
5147 	fp = popen(zdb, "r");
5148 
5149 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5150 		if (ztest_opts.zo_verbose >= 3)
5151 			(void) printf("%s", zbuf);
5152 
5153 	status = pclose(fp);
5154 
5155 	if (status == 0)
5156 		return;
5157 
5158 	ztest_dump_core = 0;
5159 	if (WIFEXITED(status))
5160 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5161 	else
5162 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5163 }
5164 
5165 static void
5166 ztest_walk_pool_directory(char *header)
5167 {
5168 	spa_t *spa = NULL;
5169 
5170 	if (ztest_opts.zo_verbose >= 6)
5171 		(void) printf("%s\n", header);
5172 
5173 	mutex_enter(&spa_namespace_lock);
5174 	while ((spa = spa_next(spa)) != NULL)
5175 		if (ztest_opts.zo_verbose >= 6)
5176 			(void) printf("\t%s\n", spa_name(spa));
5177 	mutex_exit(&spa_namespace_lock);
5178 }
5179 
5180 static void
5181 ztest_spa_import_export(char *oldname, char *newname)
5182 {
5183 	nvlist_t *config, *newconfig;
5184 	uint64_t pool_guid;
5185 	spa_t *spa;
5186 	int error;
5187 
5188 	if (ztest_opts.zo_verbose >= 4) {
5189 		(void) printf("import/export: old = %s, new = %s\n",
5190 		    oldname, newname);
5191 	}
5192 
5193 	/*
5194 	 * Clean up from previous runs.
5195 	 */
5196 	(void) spa_destroy(newname);
5197 
5198 	/*
5199 	 * Get the pool's configuration and guid.
5200 	 */
5201 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5202 
5203 	/*
5204 	 * Kick off a scrub to tickle scrub/export races.
5205 	 */
5206 	if (ztest_random(2) == 0)
5207 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5208 
5209 	pool_guid = spa_guid(spa);
5210 	spa_close(spa, FTAG);
5211 
5212 	ztest_walk_pool_directory("pools before export");
5213 
5214 	/*
5215 	 * Export it.
5216 	 */
5217 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5218 
5219 	ztest_walk_pool_directory("pools after export");
5220 
5221 	/*
5222 	 * Try to import it.
5223 	 */
5224 	newconfig = spa_tryimport(config);
5225 	ASSERT(newconfig != NULL);
5226 	nvlist_free(newconfig);
5227 
5228 	/*
5229 	 * Import it under the new name.
5230 	 */
5231 	error = spa_import(newname, config, NULL, 0);
5232 	if (error != 0) {
5233 		dump_nvlist(config, 0);
5234 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5235 		    oldname, newname, error);
5236 	}
5237 
5238 	ztest_walk_pool_directory("pools after import");
5239 
5240 	/*
5241 	 * Try to import it again -- should fail with EEXIST.
5242 	 */
5243 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5244 
5245 	/*
5246 	 * Try to import it under a different name -- should fail with EEXIST.
5247 	 */
5248 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5249 
5250 	/*
5251 	 * Verify that the pool is no longer visible under the old name.
5252 	 */
5253 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5254 
5255 	/*
5256 	 * Verify that we can open and close the pool using the new name.
5257 	 */
5258 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5259 	ASSERT(pool_guid == spa_guid(spa));
5260 	spa_close(spa, FTAG);
5261 
5262 	nvlist_free(config);
5263 }
5264 
5265 static void
5266 ztest_resume(spa_t *spa)
5267 {
5268 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5269 		(void) printf("resuming from suspended state\n");
5270 	spa_vdev_state_enter(spa, SCL_NONE);
5271 	vdev_clear(spa, NULL);
5272 	(void) spa_vdev_state_exit(spa, NULL, 0);
5273 	(void) zio_resume(spa);
5274 }
5275 
5276 static void *
5277 ztest_resume_thread(void *arg)
5278 {
5279 	spa_t *spa = arg;
5280 
5281 	while (!ztest_exiting) {
5282 		if (spa_suspended(spa))
5283 			ztest_resume(spa);
5284 		(void) poll(NULL, 0, 100);
5285 	}
5286 	return (NULL);
5287 }
5288 
5289 static void *
5290 ztest_deadman_thread(void *arg)
5291 {
5292 	ztest_shared_t *zs = arg;
5293 	int grace = 300;
5294 	hrtime_t delta;
5295 
5296 	delta = (zs->zs_thread_stop - zs->zs_thread_start) / NANOSEC + grace;
5297 
5298 	(void) poll(NULL, 0, (int)(1000 * delta));
5299 
5300 	fatal(0, "failed to complete within %d seconds of deadline", grace);
5301 
5302 	return (NULL);
5303 }
5304 
5305 static void
5306 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5307 {
5308 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5309 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5310 	hrtime_t functime = gethrtime();
5311 
5312 	for (int i = 0; i < zi->zi_iters; i++)
5313 		zi->zi_func(zd, id);
5314 
5315 	functime = gethrtime() - functime;
5316 
5317 	atomic_add_64(&zc->zc_count, 1);
5318 	atomic_add_64(&zc->zc_time, functime);
5319 
5320 	if (ztest_opts.zo_verbose >= 4) {
5321 		Dl_info dli;
5322 		(void) dladdr((void *)zi->zi_func, &dli);
5323 		(void) printf("%6.2f sec in %s\n",
5324 		    (double)functime / NANOSEC, dli.dli_sname);
5325 	}
5326 }
5327 
5328 static void *
5329 ztest_thread(void *arg)
5330 {
5331 	int rand;
5332 	uint64_t id = (uintptr_t)arg;
5333 	ztest_shared_t *zs = ztest_shared;
5334 	uint64_t call_next;
5335 	hrtime_t now;
5336 	ztest_info_t *zi;
5337 	ztest_shared_callstate_t *zc;
5338 
5339 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5340 		/*
5341 		 * See if it's time to force a crash.
5342 		 */
5343 		if (now > zs->zs_thread_kill)
5344 			ztest_kill(zs);
5345 
5346 		/*
5347 		 * If we're getting ENOSPC with some regularity, stop.
5348 		 */
5349 		if (zs->zs_enospc_count > 10)
5350 			break;
5351 
5352 		/*
5353 		 * Pick a random function to execute.
5354 		 */
5355 		rand = ztest_random(ZTEST_FUNCS);
5356 		zi = &ztest_info[rand];
5357 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5358 		call_next = zc->zc_next;
5359 
5360 		if (now >= call_next &&
5361 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5362 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5363 			ztest_execute(rand, zi, id);
5364 		}
5365 	}
5366 
5367 	return (NULL);
5368 }
5369 
5370 static void
5371 ztest_dataset_name(char *dsname, char *pool, int d)
5372 {
5373 	(void) snprintf(dsname, MAXNAMELEN, "%s/ds_%d", pool, d);
5374 }
5375 
5376 static void
5377 ztest_dataset_destroy(int d)
5378 {
5379 	char name[MAXNAMELEN];
5380 
5381 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5382 
5383 	if (ztest_opts.zo_verbose >= 3)
5384 		(void) printf("Destroying %s to free up space\n", name);
5385 
5386 	/*
5387 	 * Cleanup any non-standard clones and snapshots.  In general,
5388 	 * ztest thread t operates on dataset (t % zopt_datasets),
5389 	 * so there may be more than one thing to clean up.
5390 	 */
5391 	for (int t = d; t < ztest_opts.zo_threads;
5392 	    t += ztest_opts.zo_datasets) {
5393 		ztest_dsl_dataset_cleanup(name, t);
5394 	}
5395 
5396 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5397 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5398 }
5399 
5400 static void
5401 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5402 {
5403 	uint64_t usedobjs, dirobjs, scratch;
5404 
5405 	/*
5406 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5407 	 * Therefore, the number of objects in use should equal the
5408 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5409 	 * If not, we have an object leak.
5410 	 *
5411 	 * Note that we can only check this in ztest_dataset_open(),
5412 	 * when the open-context and syncing-context values agree.
5413 	 * That's because zap_count() returns the open-context value,
5414 	 * while dmu_objset_space() returns the rootbp fill count.
5415 	 */
5416 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5417 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5418 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5419 }
5420 
5421 static int
5422 ztest_dataset_open(int d)
5423 {
5424 	ztest_ds_t *zd = &ztest_ds[d];
5425 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5426 	objset_t *os;
5427 	zilog_t *zilog;
5428 	char name[MAXNAMELEN];
5429 	int error;
5430 
5431 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5432 
5433 	(void) rw_rdlock(&ztest_name_lock);
5434 
5435 	error = ztest_dataset_create(name);
5436 	if (error == ENOSPC) {
5437 		(void) rw_unlock(&ztest_name_lock);
5438 		ztest_record_enospc(FTAG);
5439 		return (error);
5440 	}
5441 	ASSERT(error == 0 || error == EEXIST);
5442 
5443 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5444 	(void) rw_unlock(&ztest_name_lock);
5445 
5446 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5447 
5448 	zilog = zd->zd_zilog;
5449 
5450 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5451 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5452 		fatal(0, "missing log records: claimed %llu < committed %llu",
5453 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5454 
5455 	ztest_dataset_dirobj_verify(zd);
5456 
5457 	zil_replay(os, zd, ztest_replay_vector);
5458 
5459 	ztest_dataset_dirobj_verify(zd);
5460 
5461 	if (ztest_opts.zo_verbose >= 6)
5462 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5463 		    zd->zd_name,
5464 		    (u_longlong_t)zilog->zl_parse_blk_count,
5465 		    (u_longlong_t)zilog->zl_parse_lr_count,
5466 		    (u_longlong_t)zilog->zl_replaying_seq);
5467 
5468 	zilog = zil_open(os, ztest_get_data);
5469 
5470 	if (zilog->zl_replaying_seq != 0 &&
5471 	    zilog->zl_replaying_seq < committed_seq)
5472 		fatal(0, "missing log records: replayed %llu < committed %llu",
5473 		    zilog->zl_replaying_seq, committed_seq);
5474 
5475 	return (0);
5476 }
5477 
5478 static void
5479 ztest_dataset_close(int d)
5480 {
5481 	ztest_ds_t *zd = &ztest_ds[d];
5482 
5483 	zil_close(zd->zd_zilog);
5484 	dmu_objset_disown(zd->zd_os, zd);
5485 
5486 	ztest_zd_fini(zd);
5487 }
5488 
5489 /*
5490  * Kick off threads to run tests on all datasets in parallel.
5491  */
5492 static void
5493 ztest_run(ztest_shared_t *zs)
5494 {
5495 	thread_t *tid;
5496 	spa_t *spa;
5497 	objset_t *os;
5498 	thread_t resume_tid;
5499 	int error;
5500 
5501 	ztest_exiting = B_FALSE;
5502 
5503 	/*
5504 	 * Initialize parent/child shared state.
5505 	 */
5506 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5507 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5508 
5509 	zs->zs_thread_start = gethrtime();
5510 	zs->zs_thread_stop =
5511 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5512 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5513 	zs->zs_thread_kill = zs->zs_thread_stop;
5514 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5515 		zs->zs_thread_kill -=
5516 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5517 	}
5518 
5519 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5520 
5521 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5522 	    offsetof(ztest_cb_data_t, zcd_node));
5523 
5524 	/*
5525 	 * Open our pool.
5526 	 */
5527 	kernel_init(FREAD | FWRITE);
5528 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5529 	spa->spa_debug = B_TRUE;
5530 	ztest_spa = spa;
5531 
5532 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5533 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
5534 	zs->zs_guid = dmu_objset_fsid_guid(os);
5535 	dmu_objset_disown(os, FTAG);
5536 
5537 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5538 
5539 	/*
5540 	 * We don't expect the pool to suspend unless maxfaults == 0,
5541 	 * in which case ztest_fault_inject() temporarily takes away
5542 	 * the only valid replica.
5543 	 */
5544 	if (MAXFAULTS() == 0)
5545 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5546 	else
5547 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5548 
5549 	/*
5550 	 * Create a thread to periodically resume suspended I/O.
5551 	 */
5552 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5553 	    &resume_tid) == 0);
5554 
5555 	/*
5556 	 * Create a deadman thread to abort() if we hang.
5557 	 */
5558 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5559 	    NULL) == 0);
5560 
5561 	/*
5562 	 * Verify that we can safely inquire about about any object,
5563 	 * whether it's allocated or not.  To make it interesting,
5564 	 * we probe a 5-wide window around each power of two.
5565 	 * This hits all edge cases, including zero and the max.
5566 	 */
5567 	for (int t = 0; t < 64; t++) {
5568 		for (int d = -5; d <= 5; d++) {
5569 			error = dmu_object_info(spa->spa_meta_objset,
5570 			    (1ULL << t) + d, NULL);
5571 			ASSERT(error == 0 || error == ENOENT ||
5572 			    error == EINVAL);
5573 		}
5574 	}
5575 
5576 	/*
5577 	 * If we got any ENOSPC errors on the previous run, destroy something.
5578 	 */
5579 	if (zs->zs_enospc_count != 0) {
5580 		int d = ztest_random(ztest_opts.zo_datasets);
5581 		ztest_dataset_destroy(d);
5582 	}
5583 	zs->zs_enospc_count = 0;
5584 
5585 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5586 	    UMEM_NOFAIL);
5587 
5588 	if (ztest_opts.zo_verbose >= 4)
5589 		(void) printf("starting main threads...\n");
5590 
5591 	/*
5592 	 * Kick off all the tests that run in parallel.
5593 	 */
5594 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
5595 		if (t < ztest_opts.zo_datasets &&
5596 		    ztest_dataset_open(t) != 0)
5597 			return;
5598 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5599 		    THR_BOUND, &tid[t]) == 0);
5600 	}
5601 
5602 	/*
5603 	 * Wait for all of the tests to complete.  We go in reverse order
5604 	 * so we don't close datasets while threads are still using them.
5605 	 */
5606 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5607 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5608 		if (t < ztest_opts.zo_datasets)
5609 			ztest_dataset_close(t);
5610 	}
5611 
5612 	txg_wait_synced(spa_get_dsl(spa), 0);
5613 
5614 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5615 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5616 
5617 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5618 
5619 	/* Kill the resume thread */
5620 	ztest_exiting = B_TRUE;
5621 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5622 	ztest_resume(spa);
5623 
5624 	/*
5625 	 * Right before closing the pool, kick off a bunch of async I/O;
5626 	 * spa_close() should wait for it to complete.
5627 	 */
5628 	for (uint64_t object = 1; object < 50; object++)
5629 		dmu_prefetch(spa->spa_meta_objset, object, 0, 1ULL << 20);
5630 
5631 	spa_close(spa, FTAG);
5632 
5633 	/*
5634 	 * Verify that we can loop over all pools.
5635 	 */
5636 	mutex_enter(&spa_namespace_lock);
5637 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5638 		if (ztest_opts.zo_verbose > 3)
5639 			(void) printf("spa_next: found %s\n", spa_name(spa));
5640 	mutex_exit(&spa_namespace_lock);
5641 
5642 	/*
5643 	 * Verify that we can export the pool and reimport it under a
5644 	 * different name.
5645 	 */
5646 	if (ztest_random(2) == 0) {
5647 		char name[MAXNAMELEN];
5648 		(void) snprintf(name, MAXNAMELEN, "%s_import",
5649 		    ztest_opts.zo_pool);
5650 		ztest_spa_import_export(ztest_opts.zo_pool, name);
5651 		ztest_spa_import_export(name, ztest_opts.zo_pool);
5652 	}
5653 
5654 	kernel_fini();
5655 
5656 	list_destroy(&zcl.zcl_callbacks);
5657 
5658 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5659 
5660 	(void) rwlock_destroy(&ztest_name_lock);
5661 	(void) _mutex_destroy(&ztest_vdev_lock);
5662 }
5663 
5664 static void
5665 ztest_freeze(void)
5666 {
5667 	ztest_ds_t *zd = &ztest_ds[0];
5668 	spa_t *spa;
5669 	int numloops = 0;
5670 
5671 	if (ztest_opts.zo_verbose >= 3)
5672 		(void) printf("testing spa_freeze()...\n");
5673 
5674 	kernel_init(FREAD | FWRITE);
5675 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5676 	VERIFY3U(0, ==, ztest_dataset_open(0));
5677 	spa->spa_debug = B_TRUE;
5678 	ztest_spa = spa;
5679 
5680 	/*
5681 	 * Force the first log block to be transactionally allocated.
5682 	 * We have to do this before we freeze the pool -- otherwise
5683 	 * the log chain won't be anchored.
5684 	 */
5685 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5686 		ztest_dmu_object_alloc_free(zd, 0);
5687 		zil_commit(zd->zd_zilog, 0);
5688 	}
5689 
5690 	txg_wait_synced(spa_get_dsl(spa), 0);
5691 
5692 	/*
5693 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5694 	 * so that the only way to record changes from now on is the ZIL.
5695 	 */
5696 	spa_freeze(spa);
5697 
5698 	/*
5699 	 * Run tests that generate log records but don't alter the pool config
5700 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5701 	 * We do a txg_wait_synced() after each iteration to force the txg
5702 	 * to increase well beyond the last synced value in the uberblock.
5703 	 * The ZIL should be OK with that.
5704 	 */
5705 	while (ztest_random(10) != 0 &&
5706 	    numloops++ < ztest_opts.zo_maxloops) {
5707 		ztest_dmu_write_parallel(zd, 0);
5708 		ztest_dmu_object_alloc_free(zd, 0);
5709 		txg_wait_synced(spa_get_dsl(spa), 0);
5710 	}
5711 
5712 	/*
5713 	 * Commit all of the changes we just generated.
5714 	 */
5715 	zil_commit(zd->zd_zilog, 0);
5716 	txg_wait_synced(spa_get_dsl(spa), 0);
5717 
5718 	/*
5719 	 * Close our dataset and close the pool.
5720 	 */
5721 	ztest_dataset_close(0);
5722 	spa_close(spa, FTAG);
5723 	kernel_fini();
5724 
5725 	/*
5726 	 * Open and close the pool and dataset to induce log replay.
5727 	 */
5728 	kernel_init(FREAD | FWRITE);
5729 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5730 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5731 	VERIFY3U(0, ==, ztest_dataset_open(0));
5732 	ztest_dataset_close(0);
5733 
5734 	spa->spa_debug = B_TRUE;
5735 	ztest_spa = spa;
5736 	txg_wait_synced(spa_get_dsl(spa), 0);
5737 	ztest_reguid(NULL, 0);
5738 
5739 	spa_close(spa, FTAG);
5740 	kernel_fini();
5741 }
5742 
5743 void
5744 print_time(hrtime_t t, char *timebuf)
5745 {
5746 	hrtime_t s = t / NANOSEC;
5747 	hrtime_t m = s / 60;
5748 	hrtime_t h = m / 60;
5749 	hrtime_t d = h / 24;
5750 
5751 	s -= m * 60;
5752 	m -= h * 60;
5753 	h -= d * 24;
5754 
5755 	timebuf[0] = '\0';
5756 
5757 	if (d)
5758 		(void) sprintf(timebuf,
5759 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5760 	else if (h)
5761 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5762 	else if (m)
5763 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5764 	else
5765 		(void) sprintf(timebuf, "%llus", s);
5766 }
5767 
5768 static nvlist_t *
5769 make_random_props()
5770 {
5771 	nvlist_t *props;
5772 
5773 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5774 	if (ztest_random(2) == 0)
5775 		return (props);
5776 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5777 
5778 	return (props);
5779 }
5780 
5781 /*
5782  * Create a storage pool with the given name and initial vdev size.
5783  * Then test spa_freeze() functionality.
5784  */
5785 static void
5786 ztest_init(ztest_shared_t *zs)
5787 {
5788 	spa_t *spa;
5789 	nvlist_t *nvroot, *props;
5790 
5791 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5792 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5793 
5794 	kernel_init(FREAD | FWRITE);
5795 
5796 	/*
5797 	 * Create the storage pool.
5798 	 */
5799 	(void) spa_destroy(ztest_opts.zo_pool);
5800 	ztest_shared->zs_vdev_next_leaf = 0;
5801 	zs->zs_splits = 0;
5802 	zs->zs_mirrors = ztest_opts.zo_mirrors;
5803 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5804 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5805 	props = make_random_props();
5806 	for (int i = 0; i < SPA_FEATURES; i++) {
5807 		char buf[1024];
5808 		(void) snprintf(buf, sizeof (buf), "feature@%s",
5809 		    spa_feature_table[i].fi_uname);
5810 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5811 	}
5812 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5813 	nvlist_free(nvroot);
5814 
5815 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5816 	zs->zs_metaslab_sz =
5817 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5818 
5819 	spa_close(spa, FTAG);
5820 
5821 	kernel_fini();
5822 
5823 	ztest_run_zdb(ztest_opts.zo_pool);
5824 
5825 	ztest_freeze();
5826 
5827 	ztest_run_zdb(ztest_opts.zo_pool);
5828 
5829 	(void) rwlock_destroy(&ztest_name_lock);
5830 	(void) _mutex_destroy(&ztest_vdev_lock);
5831 }
5832 
5833 static void
5834 setup_data_fd(void)
5835 {
5836 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5837 
5838 	ztest_fd_data = mkstemp(ztest_name_data);
5839 	ASSERT3S(ztest_fd_data, >=, 0);
5840 	(void) unlink(ztest_name_data);
5841 }
5842 
5843 
5844 static int
5845 shared_data_size(ztest_shared_hdr_t *hdr)
5846 {
5847 	int size;
5848 
5849 	size = hdr->zh_hdr_size;
5850 	size += hdr->zh_opts_size;
5851 	size += hdr->zh_size;
5852 	size += hdr->zh_stats_size * hdr->zh_stats_count;
5853 	size += hdr->zh_ds_size * hdr->zh_ds_count;
5854 
5855 	return (size);
5856 }
5857 
5858 static void
5859 setup_hdr(void)
5860 {
5861 	int size;
5862 	ztest_shared_hdr_t *hdr;
5863 
5864 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5865 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5866 	ASSERT(hdr != MAP_FAILED);
5867 
5868 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5869 
5870 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5871 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5872 	hdr->zh_size = sizeof (ztest_shared_t);
5873 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5874 	hdr->zh_stats_count = ZTEST_FUNCS;
5875 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5876 	hdr->zh_ds_count = ztest_opts.zo_datasets;
5877 
5878 	size = shared_data_size(hdr);
5879 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5880 
5881 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5882 }
5883 
5884 static void
5885 setup_data(void)
5886 {
5887 	int size, offset;
5888 	ztest_shared_hdr_t *hdr;
5889 	uint8_t *buf;
5890 
5891 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5892 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5893 	ASSERT(hdr != MAP_FAILED);
5894 
5895 	size = shared_data_size(hdr);
5896 
5897 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5898 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
5899 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5900 	ASSERT(hdr != MAP_FAILED);
5901 	buf = (uint8_t *)hdr;
5902 
5903 	offset = hdr->zh_hdr_size;
5904 	ztest_shared_opts = (void *)&buf[offset];
5905 	offset += hdr->zh_opts_size;
5906 	ztest_shared = (void *)&buf[offset];
5907 	offset += hdr->zh_size;
5908 	ztest_shared_callstate = (void *)&buf[offset];
5909 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
5910 	ztest_shared_ds = (void *)&buf[offset];
5911 }
5912 
5913 static boolean_t
5914 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
5915 {
5916 	pid_t pid;
5917 	int status;
5918 	char *cmdbuf = NULL;
5919 
5920 	pid = fork();
5921 
5922 	if (cmd == NULL) {
5923 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
5924 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
5925 		cmd = cmdbuf;
5926 	}
5927 
5928 	if (pid == -1)
5929 		fatal(1, "fork failed");
5930 
5931 	if (pid == 0) {	/* child */
5932 		char *emptyargv[2] = { cmd, NULL };
5933 		char fd_data_str[12];
5934 
5935 		struct rlimit rl = { 1024, 1024 };
5936 		(void) setrlimit(RLIMIT_NOFILE, &rl);
5937 
5938 		(void) close(ztest_fd_rand);
5939 		VERIFY3U(11, >=,
5940 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
5941 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
5942 
5943 		(void) enable_extended_FILE_stdio(-1, -1);
5944 		if (libpath != NULL)
5945 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
5946 		(void) execv(cmd, emptyargv);
5947 		ztest_dump_core = B_FALSE;
5948 		fatal(B_TRUE, "exec failed: %s", cmd);
5949 	}
5950 
5951 	if (cmdbuf != NULL) {
5952 		umem_free(cmdbuf, MAXPATHLEN);
5953 		cmd = NULL;
5954 	}
5955 
5956 	while (waitpid(pid, &status, 0) != pid)
5957 		continue;
5958 	if (statusp != NULL)
5959 		*statusp = status;
5960 
5961 	if (WIFEXITED(status)) {
5962 		if (WEXITSTATUS(status) != 0) {
5963 			(void) fprintf(stderr, "child exited with code %d\n",
5964 			    WEXITSTATUS(status));
5965 			exit(2);
5966 		}
5967 		return (B_FALSE);
5968 	} else if (WIFSIGNALED(status)) {
5969 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
5970 			(void) fprintf(stderr, "child died with signal %d\n",
5971 			    WTERMSIG(status));
5972 			exit(3);
5973 		}
5974 		return (B_TRUE);
5975 	} else {
5976 		(void) fprintf(stderr, "something strange happened to child\n");
5977 		exit(4);
5978 		/* NOTREACHED */
5979 	}
5980 }
5981 
5982 static void
5983 ztest_run_init(void)
5984 {
5985 	ztest_shared_t *zs = ztest_shared;
5986 
5987 	ASSERT(ztest_opts.zo_init != 0);
5988 
5989 	/*
5990 	 * Blow away any existing copy of zpool.cache
5991 	 */
5992 	(void) remove(spa_config_path);
5993 
5994 	/*
5995 	 * Create and initialize our storage pool.
5996 	 */
5997 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
5998 		bzero(zs, sizeof (ztest_shared_t));
5999 		if (ztest_opts.zo_verbose >= 3 &&
6000 		    ztest_opts.zo_init != 1) {
6001 			(void) printf("ztest_init(), pass %d\n", i);
6002 		}
6003 		ztest_init(zs);
6004 	}
6005 }
6006 
6007 int
6008 main(int argc, char **argv)
6009 {
6010 	int kills = 0;
6011 	int iters = 0;
6012 	int older = 0;
6013 	int newer = 0;
6014 	ztest_shared_t *zs;
6015 	ztest_info_t *zi;
6016 	ztest_shared_callstate_t *zc;
6017 	char timebuf[100];
6018 	char numbuf[6];
6019 	spa_t *spa;
6020 	char *cmd;
6021 	boolean_t hasalt;
6022 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6023 
6024 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6025 
6026 	dprintf_setup(&argc, argv);
6027 
6028 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6029 	ASSERT3S(ztest_fd_rand, >=, 0);
6030 
6031 	if (!fd_data_str) {
6032 		process_options(argc, argv);
6033 
6034 		setup_data_fd();
6035 		setup_hdr();
6036 		setup_data();
6037 		bcopy(&ztest_opts, ztest_shared_opts,
6038 		    sizeof (*ztest_shared_opts));
6039 	} else {
6040 		ztest_fd_data = atoi(fd_data_str);
6041 		setup_data();
6042 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6043 	}
6044 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6045 
6046 	/* Override location of zpool.cache */
6047 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6048 	    ztest_opts.zo_dir), !=, -1);
6049 
6050 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6051 	    UMEM_NOFAIL);
6052 	zs = ztest_shared;
6053 
6054 	if (fd_data_str) {
6055 		metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6056 		metaslab_df_alloc_threshold =
6057 		    zs->zs_metaslab_df_alloc_threshold;
6058 
6059 		if (zs->zs_do_init)
6060 			ztest_run_init();
6061 		else
6062 			ztest_run(zs);
6063 		exit(0);
6064 	}
6065 
6066 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6067 
6068 	if (ztest_opts.zo_verbose >= 1) {
6069 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6070 		    " %llu seconds...\n",
6071 		    (u_longlong_t)ztest_opts.zo_vdevs,
6072 		    ztest_opts.zo_datasets,
6073 		    ztest_opts.zo_threads,
6074 		    (u_longlong_t)ztest_opts.zo_time);
6075 	}
6076 
6077 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6078 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6079 
6080 	zs->zs_do_init = B_TRUE;
6081 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6082 		if (ztest_opts.zo_verbose >= 1) {
6083 			(void) printf("Executing older ztest for "
6084 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6085 		}
6086 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6087 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6088 	} else {
6089 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6090 	}
6091 	zs->zs_do_init = B_FALSE;
6092 
6093 	zs->zs_proc_start = gethrtime();
6094 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6095 
6096 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6097 		zi = &ztest_info[f];
6098 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6099 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6100 			zc->zc_next = UINT64_MAX;
6101 		else
6102 			zc->zc_next = zs->zs_proc_start +
6103 			    ztest_random(2 * zi->zi_interval[0] + 1);
6104 	}
6105 
6106 	/*
6107 	 * Run the tests in a loop.  These tests include fault injection
6108 	 * to verify that self-healing data works, and forced crashes
6109 	 * to verify that we never lose on-disk consistency.
6110 	 */
6111 	while (gethrtime() < zs->zs_proc_stop) {
6112 		int status;
6113 		boolean_t killed;
6114 
6115 		/*
6116 		 * Initialize the workload counters for each function.
6117 		 */
6118 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6119 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6120 			zc->zc_count = 0;
6121 			zc->zc_time = 0;
6122 		}
6123 
6124 		/* Set the allocation switch size */
6125 		zs->zs_metaslab_df_alloc_threshold =
6126 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6127 
6128 		if (!hasalt || ztest_random(2) == 0) {
6129 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6130 				(void) printf("Executing newer ztest: %s\n",
6131 				    cmd);
6132 			}
6133 			newer++;
6134 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6135 		} else {
6136 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6137 				(void) printf("Executing older ztest: %s\n",
6138 				    ztest_opts.zo_alt_ztest);
6139 			}
6140 			older++;
6141 			killed = exec_child(ztest_opts.zo_alt_ztest,
6142 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6143 		}
6144 
6145 		if (killed)
6146 			kills++;
6147 		iters++;
6148 
6149 		if (ztest_opts.zo_verbose >= 1) {
6150 			hrtime_t now = gethrtime();
6151 
6152 			now = MIN(now, zs->zs_proc_stop);
6153 			print_time(zs->zs_proc_stop - now, timebuf);
6154 			nicenum(zs->zs_space, numbuf);
6155 
6156 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6157 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6158 			    iters,
6159 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6160 			    (u_longlong_t)zs->zs_enospc_count,
6161 			    100.0 * zs->zs_alloc / zs->zs_space,
6162 			    numbuf,
6163 			    100.0 * (now - zs->zs_proc_start) /
6164 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6165 		}
6166 
6167 		if (ztest_opts.zo_verbose >= 2) {
6168 			(void) printf("\nWorkload summary:\n\n");
6169 			(void) printf("%7s %9s   %s\n",
6170 			    "Calls", "Time", "Function");
6171 			(void) printf("%7s %9s   %s\n",
6172 			    "-----", "----", "--------");
6173 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6174 				Dl_info dli;
6175 
6176 				zi = &ztest_info[f];
6177 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6178 				print_time(zc->zc_time, timebuf);
6179 				(void) dladdr((void *)zi->zi_func, &dli);
6180 				(void) printf("%7llu %9s   %s\n",
6181 				    (u_longlong_t)zc->zc_count, timebuf,
6182 				    dli.dli_sname);
6183 			}
6184 			(void) printf("\n");
6185 		}
6186 
6187 		/*
6188 		 * It's possible that we killed a child during a rename test,
6189 		 * in which case we'll have a 'ztest_tmp' pool lying around
6190 		 * instead of 'ztest'.  Do a blind rename in case this happened.
6191 		 */
6192 		kernel_init(FREAD);
6193 		if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6194 			spa_close(spa, FTAG);
6195 		} else {
6196 			char tmpname[MAXNAMELEN];
6197 			kernel_fini();
6198 			kernel_init(FREAD | FWRITE);
6199 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6200 			    ztest_opts.zo_pool);
6201 			(void) spa_rename(tmpname, ztest_opts.zo_pool);
6202 		}
6203 		kernel_fini();
6204 
6205 		ztest_run_zdb(ztest_opts.zo_pool);
6206 	}
6207 
6208 	if (ztest_opts.zo_verbose >= 1) {
6209 		if (hasalt) {
6210 			(void) printf("%d runs of older ztest: %s\n", older,
6211 			    ztest_opts.zo_alt_ztest);
6212 			(void) printf("%d runs of newer ztest: %s\n", newer,
6213 			    cmd);
6214 		}
6215 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6216 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6217 	}
6218 
6219 	umem_free(cmd, MAXNAMELEN);
6220 
6221 	return (0);
6222 }
6223