xref: /titanic_44/usr/src/cmd/ztest/ztest.c (revision fd435bccec40cb841f712413108d8577422fd9ab)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  */
27 
28 /*
29  * The objective of this program is to provide a DMU/ZAP/SPA stress test
30  * that runs entirely in userland, is easy to use, and easy to extend.
31  *
32  * The overall design of the ztest program is as follows:
33  *
34  * (1) For each major functional area (e.g. adding vdevs to a pool,
35  *     creating and destroying datasets, reading and writing objects, etc)
36  *     we have a simple routine to test that functionality.  These
37  *     individual routines do not have to do anything "stressful".
38  *
39  * (2) We turn these simple functionality tests into a stress test by
40  *     running them all in parallel, with as many threads as desired,
41  *     and spread across as many datasets, objects, and vdevs as desired.
42  *
43  * (3) While all this is happening, we inject faults into the pool to
44  *     verify that self-healing data really works.
45  *
46  * (4) Every time we open a dataset, we change its checksum and compression
47  *     functions.  Thus even individual objects vary from block to block
48  *     in which checksum they use and whether they're compressed.
49  *
50  * (5) To verify that we never lose on-disk consistency after a crash,
51  *     we run the entire test in a child of the main process.
52  *     At random times, the child self-immolates with a SIGKILL.
53  *     This is the software equivalent of pulling the power cord.
54  *     The parent then runs the test again, using the existing
55  *     storage pool, as many times as desired. If backwards compatibility
56  *     testing is enabled ztest will sometimes run the "older" version
57  *     of ztest after a SIGKILL.
58  *
59  * (6) To verify that we don't have future leaks or temporal incursions,
60  *     many of the functional tests record the transaction group number
61  *     as part of their data.  When reading old data, they verify that
62  *     the transaction group number is less than the current, open txg.
63  *     If you add a new test, please do this if applicable.
64  *
65  * When run with no arguments, ztest runs for about five minutes and
66  * produces no output if successful.  To get a little bit of information,
67  * specify -V.  To get more information, specify -VV, and so on.
68  *
69  * To turn this into an overnight stress test, use -T to specify run time.
70  *
71  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
72  * to increase the pool capacity, fanout, and overall stress level.
73  *
74  * Use the -k option to set the desired frequency of kills.
75  *
76  * When ztest invokes itself it passes all relevant information through a
77  * temporary file which is mmap-ed in the child process. This allows shared
78  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
79  * stored at offset 0 of this file and contains information on the size and
80  * number of shared structures in the file. The information stored in this file
81  * must remain backwards compatible with older versions of ztest so that
82  * ztest can invoke them during backwards compatibility testing (-B).
83  */
84 
85 #include <sys/zfs_context.h>
86 #include <sys/spa.h>
87 #include <sys/dmu.h>
88 #include <sys/txg.h>
89 #include <sys/dbuf.h>
90 #include <sys/zap.h>
91 #include <sys/dmu_objset.h>
92 #include <sys/poll.h>
93 #include <sys/stat.h>
94 #include <sys/time.h>
95 #include <sys/wait.h>
96 #include <sys/mman.h>
97 #include <sys/resource.h>
98 #include <sys/zio.h>
99 #include <sys/zil.h>
100 #include <sys/zil_impl.h>
101 #include <sys/vdev_impl.h>
102 #include <sys/vdev_file.h>
103 #include <sys/spa_impl.h>
104 #include <sys/metaslab_impl.h>
105 #include <sys/dsl_prop.h>
106 #include <sys/dsl_dataset.h>
107 #include <sys/dsl_destroy.h>
108 #include <sys/dsl_scan.h>
109 #include <sys/zio_checksum.h>
110 #include <sys/refcount.h>
111 #include <sys/zfeature.h>
112 #include <sys/dsl_userhold.h>
113 #include <stdio.h>
114 #include <stdio_ext.h>
115 #include <stdlib.h>
116 #include <unistd.h>
117 #include <signal.h>
118 #include <umem.h>
119 #include <dlfcn.h>
120 #include <ctype.h>
121 #include <math.h>
122 #include <sys/fs/zfs.h>
123 #include <libnvpair.h>
124 
125 static int ztest_fd_data = -1;
126 static int ztest_fd_rand = -1;
127 
128 typedef struct ztest_shared_hdr {
129 	uint64_t	zh_hdr_size;
130 	uint64_t	zh_opts_size;
131 	uint64_t	zh_size;
132 	uint64_t	zh_stats_size;
133 	uint64_t	zh_stats_count;
134 	uint64_t	zh_ds_size;
135 	uint64_t	zh_ds_count;
136 } ztest_shared_hdr_t;
137 
138 static ztest_shared_hdr_t *ztest_shared_hdr;
139 
140 typedef struct ztest_shared_opts {
141 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
142 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
143 	char zo_alt_ztest[MAXNAMELEN];
144 	char zo_alt_libpath[MAXNAMELEN];
145 	uint64_t zo_vdevs;
146 	uint64_t zo_vdevtime;
147 	size_t zo_vdev_size;
148 	int zo_ashift;
149 	int zo_mirrors;
150 	int zo_raidz;
151 	int zo_raidz_parity;
152 	int zo_datasets;
153 	int zo_threads;
154 	uint64_t zo_passtime;
155 	uint64_t zo_killrate;
156 	int zo_verbose;
157 	int zo_init;
158 	uint64_t zo_time;
159 	uint64_t zo_maxloops;
160 	uint64_t zo_metaslab_gang_bang;
161 } ztest_shared_opts_t;
162 
163 static const ztest_shared_opts_t ztest_opts_defaults = {
164 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
165 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
166 	.zo_alt_ztest = { '\0' },
167 	.zo_alt_libpath = { '\0' },
168 	.zo_vdevs = 5,
169 	.zo_ashift = SPA_MINBLOCKSHIFT,
170 	.zo_mirrors = 2,
171 	.zo_raidz = 4,
172 	.zo_raidz_parity = 1,
173 	.zo_vdev_size = SPA_MINDEVSIZE * 2,
174 	.zo_datasets = 7,
175 	.zo_threads = 23,
176 	.zo_passtime = 60,		/* 60 seconds */
177 	.zo_killrate = 70,		/* 70% kill rate */
178 	.zo_verbose = 0,
179 	.zo_init = 1,
180 	.zo_time = 300,			/* 5 minutes */
181 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
182 	.zo_metaslab_gang_bang = 32 << 10
183 };
184 
185 extern uint64_t metaslab_gang_bang;
186 extern uint64_t metaslab_df_alloc_threshold;
187 extern uint64_t zfs_deadman_synctime_ms;
188 extern int metaslab_preload_limit;
189 
190 static ztest_shared_opts_t *ztest_shared_opts;
191 static ztest_shared_opts_t ztest_opts;
192 
193 typedef struct ztest_shared_ds {
194 	uint64_t	zd_seq;
195 } ztest_shared_ds_t;
196 
197 static ztest_shared_ds_t *ztest_shared_ds;
198 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
199 
200 #define	BT_MAGIC	0x123456789abcdefULL
201 #define	MAXFAULTS() \
202 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
203 
204 enum ztest_io_type {
205 	ZTEST_IO_WRITE_TAG,
206 	ZTEST_IO_WRITE_PATTERN,
207 	ZTEST_IO_WRITE_ZEROES,
208 	ZTEST_IO_TRUNCATE,
209 	ZTEST_IO_SETATTR,
210 	ZTEST_IO_REWRITE,
211 	ZTEST_IO_TYPES
212 };
213 
214 typedef struct ztest_block_tag {
215 	uint64_t	bt_magic;
216 	uint64_t	bt_objset;
217 	uint64_t	bt_object;
218 	uint64_t	bt_offset;
219 	uint64_t	bt_gen;
220 	uint64_t	bt_txg;
221 	uint64_t	bt_crtxg;
222 } ztest_block_tag_t;
223 
224 typedef struct bufwad {
225 	uint64_t	bw_index;
226 	uint64_t	bw_txg;
227 	uint64_t	bw_data;
228 } bufwad_t;
229 
230 /*
231  * XXX -- fix zfs range locks to be generic so we can use them here.
232  */
233 typedef enum {
234 	RL_READER,
235 	RL_WRITER,
236 	RL_APPEND
237 } rl_type_t;
238 
239 typedef struct rll {
240 	void		*rll_writer;
241 	int		rll_readers;
242 	mutex_t		rll_lock;
243 	cond_t		rll_cv;
244 } rll_t;
245 
246 typedef struct rl {
247 	uint64_t	rl_object;
248 	uint64_t	rl_offset;
249 	uint64_t	rl_size;
250 	rll_t		*rl_lock;
251 } rl_t;
252 
253 #define	ZTEST_RANGE_LOCKS	64
254 #define	ZTEST_OBJECT_LOCKS	64
255 
256 /*
257  * Object descriptor.  Used as a template for object lookup/create/remove.
258  */
259 typedef struct ztest_od {
260 	uint64_t	od_dir;
261 	uint64_t	od_object;
262 	dmu_object_type_t od_type;
263 	dmu_object_type_t od_crtype;
264 	uint64_t	od_blocksize;
265 	uint64_t	od_crblocksize;
266 	uint64_t	od_gen;
267 	uint64_t	od_crgen;
268 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
269 } ztest_od_t;
270 
271 /*
272  * Per-dataset state.
273  */
274 typedef struct ztest_ds {
275 	ztest_shared_ds_t *zd_shared;
276 	objset_t	*zd_os;
277 	rwlock_t	zd_zilog_lock;
278 	zilog_t		*zd_zilog;
279 	ztest_od_t	*zd_od;		/* debugging aid */
280 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
281 	mutex_t		zd_dirobj_lock;
282 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
283 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
284 } ztest_ds_t;
285 
286 /*
287  * Per-iteration state.
288  */
289 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
290 
291 typedef struct ztest_info {
292 	ztest_func_t	*zi_func;	/* test function */
293 	uint64_t	zi_iters;	/* iterations per execution */
294 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
295 } ztest_info_t;
296 
297 typedef struct ztest_shared_callstate {
298 	uint64_t	zc_count;	/* per-pass count */
299 	uint64_t	zc_time;	/* per-pass time */
300 	uint64_t	zc_next;	/* next time to call this function */
301 } ztest_shared_callstate_t;
302 
303 static ztest_shared_callstate_t *ztest_shared_callstate;
304 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
305 
306 /*
307  * Note: these aren't static because we want dladdr() to work.
308  */
309 ztest_func_t ztest_dmu_read_write;
310 ztest_func_t ztest_dmu_write_parallel;
311 ztest_func_t ztest_dmu_object_alloc_free;
312 ztest_func_t ztest_dmu_commit_callbacks;
313 ztest_func_t ztest_zap;
314 ztest_func_t ztest_zap_parallel;
315 ztest_func_t ztest_zil_commit;
316 ztest_func_t ztest_zil_remount;
317 ztest_func_t ztest_dmu_read_write_zcopy;
318 ztest_func_t ztest_dmu_objset_create_destroy;
319 ztest_func_t ztest_dmu_prealloc;
320 ztest_func_t ztest_fzap;
321 ztest_func_t ztest_dmu_snapshot_create_destroy;
322 ztest_func_t ztest_dsl_prop_get_set;
323 ztest_func_t ztest_spa_prop_get_set;
324 ztest_func_t ztest_spa_create_destroy;
325 ztest_func_t ztest_fault_inject;
326 ztest_func_t ztest_ddt_repair;
327 ztest_func_t ztest_dmu_snapshot_hold;
328 ztest_func_t ztest_spa_rename;
329 ztest_func_t ztest_scrub;
330 ztest_func_t ztest_dsl_dataset_promote_busy;
331 ztest_func_t ztest_vdev_attach_detach;
332 ztest_func_t ztest_vdev_LUN_growth;
333 ztest_func_t ztest_vdev_add_remove;
334 ztest_func_t ztest_vdev_aux_add_remove;
335 ztest_func_t ztest_split_pool;
336 ztest_func_t ztest_reguid;
337 ztest_func_t ztest_spa_upgrade;
338 
339 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
340 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
341 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
342 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
343 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
344 
345 ztest_info_t ztest_info[] = {
346 	{ ztest_dmu_read_write,			1,	&zopt_always	},
347 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
348 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
349 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
350 	{ ztest_zap,				30,	&zopt_always	},
351 	{ ztest_zap_parallel,			100,	&zopt_always	},
352 	{ ztest_split_pool,			1,	&zopt_always	},
353 	{ ztest_zil_commit,			1,	&zopt_incessant	},
354 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
355 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
356 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
357 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
358 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
359 #if 0
360 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
361 #endif
362 	{ ztest_fzap,				1,	&zopt_sometimes	},
363 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
364 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
365 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
366 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
367 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
368 	{ ztest_reguid,				1,	&zopt_rarely	},
369 	{ ztest_spa_rename,			1,	&zopt_rarely	},
370 	{ ztest_scrub,				1,	&zopt_rarely	},
371 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
372 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
373 	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
374 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
375 	{ ztest_vdev_add_remove,		1,
376 	    &ztest_opts.zo_vdevtime				},
377 	{ ztest_vdev_aux_add_remove,		1,
378 	    &ztest_opts.zo_vdevtime				},
379 };
380 
381 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
382 
383 /*
384  * The following struct is used to hold a list of uncalled commit callbacks.
385  * The callbacks are ordered by txg number.
386  */
387 typedef struct ztest_cb_list {
388 	mutex_t	zcl_callbacks_lock;
389 	list_t	zcl_callbacks;
390 } ztest_cb_list_t;
391 
392 /*
393  * Stuff we need to share writably between parent and child.
394  */
395 typedef struct ztest_shared {
396 	boolean_t	zs_do_init;
397 	hrtime_t	zs_proc_start;
398 	hrtime_t	zs_proc_stop;
399 	hrtime_t	zs_thread_start;
400 	hrtime_t	zs_thread_stop;
401 	hrtime_t	zs_thread_kill;
402 	uint64_t	zs_enospc_count;
403 	uint64_t	zs_vdev_next_leaf;
404 	uint64_t	zs_vdev_aux;
405 	uint64_t	zs_alloc;
406 	uint64_t	zs_space;
407 	uint64_t	zs_splits;
408 	uint64_t	zs_mirrors;
409 	uint64_t	zs_metaslab_sz;
410 	uint64_t	zs_metaslab_df_alloc_threshold;
411 	uint64_t	zs_guid;
412 } ztest_shared_t;
413 
414 #define	ID_PARALLEL	-1ULL
415 
416 static char ztest_dev_template[] = "%s/%s.%llua";
417 static char ztest_aux_template[] = "%s/%s.%s.%llu";
418 ztest_shared_t *ztest_shared;
419 
420 static spa_t *ztest_spa = NULL;
421 static ztest_ds_t *ztest_ds;
422 
423 static mutex_t ztest_vdev_lock;
424 
425 /*
426  * The ztest_name_lock protects the pool and dataset namespace used by
427  * the individual tests. To modify the namespace, consumers must grab
428  * this lock as writer. Grabbing the lock as reader will ensure that the
429  * namespace does not change while the lock is held.
430  */
431 static rwlock_t ztest_name_lock;
432 
433 static boolean_t ztest_dump_core = B_TRUE;
434 static boolean_t ztest_exiting;
435 
436 /* Global commit callback list */
437 static ztest_cb_list_t zcl;
438 
439 enum ztest_object {
440 	ZTEST_META_DNODE = 0,
441 	ZTEST_DIROBJ,
442 	ZTEST_OBJECTS
443 };
444 
445 static void usage(boolean_t) __NORETURN;
446 
447 /*
448  * These libumem hooks provide a reasonable set of defaults for the allocator's
449  * debugging facilities.
450  */
451 const char *
452 _umem_debug_init()
453 {
454 	return ("default,verbose"); /* $UMEM_DEBUG setting */
455 }
456 
457 const char *
458 _umem_logging_init(void)
459 {
460 	return ("fail,contents"); /* $UMEM_LOGGING setting */
461 }
462 
463 #define	FATAL_MSG_SZ	1024
464 
465 char *fatal_msg;
466 
467 static void
468 fatal(int do_perror, char *message, ...)
469 {
470 	va_list args;
471 	int save_errno = errno;
472 	char buf[FATAL_MSG_SZ];
473 
474 	(void) fflush(stdout);
475 
476 	va_start(args, message);
477 	(void) sprintf(buf, "ztest: ");
478 	/* LINTED */
479 	(void) vsprintf(buf + strlen(buf), message, args);
480 	va_end(args);
481 	if (do_perror) {
482 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
483 		    ": %s", strerror(save_errno));
484 	}
485 	(void) fprintf(stderr, "%s\n", buf);
486 	fatal_msg = buf;			/* to ease debugging */
487 	if (ztest_dump_core)
488 		abort();
489 	exit(3);
490 }
491 
492 static int
493 str2shift(const char *buf)
494 {
495 	const char *ends = "BKMGTPEZ";
496 	int i;
497 
498 	if (buf[0] == '\0')
499 		return (0);
500 	for (i = 0; i < strlen(ends); i++) {
501 		if (toupper(buf[0]) == ends[i])
502 			break;
503 	}
504 	if (i == strlen(ends)) {
505 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
506 		    buf);
507 		usage(B_FALSE);
508 	}
509 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
510 		return (10*i);
511 	}
512 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
513 	usage(B_FALSE);
514 	/* NOTREACHED */
515 }
516 
517 static uint64_t
518 nicenumtoull(const char *buf)
519 {
520 	char *end;
521 	uint64_t val;
522 
523 	val = strtoull(buf, &end, 0);
524 	if (end == buf) {
525 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
526 		usage(B_FALSE);
527 	} else if (end[0] == '.') {
528 		double fval = strtod(buf, &end);
529 		fval *= pow(2, str2shift(end));
530 		if (fval > UINT64_MAX) {
531 			(void) fprintf(stderr, "ztest: value too large: %s\n",
532 			    buf);
533 			usage(B_FALSE);
534 		}
535 		val = (uint64_t)fval;
536 	} else {
537 		int shift = str2shift(end);
538 		if (shift >= 64 || (val << shift) >> shift != val) {
539 			(void) fprintf(stderr, "ztest: value too large: %s\n",
540 			    buf);
541 			usage(B_FALSE);
542 		}
543 		val <<= shift;
544 	}
545 	return (val);
546 }
547 
548 static void
549 usage(boolean_t requested)
550 {
551 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
552 
553 	char nice_vdev_size[10];
554 	char nice_gang_bang[10];
555 	FILE *fp = requested ? stdout : stderr;
556 
557 	nicenum(zo->zo_vdev_size, nice_vdev_size);
558 	nicenum(zo->zo_metaslab_gang_bang, nice_gang_bang);
559 
560 	(void) fprintf(fp, "Usage: %s\n"
561 	    "\t[-v vdevs (default: %llu)]\n"
562 	    "\t[-s size_of_each_vdev (default: %s)]\n"
563 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
564 	    "\t[-m mirror_copies (default: %d)]\n"
565 	    "\t[-r raidz_disks (default: %d)]\n"
566 	    "\t[-R raidz_parity (default: %d)]\n"
567 	    "\t[-d datasets (default: %d)]\n"
568 	    "\t[-t threads (default: %d)]\n"
569 	    "\t[-g gang_block_threshold (default: %s)]\n"
570 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
571 	    "\t[-k kill_percentage (default: %llu%%)]\n"
572 	    "\t[-p pool_name (default: %s)]\n"
573 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
574 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
575 	    "\t[-E] use existing pool instead of creating new one\n"
576 	    "\t[-T time (default: %llu sec)] total run time\n"
577 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
578 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
579 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
580 	    "\t[-h] (print help)\n"
581 	    "",
582 	    zo->zo_pool,
583 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
584 	    nice_vdev_size,				/* -s */
585 	    zo->zo_ashift,				/* -a */
586 	    zo->zo_mirrors,				/* -m */
587 	    zo->zo_raidz,				/* -r */
588 	    zo->zo_raidz_parity,			/* -R */
589 	    zo->zo_datasets,				/* -d */
590 	    zo->zo_threads,				/* -t */
591 	    nice_gang_bang,				/* -g */
592 	    zo->zo_init,				/* -i */
593 	    (u_longlong_t)zo->zo_killrate,		/* -k */
594 	    zo->zo_pool,				/* -p */
595 	    zo->zo_dir,					/* -f */
596 	    (u_longlong_t)zo->zo_time,			/* -T */
597 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
598 	    (u_longlong_t)zo->zo_passtime);
599 	exit(requested ? 0 : 1);
600 }
601 
602 static void
603 process_options(int argc, char **argv)
604 {
605 	char *path;
606 	ztest_shared_opts_t *zo = &ztest_opts;
607 
608 	int opt;
609 	uint64_t value;
610 	char altdir[MAXNAMELEN] = { 0 };
611 
612 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
613 
614 	while ((opt = getopt(argc, argv,
615 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:VET:P:hF:B:")) != EOF) {
616 		value = 0;
617 		switch (opt) {
618 		case 'v':
619 		case 's':
620 		case 'a':
621 		case 'm':
622 		case 'r':
623 		case 'R':
624 		case 'd':
625 		case 't':
626 		case 'g':
627 		case 'i':
628 		case 'k':
629 		case 'T':
630 		case 'P':
631 		case 'F':
632 			value = nicenumtoull(optarg);
633 		}
634 		switch (opt) {
635 		case 'v':
636 			zo->zo_vdevs = value;
637 			break;
638 		case 's':
639 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
640 			break;
641 		case 'a':
642 			zo->zo_ashift = value;
643 			break;
644 		case 'm':
645 			zo->zo_mirrors = value;
646 			break;
647 		case 'r':
648 			zo->zo_raidz = MAX(1, value);
649 			break;
650 		case 'R':
651 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
652 			break;
653 		case 'd':
654 			zo->zo_datasets = MAX(1, value);
655 			break;
656 		case 't':
657 			zo->zo_threads = MAX(1, value);
658 			break;
659 		case 'g':
660 			zo->zo_metaslab_gang_bang = MAX(SPA_MINBLOCKSIZE << 1,
661 			    value);
662 			break;
663 		case 'i':
664 			zo->zo_init = value;
665 			break;
666 		case 'k':
667 			zo->zo_killrate = value;
668 			break;
669 		case 'p':
670 			(void) strlcpy(zo->zo_pool, optarg,
671 			    sizeof (zo->zo_pool));
672 			break;
673 		case 'f':
674 			path = realpath(optarg, NULL);
675 			if (path == NULL) {
676 				(void) fprintf(stderr, "error: %s: %s\n",
677 				    optarg, strerror(errno));
678 				usage(B_FALSE);
679 			} else {
680 				(void) strlcpy(zo->zo_dir, path,
681 				    sizeof (zo->zo_dir));
682 			}
683 			break;
684 		case 'V':
685 			zo->zo_verbose++;
686 			break;
687 		case 'E':
688 			zo->zo_init = 0;
689 			break;
690 		case 'T':
691 			zo->zo_time = value;
692 			break;
693 		case 'P':
694 			zo->zo_passtime = MAX(1, value);
695 			break;
696 		case 'F':
697 			zo->zo_maxloops = MAX(1, value);
698 			break;
699 		case 'B':
700 			(void) strlcpy(altdir, optarg, sizeof (altdir));
701 			break;
702 		case 'h':
703 			usage(B_TRUE);
704 			break;
705 		case '?':
706 		default:
707 			usage(B_FALSE);
708 			break;
709 		}
710 	}
711 
712 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
713 
714 	zo->zo_vdevtime =
715 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
716 	    UINT64_MAX >> 2);
717 
718 	if (strlen(altdir) > 0) {
719 		char *cmd;
720 		char *realaltdir;
721 		char *bin;
722 		char *ztest;
723 		char *isa;
724 		int isalen;
725 
726 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
727 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
728 
729 		VERIFY(NULL != realpath(getexecname(), cmd));
730 		if (0 != access(altdir, F_OK)) {
731 			ztest_dump_core = B_FALSE;
732 			fatal(B_TRUE, "invalid alternate ztest path: %s",
733 			    altdir);
734 		}
735 		VERIFY(NULL != realpath(altdir, realaltdir));
736 
737 		/*
738 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
739 		 * We want to extract <isa> to determine if we should use
740 		 * 32 or 64 bit binaries.
741 		 */
742 		bin = strstr(cmd, "/usr/bin/");
743 		ztest = strstr(bin, "/ztest");
744 		isa = bin + 9;
745 		isalen = ztest - isa;
746 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
747 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
748 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
749 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
750 
751 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
752 			ztest_dump_core = B_FALSE;
753 			fatal(B_TRUE, "invalid alternate ztest: %s",
754 			    zo->zo_alt_ztest);
755 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
756 			ztest_dump_core = B_FALSE;
757 			fatal(B_TRUE, "invalid alternate lib directory %s",
758 			    zo->zo_alt_libpath);
759 		}
760 
761 		umem_free(cmd, MAXPATHLEN);
762 		umem_free(realaltdir, MAXPATHLEN);
763 	}
764 }
765 
766 static void
767 ztest_kill(ztest_shared_t *zs)
768 {
769 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
770 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
771 
772 	/*
773 	 * Before we kill off ztest, make sure that the config is updated.
774 	 * See comment above spa_config_sync().
775 	 */
776 	mutex_enter(&spa_namespace_lock);
777 	spa_config_sync(ztest_spa, B_FALSE, B_FALSE);
778 	mutex_exit(&spa_namespace_lock);
779 
780 	zfs_dbgmsg_print(FTAG);
781 	(void) kill(getpid(), SIGKILL);
782 }
783 
784 static uint64_t
785 ztest_random(uint64_t range)
786 {
787 	uint64_t r;
788 
789 	ASSERT3S(ztest_fd_rand, >=, 0);
790 
791 	if (range == 0)
792 		return (0);
793 
794 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
795 		fatal(1, "short read from /dev/urandom");
796 
797 	return (r % range);
798 }
799 
800 /* ARGSUSED */
801 static void
802 ztest_record_enospc(const char *s)
803 {
804 	ztest_shared->zs_enospc_count++;
805 }
806 
807 static uint64_t
808 ztest_get_ashift(void)
809 {
810 	if (ztest_opts.zo_ashift == 0)
811 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
812 	return (ztest_opts.zo_ashift);
813 }
814 
815 static nvlist_t *
816 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
817 {
818 	char pathbuf[MAXPATHLEN];
819 	uint64_t vdev;
820 	nvlist_t *file;
821 
822 	if (ashift == 0)
823 		ashift = ztest_get_ashift();
824 
825 	if (path == NULL) {
826 		path = pathbuf;
827 
828 		if (aux != NULL) {
829 			vdev = ztest_shared->zs_vdev_aux;
830 			(void) snprintf(path, sizeof (pathbuf),
831 			    ztest_aux_template, ztest_opts.zo_dir,
832 			    pool == NULL ? ztest_opts.zo_pool : pool,
833 			    aux, vdev);
834 		} else {
835 			vdev = ztest_shared->zs_vdev_next_leaf++;
836 			(void) snprintf(path, sizeof (pathbuf),
837 			    ztest_dev_template, ztest_opts.zo_dir,
838 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
839 		}
840 	}
841 
842 	if (size != 0) {
843 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
844 		if (fd == -1)
845 			fatal(1, "can't open %s", path);
846 		if (ftruncate(fd, size) != 0)
847 			fatal(1, "can't ftruncate %s", path);
848 		(void) close(fd);
849 	}
850 
851 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
852 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
853 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
854 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
855 
856 	return (file);
857 }
858 
859 static nvlist_t *
860 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
861     uint64_t ashift, int r)
862 {
863 	nvlist_t *raidz, **child;
864 	int c;
865 
866 	if (r < 2)
867 		return (make_vdev_file(path, aux, pool, size, ashift));
868 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
869 
870 	for (c = 0; c < r; c++)
871 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
872 
873 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
874 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
875 	    VDEV_TYPE_RAIDZ) == 0);
876 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
877 	    ztest_opts.zo_raidz_parity) == 0);
878 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
879 	    child, r) == 0);
880 
881 	for (c = 0; c < r; c++)
882 		nvlist_free(child[c]);
883 
884 	umem_free(child, r * sizeof (nvlist_t *));
885 
886 	return (raidz);
887 }
888 
889 static nvlist_t *
890 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
891     uint64_t ashift, int r, int m)
892 {
893 	nvlist_t *mirror, **child;
894 	int c;
895 
896 	if (m < 1)
897 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
898 
899 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
900 
901 	for (c = 0; c < m; c++)
902 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
903 
904 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
905 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
906 	    VDEV_TYPE_MIRROR) == 0);
907 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
908 	    child, m) == 0);
909 
910 	for (c = 0; c < m; c++)
911 		nvlist_free(child[c]);
912 
913 	umem_free(child, m * sizeof (nvlist_t *));
914 
915 	return (mirror);
916 }
917 
918 static nvlist_t *
919 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
920     int log, int r, int m, int t)
921 {
922 	nvlist_t *root, **child;
923 	int c;
924 
925 	ASSERT(t > 0);
926 
927 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
928 
929 	for (c = 0; c < t; c++) {
930 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
931 		    r, m);
932 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
933 		    log) == 0);
934 	}
935 
936 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
937 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
938 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
939 	    child, t) == 0);
940 
941 	for (c = 0; c < t; c++)
942 		nvlist_free(child[c]);
943 
944 	umem_free(child, t * sizeof (nvlist_t *));
945 
946 	return (root);
947 }
948 
949 /*
950  * Find a random spa version. Returns back a random spa version in the
951  * range [initial_version, SPA_VERSION_FEATURES].
952  */
953 static uint64_t
954 ztest_random_spa_version(uint64_t initial_version)
955 {
956 	uint64_t version = initial_version;
957 
958 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
959 		version = version +
960 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
961 	}
962 
963 	if (version > SPA_VERSION_BEFORE_FEATURES)
964 		version = SPA_VERSION_FEATURES;
965 
966 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
967 	return (version);
968 }
969 
970 static int
971 ztest_random_blocksize(void)
972 {
973 	uint64_t block_shift;
974 	/*
975 	 * Choose a block size >= the ashift.
976 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
977 	 */
978 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
979 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
980 		maxbs = 20;
981 	block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
982 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
983 }
984 
985 static int
986 ztest_random_ibshift(void)
987 {
988 	return (DN_MIN_INDBLKSHIFT +
989 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
990 }
991 
992 static uint64_t
993 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
994 {
995 	uint64_t top;
996 	vdev_t *rvd = spa->spa_root_vdev;
997 	vdev_t *tvd;
998 
999 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1000 
1001 	do {
1002 		top = ztest_random(rvd->vdev_children);
1003 		tvd = rvd->vdev_child[top];
1004 	} while (tvd->vdev_ishole || (tvd->vdev_islog && !log_ok) ||
1005 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1006 
1007 	return (top);
1008 }
1009 
1010 static uint64_t
1011 ztest_random_dsl_prop(zfs_prop_t prop)
1012 {
1013 	uint64_t value;
1014 
1015 	do {
1016 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1017 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1018 
1019 	return (value);
1020 }
1021 
1022 static int
1023 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1024     boolean_t inherit)
1025 {
1026 	const char *propname = zfs_prop_to_name(prop);
1027 	const char *valname;
1028 	char setpoint[MAXPATHLEN];
1029 	uint64_t curval;
1030 	int error;
1031 
1032 	error = dsl_prop_set_int(osname, propname,
1033 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1034 
1035 	if (error == ENOSPC) {
1036 		ztest_record_enospc(FTAG);
1037 		return (error);
1038 	}
1039 	ASSERT0(error);
1040 
1041 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1042 
1043 	if (ztest_opts.zo_verbose >= 6) {
1044 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1045 		(void) printf("%s %s = %s at '%s'\n",
1046 		    osname, propname, valname, setpoint);
1047 	}
1048 
1049 	return (error);
1050 }
1051 
1052 static int
1053 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1054 {
1055 	spa_t *spa = ztest_spa;
1056 	nvlist_t *props = NULL;
1057 	int error;
1058 
1059 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1060 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1061 
1062 	error = spa_prop_set(spa, props);
1063 
1064 	nvlist_free(props);
1065 
1066 	if (error == ENOSPC) {
1067 		ztest_record_enospc(FTAG);
1068 		return (error);
1069 	}
1070 	ASSERT0(error);
1071 
1072 	return (error);
1073 }
1074 
1075 static void
1076 ztest_rll_init(rll_t *rll)
1077 {
1078 	rll->rll_writer = NULL;
1079 	rll->rll_readers = 0;
1080 	VERIFY(_mutex_init(&rll->rll_lock, USYNC_THREAD, NULL) == 0);
1081 	VERIFY(cond_init(&rll->rll_cv, USYNC_THREAD, NULL) == 0);
1082 }
1083 
1084 static void
1085 ztest_rll_destroy(rll_t *rll)
1086 {
1087 	ASSERT(rll->rll_writer == NULL);
1088 	ASSERT(rll->rll_readers == 0);
1089 	VERIFY(_mutex_destroy(&rll->rll_lock) == 0);
1090 	VERIFY(cond_destroy(&rll->rll_cv) == 0);
1091 }
1092 
1093 static void
1094 ztest_rll_lock(rll_t *rll, rl_type_t type)
1095 {
1096 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1097 
1098 	if (type == RL_READER) {
1099 		while (rll->rll_writer != NULL)
1100 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1101 		rll->rll_readers++;
1102 	} else {
1103 		while (rll->rll_writer != NULL || rll->rll_readers)
1104 			(void) cond_wait(&rll->rll_cv, &rll->rll_lock);
1105 		rll->rll_writer = curthread;
1106 	}
1107 
1108 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1109 }
1110 
1111 static void
1112 ztest_rll_unlock(rll_t *rll)
1113 {
1114 	VERIFY(mutex_lock(&rll->rll_lock) == 0);
1115 
1116 	if (rll->rll_writer) {
1117 		ASSERT(rll->rll_readers == 0);
1118 		rll->rll_writer = NULL;
1119 	} else {
1120 		ASSERT(rll->rll_readers != 0);
1121 		ASSERT(rll->rll_writer == NULL);
1122 		rll->rll_readers--;
1123 	}
1124 
1125 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1126 		VERIFY(cond_broadcast(&rll->rll_cv) == 0);
1127 
1128 	VERIFY(mutex_unlock(&rll->rll_lock) == 0);
1129 }
1130 
1131 static void
1132 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1133 {
1134 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1135 
1136 	ztest_rll_lock(rll, type);
1137 }
1138 
1139 static void
1140 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1141 {
1142 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1143 
1144 	ztest_rll_unlock(rll);
1145 }
1146 
1147 static rl_t *
1148 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1149     uint64_t size, rl_type_t type)
1150 {
1151 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1152 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1153 	rl_t *rl;
1154 
1155 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1156 	rl->rl_object = object;
1157 	rl->rl_offset = offset;
1158 	rl->rl_size = size;
1159 	rl->rl_lock = rll;
1160 
1161 	ztest_rll_lock(rll, type);
1162 
1163 	return (rl);
1164 }
1165 
1166 static void
1167 ztest_range_unlock(rl_t *rl)
1168 {
1169 	rll_t *rll = rl->rl_lock;
1170 
1171 	ztest_rll_unlock(rll);
1172 
1173 	umem_free(rl, sizeof (*rl));
1174 }
1175 
1176 static void
1177 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1178 {
1179 	zd->zd_os = os;
1180 	zd->zd_zilog = dmu_objset_zil(os);
1181 	zd->zd_shared = szd;
1182 	dmu_objset_name(os, zd->zd_name);
1183 
1184 	if (zd->zd_shared != NULL)
1185 		zd->zd_shared->zd_seq = 0;
1186 
1187 	VERIFY(rwlock_init(&zd->zd_zilog_lock, USYNC_THREAD, NULL) == 0);
1188 	VERIFY(_mutex_init(&zd->zd_dirobj_lock, USYNC_THREAD, NULL) == 0);
1189 
1190 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1191 		ztest_rll_init(&zd->zd_object_lock[l]);
1192 
1193 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1194 		ztest_rll_init(&zd->zd_range_lock[l]);
1195 }
1196 
1197 static void
1198 ztest_zd_fini(ztest_ds_t *zd)
1199 {
1200 	VERIFY(_mutex_destroy(&zd->zd_dirobj_lock) == 0);
1201 
1202 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1203 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1204 
1205 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1206 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1207 }
1208 
1209 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1210 
1211 static uint64_t
1212 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1213 {
1214 	uint64_t txg;
1215 	int error;
1216 
1217 	/*
1218 	 * Attempt to assign tx to some transaction group.
1219 	 */
1220 	error = dmu_tx_assign(tx, txg_how);
1221 	if (error) {
1222 		if (error == ERESTART) {
1223 			ASSERT(txg_how == TXG_NOWAIT);
1224 			dmu_tx_wait(tx);
1225 		} else {
1226 			ASSERT3U(error, ==, ENOSPC);
1227 			ztest_record_enospc(tag);
1228 		}
1229 		dmu_tx_abort(tx);
1230 		return (0);
1231 	}
1232 	txg = dmu_tx_get_txg(tx);
1233 	ASSERT(txg != 0);
1234 	return (txg);
1235 }
1236 
1237 static void
1238 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1239 {
1240 	uint64_t *ip = buf;
1241 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1242 
1243 	while (ip < ip_end)
1244 		*ip++ = value;
1245 }
1246 
1247 static boolean_t
1248 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1249 {
1250 	uint64_t *ip = buf;
1251 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1252 	uint64_t diff = 0;
1253 
1254 	while (ip < ip_end)
1255 		diff |= (value - *ip++);
1256 
1257 	return (diff == 0);
1258 }
1259 
1260 static void
1261 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1262     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1263 {
1264 	bt->bt_magic = BT_MAGIC;
1265 	bt->bt_objset = dmu_objset_id(os);
1266 	bt->bt_object = object;
1267 	bt->bt_offset = offset;
1268 	bt->bt_gen = gen;
1269 	bt->bt_txg = txg;
1270 	bt->bt_crtxg = crtxg;
1271 }
1272 
1273 static void
1274 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1275     uint64_t offset, uint64_t gen, uint64_t txg, uint64_t crtxg)
1276 {
1277 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1278 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1279 	ASSERT3U(bt->bt_object, ==, object);
1280 	ASSERT3U(bt->bt_offset, ==, offset);
1281 	ASSERT3U(bt->bt_gen, <=, gen);
1282 	ASSERT3U(bt->bt_txg, <=, txg);
1283 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1284 }
1285 
1286 static ztest_block_tag_t *
1287 ztest_bt_bonus(dmu_buf_t *db)
1288 {
1289 	dmu_object_info_t doi;
1290 	ztest_block_tag_t *bt;
1291 
1292 	dmu_object_info_from_db(db, &doi);
1293 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1294 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1295 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1296 
1297 	return (bt);
1298 }
1299 
1300 /*
1301  * ZIL logging ops
1302  */
1303 
1304 #define	lrz_type	lr_mode
1305 #define	lrz_blocksize	lr_uid
1306 #define	lrz_ibshift	lr_gid
1307 #define	lrz_bonustype	lr_rdev
1308 #define	lrz_bonuslen	lr_crtime[1]
1309 
1310 static void
1311 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1312 {
1313 	char *name = (void *)(lr + 1);		/* name follows lr */
1314 	size_t namesize = strlen(name) + 1;
1315 	itx_t *itx;
1316 
1317 	if (zil_replaying(zd->zd_zilog, tx))
1318 		return;
1319 
1320 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1321 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1322 	    sizeof (*lr) + namesize - sizeof (lr_t));
1323 
1324 	zil_itx_assign(zd->zd_zilog, itx, tx);
1325 }
1326 
1327 static void
1328 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1329 {
1330 	char *name = (void *)(lr + 1);		/* name follows lr */
1331 	size_t namesize = strlen(name) + 1;
1332 	itx_t *itx;
1333 
1334 	if (zil_replaying(zd->zd_zilog, tx))
1335 		return;
1336 
1337 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1338 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1339 	    sizeof (*lr) + namesize - sizeof (lr_t));
1340 
1341 	itx->itx_oid = object;
1342 	zil_itx_assign(zd->zd_zilog, itx, tx);
1343 }
1344 
1345 static void
1346 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1347 {
1348 	itx_t *itx;
1349 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1350 
1351 	if (zil_replaying(zd->zd_zilog, tx))
1352 		return;
1353 
1354 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1355 		write_state = WR_INDIRECT;
1356 
1357 	itx = zil_itx_create(TX_WRITE,
1358 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1359 
1360 	if (write_state == WR_COPIED &&
1361 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1362 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1363 		zil_itx_destroy(itx);
1364 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1365 		write_state = WR_NEED_COPY;
1366 	}
1367 	itx->itx_private = zd;
1368 	itx->itx_wr_state = write_state;
1369 	itx->itx_sync = (ztest_random(8) == 0);
1370 	itx->itx_sod += (write_state == WR_NEED_COPY ? lr->lr_length : 0);
1371 
1372 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1373 	    sizeof (*lr) - sizeof (lr_t));
1374 
1375 	zil_itx_assign(zd->zd_zilog, itx, tx);
1376 }
1377 
1378 static void
1379 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1380 {
1381 	itx_t *itx;
1382 
1383 	if (zil_replaying(zd->zd_zilog, tx))
1384 		return;
1385 
1386 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1387 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1388 	    sizeof (*lr) - sizeof (lr_t));
1389 
1390 	itx->itx_sync = B_FALSE;
1391 	zil_itx_assign(zd->zd_zilog, itx, tx);
1392 }
1393 
1394 static void
1395 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1396 {
1397 	itx_t *itx;
1398 
1399 	if (zil_replaying(zd->zd_zilog, tx))
1400 		return;
1401 
1402 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1403 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1404 	    sizeof (*lr) - sizeof (lr_t));
1405 
1406 	itx->itx_sync = B_FALSE;
1407 	zil_itx_assign(zd->zd_zilog, itx, tx);
1408 }
1409 
1410 /*
1411  * ZIL replay ops
1412  */
1413 static int
1414 ztest_replay_create(ztest_ds_t *zd, lr_create_t *lr, boolean_t byteswap)
1415 {
1416 	char *name = (void *)(lr + 1);		/* name follows lr */
1417 	objset_t *os = zd->zd_os;
1418 	ztest_block_tag_t *bbt;
1419 	dmu_buf_t *db;
1420 	dmu_tx_t *tx;
1421 	uint64_t txg;
1422 	int error = 0;
1423 
1424 	if (byteswap)
1425 		byteswap_uint64_array(lr, sizeof (*lr));
1426 
1427 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1428 	ASSERT(name[0] != '\0');
1429 
1430 	tx = dmu_tx_create(os);
1431 
1432 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1433 
1434 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1435 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1436 	} else {
1437 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1438 	}
1439 
1440 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1441 	if (txg == 0)
1442 		return (ENOSPC);
1443 
1444 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1445 
1446 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1447 		if (lr->lr_foid == 0) {
1448 			lr->lr_foid = zap_create(os,
1449 			    lr->lrz_type, lr->lrz_bonustype,
1450 			    lr->lrz_bonuslen, tx);
1451 		} else {
1452 			error = zap_create_claim(os, lr->lr_foid,
1453 			    lr->lrz_type, lr->lrz_bonustype,
1454 			    lr->lrz_bonuslen, tx);
1455 		}
1456 	} else {
1457 		if (lr->lr_foid == 0) {
1458 			lr->lr_foid = dmu_object_alloc(os,
1459 			    lr->lrz_type, 0, lr->lrz_bonustype,
1460 			    lr->lrz_bonuslen, tx);
1461 		} else {
1462 			error = dmu_object_claim(os, lr->lr_foid,
1463 			    lr->lrz_type, 0, lr->lrz_bonustype,
1464 			    lr->lrz_bonuslen, tx);
1465 		}
1466 	}
1467 
1468 	if (error) {
1469 		ASSERT3U(error, ==, EEXIST);
1470 		ASSERT(zd->zd_zilog->zl_replay);
1471 		dmu_tx_commit(tx);
1472 		return (error);
1473 	}
1474 
1475 	ASSERT(lr->lr_foid != 0);
1476 
1477 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1478 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1479 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1480 
1481 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1482 	bbt = ztest_bt_bonus(db);
1483 	dmu_buf_will_dirty(db, tx);
1484 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_gen, txg, txg);
1485 	dmu_buf_rele(db, FTAG);
1486 
1487 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1488 	    &lr->lr_foid, tx));
1489 
1490 	(void) ztest_log_create(zd, tx, lr);
1491 
1492 	dmu_tx_commit(tx);
1493 
1494 	return (0);
1495 }
1496 
1497 static int
1498 ztest_replay_remove(ztest_ds_t *zd, lr_remove_t *lr, boolean_t byteswap)
1499 {
1500 	char *name = (void *)(lr + 1);		/* name follows lr */
1501 	objset_t *os = zd->zd_os;
1502 	dmu_object_info_t doi;
1503 	dmu_tx_t *tx;
1504 	uint64_t object, txg;
1505 
1506 	if (byteswap)
1507 		byteswap_uint64_array(lr, sizeof (*lr));
1508 
1509 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1510 	ASSERT(name[0] != '\0');
1511 
1512 	VERIFY3U(0, ==,
1513 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1514 	ASSERT(object != 0);
1515 
1516 	ztest_object_lock(zd, object, RL_WRITER);
1517 
1518 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1519 
1520 	tx = dmu_tx_create(os);
1521 
1522 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1523 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1524 
1525 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1526 	if (txg == 0) {
1527 		ztest_object_unlock(zd, object);
1528 		return (ENOSPC);
1529 	}
1530 
1531 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1532 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1533 	} else {
1534 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1535 	}
1536 
1537 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1538 
1539 	(void) ztest_log_remove(zd, tx, lr, object);
1540 
1541 	dmu_tx_commit(tx);
1542 
1543 	ztest_object_unlock(zd, object);
1544 
1545 	return (0);
1546 }
1547 
1548 static int
1549 ztest_replay_write(ztest_ds_t *zd, lr_write_t *lr, boolean_t byteswap)
1550 {
1551 	objset_t *os = zd->zd_os;
1552 	void *data = lr + 1;			/* data follows lr */
1553 	uint64_t offset, length;
1554 	ztest_block_tag_t *bt = data;
1555 	ztest_block_tag_t *bbt;
1556 	uint64_t gen, txg, lrtxg, crtxg;
1557 	dmu_object_info_t doi;
1558 	dmu_tx_t *tx;
1559 	dmu_buf_t *db;
1560 	arc_buf_t *abuf = NULL;
1561 	rl_t *rl;
1562 
1563 	if (byteswap)
1564 		byteswap_uint64_array(lr, sizeof (*lr));
1565 
1566 	offset = lr->lr_offset;
1567 	length = lr->lr_length;
1568 
1569 	/* If it's a dmu_sync() block, write the whole block */
1570 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1571 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1572 		if (length < blocksize) {
1573 			offset -= offset % blocksize;
1574 			length = blocksize;
1575 		}
1576 	}
1577 
1578 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1579 		byteswap_uint64_array(bt, sizeof (*bt));
1580 
1581 	if (bt->bt_magic != BT_MAGIC)
1582 		bt = NULL;
1583 
1584 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1585 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1586 
1587 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1588 
1589 	dmu_object_info_from_db(db, &doi);
1590 
1591 	bbt = ztest_bt_bonus(db);
1592 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1593 	gen = bbt->bt_gen;
1594 	crtxg = bbt->bt_crtxg;
1595 	lrtxg = lr->lr_common.lrc_txg;
1596 
1597 	tx = dmu_tx_create(os);
1598 
1599 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1600 
1601 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1602 	    P2PHASE(offset, length) == 0)
1603 		abuf = dmu_request_arcbuf(db, length);
1604 
1605 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1606 	if (txg == 0) {
1607 		if (abuf != NULL)
1608 			dmu_return_arcbuf(abuf);
1609 		dmu_buf_rele(db, FTAG);
1610 		ztest_range_unlock(rl);
1611 		ztest_object_unlock(zd, lr->lr_foid);
1612 		return (ENOSPC);
1613 	}
1614 
1615 	if (bt != NULL) {
1616 		/*
1617 		 * Usually, verify the old data before writing new data --
1618 		 * but not always, because we also want to verify correct
1619 		 * behavior when the data was not recently read into cache.
1620 		 */
1621 		ASSERT(offset % doi.doi_data_block_size == 0);
1622 		if (ztest_random(4) != 0) {
1623 			int prefetch = ztest_random(2) ?
1624 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1625 			ztest_block_tag_t rbt;
1626 
1627 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1628 			    sizeof (rbt), &rbt, prefetch) == 0);
1629 			if (rbt.bt_magic == BT_MAGIC) {
1630 				ztest_bt_verify(&rbt, os, lr->lr_foid,
1631 				    offset, gen, txg, crtxg);
1632 			}
1633 		}
1634 
1635 		/*
1636 		 * Writes can appear to be newer than the bonus buffer because
1637 		 * the ztest_get_data() callback does a dmu_read() of the
1638 		 * open-context data, which may be different than the data
1639 		 * as it was when the write was generated.
1640 		 */
1641 		if (zd->zd_zilog->zl_replay) {
1642 			ztest_bt_verify(bt, os, lr->lr_foid, offset,
1643 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1644 			    bt->bt_crtxg);
1645 		}
1646 
1647 		/*
1648 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1649 		 * so that all of the usual ASSERTs will work.
1650 		 */
1651 		ztest_bt_generate(bt, os, lr->lr_foid, offset, gen, txg, crtxg);
1652 	}
1653 
1654 	if (abuf == NULL) {
1655 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1656 	} else {
1657 		bcopy(data, abuf->b_data, length);
1658 		dmu_assign_arcbuf(db, offset, abuf, tx);
1659 	}
1660 
1661 	(void) ztest_log_write(zd, tx, lr);
1662 
1663 	dmu_buf_rele(db, FTAG);
1664 
1665 	dmu_tx_commit(tx);
1666 
1667 	ztest_range_unlock(rl);
1668 	ztest_object_unlock(zd, lr->lr_foid);
1669 
1670 	return (0);
1671 }
1672 
1673 static int
1674 ztest_replay_truncate(ztest_ds_t *zd, lr_truncate_t *lr, boolean_t byteswap)
1675 {
1676 	objset_t *os = zd->zd_os;
1677 	dmu_tx_t *tx;
1678 	uint64_t txg;
1679 	rl_t *rl;
1680 
1681 	if (byteswap)
1682 		byteswap_uint64_array(lr, sizeof (*lr));
1683 
1684 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1685 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1686 	    RL_WRITER);
1687 
1688 	tx = dmu_tx_create(os);
1689 
1690 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1691 
1692 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1693 	if (txg == 0) {
1694 		ztest_range_unlock(rl);
1695 		ztest_object_unlock(zd, lr->lr_foid);
1696 		return (ENOSPC);
1697 	}
1698 
1699 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1700 	    lr->lr_length, tx) == 0);
1701 
1702 	(void) ztest_log_truncate(zd, tx, lr);
1703 
1704 	dmu_tx_commit(tx);
1705 
1706 	ztest_range_unlock(rl);
1707 	ztest_object_unlock(zd, lr->lr_foid);
1708 
1709 	return (0);
1710 }
1711 
1712 static int
1713 ztest_replay_setattr(ztest_ds_t *zd, lr_setattr_t *lr, boolean_t byteswap)
1714 {
1715 	objset_t *os = zd->zd_os;
1716 	dmu_tx_t *tx;
1717 	dmu_buf_t *db;
1718 	ztest_block_tag_t *bbt;
1719 	uint64_t txg, lrtxg, crtxg;
1720 
1721 	if (byteswap)
1722 		byteswap_uint64_array(lr, sizeof (*lr));
1723 
1724 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1725 
1726 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1727 
1728 	tx = dmu_tx_create(os);
1729 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1730 
1731 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1732 	if (txg == 0) {
1733 		dmu_buf_rele(db, FTAG);
1734 		ztest_object_unlock(zd, lr->lr_foid);
1735 		return (ENOSPC);
1736 	}
1737 
1738 	bbt = ztest_bt_bonus(db);
1739 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1740 	crtxg = bbt->bt_crtxg;
1741 	lrtxg = lr->lr_common.lrc_txg;
1742 
1743 	if (zd->zd_zilog->zl_replay) {
1744 		ASSERT(lr->lr_size != 0);
1745 		ASSERT(lr->lr_mode != 0);
1746 		ASSERT(lrtxg != 0);
1747 	} else {
1748 		/*
1749 		 * Randomly change the size and increment the generation.
1750 		 */
1751 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1752 		    sizeof (*bbt);
1753 		lr->lr_mode = bbt->bt_gen + 1;
1754 		ASSERT(lrtxg == 0);
1755 	}
1756 
1757 	/*
1758 	 * Verify that the current bonus buffer is not newer than our txg.
1759 	 */
1760 	ztest_bt_verify(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode,
1761 	    MAX(txg, lrtxg), crtxg);
1762 
1763 	dmu_buf_will_dirty(db, tx);
1764 
1765 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1766 	ASSERT3U(lr->lr_size, <=, db->db_size);
1767 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1768 	bbt = ztest_bt_bonus(db);
1769 
1770 	ztest_bt_generate(bbt, os, lr->lr_foid, -1ULL, lr->lr_mode, txg, crtxg);
1771 
1772 	dmu_buf_rele(db, FTAG);
1773 
1774 	(void) ztest_log_setattr(zd, tx, lr);
1775 
1776 	dmu_tx_commit(tx);
1777 
1778 	ztest_object_unlock(zd, lr->lr_foid);
1779 
1780 	return (0);
1781 }
1782 
1783 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1784 	NULL,			/* 0 no such transaction type */
1785 	ztest_replay_create,	/* TX_CREATE */
1786 	NULL,			/* TX_MKDIR */
1787 	NULL,			/* TX_MKXATTR */
1788 	NULL,			/* TX_SYMLINK */
1789 	ztest_replay_remove,	/* TX_REMOVE */
1790 	NULL,			/* TX_RMDIR */
1791 	NULL,			/* TX_LINK */
1792 	NULL,			/* TX_RENAME */
1793 	ztest_replay_write,	/* TX_WRITE */
1794 	ztest_replay_truncate,	/* TX_TRUNCATE */
1795 	ztest_replay_setattr,	/* TX_SETATTR */
1796 	NULL,			/* TX_ACL */
1797 	NULL,			/* TX_CREATE_ACL */
1798 	NULL,			/* TX_CREATE_ATTR */
1799 	NULL,			/* TX_CREATE_ACL_ATTR */
1800 	NULL,			/* TX_MKDIR_ACL */
1801 	NULL,			/* TX_MKDIR_ATTR */
1802 	NULL,			/* TX_MKDIR_ACL_ATTR */
1803 	NULL,			/* TX_WRITE2 */
1804 };
1805 
1806 /*
1807  * ZIL get_data callbacks
1808  */
1809 
1810 static void
1811 ztest_get_done(zgd_t *zgd, int error)
1812 {
1813 	ztest_ds_t *zd = zgd->zgd_private;
1814 	uint64_t object = zgd->zgd_rl->rl_object;
1815 
1816 	if (zgd->zgd_db)
1817 		dmu_buf_rele(zgd->zgd_db, zgd);
1818 
1819 	ztest_range_unlock(zgd->zgd_rl);
1820 	ztest_object_unlock(zd, object);
1821 
1822 	if (error == 0 && zgd->zgd_bp)
1823 		zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
1824 
1825 	umem_free(zgd, sizeof (*zgd));
1826 }
1827 
1828 static int
1829 ztest_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
1830 {
1831 	ztest_ds_t *zd = arg;
1832 	objset_t *os = zd->zd_os;
1833 	uint64_t object = lr->lr_foid;
1834 	uint64_t offset = lr->lr_offset;
1835 	uint64_t size = lr->lr_length;
1836 	blkptr_t *bp = &lr->lr_blkptr;
1837 	uint64_t txg = lr->lr_common.lrc_txg;
1838 	uint64_t crtxg;
1839 	dmu_object_info_t doi;
1840 	dmu_buf_t *db;
1841 	zgd_t *zgd;
1842 	int error;
1843 
1844 	ztest_object_lock(zd, object, RL_READER);
1845 	error = dmu_bonus_hold(os, object, FTAG, &db);
1846 	if (error) {
1847 		ztest_object_unlock(zd, object);
1848 		return (error);
1849 	}
1850 
1851 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1852 
1853 	if (crtxg == 0 || crtxg > txg) {
1854 		dmu_buf_rele(db, FTAG);
1855 		ztest_object_unlock(zd, object);
1856 		return (ENOENT);
1857 	}
1858 
1859 	dmu_object_info_from_db(db, &doi);
1860 	dmu_buf_rele(db, FTAG);
1861 	db = NULL;
1862 
1863 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
1864 	zgd->zgd_zilog = zd->zd_zilog;
1865 	zgd->zgd_private = zd;
1866 
1867 	if (buf != NULL) {	/* immediate write */
1868 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1869 		    RL_READER);
1870 
1871 		error = dmu_read(os, object, offset, size, buf,
1872 		    DMU_READ_NO_PREFETCH);
1873 		ASSERT(error == 0);
1874 	} else {
1875 		size = doi.doi_data_block_size;
1876 		if (ISP2(size)) {
1877 			offset = P2ALIGN(offset, size);
1878 		} else {
1879 			ASSERT(offset < size);
1880 			offset = 0;
1881 		}
1882 
1883 		zgd->zgd_rl = ztest_range_lock(zd, object, offset, size,
1884 		    RL_READER);
1885 
1886 		error = dmu_buf_hold(os, object, offset, zgd, &db,
1887 		    DMU_READ_NO_PREFETCH);
1888 
1889 		if (error == 0) {
1890 			blkptr_t *obp = dmu_buf_get_blkptr(db);
1891 			if (obp) {
1892 				ASSERT(BP_IS_HOLE(bp));
1893 				*bp = *obp;
1894 			}
1895 
1896 			zgd->zgd_db = db;
1897 			zgd->zgd_bp = bp;
1898 
1899 			ASSERT(db->db_offset == offset);
1900 			ASSERT(db->db_size == size);
1901 
1902 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
1903 			    ztest_get_done, zgd);
1904 
1905 			if (error == 0)
1906 				return (0);
1907 		}
1908 	}
1909 
1910 	ztest_get_done(zgd, error);
1911 
1912 	return (error);
1913 }
1914 
1915 static void *
1916 ztest_lr_alloc(size_t lrsize, char *name)
1917 {
1918 	char *lr;
1919 	size_t namesize = name ? strlen(name) + 1 : 0;
1920 
1921 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
1922 
1923 	if (name)
1924 		bcopy(name, lr + lrsize, namesize);
1925 
1926 	return (lr);
1927 }
1928 
1929 void
1930 ztest_lr_free(void *lr, size_t lrsize, char *name)
1931 {
1932 	size_t namesize = name ? strlen(name) + 1 : 0;
1933 
1934 	umem_free(lr, lrsize + namesize);
1935 }
1936 
1937 /*
1938  * Lookup a bunch of objects.  Returns the number of objects not found.
1939  */
1940 static int
1941 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
1942 {
1943 	int missing = 0;
1944 	int error;
1945 
1946 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1947 
1948 	for (int i = 0; i < count; i++, od++) {
1949 		od->od_object = 0;
1950 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
1951 		    sizeof (uint64_t), 1, &od->od_object);
1952 		if (error) {
1953 			ASSERT(error == ENOENT);
1954 			ASSERT(od->od_object == 0);
1955 			missing++;
1956 		} else {
1957 			dmu_buf_t *db;
1958 			ztest_block_tag_t *bbt;
1959 			dmu_object_info_t doi;
1960 
1961 			ASSERT(od->od_object != 0);
1962 			ASSERT(missing == 0);	/* there should be no gaps */
1963 
1964 			ztest_object_lock(zd, od->od_object, RL_READER);
1965 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
1966 			    od->od_object, FTAG, &db));
1967 			dmu_object_info_from_db(db, &doi);
1968 			bbt = ztest_bt_bonus(db);
1969 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1970 			od->od_type = doi.doi_type;
1971 			od->od_blocksize = doi.doi_data_block_size;
1972 			od->od_gen = bbt->bt_gen;
1973 			dmu_buf_rele(db, FTAG);
1974 			ztest_object_unlock(zd, od->od_object);
1975 		}
1976 	}
1977 
1978 	return (missing);
1979 }
1980 
1981 static int
1982 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
1983 {
1984 	int missing = 0;
1985 
1986 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
1987 
1988 	for (int i = 0; i < count; i++, od++) {
1989 		if (missing) {
1990 			od->od_object = 0;
1991 			missing++;
1992 			continue;
1993 		}
1994 
1995 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
1996 
1997 		lr->lr_doid = od->od_dir;
1998 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
1999 		lr->lrz_type = od->od_crtype;
2000 		lr->lrz_blocksize = od->od_crblocksize;
2001 		lr->lrz_ibshift = ztest_random_ibshift();
2002 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2003 		lr->lrz_bonuslen = dmu_bonus_max();
2004 		lr->lr_gen = od->od_crgen;
2005 		lr->lr_crtime[0] = time(NULL);
2006 
2007 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2008 			ASSERT(missing == 0);
2009 			od->od_object = 0;
2010 			missing++;
2011 		} else {
2012 			od->od_object = lr->lr_foid;
2013 			od->od_type = od->od_crtype;
2014 			od->od_blocksize = od->od_crblocksize;
2015 			od->od_gen = od->od_crgen;
2016 			ASSERT(od->od_object != 0);
2017 		}
2018 
2019 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2020 	}
2021 
2022 	return (missing);
2023 }
2024 
2025 static int
2026 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2027 {
2028 	int missing = 0;
2029 	int error;
2030 
2031 	ASSERT(_mutex_held(&zd->zd_dirobj_lock));
2032 
2033 	od += count - 1;
2034 
2035 	for (int i = count - 1; i >= 0; i--, od--) {
2036 		if (missing) {
2037 			missing++;
2038 			continue;
2039 		}
2040 
2041 		/*
2042 		 * No object was found.
2043 		 */
2044 		if (od->od_object == 0)
2045 			continue;
2046 
2047 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2048 
2049 		lr->lr_doid = od->od_dir;
2050 
2051 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2052 			ASSERT3U(error, ==, ENOSPC);
2053 			missing++;
2054 		} else {
2055 			od->od_object = 0;
2056 		}
2057 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2058 	}
2059 
2060 	return (missing);
2061 }
2062 
2063 static int
2064 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2065     void *data)
2066 {
2067 	lr_write_t *lr;
2068 	int error;
2069 
2070 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2071 
2072 	lr->lr_foid = object;
2073 	lr->lr_offset = offset;
2074 	lr->lr_length = size;
2075 	lr->lr_blkoff = 0;
2076 	BP_ZERO(&lr->lr_blkptr);
2077 
2078 	bcopy(data, lr + 1, size);
2079 
2080 	error = ztest_replay_write(zd, lr, B_FALSE);
2081 
2082 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2083 
2084 	return (error);
2085 }
2086 
2087 static int
2088 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2089 {
2090 	lr_truncate_t *lr;
2091 	int error;
2092 
2093 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2094 
2095 	lr->lr_foid = object;
2096 	lr->lr_offset = offset;
2097 	lr->lr_length = size;
2098 
2099 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2100 
2101 	ztest_lr_free(lr, sizeof (*lr), NULL);
2102 
2103 	return (error);
2104 }
2105 
2106 static int
2107 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2108 {
2109 	lr_setattr_t *lr;
2110 	int error;
2111 
2112 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2113 
2114 	lr->lr_foid = object;
2115 	lr->lr_size = 0;
2116 	lr->lr_mode = 0;
2117 
2118 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2119 
2120 	ztest_lr_free(lr, sizeof (*lr), NULL);
2121 
2122 	return (error);
2123 }
2124 
2125 static void
2126 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2127 {
2128 	objset_t *os = zd->zd_os;
2129 	dmu_tx_t *tx;
2130 	uint64_t txg;
2131 	rl_t *rl;
2132 
2133 	txg_wait_synced(dmu_objset_pool(os), 0);
2134 
2135 	ztest_object_lock(zd, object, RL_READER);
2136 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2137 
2138 	tx = dmu_tx_create(os);
2139 
2140 	dmu_tx_hold_write(tx, object, offset, size);
2141 
2142 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2143 
2144 	if (txg != 0) {
2145 		dmu_prealloc(os, object, offset, size, tx);
2146 		dmu_tx_commit(tx);
2147 		txg_wait_synced(dmu_objset_pool(os), txg);
2148 	} else {
2149 		(void) dmu_free_long_range(os, object, offset, size);
2150 	}
2151 
2152 	ztest_range_unlock(rl);
2153 	ztest_object_unlock(zd, object);
2154 }
2155 
2156 static void
2157 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2158 {
2159 	int err;
2160 	ztest_block_tag_t wbt;
2161 	dmu_object_info_t doi;
2162 	enum ztest_io_type io_type;
2163 	uint64_t blocksize;
2164 	void *data;
2165 
2166 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2167 	blocksize = doi.doi_data_block_size;
2168 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2169 
2170 	/*
2171 	 * Pick an i/o type at random, biased toward writing block tags.
2172 	 */
2173 	io_type = ztest_random(ZTEST_IO_TYPES);
2174 	if (ztest_random(2) == 0)
2175 		io_type = ZTEST_IO_WRITE_TAG;
2176 
2177 	(void) rw_rdlock(&zd->zd_zilog_lock);
2178 
2179 	switch (io_type) {
2180 
2181 	case ZTEST_IO_WRITE_TAG:
2182 		ztest_bt_generate(&wbt, zd->zd_os, object, offset, 0, 0, 0);
2183 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2184 		break;
2185 
2186 	case ZTEST_IO_WRITE_PATTERN:
2187 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2188 		if (ztest_random(2) == 0) {
2189 			/*
2190 			 * Induce fletcher2 collisions to ensure that
2191 			 * zio_ddt_collision() detects and resolves them
2192 			 * when using fletcher2-verify for deduplication.
2193 			 */
2194 			((uint64_t *)data)[0] ^= 1ULL << 63;
2195 			((uint64_t *)data)[4] ^= 1ULL << 63;
2196 		}
2197 		(void) ztest_write(zd, object, offset, blocksize, data);
2198 		break;
2199 
2200 	case ZTEST_IO_WRITE_ZEROES:
2201 		bzero(data, blocksize);
2202 		(void) ztest_write(zd, object, offset, blocksize, data);
2203 		break;
2204 
2205 	case ZTEST_IO_TRUNCATE:
2206 		(void) ztest_truncate(zd, object, offset, blocksize);
2207 		break;
2208 
2209 	case ZTEST_IO_SETATTR:
2210 		(void) ztest_setattr(zd, object);
2211 		break;
2212 
2213 	case ZTEST_IO_REWRITE:
2214 		(void) rw_rdlock(&ztest_name_lock);
2215 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2216 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2217 		    B_FALSE);
2218 		VERIFY(err == 0 || err == ENOSPC);
2219 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2220 		    ZFS_PROP_COMPRESSION,
2221 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2222 		    B_FALSE);
2223 		VERIFY(err == 0 || err == ENOSPC);
2224 		(void) rw_unlock(&ztest_name_lock);
2225 
2226 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2227 		    DMU_READ_NO_PREFETCH));
2228 
2229 		(void) ztest_write(zd, object, offset, blocksize, data);
2230 		break;
2231 	}
2232 
2233 	(void) rw_unlock(&zd->zd_zilog_lock);
2234 
2235 	umem_free(data, blocksize);
2236 }
2237 
2238 /*
2239  * Initialize an object description template.
2240  */
2241 static void
2242 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2243     dmu_object_type_t type, uint64_t blocksize, uint64_t gen)
2244 {
2245 	od->od_dir = ZTEST_DIROBJ;
2246 	od->od_object = 0;
2247 
2248 	od->od_crtype = type;
2249 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2250 	od->od_crgen = gen;
2251 
2252 	od->od_type = DMU_OT_NONE;
2253 	od->od_blocksize = 0;
2254 	od->od_gen = 0;
2255 
2256 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2257 	    tag, (int64_t)id, index);
2258 }
2259 
2260 /*
2261  * Lookup or create the objects for a test using the od template.
2262  * If the objects do not all exist, or if 'remove' is specified,
2263  * remove any existing objects and create new ones.  Otherwise,
2264  * use the existing objects.
2265  */
2266 static int
2267 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2268 {
2269 	int count = size / sizeof (*od);
2270 	int rv = 0;
2271 
2272 	VERIFY(mutex_lock(&zd->zd_dirobj_lock) == 0);
2273 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2274 	    (ztest_remove(zd, od, count) != 0 ||
2275 	    ztest_create(zd, od, count) != 0))
2276 		rv = -1;
2277 	zd->zd_od = od;
2278 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2279 
2280 	return (rv);
2281 }
2282 
2283 /* ARGSUSED */
2284 void
2285 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2286 {
2287 	zilog_t *zilog = zd->zd_zilog;
2288 
2289 	(void) rw_rdlock(&zd->zd_zilog_lock);
2290 
2291 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2292 
2293 	/*
2294 	 * Remember the committed values in zd, which is in parent/child
2295 	 * shared memory.  If we die, the next iteration of ztest_run()
2296 	 * will verify that the log really does contain this record.
2297 	 */
2298 	mutex_enter(&zilog->zl_lock);
2299 	ASSERT(zd->zd_shared != NULL);
2300 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2301 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2302 	mutex_exit(&zilog->zl_lock);
2303 
2304 	(void) rw_unlock(&zd->zd_zilog_lock);
2305 }
2306 
2307 /*
2308  * This function is designed to simulate the operations that occur during a
2309  * mount/unmount operation.  We hold the dataset across these operations in an
2310  * attempt to expose any implicit assumptions about ZIL management.
2311  */
2312 /* ARGSUSED */
2313 void
2314 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2315 {
2316 	objset_t *os = zd->zd_os;
2317 
2318 	/*
2319 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2320 	 * updating the zil (i.e. adding in-memory log records) and the
2321 	 * zd_zilog_lock to block any I/O.
2322 	 */
2323 	VERIFY0(mutex_lock(&zd->zd_dirobj_lock));
2324 	(void) rw_wrlock(&zd->zd_zilog_lock);
2325 
2326 	/* zfsvfs_teardown() */
2327 	zil_close(zd->zd_zilog);
2328 
2329 	/* zfsvfs_setup() */
2330 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2331 	zil_replay(os, zd, ztest_replay_vector);
2332 
2333 	(void) rw_unlock(&zd->zd_zilog_lock);
2334 	VERIFY(mutex_unlock(&zd->zd_dirobj_lock) == 0);
2335 }
2336 
2337 /*
2338  * Verify that we can't destroy an active pool, create an existing pool,
2339  * or create a pool with a bad vdev spec.
2340  */
2341 /* ARGSUSED */
2342 void
2343 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2344 {
2345 	ztest_shared_opts_t *zo = &ztest_opts;
2346 	spa_t *spa;
2347 	nvlist_t *nvroot;
2348 
2349 	/*
2350 	 * Attempt to create using a bad file.
2351 	 */
2352 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2353 	VERIFY3U(ENOENT, ==,
2354 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2355 	nvlist_free(nvroot);
2356 
2357 	/*
2358 	 * Attempt to create using a bad mirror.
2359 	 */
2360 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2361 	VERIFY3U(ENOENT, ==,
2362 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2363 	nvlist_free(nvroot);
2364 
2365 	/*
2366 	 * Attempt to create an existing pool.  It shouldn't matter
2367 	 * what's in the nvroot; we should fail with EEXIST.
2368 	 */
2369 	(void) rw_rdlock(&ztest_name_lock);
2370 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2371 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2372 	nvlist_free(nvroot);
2373 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2374 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2375 	spa_close(spa, FTAG);
2376 
2377 	(void) rw_unlock(&ztest_name_lock);
2378 }
2379 
2380 /* ARGSUSED */
2381 void
2382 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2383 {
2384 	spa_t *spa;
2385 	uint64_t initial_version = SPA_VERSION_INITIAL;
2386 	uint64_t version, newversion;
2387 	nvlist_t *nvroot, *props;
2388 	char *name;
2389 
2390 	VERIFY0(mutex_lock(&ztest_vdev_lock));
2391 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2392 
2393 	/*
2394 	 * Clean up from previous runs.
2395 	 */
2396 	(void) spa_destroy(name);
2397 
2398 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2399 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2400 
2401 	/*
2402 	 * If we're configuring a RAIDZ device then make sure that the
2403 	 * the initial version is capable of supporting that feature.
2404 	 */
2405 	switch (ztest_opts.zo_raidz_parity) {
2406 	case 0:
2407 	case 1:
2408 		initial_version = SPA_VERSION_INITIAL;
2409 		break;
2410 	case 2:
2411 		initial_version = SPA_VERSION_RAIDZ2;
2412 		break;
2413 	case 3:
2414 		initial_version = SPA_VERSION_RAIDZ3;
2415 		break;
2416 	}
2417 
2418 	/*
2419 	 * Create a pool with a spa version that can be upgraded. Pick
2420 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2421 	 */
2422 	do {
2423 		version = ztest_random_spa_version(initial_version);
2424 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2425 
2426 	props = fnvlist_alloc();
2427 	fnvlist_add_uint64(props,
2428 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2429 	VERIFY0(spa_create(name, nvroot, props, NULL));
2430 	fnvlist_free(nvroot);
2431 	fnvlist_free(props);
2432 
2433 	VERIFY0(spa_open(name, &spa, FTAG));
2434 	VERIFY3U(spa_version(spa), ==, version);
2435 	newversion = ztest_random_spa_version(version + 1);
2436 
2437 	if (ztest_opts.zo_verbose >= 4) {
2438 		(void) printf("upgrading spa version from %llu to %llu\n",
2439 		    (u_longlong_t)version, (u_longlong_t)newversion);
2440 	}
2441 
2442 	spa_upgrade(spa, newversion);
2443 	VERIFY3U(spa_version(spa), >, version);
2444 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2445 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2446 	spa_close(spa, FTAG);
2447 
2448 	strfree(name);
2449 	VERIFY0(mutex_unlock(&ztest_vdev_lock));
2450 }
2451 
2452 static vdev_t *
2453 vdev_lookup_by_path(vdev_t *vd, const char *path)
2454 {
2455 	vdev_t *mvd;
2456 
2457 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2458 		return (vd);
2459 
2460 	for (int c = 0; c < vd->vdev_children; c++)
2461 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2462 		    NULL)
2463 			return (mvd);
2464 
2465 	return (NULL);
2466 }
2467 
2468 /*
2469  * Find the first available hole which can be used as a top-level.
2470  */
2471 int
2472 find_vdev_hole(spa_t *spa)
2473 {
2474 	vdev_t *rvd = spa->spa_root_vdev;
2475 	int c;
2476 
2477 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2478 
2479 	for (c = 0; c < rvd->vdev_children; c++) {
2480 		vdev_t *cvd = rvd->vdev_child[c];
2481 
2482 		if (cvd->vdev_ishole)
2483 			break;
2484 	}
2485 	return (c);
2486 }
2487 
2488 /*
2489  * Verify that vdev_add() works as expected.
2490  */
2491 /* ARGSUSED */
2492 void
2493 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2494 {
2495 	ztest_shared_t *zs = ztest_shared;
2496 	spa_t *spa = ztest_spa;
2497 	uint64_t leaves;
2498 	uint64_t guid;
2499 	nvlist_t *nvroot;
2500 	int error;
2501 
2502 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2503 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2504 
2505 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2506 
2507 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2508 
2509 	/*
2510 	 * If we have slogs then remove them 1/4 of the time.
2511 	 */
2512 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2513 		/*
2514 		 * Grab the guid from the head of the log class rotor.
2515 		 */
2516 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2517 
2518 		spa_config_exit(spa, SCL_VDEV, FTAG);
2519 
2520 		/*
2521 		 * We have to grab the zs_name_lock as writer to
2522 		 * prevent a race between removing a slog (dmu_objset_find)
2523 		 * and destroying a dataset. Removing the slog will
2524 		 * grab a reference on the dataset which may cause
2525 		 * dmu_objset_destroy() to fail with EBUSY thus
2526 		 * leaving the dataset in an inconsistent state.
2527 		 */
2528 		VERIFY(rw_wrlock(&ztest_name_lock) == 0);
2529 		error = spa_vdev_remove(spa, guid, B_FALSE);
2530 		VERIFY(rw_unlock(&ztest_name_lock) == 0);
2531 
2532 		if (error && error != EEXIST)
2533 			fatal(0, "spa_vdev_remove() = %d", error);
2534 	} else {
2535 		spa_config_exit(spa, SCL_VDEV, FTAG);
2536 
2537 		/*
2538 		 * Make 1/4 of the devices be log devices.
2539 		 */
2540 		nvroot = make_vdev_root(NULL, NULL, NULL,
2541 		    ztest_opts.zo_vdev_size, 0,
2542 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2543 		    zs->zs_mirrors, 1);
2544 
2545 		error = spa_vdev_add(spa, nvroot);
2546 		nvlist_free(nvroot);
2547 
2548 		if (error == ENOSPC)
2549 			ztest_record_enospc("spa_vdev_add");
2550 		else if (error != 0)
2551 			fatal(0, "spa_vdev_add() = %d", error);
2552 	}
2553 
2554 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2555 }
2556 
2557 /*
2558  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2559  */
2560 /* ARGSUSED */
2561 void
2562 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2563 {
2564 	ztest_shared_t *zs = ztest_shared;
2565 	spa_t *spa = ztest_spa;
2566 	vdev_t *rvd = spa->spa_root_vdev;
2567 	spa_aux_vdev_t *sav;
2568 	char *aux;
2569 	uint64_t guid = 0;
2570 	int error;
2571 
2572 	if (ztest_random(2) == 0) {
2573 		sav = &spa->spa_spares;
2574 		aux = ZPOOL_CONFIG_SPARES;
2575 	} else {
2576 		sav = &spa->spa_l2cache;
2577 		aux = ZPOOL_CONFIG_L2CACHE;
2578 	}
2579 
2580 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2581 
2582 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2583 
2584 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2585 		/*
2586 		 * Pick a random device to remove.
2587 		 */
2588 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2589 	} else {
2590 		/*
2591 		 * Find an unused device we can add.
2592 		 */
2593 		zs->zs_vdev_aux = 0;
2594 		for (;;) {
2595 			char path[MAXPATHLEN];
2596 			int c;
2597 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2598 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2599 			    zs->zs_vdev_aux);
2600 			for (c = 0; c < sav->sav_count; c++)
2601 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2602 				    path) == 0)
2603 					break;
2604 			if (c == sav->sav_count &&
2605 			    vdev_lookup_by_path(rvd, path) == NULL)
2606 				break;
2607 			zs->zs_vdev_aux++;
2608 		}
2609 	}
2610 
2611 	spa_config_exit(spa, SCL_VDEV, FTAG);
2612 
2613 	if (guid == 0) {
2614 		/*
2615 		 * Add a new device.
2616 		 */
2617 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2618 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2619 		error = spa_vdev_add(spa, nvroot);
2620 		if (error != 0)
2621 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2622 		nvlist_free(nvroot);
2623 	} else {
2624 		/*
2625 		 * Remove an existing device.  Sometimes, dirty its
2626 		 * vdev state first to make sure we handle removal
2627 		 * of devices that have pending state changes.
2628 		 */
2629 		if (ztest_random(2) == 0)
2630 			(void) vdev_online(spa, guid, 0, NULL);
2631 
2632 		error = spa_vdev_remove(spa, guid, B_FALSE);
2633 		if (error != 0 && error != EBUSY)
2634 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2635 	}
2636 
2637 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2638 }
2639 
2640 /*
2641  * split a pool if it has mirror tlvdevs
2642  */
2643 /* ARGSUSED */
2644 void
2645 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2646 {
2647 	ztest_shared_t *zs = ztest_shared;
2648 	spa_t *spa = ztest_spa;
2649 	vdev_t *rvd = spa->spa_root_vdev;
2650 	nvlist_t *tree, **child, *config, *split, **schild;
2651 	uint_t c, children, schildren = 0, lastlogid = 0;
2652 	int error = 0;
2653 
2654 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2655 
2656 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2657 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2658 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2659 		return;
2660 	}
2661 
2662 	/* clean up the old pool, if any */
2663 	(void) spa_destroy("splitp");
2664 
2665 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2666 
2667 	/* generate a config from the existing config */
2668 	mutex_enter(&spa->spa_props_lock);
2669 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2670 	    &tree) == 0);
2671 	mutex_exit(&spa->spa_props_lock);
2672 
2673 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2674 	    &children) == 0);
2675 
2676 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2677 	for (c = 0; c < children; c++) {
2678 		vdev_t *tvd = rvd->vdev_child[c];
2679 		nvlist_t **mchild;
2680 		uint_t mchildren;
2681 
2682 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2683 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2684 			    0) == 0);
2685 			VERIFY(nvlist_add_string(schild[schildren],
2686 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2687 			VERIFY(nvlist_add_uint64(schild[schildren],
2688 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2689 			if (lastlogid == 0)
2690 				lastlogid = schildren;
2691 			++schildren;
2692 			continue;
2693 		}
2694 		lastlogid = 0;
2695 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2696 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2697 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2698 	}
2699 
2700 	/* OK, create a config that can be used to split */
2701 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2702 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2703 	    VDEV_TYPE_ROOT) == 0);
2704 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2705 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2706 
2707 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2708 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2709 
2710 	for (c = 0; c < schildren; c++)
2711 		nvlist_free(schild[c]);
2712 	free(schild);
2713 	nvlist_free(split);
2714 
2715 	spa_config_exit(spa, SCL_VDEV, FTAG);
2716 
2717 	(void) rw_wrlock(&ztest_name_lock);
2718 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
2719 	(void) rw_unlock(&ztest_name_lock);
2720 
2721 	nvlist_free(config);
2722 
2723 	if (error == 0) {
2724 		(void) printf("successful split - results:\n");
2725 		mutex_enter(&spa_namespace_lock);
2726 		show_pool_stats(spa);
2727 		show_pool_stats(spa_lookup("splitp"));
2728 		mutex_exit(&spa_namespace_lock);
2729 		++zs->zs_splits;
2730 		--zs->zs_mirrors;
2731 	}
2732 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2733 
2734 }
2735 
2736 /*
2737  * Verify that we can attach and detach devices.
2738  */
2739 /* ARGSUSED */
2740 void
2741 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
2742 {
2743 	ztest_shared_t *zs = ztest_shared;
2744 	spa_t *spa = ztest_spa;
2745 	spa_aux_vdev_t *sav = &spa->spa_spares;
2746 	vdev_t *rvd = spa->spa_root_vdev;
2747 	vdev_t *oldvd, *newvd, *pvd;
2748 	nvlist_t *root;
2749 	uint64_t leaves;
2750 	uint64_t leaf, top;
2751 	uint64_t ashift = ztest_get_ashift();
2752 	uint64_t oldguid, pguid;
2753 	uint64_t oldsize, newsize;
2754 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
2755 	int replacing;
2756 	int oldvd_has_siblings = B_FALSE;
2757 	int newvd_is_spare = B_FALSE;
2758 	int oldvd_is_log;
2759 	int error, expected_error;
2760 
2761 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
2762 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
2763 
2764 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2765 
2766 	/*
2767 	 * Decide whether to do an attach or a replace.
2768 	 */
2769 	replacing = ztest_random(2);
2770 
2771 	/*
2772 	 * Pick a random top-level vdev.
2773 	 */
2774 	top = ztest_random_vdev_top(spa, B_TRUE);
2775 
2776 	/*
2777 	 * Pick a random leaf within it.
2778 	 */
2779 	leaf = ztest_random(leaves);
2780 
2781 	/*
2782 	 * Locate this vdev.
2783 	 */
2784 	oldvd = rvd->vdev_child[top];
2785 	if (zs->zs_mirrors >= 1) {
2786 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
2787 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
2788 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
2789 	}
2790 	if (ztest_opts.zo_raidz > 1) {
2791 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
2792 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
2793 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
2794 	}
2795 
2796 	/*
2797 	 * If we're already doing an attach or replace, oldvd may be a
2798 	 * mirror vdev -- in which case, pick a random child.
2799 	 */
2800 	while (oldvd->vdev_children != 0) {
2801 		oldvd_has_siblings = B_TRUE;
2802 		ASSERT(oldvd->vdev_children >= 2);
2803 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
2804 	}
2805 
2806 	oldguid = oldvd->vdev_guid;
2807 	oldsize = vdev_get_min_asize(oldvd);
2808 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
2809 	(void) strcpy(oldpath, oldvd->vdev_path);
2810 	pvd = oldvd->vdev_parent;
2811 	pguid = pvd->vdev_guid;
2812 
2813 	/*
2814 	 * If oldvd has siblings, then half of the time, detach it.
2815 	 */
2816 	if (oldvd_has_siblings && ztest_random(2) == 0) {
2817 		spa_config_exit(spa, SCL_VDEV, FTAG);
2818 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
2819 		if (error != 0 && error != ENODEV && error != EBUSY &&
2820 		    error != ENOTSUP)
2821 			fatal(0, "detach (%s) returned %d", oldpath, error);
2822 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2823 		return;
2824 	}
2825 
2826 	/*
2827 	 * For the new vdev, choose with equal probability between the two
2828 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
2829 	 */
2830 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
2831 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
2832 		newvd_is_spare = B_TRUE;
2833 		(void) strcpy(newpath, newvd->vdev_path);
2834 	} else {
2835 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
2836 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
2837 		    top * leaves + leaf);
2838 		if (ztest_random(2) == 0)
2839 			newpath[strlen(newpath) - 1] = 'b';
2840 		newvd = vdev_lookup_by_path(rvd, newpath);
2841 	}
2842 
2843 	if (newvd) {
2844 		newsize = vdev_get_min_asize(newvd);
2845 	} else {
2846 		/*
2847 		 * Make newsize a little bigger or smaller than oldsize.
2848 		 * If it's smaller, the attach should fail.
2849 		 * If it's larger, and we're doing a replace,
2850 		 * we should get dynamic LUN growth when we're done.
2851 		 */
2852 		newsize = 10 * oldsize / (9 + ztest_random(3));
2853 	}
2854 
2855 	/*
2856 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
2857 	 * unless it's a replace; in that case any non-replacing parent is OK.
2858 	 *
2859 	 * If newvd is already part of the pool, it should fail with EBUSY.
2860 	 *
2861 	 * If newvd is too small, it should fail with EOVERFLOW.
2862 	 */
2863 	if (pvd->vdev_ops != &vdev_mirror_ops &&
2864 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
2865 	    pvd->vdev_ops == &vdev_replacing_ops ||
2866 	    pvd->vdev_ops == &vdev_spare_ops))
2867 		expected_error = ENOTSUP;
2868 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
2869 		expected_error = ENOTSUP;
2870 	else if (newvd == oldvd)
2871 		expected_error = replacing ? 0 : EBUSY;
2872 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
2873 		expected_error = EBUSY;
2874 	else if (newsize < oldsize)
2875 		expected_error = EOVERFLOW;
2876 	else if (ashift > oldvd->vdev_top->vdev_ashift)
2877 		expected_error = EDOM;
2878 	else
2879 		expected_error = 0;
2880 
2881 	spa_config_exit(spa, SCL_VDEV, FTAG);
2882 
2883 	/*
2884 	 * Build the nvlist describing newpath.
2885 	 */
2886 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
2887 	    ashift, 0, 0, 0, 1);
2888 
2889 	error = spa_vdev_attach(spa, oldguid, root, replacing);
2890 
2891 	nvlist_free(root);
2892 
2893 	/*
2894 	 * If our parent was the replacing vdev, but the replace completed,
2895 	 * then instead of failing with ENOTSUP we may either succeed,
2896 	 * fail with ENODEV, or fail with EOVERFLOW.
2897 	 */
2898 	if (expected_error == ENOTSUP &&
2899 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
2900 		expected_error = error;
2901 
2902 	/*
2903 	 * If someone grew the LUN, the replacement may be too small.
2904 	 */
2905 	if (error == EOVERFLOW || error == EBUSY)
2906 		expected_error = error;
2907 
2908 	/* XXX workaround 6690467 */
2909 	if (error != expected_error && expected_error != EBUSY) {
2910 		fatal(0, "attach (%s %llu, %s %llu, %d) "
2911 		    "returned %d, expected %d",
2912 		    oldpath, oldsize, newpath,
2913 		    newsize, replacing, error, expected_error);
2914 	}
2915 
2916 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
2917 }
2918 
2919 /*
2920  * Callback function which expands the physical size of the vdev.
2921  */
2922 vdev_t *
2923 grow_vdev(vdev_t *vd, void *arg)
2924 {
2925 	spa_t *spa = vd->vdev_spa;
2926 	size_t *newsize = arg;
2927 	size_t fsize;
2928 	int fd;
2929 
2930 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2931 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2932 
2933 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
2934 		return (vd);
2935 
2936 	fsize = lseek(fd, 0, SEEK_END);
2937 	(void) ftruncate(fd, *newsize);
2938 
2939 	if (ztest_opts.zo_verbose >= 6) {
2940 		(void) printf("%s grew from %lu to %lu bytes\n",
2941 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
2942 	}
2943 	(void) close(fd);
2944 	return (NULL);
2945 }
2946 
2947 /*
2948  * Callback function which expands a given vdev by calling vdev_online().
2949  */
2950 /* ARGSUSED */
2951 vdev_t *
2952 online_vdev(vdev_t *vd, void *arg)
2953 {
2954 	spa_t *spa = vd->vdev_spa;
2955 	vdev_t *tvd = vd->vdev_top;
2956 	uint64_t guid = vd->vdev_guid;
2957 	uint64_t generation = spa->spa_config_generation + 1;
2958 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
2959 	int error;
2960 
2961 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
2962 	ASSERT(vd->vdev_ops->vdev_op_leaf);
2963 
2964 	/* Calling vdev_online will initialize the new metaslabs */
2965 	spa_config_exit(spa, SCL_STATE, spa);
2966 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
2967 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
2968 
2969 	/*
2970 	 * If vdev_online returned an error or the underlying vdev_open
2971 	 * failed then we abort the expand. The only way to know that
2972 	 * vdev_open fails is by checking the returned newstate.
2973 	 */
2974 	if (error || newstate != VDEV_STATE_HEALTHY) {
2975 		if (ztest_opts.zo_verbose >= 5) {
2976 			(void) printf("Unable to expand vdev, state %llu, "
2977 			    "error %d\n", (u_longlong_t)newstate, error);
2978 		}
2979 		return (vd);
2980 	}
2981 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
2982 
2983 	/*
2984 	 * Since we dropped the lock we need to ensure that we're
2985 	 * still talking to the original vdev. It's possible this
2986 	 * vdev may have been detached/replaced while we were
2987 	 * trying to online it.
2988 	 */
2989 	if (generation != spa->spa_config_generation) {
2990 		if (ztest_opts.zo_verbose >= 5) {
2991 			(void) printf("vdev configuration has changed, "
2992 			    "guid %llu, state %llu, expected gen %llu, "
2993 			    "got gen %llu\n",
2994 			    (u_longlong_t)guid,
2995 			    (u_longlong_t)tvd->vdev_state,
2996 			    (u_longlong_t)generation,
2997 			    (u_longlong_t)spa->spa_config_generation);
2998 		}
2999 		return (vd);
3000 	}
3001 	return (NULL);
3002 }
3003 
3004 /*
3005  * Traverse the vdev tree calling the supplied function.
3006  * We continue to walk the tree until we either have walked all
3007  * children or we receive a non-NULL return from the callback.
3008  * If a NULL callback is passed, then we just return back the first
3009  * leaf vdev we encounter.
3010  */
3011 vdev_t *
3012 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3013 {
3014 	if (vd->vdev_ops->vdev_op_leaf) {
3015 		if (func == NULL)
3016 			return (vd);
3017 		else
3018 			return (func(vd, arg));
3019 	}
3020 
3021 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3022 		vdev_t *cvd = vd->vdev_child[c];
3023 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3024 			return (cvd);
3025 	}
3026 	return (NULL);
3027 }
3028 
3029 /*
3030  * Verify that dynamic LUN growth works as expected.
3031  */
3032 /* ARGSUSED */
3033 void
3034 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3035 {
3036 	spa_t *spa = ztest_spa;
3037 	vdev_t *vd, *tvd;
3038 	metaslab_class_t *mc;
3039 	metaslab_group_t *mg;
3040 	size_t psize, newsize;
3041 	uint64_t top;
3042 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3043 
3044 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
3045 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3046 
3047 	top = ztest_random_vdev_top(spa, B_TRUE);
3048 
3049 	tvd = spa->spa_root_vdev->vdev_child[top];
3050 	mg = tvd->vdev_mg;
3051 	mc = mg->mg_class;
3052 	old_ms_count = tvd->vdev_ms_count;
3053 	old_class_space = metaslab_class_get_space(mc);
3054 
3055 	/*
3056 	 * Determine the size of the first leaf vdev associated with
3057 	 * our top-level device.
3058 	 */
3059 	vd = vdev_walk_tree(tvd, NULL, NULL);
3060 	ASSERT3P(vd, !=, NULL);
3061 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3062 
3063 	psize = vd->vdev_psize;
3064 
3065 	/*
3066 	 * We only try to expand the vdev if it's healthy, less than 4x its
3067 	 * original size, and it has a valid psize.
3068 	 */
3069 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3070 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3071 		spa_config_exit(spa, SCL_STATE, spa);
3072 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3073 		return;
3074 	}
3075 	ASSERT(psize > 0);
3076 	newsize = psize + psize / 8;
3077 	ASSERT3U(newsize, >, psize);
3078 
3079 	if (ztest_opts.zo_verbose >= 6) {
3080 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3081 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3082 	}
3083 
3084 	/*
3085 	 * Growing the vdev is a two step process:
3086 	 *	1). expand the physical size (i.e. relabel)
3087 	 *	2). online the vdev to create the new metaslabs
3088 	 */
3089 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3090 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3091 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3092 		if (ztest_opts.zo_verbose >= 5) {
3093 			(void) printf("Could not expand LUN because "
3094 			    "the vdev configuration changed.\n");
3095 		}
3096 		spa_config_exit(spa, SCL_STATE, spa);
3097 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3098 		return;
3099 	}
3100 
3101 	spa_config_exit(spa, SCL_STATE, spa);
3102 
3103 	/*
3104 	 * Expanding the LUN will update the config asynchronously,
3105 	 * thus we must wait for the async thread to complete any
3106 	 * pending tasks before proceeding.
3107 	 */
3108 	for (;;) {
3109 		boolean_t done;
3110 		mutex_enter(&spa->spa_async_lock);
3111 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3112 		mutex_exit(&spa->spa_async_lock);
3113 		if (done)
3114 			break;
3115 		txg_wait_synced(spa_get_dsl(spa), 0);
3116 		(void) poll(NULL, 0, 100);
3117 	}
3118 
3119 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3120 
3121 	tvd = spa->spa_root_vdev->vdev_child[top];
3122 	new_ms_count = tvd->vdev_ms_count;
3123 	new_class_space = metaslab_class_get_space(mc);
3124 
3125 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3126 		if (ztest_opts.zo_verbose >= 5) {
3127 			(void) printf("Could not verify LUN expansion due to "
3128 			    "intervening vdev offline or remove.\n");
3129 		}
3130 		spa_config_exit(spa, SCL_STATE, spa);
3131 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3132 		return;
3133 	}
3134 
3135 	/*
3136 	 * Make sure we were able to grow the vdev.
3137 	 */
3138 	if (new_ms_count <= old_ms_count)
3139 		fatal(0, "LUN expansion failed: ms_count %llu <= %llu\n",
3140 		    old_ms_count, new_ms_count);
3141 
3142 	/*
3143 	 * Make sure we were able to grow the pool.
3144 	 */
3145 	if (new_class_space <= old_class_space)
3146 		fatal(0, "LUN expansion failed: class_space %llu <= %llu\n",
3147 		    old_class_space, new_class_space);
3148 
3149 	if (ztest_opts.zo_verbose >= 5) {
3150 		char oldnumbuf[6], newnumbuf[6];
3151 
3152 		nicenum(old_class_space, oldnumbuf);
3153 		nicenum(new_class_space, newnumbuf);
3154 		(void) printf("%s grew from %s to %s\n",
3155 		    spa->spa_name, oldnumbuf, newnumbuf);
3156 	}
3157 
3158 	spa_config_exit(spa, SCL_STATE, spa);
3159 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
3160 }
3161 
3162 /*
3163  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3164  */
3165 /* ARGSUSED */
3166 static void
3167 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3168 {
3169 	/*
3170 	 * Create the objects common to all ztest datasets.
3171 	 */
3172 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3173 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3174 }
3175 
3176 static int
3177 ztest_dataset_create(char *dsname)
3178 {
3179 	uint64_t zilset = ztest_random(100);
3180 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3181 	    ztest_objset_create_cb, NULL);
3182 
3183 	if (err || zilset < 80)
3184 		return (err);
3185 
3186 	if (ztest_opts.zo_verbose >= 6)
3187 		(void) printf("Setting dataset %s to sync always\n", dsname);
3188 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3189 	    ZFS_SYNC_ALWAYS, B_FALSE));
3190 }
3191 
3192 /* ARGSUSED */
3193 static int
3194 ztest_objset_destroy_cb(const char *name, void *arg)
3195 {
3196 	objset_t *os;
3197 	dmu_object_info_t doi;
3198 	int error;
3199 
3200 	/*
3201 	 * Verify that the dataset contains a directory object.
3202 	 */
3203 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3204 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3205 	if (error != ENOENT) {
3206 		/* We could have crashed in the middle of destroying it */
3207 		ASSERT0(error);
3208 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3209 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3210 	}
3211 	dmu_objset_disown(os, FTAG);
3212 
3213 	/*
3214 	 * Destroy the dataset.
3215 	 */
3216 	if (strchr(name, '@') != NULL) {
3217 		VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3218 	} else {
3219 		VERIFY0(dsl_destroy_head(name));
3220 	}
3221 	return (0);
3222 }
3223 
3224 static boolean_t
3225 ztest_snapshot_create(char *osname, uint64_t id)
3226 {
3227 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3228 	int error;
3229 
3230 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3231 
3232 	error = dmu_objset_snapshot_one(osname, snapname);
3233 	if (error == ENOSPC) {
3234 		ztest_record_enospc(FTAG);
3235 		return (B_FALSE);
3236 	}
3237 	if (error != 0 && error != EEXIST) {
3238 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3239 		    snapname, error);
3240 	}
3241 	return (B_TRUE);
3242 }
3243 
3244 static boolean_t
3245 ztest_snapshot_destroy(char *osname, uint64_t id)
3246 {
3247 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3248 	int error;
3249 
3250 	(void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3251 	    (u_longlong_t)id);
3252 
3253 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3254 	if (error != 0 && error != ENOENT)
3255 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3256 	return (B_TRUE);
3257 }
3258 
3259 /* ARGSUSED */
3260 void
3261 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3262 {
3263 	ztest_ds_t zdtmp;
3264 	int iters;
3265 	int error;
3266 	objset_t *os, *os2;
3267 	char name[ZFS_MAX_DATASET_NAME_LEN];
3268 	zilog_t *zilog;
3269 
3270 	(void) rw_rdlock(&ztest_name_lock);
3271 
3272 	(void) snprintf(name, sizeof (name), "%s/temp_%llu",
3273 	    ztest_opts.zo_pool, (u_longlong_t)id);
3274 
3275 	/*
3276 	 * If this dataset exists from a previous run, process its replay log
3277 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3278 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3279 	 */
3280 	if (ztest_random(2) == 0 &&
3281 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3282 		ztest_zd_init(&zdtmp, NULL, os);
3283 		zil_replay(os, &zdtmp, ztest_replay_vector);
3284 		ztest_zd_fini(&zdtmp);
3285 		dmu_objset_disown(os, FTAG);
3286 	}
3287 
3288 	/*
3289 	 * There may be an old instance of the dataset we're about to
3290 	 * create lying around from a previous run.  If so, destroy it
3291 	 * and all of its snapshots.
3292 	 */
3293 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3294 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3295 
3296 	/*
3297 	 * Verify that the destroyed dataset is no longer in the namespace.
3298 	 */
3299 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3300 	    FTAG, &os));
3301 
3302 	/*
3303 	 * Verify that we can create a new dataset.
3304 	 */
3305 	error = ztest_dataset_create(name);
3306 	if (error) {
3307 		if (error == ENOSPC) {
3308 			ztest_record_enospc(FTAG);
3309 			(void) rw_unlock(&ztest_name_lock);
3310 			return;
3311 		}
3312 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3313 	}
3314 
3315 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3316 
3317 	ztest_zd_init(&zdtmp, NULL, os);
3318 
3319 	/*
3320 	 * Open the intent log for it.
3321 	 */
3322 	zilog = zil_open(os, ztest_get_data);
3323 
3324 	/*
3325 	 * Put some objects in there, do a little I/O to them,
3326 	 * and randomly take a couple of snapshots along the way.
3327 	 */
3328 	iters = ztest_random(5);
3329 	for (int i = 0; i < iters; i++) {
3330 		ztest_dmu_object_alloc_free(&zdtmp, id);
3331 		if (ztest_random(iters) == 0)
3332 			(void) ztest_snapshot_create(name, i);
3333 	}
3334 
3335 	/*
3336 	 * Verify that we cannot create an existing dataset.
3337 	 */
3338 	VERIFY3U(EEXIST, ==,
3339 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3340 
3341 	/*
3342 	 * Verify that we can hold an objset that is also owned.
3343 	 */
3344 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3345 	dmu_objset_rele(os2, FTAG);
3346 
3347 	/*
3348 	 * Verify that we cannot own an objset that is already owned.
3349 	 */
3350 	VERIFY3U(EBUSY, ==,
3351 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3352 
3353 	zil_close(zilog);
3354 	dmu_objset_disown(os, FTAG);
3355 	ztest_zd_fini(&zdtmp);
3356 
3357 	(void) rw_unlock(&ztest_name_lock);
3358 }
3359 
3360 /*
3361  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3362  */
3363 void
3364 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3365 {
3366 	(void) rw_rdlock(&ztest_name_lock);
3367 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3368 	(void) ztest_snapshot_create(zd->zd_name, id);
3369 	(void) rw_unlock(&ztest_name_lock);
3370 }
3371 
3372 /*
3373  * Cleanup non-standard snapshots and clones.
3374  */
3375 void
3376 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3377 {
3378 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3379 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3380 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3381 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3382 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3383 	int error;
3384 
3385 	(void) snprintf(snap1name, sizeof (snap1name),
3386 	    "%s@s1_%llu", osname, id);
3387 	(void) snprintf(clone1name, sizeof (clone1name),
3388 	    "%s/c1_%llu", osname, id);
3389 	(void) snprintf(snap2name, sizeof (snap2name),
3390 	    "%s@s2_%llu", clone1name, id);
3391 	(void) snprintf(clone2name, sizeof (clone2name),
3392 	    "%s/c2_%llu", osname, id);
3393 	(void) snprintf(snap3name, sizeof (snap3name),
3394 	    "%s@s3_%llu", clone1name, id);
3395 
3396 	error = dsl_destroy_head(clone2name);
3397 	if (error && error != ENOENT)
3398 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3399 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3400 	if (error && error != ENOENT)
3401 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3402 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3403 	if (error && error != ENOENT)
3404 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3405 	error = dsl_destroy_head(clone1name);
3406 	if (error && error != ENOENT)
3407 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3408 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3409 	if (error && error != ENOENT)
3410 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3411 }
3412 
3413 /*
3414  * Verify dsl_dataset_promote handles EBUSY
3415  */
3416 void
3417 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3418 {
3419 	objset_t *os;
3420 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3421 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3422 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3423 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3424 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3425 	char *osname = zd->zd_name;
3426 	int error;
3427 
3428 	(void) rw_rdlock(&ztest_name_lock);
3429 
3430 	ztest_dsl_dataset_cleanup(osname, id);
3431 
3432 	(void) snprintf(snap1name, sizeof (snap1name),
3433 	    "%s@s1_%llu", osname, id);
3434 	(void) snprintf(clone1name, sizeof (clone1name),
3435 	    "%s/c1_%llu", osname, id);
3436 	(void) snprintf(snap2name, sizeof (snap2name),
3437 	    "%s@s2_%llu", clone1name, id);
3438 	(void) snprintf(clone2name, sizeof (clone2name),
3439 	    "%s/c2_%llu", osname, id);
3440 	(void) snprintf(snap3name, sizeof (snap3name),
3441 	    "%s@s3_%llu", clone1name, id);
3442 
3443 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3444 	if (error && error != EEXIST) {
3445 		if (error == ENOSPC) {
3446 			ztest_record_enospc(FTAG);
3447 			goto out;
3448 		}
3449 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3450 	}
3451 
3452 	error = dmu_objset_clone(clone1name, snap1name);
3453 	if (error) {
3454 		if (error == ENOSPC) {
3455 			ztest_record_enospc(FTAG);
3456 			goto out;
3457 		}
3458 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3459 	}
3460 
3461 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3462 	if (error && error != EEXIST) {
3463 		if (error == ENOSPC) {
3464 			ztest_record_enospc(FTAG);
3465 			goto out;
3466 		}
3467 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3468 	}
3469 
3470 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3471 	if (error && error != EEXIST) {
3472 		if (error == ENOSPC) {
3473 			ztest_record_enospc(FTAG);
3474 			goto out;
3475 		}
3476 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3477 	}
3478 
3479 	error = dmu_objset_clone(clone2name, snap3name);
3480 	if (error) {
3481 		if (error == ENOSPC) {
3482 			ztest_record_enospc(FTAG);
3483 			goto out;
3484 		}
3485 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3486 	}
3487 
3488 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3489 	if (error)
3490 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3491 	error = dsl_dataset_promote(clone2name, NULL);
3492 	if (error == ENOSPC) {
3493 		dmu_objset_disown(os, FTAG);
3494 		ztest_record_enospc(FTAG);
3495 		goto out;
3496 	}
3497 	if (error != EBUSY)
3498 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3499 		    error);
3500 	dmu_objset_disown(os, FTAG);
3501 
3502 out:
3503 	ztest_dsl_dataset_cleanup(osname, id);
3504 
3505 	(void) rw_unlock(&ztest_name_lock);
3506 }
3507 
3508 /*
3509  * Verify that dmu_object_{alloc,free} work as expected.
3510  */
3511 void
3512 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3513 {
3514 	ztest_od_t od[4];
3515 	int batchsize = sizeof (od) / sizeof (od[0]);
3516 
3517 	for (int b = 0; b < batchsize; b++)
3518 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER, 0, 0);
3519 
3520 	/*
3521 	 * Destroy the previous batch of objects, create a new batch,
3522 	 * and do some I/O on the new objects.
3523 	 */
3524 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3525 		return;
3526 
3527 	while (ztest_random(4 * batchsize) != 0)
3528 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3529 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3530 }
3531 
3532 /*
3533  * Verify that dmu_{read,write} work as expected.
3534  */
3535 void
3536 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3537 {
3538 	objset_t *os = zd->zd_os;
3539 	ztest_od_t od[2];
3540 	dmu_tx_t *tx;
3541 	int i, freeit, error;
3542 	uint64_t n, s, txg;
3543 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3544 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3545 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3546 	uint64_t regions = 997;
3547 	uint64_t stride = 123456789ULL;
3548 	uint64_t width = 40;
3549 	int free_percent = 5;
3550 
3551 	/*
3552 	 * This test uses two objects, packobj and bigobj, that are always
3553 	 * updated together (i.e. in the same tx) so that their contents are
3554 	 * in sync and can be compared.  Their contents relate to each other
3555 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3556 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3557 	 * for any index n, there are three bufwads that should be identical:
3558 	 *
3559 	 *	packobj, at offset n * sizeof (bufwad_t)
3560 	 *	bigobj, at the head of the nth chunk
3561 	 *	bigobj, at the tail of the nth chunk
3562 	 *
3563 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3564 	 * and it doesn't have any relation to the object blocksize.
3565 	 * The only requirement is that it can hold at least two bufwads.
3566 	 *
3567 	 * Normally, we write the bufwad to each of these locations.
3568 	 * However, free_percent of the time we instead write zeroes to
3569 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3570 	 * bigobj to packobj, we can verify that the DMU is correctly
3571 	 * tracking which parts of an object are allocated and free,
3572 	 * and that the contents of the allocated blocks are correct.
3573 	 */
3574 
3575 	/*
3576 	 * Read the directory info.  If it's the first time, set things up.
3577 	 */
3578 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, chunksize);
3579 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3580 
3581 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3582 		return;
3583 
3584 	bigobj = od[0].od_object;
3585 	packobj = od[1].od_object;
3586 	chunksize = od[0].od_gen;
3587 	ASSERT(chunksize == od[1].od_gen);
3588 
3589 	/*
3590 	 * Prefetch a random chunk of the big object.
3591 	 * Our aim here is to get some async reads in flight
3592 	 * for blocks that we may free below; the DMU should
3593 	 * handle this race correctly.
3594 	 */
3595 	n = ztest_random(regions) * stride + ztest_random(width);
3596 	s = 1 + ztest_random(2 * width - 1);
3597 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
3598 	    ZIO_PRIORITY_SYNC_READ);
3599 
3600 	/*
3601 	 * Pick a random index and compute the offsets into packobj and bigobj.
3602 	 */
3603 	n = ztest_random(regions) * stride + ztest_random(width);
3604 	s = 1 + ztest_random(width - 1);
3605 
3606 	packoff = n * sizeof (bufwad_t);
3607 	packsize = s * sizeof (bufwad_t);
3608 
3609 	bigoff = n * chunksize;
3610 	bigsize = s * chunksize;
3611 
3612 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
3613 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
3614 
3615 	/*
3616 	 * free_percent of the time, free a range of bigobj rather than
3617 	 * overwriting it.
3618 	 */
3619 	freeit = (ztest_random(100) < free_percent);
3620 
3621 	/*
3622 	 * Read the current contents of our objects.
3623 	 */
3624 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
3625 	    DMU_READ_PREFETCH);
3626 	ASSERT0(error);
3627 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
3628 	    DMU_READ_PREFETCH);
3629 	ASSERT0(error);
3630 
3631 	/*
3632 	 * Get a tx for the mods to both packobj and bigobj.
3633 	 */
3634 	tx = dmu_tx_create(os);
3635 
3636 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
3637 
3638 	if (freeit)
3639 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
3640 	else
3641 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3642 
3643 	/* This accounts for setting the checksum/compression. */
3644 	dmu_tx_hold_bonus(tx, bigobj);
3645 
3646 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3647 	if (txg == 0) {
3648 		umem_free(packbuf, packsize);
3649 		umem_free(bigbuf, bigsize);
3650 		return;
3651 	}
3652 
3653 	enum zio_checksum cksum;
3654 	do {
3655 		cksum = (enum zio_checksum)
3656 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
3657 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
3658 	dmu_object_set_checksum(os, bigobj, cksum, tx);
3659 
3660 	enum zio_compress comp;
3661 	do {
3662 		comp = (enum zio_compress)
3663 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
3664 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
3665 	dmu_object_set_compress(os, bigobj, comp, tx);
3666 
3667 	/*
3668 	 * For each index from n to n + s, verify that the existing bufwad
3669 	 * in packobj matches the bufwads at the head and tail of the
3670 	 * corresponding chunk in bigobj.  Then update all three bufwads
3671 	 * with the new values we want to write out.
3672 	 */
3673 	for (i = 0; i < s; i++) {
3674 		/* LINTED */
3675 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3676 		/* LINTED */
3677 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3678 		/* LINTED */
3679 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3680 
3681 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3682 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3683 
3684 		if (pack->bw_txg > txg)
3685 			fatal(0, "future leak: got %llx, open txg is %llx",
3686 			    pack->bw_txg, txg);
3687 
3688 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3689 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3690 			    pack->bw_index, n, i);
3691 
3692 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3693 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3694 
3695 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3696 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3697 
3698 		if (freeit) {
3699 			bzero(pack, sizeof (bufwad_t));
3700 		} else {
3701 			pack->bw_index = n + i;
3702 			pack->bw_txg = txg;
3703 			pack->bw_data = 1 + ztest_random(-2ULL);
3704 		}
3705 		*bigH = *pack;
3706 		*bigT = *pack;
3707 	}
3708 
3709 	/*
3710 	 * We've verified all the old bufwads, and made new ones.
3711 	 * Now write them out.
3712 	 */
3713 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3714 
3715 	if (freeit) {
3716 		if (ztest_opts.zo_verbose >= 7) {
3717 			(void) printf("freeing offset %llx size %llx"
3718 			    " txg %llx\n",
3719 			    (u_longlong_t)bigoff,
3720 			    (u_longlong_t)bigsize,
3721 			    (u_longlong_t)txg);
3722 		}
3723 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
3724 	} else {
3725 		if (ztest_opts.zo_verbose >= 7) {
3726 			(void) printf("writing offset %llx size %llx"
3727 			    " txg %llx\n",
3728 			    (u_longlong_t)bigoff,
3729 			    (u_longlong_t)bigsize,
3730 			    (u_longlong_t)txg);
3731 		}
3732 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
3733 	}
3734 
3735 	dmu_tx_commit(tx);
3736 
3737 	/*
3738 	 * Sanity check the stuff we just wrote.
3739 	 */
3740 	{
3741 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
3742 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
3743 
3744 		VERIFY(0 == dmu_read(os, packobj, packoff,
3745 		    packsize, packcheck, DMU_READ_PREFETCH));
3746 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
3747 		    bigsize, bigcheck, DMU_READ_PREFETCH));
3748 
3749 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
3750 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
3751 
3752 		umem_free(packcheck, packsize);
3753 		umem_free(bigcheck, bigsize);
3754 	}
3755 
3756 	umem_free(packbuf, packsize);
3757 	umem_free(bigbuf, bigsize);
3758 }
3759 
3760 void
3761 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
3762     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
3763 {
3764 	uint64_t i;
3765 	bufwad_t *pack;
3766 	bufwad_t *bigH;
3767 	bufwad_t *bigT;
3768 
3769 	/*
3770 	 * For each index from n to n + s, verify that the existing bufwad
3771 	 * in packobj matches the bufwads at the head and tail of the
3772 	 * corresponding chunk in bigobj.  Then update all three bufwads
3773 	 * with the new values we want to write out.
3774 	 */
3775 	for (i = 0; i < s; i++) {
3776 		/* LINTED */
3777 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
3778 		/* LINTED */
3779 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
3780 		/* LINTED */
3781 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
3782 
3783 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
3784 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
3785 
3786 		if (pack->bw_txg > txg)
3787 			fatal(0, "future leak: got %llx, open txg is %llx",
3788 			    pack->bw_txg, txg);
3789 
3790 		if (pack->bw_data != 0 && pack->bw_index != n + i)
3791 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
3792 			    pack->bw_index, n, i);
3793 
3794 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
3795 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
3796 
3797 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
3798 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
3799 
3800 		pack->bw_index = n + i;
3801 		pack->bw_txg = txg;
3802 		pack->bw_data = 1 + ztest_random(-2ULL);
3803 
3804 		*bigH = *pack;
3805 		*bigT = *pack;
3806 	}
3807 }
3808 
3809 void
3810 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
3811 {
3812 	objset_t *os = zd->zd_os;
3813 	ztest_od_t od[2];
3814 	dmu_tx_t *tx;
3815 	uint64_t i;
3816 	int error;
3817 	uint64_t n, s, txg;
3818 	bufwad_t *packbuf, *bigbuf;
3819 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3820 	uint64_t blocksize = ztest_random_blocksize();
3821 	uint64_t chunksize = blocksize;
3822 	uint64_t regions = 997;
3823 	uint64_t stride = 123456789ULL;
3824 	uint64_t width = 9;
3825 	dmu_buf_t *bonus_db;
3826 	arc_buf_t **bigbuf_arcbufs;
3827 	dmu_object_info_t doi;
3828 
3829 	/*
3830 	 * This test uses two objects, packobj and bigobj, that are always
3831 	 * updated together (i.e. in the same tx) so that their contents are
3832 	 * in sync and can be compared.  Their contents relate to each other
3833 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3834 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3835 	 * for any index n, there are three bufwads that should be identical:
3836 	 *
3837 	 *	packobj, at offset n * sizeof (bufwad_t)
3838 	 *	bigobj, at the head of the nth chunk
3839 	 *	bigobj, at the tail of the nth chunk
3840 	 *
3841 	 * The chunk size is set equal to bigobj block size so that
3842 	 * dmu_assign_arcbuf() can be tested for object updates.
3843 	 */
3844 
3845 	/*
3846 	 * Read the directory info.  If it's the first time, set things up.
3847 	 */
3848 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
3849 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, chunksize);
3850 
3851 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3852 		return;
3853 
3854 	bigobj = od[0].od_object;
3855 	packobj = od[1].od_object;
3856 	blocksize = od[0].od_blocksize;
3857 	chunksize = blocksize;
3858 	ASSERT(chunksize == od[1].od_gen);
3859 
3860 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
3861 	VERIFY(ISP2(doi.doi_data_block_size));
3862 	VERIFY(chunksize == doi.doi_data_block_size);
3863 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
3864 
3865 	/*
3866 	 * Pick a random index and compute the offsets into packobj and bigobj.
3867 	 */
3868 	n = ztest_random(regions) * stride + ztest_random(width);
3869 	s = 1 + ztest_random(width - 1);
3870 
3871 	packoff = n * sizeof (bufwad_t);
3872 	packsize = s * sizeof (bufwad_t);
3873 
3874 	bigoff = n * chunksize;
3875 	bigsize = s * chunksize;
3876 
3877 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
3878 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
3879 
3880 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
3881 
3882 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
3883 
3884 	/*
3885 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
3886 	 * Iteration 1 test zcopy to already referenced dbufs.
3887 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
3888 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
3889 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
3890 	 * Iteration 5 test zcopy when it can't be done.
3891 	 * Iteration 6 one more zcopy write.
3892 	 */
3893 	for (i = 0; i < 7; i++) {
3894 		uint64_t j;
3895 		uint64_t off;
3896 
3897 		/*
3898 		 * In iteration 5 (i == 5) use arcbufs
3899 		 * that don't match bigobj blksz to test
3900 		 * dmu_assign_arcbuf() when it can't directly
3901 		 * assign an arcbuf to a dbuf.
3902 		 */
3903 		for (j = 0; j < s; j++) {
3904 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3905 				bigbuf_arcbufs[j] =
3906 				    dmu_request_arcbuf(bonus_db, chunksize);
3907 			} else {
3908 				bigbuf_arcbufs[2 * j] =
3909 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3910 				bigbuf_arcbufs[2 * j + 1] =
3911 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
3912 			}
3913 		}
3914 
3915 		/*
3916 		 * Get a tx for the mods to both packobj and bigobj.
3917 		 */
3918 		tx = dmu_tx_create(os);
3919 
3920 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
3921 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
3922 
3923 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
3924 		if (txg == 0) {
3925 			umem_free(packbuf, packsize);
3926 			umem_free(bigbuf, bigsize);
3927 			for (j = 0; j < s; j++) {
3928 				if (i != 5 ||
3929 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
3930 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
3931 				} else {
3932 					dmu_return_arcbuf(
3933 					    bigbuf_arcbufs[2 * j]);
3934 					dmu_return_arcbuf(
3935 					    bigbuf_arcbufs[2 * j + 1]);
3936 				}
3937 			}
3938 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
3939 			dmu_buf_rele(bonus_db, FTAG);
3940 			return;
3941 		}
3942 
3943 		/*
3944 		 * 50% of the time don't read objects in the 1st iteration to
3945 		 * test dmu_assign_arcbuf() for the case when there're no
3946 		 * existing dbufs for the specified offsets.
3947 		 */
3948 		if (i != 0 || ztest_random(2) != 0) {
3949 			error = dmu_read(os, packobj, packoff,
3950 			    packsize, packbuf, DMU_READ_PREFETCH);
3951 			ASSERT0(error);
3952 			error = dmu_read(os, bigobj, bigoff, bigsize,
3953 			    bigbuf, DMU_READ_PREFETCH);
3954 			ASSERT0(error);
3955 		}
3956 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
3957 		    n, chunksize, txg);
3958 
3959 		/*
3960 		 * We've verified all the old bufwads, and made new ones.
3961 		 * Now write them out.
3962 		 */
3963 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
3964 		if (ztest_opts.zo_verbose >= 7) {
3965 			(void) printf("writing offset %llx size %llx"
3966 			    " txg %llx\n",
3967 			    (u_longlong_t)bigoff,
3968 			    (u_longlong_t)bigsize,
3969 			    (u_longlong_t)txg);
3970 		}
3971 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
3972 			dmu_buf_t *dbt;
3973 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3974 				bcopy((caddr_t)bigbuf + (off - bigoff),
3975 				    bigbuf_arcbufs[j]->b_data, chunksize);
3976 			} else {
3977 				bcopy((caddr_t)bigbuf + (off - bigoff),
3978 				    bigbuf_arcbufs[2 * j]->b_data,
3979 				    chunksize / 2);
3980 				bcopy((caddr_t)bigbuf + (off - bigoff) +
3981 				    chunksize / 2,
3982 				    bigbuf_arcbufs[2 * j + 1]->b_data,
3983 				    chunksize / 2);
3984 			}
3985 
3986 			if (i == 1) {
3987 				VERIFY(dmu_buf_hold(os, bigobj, off,
3988 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
3989 			}
3990 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
3991 				dmu_assign_arcbuf(bonus_db, off,
3992 				    bigbuf_arcbufs[j], tx);
3993 			} else {
3994 				dmu_assign_arcbuf(bonus_db, off,
3995 				    bigbuf_arcbufs[2 * j], tx);
3996 				dmu_assign_arcbuf(bonus_db,
3997 				    off + chunksize / 2,
3998 				    bigbuf_arcbufs[2 * j + 1], tx);
3999 			}
4000 			if (i == 1) {
4001 				dmu_buf_rele(dbt, FTAG);
4002 			}
4003 		}
4004 		dmu_tx_commit(tx);
4005 
4006 		/*
4007 		 * Sanity check the stuff we just wrote.
4008 		 */
4009 		{
4010 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4011 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4012 
4013 			VERIFY(0 == dmu_read(os, packobj, packoff,
4014 			    packsize, packcheck, DMU_READ_PREFETCH));
4015 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4016 			    bigsize, bigcheck, DMU_READ_PREFETCH));
4017 
4018 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4019 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4020 
4021 			umem_free(packcheck, packsize);
4022 			umem_free(bigcheck, bigsize);
4023 		}
4024 		if (i == 2) {
4025 			txg_wait_open(dmu_objset_pool(os), 0);
4026 		} else if (i == 3) {
4027 			txg_wait_synced(dmu_objset_pool(os), 0);
4028 		}
4029 	}
4030 
4031 	dmu_buf_rele(bonus_db, FTAG);
4032 	umem_free(packbuf, packsize);
4033 	umem_free(bigbuf, bigsize);
4034 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4035 }
4036 
4037 /* ARGSUSED */
4038 void
4039 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4040 {
4041 	ztest_od_t od[1];
4042 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4043 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4044 
4045 	/*
4046 	 * Have multiple threads write to large offsets in an object
4047 	 * to verify that parallel writes to an object -- even to the
4048 	 * same blocks within the object -- doesn't cause any trouble.
4049 	 */
4050 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4051 
4052 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4053 		return;
4054 
4055 	while (ztest_random(10) != 0)
4056 		ztest_io(zd, od[0].od_object, offset);
4057 }
4058 
4059 void
4060 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4061 {
4062 	ztest_od_t od[1];
4063 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4064 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4065 	uint64_t count = ztest_random(20) + 1;
4066 	uint64_t blocksize = ztest_random_blocksize();
4067 	void *data;
4068 
4069 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
4070 
4071 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4072 		return;
4073 
4074 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4075 		return;
4076 
4077 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4078 
4079 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4080 
4081 	while (ztest_random(count) != 0) {
4082 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4083 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4084 		    data) != 0)
4085 			break;
4086 		while (ztest_random(4) != 0)
4087 			ztest_io(zd, od[0].od_object, randoff);
4088 	}
4089 
4090 	umem_free(data, blocksize);
4091 }
4092 
4093 /*
4094  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4095  */
4096 #define	ZTEST_ZAP_MIN_INTS	1
4097 #define	ZTEST_ZAP_MAX_INTS	4
4098 #define	ZTEST_ZAP_MAX_PROPS	1000
4099 
4100 void
4101 ztest_zap(ztest_ds_t *zd, uint64_t id)
4102 {
4103 	objset_t *os = zd->zd_os;
4104 	ztest_od_t od[1];
4105 	uint64_t object;
4106 	uint64_t txg, last_txg;
4107 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4108 	uint64_t zl_ints, zl_intsize, prop;
4109 	int i, ints;
4110 	dmu_tx_t *tx;
4111 	char propname[100], txgname[100];
4112 	int error;
4113 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4114 
4115 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4116 
4117 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4118 		return;
4119 
4120 	object = od[0].od_object;
4121 
4122 	/*
4123 	 * Generate a known hash collision, and verify that
4124 	 * we can lookup and remove both entries.
4125 	 */
4126 	tx = dmu_tx_create(os);
4127 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4128 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4129 	if (txg == 0)
4130 		return;
4131 	for (i = 0; i < 2; i++) {
4132 		value[i] = i;
4133 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4134 		    1, &value[i], tx));
4135 	}
4136 	for (i = 0; i < 2; i++) {
4137 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4138 		    sizeof (uint64_t), 1, &value[i], tx));
4139 		VERIFY3U(0, ==,
4140 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4141 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4142 		ASSERT3U(zl_ints, ==, 1);
4143 	}
4144 	for (i = 0; i < 2; i++) {
4145 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4146 	}
4147 	dmu_tx_commit(tx);
4148 
4149 	/*
4150 	 * Generate a buch of random entries.
4151 	 */
4152 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4153 
4154 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4155 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4156 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4157 	bzero(value, sizeof (value));
4158 	last_txg = 0;
4159 
4160 	/*
4161 	 * If these zap entries already exist, validate their contents.
4162 	 */
4163 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4164 	if (error == 0) {
4165 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4166 		ASSERT3U(zl_ints, ==, 1);
4167 
4168 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4169 		    zl_ints, &last_txg) == 0);
4170 
4171 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4172 		    &zl_ints) == 0);
4173 
4174 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4175 		ASSERT3U(zl_ints, ==, ints);
4176 
4177 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4178 		    zl_ints, value) == 0);
4179 
4180 		for (i = 0; i < ints; i++) {
4181 			ASSERT3U(value[i], ==, last_txg + object + i);
4182 		}
4183 	} else {
4184 		ASSERT3U(error, ==, ENOENT);
4185 	}
4186 
4187 	/*
4188 	 * Atomically update two entries in our zap object.
4189 	 * The first is named txg_%llu, and contains the txg
4190 	 * in which the property was last updated.  The second
4191 	 * is named prop_%llu, and the nth element of its value
4192 	 * should be txg + object + n.
4193 	 */
4194 	tx = dmu_tx_create(os);
4195 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4196 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4197 	if (txg == 0)
4198 		return;
4199 
4200 	if (last_txg > txg)
4201 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4202 
4203 	for (i = 0; i < ints; i++)
4204 		value[i] = txg + object + i;
4205 
4206 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4207 	    1, &txg, tx));
4208 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4209 	    ints, value, tx));
4210 
4211 	dmu_tx_commit(tx);
4212 
4213 	/*
4214 	 * Remove a random pair of entries.
4215 	 */
4216 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4217 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4218 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4219 
4220 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4221 
4222 	if (error == ENOENT)
4223 		return;
4224 
4225 	ASSERT0(error);
4226 
4227 	tx = dmu_tx_create(os);
4228 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4229 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4230 	if (txg == 0)
4231 		return;
4232 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4233 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4234 	dmu_tx_commit(tx);
4235 }
4236 
4237 /*
4238  * Testcase to test the upgrading of a microzap to fatzap.
4239  */
4240 void
4241 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4242 {
4243 	objset_t *os = zd->zd_os;
4244 	ztest_od_t od[1];
4245 	uint64_t object, txg;
4246 
4247 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0);
4248 
4249 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4250 		return;
4251 
4252 	object = od[0].od_object;
4253 
4254 	/*
4255 	 * Add entries to this ZAP and make sure it spills over
4256 	 * and gets upgraded to a fatzap. Also, since we are adding
4257 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4258 	 */
4259 	for (int i = 0; i < 2050; i++) {
4260 		char name[ZFS_MAX_DATASET_NAME_LEN];
4261 		uint64_t value = i;
4262 		dmu_tx_t *tx;
4263 		int error;
4264 
4265 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4266 		    id, value);
4267 
4268 		tx = dmu_tx_create(os);
4269 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4270 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4271 		if (txg == 0)
4272 			return;
4273 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4274 		    &value, tx);
4275 		ASSERT(error == 0 || error == EEXIST);
4276 		dmu_tx_commit(tx);
4277 	}
4278 }
4279 
4280 /* ARGSUSED */
4281 void
4282 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4283 {
4284 	objset_t *os = zd->zd_os;
4285 	ztest_od_t od[1];
4286 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4287 	dmu_tx_t *tx;
4288 	int i, namelen, error;
4289 	int micro = ztest_random(2);
4290 	char name[20], string_value[20];
4291 	void *data;
4292 
4293 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER, 0, 0);
4294 
4295 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4296 		return;
4297 
4298 	object = od[0].od_object;
4299 
4300 	/*
4301 	 * Generate a random name of the form 'xxx.....' where each
4302 	 * x is a random printable character and the dots are dots.
4303 	 * There are 94 such characters, and the name length goes from
4304 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4305 	 */
4306 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4307 
4308 	for (i = 0; i < 3; i++)
4309 		name[i] = '!' + ztest_random('~' - '!' + 1);
4310 	for (; i < namelen - 1; i++)
4311 		name[i] = '.';
4312 	name[i] = '\0';
4313 
4314 	if ((namelen & 1) || micro) {
4315 		wsize = sizeof (txg);
4316 		wc = 1;
4317 		data = &txg;
4318 	} else {
4319 		wsize = 1;
4320 		wc = namelen;
4321 		data = string_value;
4322 	}
4323 
4324 	count = -1ULL;
4325 	VERIFY0(zap_count(os, object, &count));
4326 	ASSERT(count != -1ULL);
4327 
4328 	/*
4329 	 * Select an operation: length, lookup, add, update, remove.
4330 	 */
4331 	i = ztest_random(5);
4332 
4333 	if (i >= 2) {
4334 		tx = dmu_tx_create(os);
4335 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4336 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4337 		if (txg == 0)
4338 			return;
4339 		bcopy(name, string_value, namelen);
4340 	} else {
4341 		tx = NULL;
4342 		txg = 0;
4343 		bzero(string_value, namelen);
4344 	}
4345 
4346 	switch (i) {
4347 
4348 	case 0:
4349 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4350 		if (error == 0) {
4351 			ASSERT3U(wsize, ==, zl_wsize);
4352 			ASSERT3U(wc, ==, zl_wc);
4353 		} else {
4354 			ASSERT3U(error, ==, ENOENT);
4355 		}
4356 		break;
4357 
4358 	case 1:
4359 		error = zap_lookup(os, object, name, wsize, wc, data);
4360 		if (error == 0) {
4361 			if (data == string_value &&
4362 			    bcmp(name, data, namelen) != 0)
4363 				fatal(0, "name '%s' != val '%s' len %d",
4364 				    name, data, namelen);
4365 		} else {
4366 			ASSERT3U(error, ==, ENOENT);
4367 		}
4368 		break;
4369 
4370 	case 2:
4371 		error = zap_add(os, object, name, wsize, wc, data, tx);
4372 		ASSERT(error == 0 || error == EEXIST);
4373 		break;
4374 
4375 	case 3:
4376 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4377 		break;
4378 
4379 	case 4:
4380 		error = zap_remove(os, object, name, tx);
4381 		ASSERT(error == 0 || error == ENOENT);
4382 		break;
4383 	}
4384 
4385 	if (tx != NULL)
4386 		dmu_tx_commit(tx);
4387 }
4388 
4389 /*
4390  * Commit callback data.
4391  */
4392 typedef struct ztest_cb_data {
4393 	list_node_t		zcd_node;
4394 	uint64_t		zcd_txg;
4395 	int			zcd_expected_err;
4396 	boolean_t		zcd_added;
4397 	boolean_t		zcd_called;
4398 	spa_t			*zcd_spa;
4399 } ztest_cb_data_t;
4400 
4401 /* This is the actual commit callback function */
4402 static void
4403 ztest_commit_callback(void *arg, int error)
4404 {
4405 	ztest_cb_data_t *data = arg;
4406 	uint64_t synced_txg;
4407 
4408 	VERIFY(data != NULL);
4409 	VERIFY3S(data->zcd_expected_err, ==, error);
4410 	VERIFY(!data->zcd_called);
4411 
4412 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4413 	if (data->zcd_txg > synced_txg)
4414 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4415 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4416 		    synced_txg);
4417 
4418 	data->zcd_called = B_TRUE;
4419 
4420 	if (error == ECANCELED) {
4421 		ASSERT0(data->zcd_txg);
4422 		ASSERT(!data->zcd_added);
4423 
4424 		/*
4425 		 * The private callback data should be destroyed here, but
4426 		 * since we are going to check the zcd_called field after
4427 		 * dmu_tx_abort(), we will destroy it there.
4428 		 */
4429 		return;
4430 	}
4431 
4432 	/* Was this callback added to the global callback list? */
4433 	if (!data->zcd_added)
4434 		goto out;
4435 
4436 	ASSERT3U(data->zcd_txg, !=, 0);
4437 
4438 	/* Remove our callback from the list */
4439 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4440 	list_remove(&zcl.zcl_callbacks, data);
4441 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4442 
4443 out:
4444 	umem_free(data, sizeof (ztest_cb_data_t));
4445 }
4446 
4447 /* Allocate and initialize callback data structure */
4448 static ztest_cb_data_t *
4449 ztest_create_cb_data(objset_t *os, uint64_t txg)
4450 {
4451 	ztest_cb_data_t *cb_data;
4452 
4453 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4454 
4455 	cb_data->zcd_txg = txg;
4456 	cb_data->zcd_spa = dmu_objset_spa(os);
4457 
4458 	return (cb_data);
4459 }
4460 
4461 /*
4462  * If a number of txgs equal to this threshold have been created after a commit
4463  * callback has been registered but not called, then we assume there is an
4464  * implementation bug.
4465  */
4466 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4467 
4468 /*
4469  * Commit callback test.
4470  */
4471 void
4472 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4473 {
4474 	objset_t *os = zd->zd_os;
4475 	ztest_od_t od[1];
4476 	dmu_tx_t *tx;
4477 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4478 	uint64_t old_txg, txg;
4479 	int i, error;
4480 
4481 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
4482 
4483 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4484 		return;
4485 
4486 	tx = dmu_tx_create(os);
4487 
4488 	cb_data[0] = ztest_create_cb_data(os, 0);
4489 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4490 
4491 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4492 
4493 	/* Every once in a while, abort the transaction on purpose */
4494 	if (ztest_random(100) == 0)
4495 		error = -1;
4496 
4497 	if (!error)
4498 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4499 
4500 	txg = error ? 0 : dmu_tx_get_txg(tx);
4501 
4502 	cb_data[0]->zcd_txg = txg;
4503 	cb_data[1] = ztest_create_cb_data(os, txg);
4504 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4505 
4506 	if (error) {
4507 		/*
4508 		 * It's not a strict requirement to call the registered
4509 		 * callbacks from inside dmu_tx_abort(), but that's what
4510 		 * it's supposed to happen in the current implementation
4511 		 * so we will check for that.
4512 		 */
4513 		for (i = 0; i < 2; i++) {
4514 			cb_data[i]->zcd_expected_err = ECANCELED;
4515 			VERIFY(!cb_data[i]->zcd_called);
4516 		}
4517 
4518 		dmu_tx_abort(tx);
4519 
4520 		for (i = 0; i < 2; i++) {
4521 			VERIFY(cb_data[i]->zcd_called);
4522 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4523 		}
4524 
4525 		return;
4526 	}
4527 
4528 	cb_data[2] = ztest_create_cb_data(os, txg);
4529 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4530 
4531 	/*
4532 	 * Read existing data to make sure there isn't a future leak.
4533 	 */
4534 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4535 	    &old_txg, DMU_READ_PREFETCH));
4536 
4537 	if (old_txg > txg)
4538 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4539 		    old_txg, txg);
4540 
4541 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4542 
4543 	(void) mutex_lock(&zcl.zcl_callbacks_lock);
4544 
4545 	/*
4546 	 * Since commit callbacks don't have any ordering requirement and since
4547 	 * it is theoretically possible for a commit callback to be called
4548 	 * after an arbitrary amount of time has elapsed since its txg has been
4549 	 * synced, it is difficult to reliably determine whether a commit
4550 	 * callback hasn't been called due to high load or due to a flawed
4551 	 * implementation.
4552 	 *
4553 	 * In practice, we will assume that if after a certain number of txgs a
4554 	 * commit callback hasn't been called, then most likely there's an
4555 	 * implementation bug..
4556 	 */
4557 	tmp_cb = list_head(&zcl.zcl_callbacks);
4558 	if (tmp_cb != NULL &&
4559 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4560 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4561 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4562 	}
4563 
4564 	/*
4565 	 * Let's find the place to insert our callbacks.
4566 	 *
4567 	 * Even though the list is ordered by txg, it is possible for the
4568 	 * insertion point to not be the end because our txg may already be
4569 	 * quiescing at this point and other callbacks in the open txg
4570 	 * (from other objsets) may have sneaked in.
4571 	 */
4572 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4573 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4574 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4575 
4576 	/* Add the 3 callbacks to the list */
4577 	for (i = 0; i < 3; i++) {
4578 		if (tmp_cb == NULL)
4579 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4580 		else
4581 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4582 			    cb_data[i]);
4583 
4584 		cb_data[i]->zcd_added = B_TRUE;
4585 		VERIFY(!cb_data[i]->zcd_called);
4586 
4587 		tmp_cb = cb_data[i];
4588 	}
4589 
4590 	(void) mutex_unlock(&zcl.zcl_callbacks_lock);
4591 
4592 	dmu_tx_commit(tx);
4593 }
4594 
4595 /* ARGSUSED */
4596 void
4597 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
4598 {
4599 	zfs_prop_t proplist[] = {
4600 		ZFS_PROP_CHECKSUM,
4601 		ZFS_PROP_COMPRESSION,
4602 		ZFS_PROP_COPIES,
4603 		ZFS_PROP_DEDUP
4604 	};
4605 
4606 	(void) rw_rdlock(&ztest_name_lock);
4607 
4608 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
4609 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
4610 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
4611 
4612 	(void) rw_unlock(&ztest_name_lock);
4613 }
4614 
4615 /* ARGSUSED */
4616 void
4617 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
4618 {
4619 	nvlist_t *props = NULL;
4620 
4621 	(void) rw_rdlock(&ztest_name_lock);
4622 
4623 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
4624 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
4625 
4626 	VERIFY0(spa_prop_get(ztest_spa, &props));
4627 
4628 	if (ztest_opts.zo_verbose >= 6)
4629 		dump_nvlist(props, 4);
4630 
4631 	nvlist_free(props);
4632 
4633 	(void) rw_unlock(&ztest_name_lock);
4634 }
4635 
4636 static int
4637 user_release_one(const char *snapname, const char *holdname)
4638 {
4639 	nvlist_t *snaps, *holds;
4640 	int error;
4641 
4642 	snaps = fnvlist_alloc();
4643 	holds = fnvlist_alloc();
4644 	fnvlist_add_boolean(holds, holdname);
4645 	fnvlist_add_nvlist(snaps, snapname, holds);
4646 	fnvlist_free(holds);
4647 	error = dsl_dataset_user_release(snaps, NULL);
4648 	fnvlist_free(snaps);
4649 	return (error);
4650 }
4651 
4652 /*
4653  * Test snapshot hold/release and deferred destroy.
4654  */
4655 void
4656 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
4657 {
4658 	int error;
4659 	objset_t *os = zd->zd_os;
4660 	objset_t *origin;
4661 	char snapname[100];
4662 	char fullname[100];
4663 	char clonename[100];
4664 	char tag[100];
4665 	char osname[ZFS_MAX_DATASET_NAME_LEN];
4666 	nvlist_t *holds;
4667 
4668 	(void) rw_rdlock(&ztest_name_lock);
4669 
4670 	dmu_objset_name(os, osname);
4671 
4672 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
4673 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
4674 	(void) snprintf(clonename, sizeof (clonename),
4675 	    "%s/ch1_%llu", osname, id);
4676 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
4677 
4678 	/*
4679 	 * Clean up from any previous run.
4680 	 */
4681 	error = dsl_destroy_head(clonename);
4682 	if (error != ENOENT)
4683 		ASSERT0(error);
4684 	error = user_release_one(fullname, tag);
4685 	if (error != ESRCH && error != ENOENT)
4686 		ASSERT0(error);
4687 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4688 	if (error != ENOENT)
4689 		ASSERT0(error);
4690 
4691 	/*
4692 	 * Create snapshot, clone it, mark snap for deferred destroy,
4693 	 * destroy clone, verify snap was also destroyed.
4694 	 */
4695 	error = dmu_objset_snapshot_one(osname, snapname);
4696 	if (error) {
4697 		if (error == ENOSPC) {
4698 			ztest_record_enospc("dmu_objset_snapshot");
4699 			goto out;
4700 		}
4701 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4702 	}
4703 
4704 	error = dmu_objset_clone(clonename, fullname);
4705 	if (error) {
4706 		if (error == ENOSPC) {
4707 			ztest_record_enospc("dmu_objset_clone");
4708 			goto out;
4709 		}
4710 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
4711 	}
4712 
4713 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4714 	if (error) {
4715 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4716 		    fullname, error);
4717 	}
4718 
4719 	error = dsl_destroy_head(clonename);
4720 	if (error)
4721 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
4722 
4723 	error = dmu_objset_hold(fullname, FTAG, &origin);
4724 	if (error != ENOENT)
4725 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
4726 
4727 	/*
4728 	 * Create snapshot, add temporary hold, verify that we can't
4729 	 * destroy a held snapshot, mark for deferred destroy,
4730 	 * release hold, verify snapshot was destroyed.
4731 	 */
4732 	error = dmu_objset_snapshot_one(osname, snapname);
4733 	if (error) {
4734 		if (error == ENOSPC) {
4735 			ztest_record_enospc("dmu_objset_snapshot");
4736 			goto out;
4737 		}
4738 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
4739 	}
4740 
4741 	holds = fnvlist_alloc();
4742 	fnvlist_add_string(holds, fullname, tag);
4743 	error = dsl_dataset_user_hold(holds, 0, NULL);
4744 	fnvlist_free(holds);
4745 
4746 	if (error == ENOSPC) {
4747 		ztest_record_enospc("dsl_dataset_user_hold");
4748 		goto out;
4749 	} else if (error) {
4750 		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
4751 		    fullname, tag, error);
4752 	}
4753 
4754 	error = dsl_destroy_snapshot(fullname, B_FALSE);
4755 	if (error != EBUSY) {
4756 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
4757 		    fullname, error);
4758 	}
4759 
4760 	error = dsl_destroy_snapshot(fullname, B_TRUE);
4761 	if (error) {
4762 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
4763 		    fullname, error);
4764 	}
4765 
4766 	error = user_release_one(fullname, tag);
4767 	if (error)
4768 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
4769 
4770 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
4771 
4772 out:
4773 	(void) rw_unlock(&ztest_name_lock);
4774 }
4775 
4776 /*
4777  * Inject random faults into the on-disk data.
4778  */
4779 /* ARGSUSED */
4780 void
4781 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
4782 {
4783 	ztest_shared_t *zs = ztest_shared;
4784 	spa_t *spa = ztest_spa;
4785 	int fd;
4786 	uint64_t offset;
4787 	uint64_t leaves;
4788 	uint64_t bad = 0x1990c0ffeedecade;
4789 	uint64_t top, leaf;
4790 	char path0[MAXPATHLEN];
4791 	char pathrand[MAXPATHLEN];
4792 	size_t fsize;
4793 	int bshift = SPA_OLD_MAXBLOCKSHIFT + 2;	/* don't scrog all labels */
4794 	int iters = 1000;
4795 	int maxfaults;
4796 	int mirror_save;
4797 	vdev_t *vd0 = NULL;
4798 	uint64_t guid0 = 0;
4799 	boolean_t islog = B_FALSE;
4800 
4801 	VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4802 	maxfaults = MAXFAULTS();
4803 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
4804 	mirror_save = zs->zs_mirrors;
4805 	VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4806 
4807 	ASSERT(leaves >= 1);
4808 
4809 	/*
4810 	 * Grab the name lock as reader. There are some operations
4811 	 * which don't like to have their vdevs changed while
4812 	 * they are in progress (i.e. spa_change_guid). Those
4813 	 * operations will have grabbed the name lock as writer.
4814 	 */
4815 	(void) rw_rdlock(&ztest_name_lock);
4816 
4817 	/*
4818 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
4819 	 */
4820 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
4821 
4822 	if (ztest_random(2) == 0) {
4823 		/*
4824 		 * Inject errors on a normal data device or slog device.
4825 		 */
4826 		top = ztest_random_vdev_top(spa, B_TRUE);
4827 		leaf = ztest_random(leaves) + zs->zs_splits;
4828 
4829 		/*
4830 		 * Generate paths to the first leaf in this top-level vdev,
4831 		 * and to the random leaf we selected.  We'll induce transient
4832 		 * write failures and random online/offline activity on leaf 0,
4833 		 * and we'll write random garbage to the randomly chosen leaf.
4834 		 */
4835 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
4836 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4837 		    top * leaves + zs->zs_splits);
4838 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
4839 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
4840 		    top * leaves + leaf);
4841 
4842 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
4843 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
4844 			islog = B_TRUE;
4845 
4846 		/*
4847 		 * If the top-level vdev needs to be resilvered
4848 		 * then we only allow faults on the device that is
4849 		 * resilvering.
4850 		 */
4851 		if (vd0 != NULL && maxfaults != 1 &&
4852 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
4853 		    vd0->vdev_resilver_txg != 0)) {
4854 			/*
4855 			 * Make vd0 explicitly claim to be unreadable,
4856 			 * or unwriteable, or reach behind its back
4857 			 * and close the underlying fd.  We can do this if
4858 			 * maxfaults == 0 because we'll fail and reexecute,
4859 			 * and we can do it if maxfaults >= 2 because we'll
4860 			 * have enough redundancy.  If maxfaults == 1, the
4861 			 * combination of this with injection of random data
4862 			 * corruption below exceeds the pool's fault tolerance.
4863 			 */
4864 			vdev_file_t *vf = vd0->vdev_tsd;
4865 
4866 			if (vf != NULL && ztest_random(3) == 0) {
4867 				(void) close(vf->vf_vnode->v_fd);
4868 				vf->vf_vnode->v_fd = -1;
4869 			} else if (ztest_random(2) == 0) {
4870 				vd0->vdev_cant_read = B_TRUE;
4871 			} else {
4872 				vd0->vdev_cant_write = B_TRUE;
4873 			}
4874 			guid0 = vd0->vdev_guid;
4875 		}
4876 	} else {
4877 		/*
4878 		 * Inject errors on an l2cache device.
4879 		 */
4880 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
4881 
4882 		if (sav->sav_count == 0) {
4883 			spa_config_exit(spa, SCL_STATE, FTAG);
4884 			(void) rw_unlock(&ztest_name_lock);
4885 			return;
4886 		}
4887 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
4888 		guid0 = vd0->vdev_guid;
4889 		(void) strcpy(path0, vd0->vdev_path);
4890 		(void) strcpy(pathrand, vd0->vdev_path);
4891 
4892 		leaf = 0;
4893 		leaves = 1;
4894 		maxfaults = INT_MAX;	/* no limit on cache devices */
4895 	}
4896 
4897 	spa_config_exit(spa, SCL_STATE, FTAG);
4898 	(void) rw_unlock(&ztest_name_lock);
4899 
4900 	/*
4901 	 * If we can tolerate two or more faults, or we're dealing
4902 	 * with a slog, randomly online/offline vd0.
4903 	 */
4904 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
4905 		if (ztest_random(10) < 6) {
4906 			int flags = (ztest_random(2) == 0 ?
4907 			    ZFS_OFFLINE_TEMPORARY : 0);
4908 
4909 			/*
4910 			 * We have to grab the zs_name_lock as writer to
4911 			 * prevent a race between offlining a slog and
4912 			 * destroying a dataset. Offlining the slog will
4913 			 * grab a reference on the dataset which may cause
4914 			 * dmu_objset_destroy() to fail with EBUSY thus
4915 			 * leaving the dataset in an inconsistent state.
4916 			 */
4917 			if (islog)
4918 				(void) rw_wrlock(&ztest_name_lock);
4919 
4920 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
4921 
4922 			if (islog)
4923 				(void) rw_unlock(&ztest_name_lock);
4924 		} else {
4925 			/*
4926 			 * Ideally we would like to be able to randomly
4927 			 * call vdev_[on|off]line without holding locks
4928 			 * to force unpredictable failures but the side
4929 			 * effects of vdev_[on|off]line prevent us from
4930 			 * doing so. We grab the ztest_vdev_lock here to
4931 			 * prevent a race between injection testing and
4932 			 * aux_vdev removal.
4933 			 */
4934 			VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4935 			(void) vdev_online(spa, guid0, 0, NULL);
4936 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4937 		}
4938 	}
4939 
4940 	if (maxfaults == 0)
4941 		return;
4942 
4943 	/*
4944 	 * We have at least single-fault tolerance, so inject data corruption.
4945 	 */
4946 	fd = open(pathrand, O_RDWR);
4947 
4948 	if (fd == -1)	/* we hit a gap in the device namespace */
4949 		return;
4950 
4951 	fsize = lseek(fd, 0, SEEK_END);
4952 
4953 	while (--iters != 0) {
4954 		offset = ztest_random(fsize / (leaves << bshift)) *
4955 		    (leaves << bshift) + (leaf << bshift) +
4956 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
4957 
4958 		if (offset >= fsize)
4959 			continue;
4960 
4961 		VERIFY(mutex_lock(&ztest_vdev_lock) == 0);
4962 		if (mirror_save != zs->zs_mirrors) {
4963 			VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4964 			(void) close(fd);
4965 			return;
4966 		}
4967 
4968 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
4969 			fatal(1, "can't inject bad word at 0x%llx in %s",
4970 			    offset, pathrand);
4971 
4972 		VERIFY(mutex_unlock(&ztest_vdev_lock) == 0);
4973 
4974 		if (ztest_opts.zo_verbose >= 7)
4975 			(void) printf("injected bad word into %s,"
4976 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
4977 	}
4978 
4979 	(void) close(fd);
4980 }
4981 
4982 /*
4983  * Verify that DDT repair works as expected.
4984  */
4985 void
4986 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
4987 {
4988 	ztest_shared_t *zs = ztest_shared;
4989 	spa_t *spa = ztest_spa;
4990 	objset_t *os = zd->zd_os;
4991 	ztest_od_t od[1];
4992 	uint64_t object, blocksize, txg, pattern, psize;
4993 	enum zio_checksum checksum = spa_dedup_checksum(spa);
4994 	dmu_buf_t *db;
4995 	dmu_tx_t *tx;
4996 	void *buf;
4997 	blkptr_t blk;
4998 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
4999 
5000 	blocksize = ztest_random_blocksize();
5001 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
5002 
5003 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize, 0);
5004 
5005 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5006 		return;
5007 
5008 	/*
5009 	 * Take the name lock as writer to prevent anyone else from changing
5010 	 * the pool and dataset properies we need to maintain during this test.
5011 	 */
5012 	(void) rw_wrlock(&ztest_name_lock);
5013 
5014 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5015 	    B_FALSE) != 0 ||
5016 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5017 	    B_FALSE) != 0) {
5018 		(void) rw_unlock(&ztest_name_lock);
5019 		return;
5020 	}
5021 
5022 	object = od[0].od_object;
5023 	blocksize = od[0].od_blocksize;
5024 	pattern = zs->zs_guid ^ dmu_objset_fsid_guid(os);
5025 
5026 	ASSERT(object != 0);
5027 
5028 	tx = dmu_tx_create(os);
5029 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5030 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5031 	if (txg == 0) {
5032 		(void) rw_unlock(&ztest_name_lock);
5033 		return;
5034 	}
5035 
5036 	/*
5037 	 * Write all the copies of our block.
5038 	 */
5039 	for (int i = 0; i < copies; i++) {
5040 		uint64_t offset = i * blocksize;
5041 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5042 		    DMU_READ_NO_PREFETCH);
5043 		if (error != 0) {
5044 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5045 			    os, (long long)object, (long long) offset, error);
5046 		}
5047 		ASSERT(db->db_offset == offset);
5048 		ASSERT(db->db_size == blocksize);
5049 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5050 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5051 		dmu_buf_will_fill(db, tx);
5052 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5053 		dmu_buf_rele(db, FTAG);
5054 	}
5055 
5056 	dmu_tx_commit(tx);
5057 	txg_wait_synced(spa_get_dsl(spa), txg);
5058 
5059 	/*
5060 	 * Find out what block we got.
5061 	 */
5062 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5063 	    DMU_READ_NO_PREFETCH));
5064 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5065 	dmu_buf_rele(db, FTAG);
5066 
5067 	/*
5068 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5069 	 */
5070 	psize = BP_GET_PSIZE(&blk);
5071 	buf = zio_buf_alloc(psize);
5072 	ztest_pattern_set(buf, psize, ~pattern);
5073 
5074 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5075 	    buf, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5076 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5077 
5078 	zio_buf_free(buf, psize);
5079 
5080 	(void) rw_unlock(&ztest_name_lock);
5081 }
5082 
5083 /*
5084  * Scrub the pool.
5085  */
5086 /* ARGSUSED */
5087 void
5088 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5089 {
5090 	spa_t *spa = ztest_spa;
5091 
5092 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5093 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5094 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5095 }
5096 
5097 /*
5098  * Change the guid for the pool.
5099  */
5100 /* ARGSUSED */
5101 void
5102 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5103 {
5104 	spa_t *spa = ztest_spa;
5105 	uint64_t orig, load;
5106 	int error;
5107 
5108 	orig = spa_guid(spa);
5109 	load = spa_load_guid(spa);
5110 
5111 	(void) rw_wrlock(&ztest_name_lock);
5112 	error = spa_change_guid(spa);
5113 	(void) rw_unlock(&ztest_name_lock);
5114 
5115 	if (error != 0)
5116 		return;
5117 
5118 	if (ztest_opts.zo_verbose >= 4) {
5119 		(void) printf("Changed guid old %llu -> %llu\n",
5120 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5121 	}
5122 
5123 	VERIFY3U(orig, !=, spa_guid(spa));
5124 	VERIFY3U(load, ==, spa_load_guid(spa));
5125 }
5126 
5127 /*
5128  * Rename the pool to a different name and then rename it back.
5129  */
5130 /* ARGSUSED */
5131 void
5132 ztest_spa_rename(ztest_ds_t *zd, uint64_t id)
5133 {
5134 	char *oldname, *newname;
5135 	spa_t *spa;
5136 
5137 	(void) rw_wrlock(&ztest_name_lock);
5138 
5139 	oldname = ztest_opts.zo_pool;
5140 	newname = umem_alloc(strlen(oldname) + 5, UMEM_NOFAIL);
5141 	(void) strcpy(newname, oldname);
5142 	(void) strcat(newname, "_tmp");
5143 
5144 	/*
5145 	 * Do the rename
5146 	 */
5147 	VERIFY3U(0, ==, spa_rename(oldname, newname));
5148 
5149 	/*
5150 	 * Try to open it under the old name, which shouldn't exist
5151 	 */
5152 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5153 
5154 	/*
5155 	 * Open it under the new name and make sure it's still the same spa_t.
5156 	 */
5157 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5158 
5159 	ASSERT(spa == ztest_spa);
5160 	spa_close(spa, FTAG);
5161 
5162 	/*
5163 	 * Rename it back to the original
5164 	 */
5165 	VERIFY3U(0, ==, spa_rename(newname, oldname));
5166 
5167 	/*
5168 	 * Make sure it can still be opened
5169 	 */
5170 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5171 
5172 	ASSERT(spa == ztest_spa);
5173 	spa_close(spa, FTAG);
5174 
5175 	umem_free(newname, strlen(newname) + 1);
5176 
5177 	(void) rw_unlock(&ztest_name_lock);
5178 }
5179 
5180 /*
5181  * Verify pool integrity by running zdb.
5182  */
5183 static void
5184 ztest_run_zdb(char *pool)
5185 {
5186 	int status;
5187 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5188 	char zbuf[1024];
5189 	char *bin;
5190 	char *ztest;
5191 	char *isa;
5192 	int isalen;
5193 	FILE *fp;
5194 
5195 	(void) realpath(getexecname(), zdb);
5196 
5197 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5198 	bin = strstr(zdb, "/usr/bin/");
5199 	ztest = strstr(bin, "/ztest");
5200 	isa = bin + 8;
5201 	isalen = ztest - isa;
5202 	isa = strdup(isa);
5203 	/* LINTED */
5204 	(void) sprintf(bin,
5205 	    "/usr/sbin%.*s/zdb -bcc%s%s -d -U %s %s",
5206 	    isalen,
5207 	    isa,
5208 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5209 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5210 	    spa_config_path,
5211 	    pool);
5212 	free(isa);
5213 
5214 	if (ztest_opts.zo_verbose >= 5)
5215 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5216 
5217 	fp = popen(zdb, "r");
5218 
5219 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5220 		if (ztest_opts.zo_verbose >= 3)
5221 			(void) printf("%s", zbuf);
5222 
5223 	status = pclose(fp);
5224 
5225 	if (status == 0)
5226 		return;
5227 
5228 	ztest_dump_core = 0;
5229 	if (WIFEXITED(status))
5230 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5231 	else
5232 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5233 }
5234 
5235 static void
5236 ztest_walk_pool_directory(char *header)
5237 {
5238 	spa_t *spa = NULL;
5239 
5240 	if (ztest_opts.zo_verbose >= 6)
5241 		(void) printf("%s\n", header);
5242 
5243 	mutex_enter(&spa_namespace_lock);
5244 	while ((spa = spa_next(spa)) != NULL)
5245 		if (ztest_opts.zo_verbose >= 6)
5246 			(void) printf("\t%s\n", spa_name(spa));
5247 	mutex_exit(&spa_namespace_lock);
5248 }
5249 
5250 static void
5251 ztest_spa_import_export(char *oldname, char *newname)
5252 {
5253 	nvlist_t *config, *newconfig;
5254 	uint64_t pool_guid;
5255 	spa_t *spa;
5256 	int error;
5257 
5258 	if (ztest_opts.zo_verbose >= 4) {
5259 		(void) printf("import/export: old = %s, new = %s\n",
5260 		    oldname, newname);
5261 	}
5262 
5263 	/*
5264 	 * Clean up from previous runs.
5265 	 */
5266 	(void) spa_destroy(newname);
5267 
5268 	/*
5269 	 * Get the pool's configuration and guid.
5270 	 */
5271 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5272 
5273 	/*
5274 	 * Kick off a scrub to tickle scrub/export races.
5275 	 */
5276 	if (ztest_random(2) == 0)
5277 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5278 
5279 	pool_guid = spa_guid(spa);
5280 	spa_close(spa, FTAG);
5281 
5282 	ztest_walk_pool_directory("pools before export");
5283 
5284 	/*
5285 	 * Export it.
5286 	 */
5287 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5288 
5289 	ztest_walk_pool_directory("pools after export");
5290 
5291 	/*
5292 	 * Try to import it.
5293 	 */
5294 	newconfig = spa_tryimport(config);
5295 	ASSERT(newconfig != NULL);
5296 	nvlist_free(newconfig);
5297 
5298 	/*
5299 	 * Import it under the new name.
5300 	 */
5301 	error = spa_import(newname, config, NULL, 0);
5302 	if (error != 0) {
5303 		dump_nvlist(config, 0);
5304 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5305 		    oldname, newname, error);
5306 	}
5307 
5308 	ztest_walk_pool_directory("pools after import");
5309 
5310 	/*
5311 	 * Try to import it again -- should fail with EEXIST.
5312 	 */
5313 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5314 
5315 	/*
5316 	 * Try to import it under a different name -- should fail with EEXIST.
5317 	 */
5318 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5319 
5320 	/*
5321 	 * Verify that the pool is no longer visible under the old name.
5322 	 */
5323 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5324 
5325 	/*
5326 	 * Verify that we can open and close the pool using the new name.
5327 	 */
5328 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5329 	ASSERT(pool_guid == spa_guid(spa));
5330 	spa_close(spa, FTAG);
5331 
5332 	nvlist_free(config);
5333 }
5334 
5335 static void
5336 ztest_resume(spa_t *spa)
5337 {
5338 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5339 		(void) printf("resuming from suspended state\n");
5340 	spa_vdev_state_enter(spa, SCL_NONE);
5341 	vdev_clear(spa, NULL);
5342 	(void) spa_vdev_state_exit(spa, NULL, 0);
5343 	(void) zio_resume(spa);
5344 }
5345 
5346 static void *
5347 ztest_resume_thread(void *arg)
5348 {
5349 	spa_t *spa = arg;
5350 
5351 	while (!ztest_exiting) {
5352 		if (spa_suspended(spa))
5353 			ztest_resume(spa);
5354 		(void) poll(NULL, 0, 100);
5355 	}
5356 	return (NULL);
5357 }
5358 
5359 static void *
5360 ztest_deadman_thread(void *arg)
5361 {
5362 	ztest_shared_t *zs = arg;
5363 	spa_t *spa = ztest_spa;
5364 	hrtime_t delta, total = 0;
5365 
5366 	for (;;) {
5367 		delta = zs->zs_thread_stop - zs->zs_thread_start +
5368 		    MSEC2NSEC(zfs_deadman_synctime_ms);
5369 
5370 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5371 
5372 		/*
5373 		 * If the pool is suspended then fail immediately. Otherwise,
5374 		 * check to see if the pool is making any progress. If
5375 		 * vdev_deadman() discovers that there hasn't been any recent
5376 		 * I/Os then it will end up aborting the tests.
5377 		 */
5378 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5379 			fatal(0, "aborting test after %llu seconds because "
5380 			    "pool has transitioned to a suspended state.",
5381 			    zfs_deadman_synctime_ms / 1000);
5382 			return (NULL);
5383 		}
5384 		vdev_deadman(spa->spa_root_vdev);
5385 
5386 		total += zfs_deadman_synctime_ms/1000;
5387 		(void) printf("ztest has been running for %lld seconds\n",
5388 		    total);
5389 	}
5390 }
5391 
5392 static void
5393 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5394 {
5395 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5396 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5397 	hrtime_t functime = gethrtime();
5398 
5399 	for (int i = 0; i < zi->zi_iters; i++)
5400 		zi->zi_func(zd, id);
5401 
5402 	functime = gethrtime() - functime;
5403 
5404 	atomic_add_64(&zc->zc_count, 1);
5405 	atomic_add_64(&zc->zc_time, functime);
5406 
5407 	if (ztest_opts.zo_verbose >= 4) {
5408 		Dl_info dli;
5409 		(void) dladdr((void *)zi->zi_func, &dli);
5410 		(void) printf("%6.2f sec in %s\n",
5411 		    (double)functime / NANOSEC, dli.dli_sname);
5412 	}
5413 }
5414 
5415 static void *
5416 ztest_thread(void *arg)
5417 {
5418 	int rand;
5419 	uint64_t id = (uintptr_t)arg;
5420 	ztest_shared_t *zs = ztest_shared;
5421 	uint64_t call_next;
5422 	hrtime_t now;
5423 	ztest_info_t *zi;
5424 	ztest_shared_callstate_t *zc;
5425 
5426 	while ((now = gethrtime()) < zs->zs_thread_stop) {
5427 		/*
5428 		 * See if it's time to force a crash.
5429 		 */
5430 		if (now > zs->zs_thread_kill)
5431 			ztest_kill(zs);
5432 
5433 		/*
5434 		 * If we're getting ENOSPC with some regularity, stop.
5435 		 */
5436 		if (zs->zs_enospc_count > 10)
5437 			break;
5438 
5439 		/*
5440 		 * Pick a random function to execute.
5441 		 */
5442 		rand = ztest_random(ZTEST_FUNCS);
5443 		zi = &ztest_info[rand];
5444 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
5445 		call_next = zc->zc_next;
5446 
5447 		if (now >= call_next &&
5448 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
5449 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
5450 			ztest_execute(rand, zi, id);
5451 		}
5452 	}
5453 
5454 	return (NULL);
5455 }
5456 
5457 static void
5458 ztest_dataset_name(char *dsname, char *pool, int d)
5459 {
5460 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
5461 }
5462 
5463 static void
5464 ztest_dataset_destroy(int d)
5465 {
5466 	char name[ZFS_MAX_DATASET_NAME_LEN];
5467 
5468 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5469 
5470 	if (ztest_opts.zo_verbose >= 3)
5471 		(void) printf("Destroying %s to free up space\n", name);
5472 
5473 	/*
5474 	 * Cleanup any non-standard clones and snapshots.  In general,
5475 	 * ztest thread t operates on dataset (t % zopt_datasets),
5476 	 * so there may be more than one thing to clean up.
5477 	 */
5478 	for (int t = d; t < ztest_opts.zo_threads;
5479 	    t += ztest_opts.zo_datasets) {
5480 		ztest_dsl_dataset_cleanup(name, t);
5481 	}
5482 
5483 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
5484 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
5485 }
5486 
5487 static void
5488 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
5489 {
5490 	uint64_t usedobjs, dirobjs, scratch;
5491 
5492 	/*
5493 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
5494 	 * Therefore, the number of objects in use should equal the
5495 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
5496 	 * If not, we have an object leak.
5497 	 *
5498 	 * Note that we can only check this in ztest_dataset_open(),
5499 	 * when the open-context and syncing-context values agree.
5500 	 * That's because zap_count() returns the open-context value,
5501 	 * while dmu_objset_space() returns the rootbp fill count.
5502 	 */
5503 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
5504 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
5505 	ASSERT3U(dirobjs + 1, ==, usedobjs);
5506 }
5507 
5508 static int
5509 ztest_dataset_open(int d)
5510 {
5511 	ztest_ds_t *zd = &ztest_ds[d];
5512 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
5513 	objset_t *os;
5514 	zilog_t *zilog;
5515 	char name[ZFS_MAX_DATASET_NAME_LEN];
5516 	int error;
5517 
5518 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
5519 
5520 	(void) rw_rdlock(&ztest_name_lock);
5521 
5522 	error = ztest_dataset_create(name);
5523 	if (error == ENOSPC) {
5524 		(void) rw_unlock(&ztest_name_lock);
5525 		ztest_record_enospc(FTAG);
5526 		return (error);
5527 	}
5528 	ASSERT(error == 0 || error == EEXIST);
5529 
5530 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
5531 	(void) rw_unlock(&ztest_name_lock);
5532 
5533 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
5534 
5535 	zilog = zd->zd_zilog;
5536 
5537 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
5538 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
5539 		fatal(0, "missing log records: claimed %llu < committed %llu",
5540 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
5541 
5542 	ztest_dataset_dirobj_verify(zd);
5543 
5544 	zil_replay(os, zd, ztest_replay_vector);
5545 
5546 	ztest_dataset_dirobj_verify(zd);
5547 
5548 	if (ztest_opts.zo_verbose >= 6)
5549 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
5550 		    zd->zd_name,
5551 		    (u_longlong_t)zilog->zl_parse_blk_count,
5552 		    (u_longlong_t)zilog->zl_parse_lr_count,
5553 		    (u_longlong_t)zilog->zl_replaying_seq);
5554 
5555 	zilog = zil_open(os, ztest_get_data);
5556 
5557 	if (zilog->zl_replaying_seq != 0 &&
5558 	    zilog->zl_replaying_seq < committed_seq)
5559 		fatal(0, "missing log records: replayed %llu < committed %llu",
5560 		    zilog->zl_replaying_seq, committed_seq);
5561 
5562 	return (0);
5563 }
5564 
5565 static void
5566 ztest_dataset_close(int d)
5567 {
5568 	ztest_ds_t *zd = &ztest_ds[d];
5569 
5570 	zil_close(zd->zd_zilog);
5571 	dmu_objset_disown(zd->zd_os, zd);
5572 
5573 	ztest_zd_fini(zd);
5574 }
5575 
5576 /*
5577  * Kick off threads to run tests on all datasets in parallel.
5578  */
5579 static void
5580 ztest_run(ztest_shared_t *zs)
5581 {
5582 	thread_t *tid;
5583 	spa_t *spa;
5584 	objset_t *os;
5585 	thread_t resume_tid;
5586 	int error;
5587 
5588 	ztest_exiting = B_FALSE;
5589 
5590 	/*
5591 	 * Initialize parent/child shared state.
5592 	 */
5593 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5594 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5595 
5596 	zs->zs_thread_start = gethrtime();
5597 	zs->zs_thread_stop =
5598 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
5599 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
5600 	zs->zs_thread_kill = zs->zs_thread_stop;
5601 	if (ztest_random(100) < ztest_opts.zo_killrate) {
5602 		zs->zs_thread_kill -=
5603 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
5604 	}
5605 
5606 	(void) _mutex_init(&zcl.zcl_callbacks_lock, USYNC_THREAD, NULL);
5607 
5608 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
5609 	    offsetof(ztest_cb_data_t, zcd_node));
5610 
5611 	/*
5612 	 * Open our pool.
5613 	 */
5614 	kernel_init(FREAD | FWRITE);
5615 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
5616 	spa->spa_debug = B_TRUE;
5617 	metaslab_preload_limit = ztest_random(20) + 1;
5618 	ztest_spa = spa;
5619 
5620 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
5621 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
5622 	zs->zs_guid = dmu_objset_fsid_guid(os);
5623 	dmu_objset_disown(os, FTAG);
5624 
5625 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
5626 
5627 	/*
5628 	 * We don't expect the pool to suspend unless maxfaults == 0,
5629 	 * in which case ztest_fault_inject() temporarily takes away
5630 	 * the only valid replica.
5631 	 */
5632 	if (MAXFAULTS() == 0)
5633 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
5634 	else
5635 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
5636 
5637 	/*
5638 	 * Create a thread to periodically resume suspended I/O.
5639 	 */
5640 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
5641 	    &resume_tid) == 0);
5642 
5643 	/*
5644 	 * Create a deadman thread to abort() if we hang.
5645 	 */
5646 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
5647 	    NULL) == 0);
5648 
5649 	/*
5650 	 * Verify that we can safely inquire about about any object,
5651 	 * whether it's allocated or not.  To make it interesting,
5652 	 * we probe a 5-wide window around each power of two.
5653 	 * This hits all edge cases, including zero and the max.
5654 	 */
5655 	for (int t = 0; t < 64; t++) {
5656 		for (int d = -5; d <= 5; d++) {
5657 			error = dmu_object_info(spa->spa_meta_objset,
5658 			    (1ULL << t) + d, NULL);
5659 			ASSERT(error == 0 || error == ENOENT ||
5660 			    error == EINVAL);
5661 		}
5662 	}
5663 
5664 	/*
5665 	 * If we got any ENOSPC errors on the previous run, destroy something.
5666 	 */
5667 	if (zs->zs_enospc_count != 0) {
5668 		int d = ztest_random(ztest_opts.zo_datasets);
5669 		ztest_dataset_destroy(d);
5670 	}
5671 	zs->zs_enospc_count = 0;
5672 
5673 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
5674 	    UMEM_NOFAIL);
5675 
5676 	if (ztest_opts.zo_verbose >= 4)
5677 		(void) printf("starting main threads...\n");
5678 
5679 	/*
5680 	 * Kick off all the tests that run in parallel.
5681 	 */
5682 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
5683 		if (t < ztest_opts.zo_datasets &&
5684 		    ztest_dataset_open(t) != 0)
5685 			return;
5686 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
5687 		    THR_BOUND, &tid[t]) == 0);
5688 	}
5689 
5690 	/*
5691 	 * Wait for all of the tests to complete.  We go in reverse order
5692 	 * so we don't close datasets while threads are still using them.
5693 	 */
5694 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
5695 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
5696 		if (t < ztest_opts.zo_datasets)
5697 			ztest_dataset_close(t);
5698 	}
5699 
5700 	txg_wait_synced(spa_get_dsl(spa), 0);
5701 
5702 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
5703 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
5704 	zfs_dbgmsg_print(FTAG);
5705 
5706 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
5707 
5708 	/* Kill the resume thread */
5709 	ztest_exiting = B_TRUE;
5710 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
5711 	ztest_resume(spa);
5712 
5713 	/*
5714 	 * Right before closing the pool, kick off a bunch of async I/O;
5715 	 * spa_close() should wait for it to complete.
5716 	 */
5717 	for (uint64_t object = 1; object < 50; object++) {
5718 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
5719 		    ZIO_PRIORITY_SYNC_READ);
5720 	}
5721 
5722 	spa_close(spa, FTAG);
5723 
5724 	/*
5725 	 * Verify that we can loop over all pools.
5726 	 */
5727 	mutex_enter(&spa_namespace_lock);
5728 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
5729 		if (ztest_opts.zo_verbose > 3)
5730 			(void) printf("spa_next: found %s\n", spa_name(spa));
5731 	mutex_exit(&spa_namespace_lock);
5732 
5733 	/*
5734 	 * Verify that we can export the pool and reimport it under a
5735 	 * different name.
5736 	 */
5737 	if (ztest_random(2) == 0) {
5738 		char name[ZFS_MAX_DATASET_NAME_LEN];
5739 		(void) snprintf(name, sizeof (name), "%s_import",
5740 		    ztest_opts.zo_pool);
5741 		ztest_spa_import_export(ztest_opts.zo_pool, name);
5742 		ztest_spa_import_export(name, ztest_opts.zo_pool);
5743 	}
5744 
5745 	kernel_fini();
5746 
5747 	list_destroy(&zcl.zcl_callbacks);
5748 
5749 	(void) _mutex_destroy(&zcl.zcl_callbacks_lock);
5750 
5751 	(void) rwlock_destroy(&ztest_name_lock);
5752 	(void) _mutex_destroy(&ztest_vdev_lock);
5753 }
5754 
5755 static void
5756 ztest_freeze(void)
5757 {
5758 	ztest_ds_t *zd = &ztest_ds[0];
5759 	spa_t *spa;
5760 	int numloops = 0;
5761 
5762 	if (ztest_opts.zo_verbose >= 3)
5763 		(void) printf("testing spa_freeze()...\n");
5764 
5765 	kernel_init(FREAD | FWRITE);
5766 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5767 	VERIFY3U(0, ==, ztest_dataset_open(0));
5768 	spa->spa_debug = B_TRUE;
5769 	ztest_spa = spa;
5770 
5771 	/*
5772 	 * Force the first log block to be transactionally allocated.
5773 	 * We have to do this before we freeze the pool -- otherwise
5774 	 * the log chain won't be anchored.
5775 	 */
5776 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
5777 		ztest_dmu_object_alloc_free(zd, 0);
5778 		zil_commit(zd->zd_zilog, 0);
5779 	}
5780 
5781 	txg_wait_synced(spa_get_dsl(spa), 0);
5782 
5783 	/*
5784 	 * Freeze the pool.  This stops spa_sync() from doing anything,
5785 	 * so that the only way to record changes from now on is the ZIL.
5786 	 */
5787 	spa_freeze(spa);
5788 
5789 	/*
5790 	 * Because it is hard to predict how much space a write will actually
5791 	 * require beforehand, we leave ourselves some fudge space to write over
5792 	 * capacity.
5793 	 */
5794 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
5795 
5796 	/*
5797 	 * Run tests that generate log records but don't alter the pool config
5798 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
5799 	 * We do a txg_wait_synced() after each iteration to force the txg
5800 	 * to increase well beyond the last synced value in the uberblock.
5801 	 * The ZIL should be OK with that.
5802 	 *
5803 	 * Run a random number of times less than zo_maxloops and ensure we do
5804 	 * not run out of space on the pool.
5805 	 */
5806 	while (ztest_random(10) != 0 &&
5807 	    numloops++ < ztest_opts.zo_maxloops &&
5808 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
5809 		ztest_od_t od;
5810 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0);
5811 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
5812 		ztest_io(zd, od.od_object,
5813 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
5814 		txg_wait_synced(spa_get_dsl(spa), 0);
5815 	}
5816 
5817 	/*
5818 	 * Commit all of the changes we just generated.
5819 	 */
5820 	zil_commit(zd->zd_zilog, 0);
5821 	txg_wait_synced(spa_get_dsl(spa), 0);
5822 
5823 	/*
5824 	 * Close our dataset and close the pool.
5825 	 */
5826 	ztest_dataset_close(0);
5827 	spa_close(spa, FTAG);
5828 	kernel_fini();
5829 
5830 	/*
5831 	 * Open and close the pool and dataset to induce log replay.
5832 	 */
5833 	kernel_init(FREAD | FWRITE);
5834 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5835 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
5836 	VERIFY3U(0, ==, ztest_dataset_open(0));
5837 	ztest_dataset_close(0);
5838 
5839 	spa->spa_debug = B_TRUE;
5840 	ztest_spa = spa;
5841 	txg_wait_synced(spa_get_dsl(spa), 0);
5842 	ztest_reguid(NULL, 0);
5843 
5844 	spa_close(spa, FTAG);
5845 	kernel_fini();
5846 }
5847 
5848 void
5849 print_time(hrtime_t t, char *timebuf)
5850 {
5851 	hrtime_t s = t / NANOSEC;
5852 	hrtime_t m = s / 60;
5853 	hrtime_t h = m / 60;
5854 	hrtime_t d = h / 24;
5855 
5856 	s -= m * 60;
5857 	m -= h * 60;
5858 	h -= d * 24;
5859 
5860 	timebuf[0] = '\0';
5861 
5862 	if (d)
5863 		(void) sprintf(timebuf,
5864 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
5865 	else if (h)
5866 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
5867 	else if (m)
5868 		(void) sprintf(timebuf, "%llum%02llus", m, s);
5869 	else
5870 		(void) sprintf(timebuf, "%llus", s);
5871 }
5872 
5873 static nvlist_t *
5874 make_random_props()
5875 {
5876 	nvlist_t *props;
5877 
5878 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
5879 	if (ztest_random(2) == 0)
5880 		return (props);
5881 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
5882 
5883 	return (props);
5884 }
5885 
5886 /*
5887  * Create a storage pool with the given name and initial vdev size.
5888  * Then test spa_freeze() functionality.
5889  */
5890 static void
5891 ztest_init(ztest_shared_t *zs)
5892 {
5893 	spa_t *spa;
5894 	nvlist_t *nvroot, *props;
5895 
5896 	VERIFY(_mutex_init(&ztest_vdev_lock, USYNC_THREAD, NULL) == 0);
5897 	VERIFY(rwlock_init(&ztest_name_lock, USYNC_THREAD, NULL) == 0);
5898 
5899 	kernel_init(FREAD | FWRITE);
5900 
5901 	/*
5902 	 * Create the storage pool.
5903 	 */
5904 	(void) spa_destroy(ztest_opts.zo_pool);
5905 	ztest_shared->zs_vdev_next_leaf = 0;
5906 	zs->zs_splits = 0;
5907 	zs->zs_mirrors = ztest_opts.zo_mirrors;
5908 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
5909 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
5910 	props = make_random_props();
5911 	for (int i = 0; i < SPA_FEATURES; i++) {
5912 		char buf[1024];
5913 		(void) snprintf(buf, sizeof (buf), "feature@%s",
5914 		    spa_feature_table[i].fi_uname);
5915 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
5916 	}
5917 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
5918 	nvlist_free(nvroot);
5919 
5920 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
5921 	zs->zs_metaslab_sz =
5922 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
5923 
5924 	spa_close(spa, FTAG);
5925 
5926 	kernel_fini();
5927 
5928 	ztest_run_zdb(ztest_opts.zo_pool);
5929 
5930 	ztest_freeze();
5931 
5932 	ztest_run_zdb(ztest_opts.zo_pool);
5933 
5934 	(void) rwlock_destroy(&ztest_name_lock);
5935 	(void) _mutex_destroy(&ztest_vdev_lock);
5936 }
5937 
5938 static void
5939 setup_data_fd(void)
5940 {
5941 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
5942 
5943 	ztest_fd_data = mkstemp(ztest_name_data);
5944 	ASSERT3S(ztest_fd_data, >=, 0);
5945 	(void) unlink(ztest_name_data);
5946 }
5947 
5948 
5949 static int
5950 shared_data_size(ztest_shared_hdr_t *hdr)
5951 {
5952 	int size;
5953 
5954 	size = hdr->zh_hdr_size;
5955 	size += hdr->zh_opts_size;
5956 	size += hdr->zh_size;
5957 	size += hdr->zh_stats_size * hdr->zh_stats_count;
5958 	size += hdr->zh_ds_size * hdr->zh_ds_count;
5959 
5960 	return (size);
5961 }
5962 
5963 static void
5964 setup_hdr(void)
5965 {
5966 	int size;
5967 	ztest_shared_hdr_t *hdr;
5968 
5969 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5970 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
5971 	ASSERT(hdr != MAP_FAILED);
5972 
5973 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
5974 
5975 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
5976 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
5977 	hdr->zh_size = sizeof (ztest_shared_t);
5978 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
5979 	hdr->zh_stats_count = ZTEST_FUNCS;
5980 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
5981 	hdr->zh_ds_count = ztest_opts.zo_datasets;
5982 
5983 	size = shared_data_size(hdr);
5984 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
5985 
5986 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
5987 }
5988 
5989 static void
5990 setup_data(void)
5991 {
5992 	int size, offset;
5993 	ztest_shared_hdr_t *hdr;
5994 	uint8_t *buf;
5995 
5996 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
5997 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
5998 	ASSERT(hdr != MAP_FAILED);
5999 
6000 	size = shared_data_size(hdr);
6001 
6002 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6003 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6004 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6005 	ASSERT(hdr != MAP_FAILED);
6006 	buf = (uint8_t *)hdr;
6007 
6008 	offset = hdr->zh_hdr_size;
6009 	ztest_shared_opts = (void *)&buf[offset];
6010 	offset += hdr->zh_opts_size;
6011 	ztest_shared = (void *)&buf[offset];
6012 	offset += hdr->zh_size;
6013 	ztest_shared_callstate = (void *)&buf[offset];
6014 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
6015 	ztest_shared_ds = (void *)&buf[offset];
6016 }
6017 
6018 static boolean_t
6019 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6020 {
6021 	pid_t pid;
6022 	int status;
6023 	char *cmdbuf = NULL;
6024 
6025 	pid = fork();
6026 
6027 	if (cmd == NULL) {
6028 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6029 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6030 		cmd = cmdbuf;
6031 	}
6032 
6033 	if (pid == -1)
6034 		fatal(1, "fork failed");
6035 
6036 	if (pid == 0) {	/* child */
6037 		char *emptyargv[2] = { cmd, NULL };
6038 		char fd_data_str[12];
6039 
6040 		struct rlimit rl = { 1024, 1024 };
6041 		(void) setrlimit(RLIMIT_NOFILE, &rl);
6042 
6043 		(void) close(ztest_fd_rand);
6044 		VERIFY3U(11, >=,
6045 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6046 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6047 
6048 		(void) enable_extended_FILE_stdio(-1, -1);
6049 		if (libpath != NULL)
6050 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6051 		(void) execv(cmd, emptyargv);
6052 		ztest_dump_core = B_FALSE;
6053 		fatal(B_TRUE, "exec failed: %s", cmd);
6054 	}
6055 
6056 	if (cmdbuf != NULL) {
6057 		umem_free(cmdbuf, MAXPATHLEN);
6058 		cmd = NULL;
6059 	}
6060 
6061 	while (waitpid(pid, &status, 0) != pid)
6062 		continue;
6063 	if (statusp != NULL)
6064 		*statusp = status;
6065 
6066 	if (WIFEXITED(status)) {
6067 		if (WEXITSTATUS(status) != 0) {
6068 			(void) fprintf(stderr, "child exited with code %d\n",
6069 			    WEXITSTATUS(status));
6070 			exit(2);
6071 		}
6072 		return (B_FALSE);
6073 	} else if (WIFSIGNALED(status)) {
6074 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6075 			(void) fprintf(stderr, "child died with signal %d\n",
6076 			    WTERMSIG(status));
6077 			exit(3);
6078 		}
6079 		return (B_TRUE);
6080 	} else {
6081 		(void) fprintf(stderr, "something strange happened to child\n");
6082 		exit(4);
6083 		/* NOTREACHED */
6084 	}
6085 }
6086 
6087 static void
6088 ztest_run_init(void)
6089 {
6090 	ztest_shared_t *zs = ztest_shared;
6091 
6092 	ASSERT(ztest_opts.zo_init != 0);
6093 
6094 	/*
6095 	 * Blow away any existing copy of zpool.cache
6096 	 */
6097 	(void) remove(spa_config_path);
6098 
6099 	/*
6100 	 * Create and initialize our storage pool.
6101 	 */
6102 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
6103 		bzero(zs, sizeof (ztest_shared_t));
6104 		if (ztest_opts.zo_verbose >= 3 &&
6105 		    ztest_opts.zo_init != 1) {
6106 			(void) printf("ztest_init(), pass %d\n", i);
6107 		}
6108 		ztest_init(zs);
6109 	}
6110 }
6111 
6112 int
6113 main(int argc, char **argv)
6114 {
6115 	int kills = 0;
6116 	int iters = 0;
6117 	int older = 0;
6118 	int newer = 0;
6119 	ztest_shared_t *zs;
6120 	ztest_info_t *zi;
6121 	ztest_shared_callstate_t *zc;
6122 	char timebuf[100];
6123 	char numbuf[6];
6124 	spa_t *spa;
6125 	char *cmd;
6126 	boolean_t hasalt;
6127 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6128 
6129 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6130 
6131 	dprintf_setup(&argc, argv);
6132 	zfs_deadman_synctime_ms = 300000;
6133 
6134 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6135 	ASSERT3S(ztest_fd_rand, >=, 0);
6136 
6137 	if (!fd_data_str) {
6138 		process_options(argc, argv);
6139 
6140 		setup_data_fd();
6141 		setup_hdr();
6142 		setup_data();
6143 		bcopy(&ztest_opts, ztest_shared_opts,
6144 		    sizeof (*ztest_shared_opts));
6145 	} else {
6146 		ztest_fd_data = atoi(fd_data_str);
6147 		setup_data();
6148 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6149 	}
6150 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6151 
6152 	/* Override location of zpool.cache */
6153 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6154 	    ztest_opts.zo_dir), !=, -1);
6155 
6156 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6157 	    UMEM_NOFAIL);
6158 	zs = ztest_shared;
6159 
6160 	if (fd_data_str) {
6161 		metaslab_gang_bang = ztest_opts.zo_metaslab_gang_bang;
6162 		metaslab_df_alloc_threshold =
6163 		    zs->zs_metaslab_df_alloc_threshold;
6164 
6165 		if (zs->zs_do_init)
6166 			ztest_run_init();
6167 		else
6168 			ztest_run(zs);
6169 		exit(0);
6170 	}
6171 
6172 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6173 
6174 	if (ztest_opts.zo_verbose >= 1) {
6175 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6176 		    " %llu seconds...\n",
6177 		    (u_longlong_t)ztest_opts.zo_vdevs,
6178 		    ztest_opts.zo_datasets,
6179 		    ztest_opts.zo_threads,
6180 		    (u_longlong_t)ztest_opts.zo_time);
6181 	}
6182 
6183 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6184 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6185 
6186 	zs->zs_do_init = B_TRUE;
6187 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6188 		if (ztest_opts.zo_verbose >= 1) {
6189 			(void) printf("Executing older ztest for "
6190 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6191 		}
6192 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6193 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6194 	} else {
6195 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6196 	}
6197 	zs->zs_do_init = B_FALSE;
6198 
6199 	zs->zs_proc_start = gethrtime();
6200 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6201 
6202 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6203 		zi = &ztest_info[f];
6204 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6205 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6206 			zc->zc_next = UINT64_MAX;
6207 		else
6208 			zc->zc_next = zs->zs_proc_start +
6209 			    ztest_random(2 * zi->zi_interval[0] + 1);
6210 	}
6211 
6212 	/*
6213 	 * Run the tests in a loop.  These tests include fault injection
6214 	 * to verify that self-healing data works, and forced crashes
6215 	 * to verify that we never lose on-disk consistency.
6216 	 */
6217 	while (gethrtime() < zs->zs_proc_stop) {
6218 		int status;
6219 		boolean_t killed;
6220 
6221 		/*
6222 		 * Initialize the workload counters for each function.
6223 		 */
6224 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6225 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6226 			zc->zc_count = 0;
6227 			zc->zc_time = 0;
6228 		}
6229 
6230 		/* Set the allocation switch size */
6231 		zs->zs_metaslab_df_alloc_threshold =
6232 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6233 
6234 		if (!hasalt || ztest_random(2) == 0) {
6235 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6236 				(void) printf("Executing newer ztest: %s\n",
6237 				    cmd);
6238 			}
6239 			newer++;
6240 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6241 		} else {
6242 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6243 				(void) printf("Executing older ztest: %s\n",
6244 				    ztest_opts.zo_alt_ztest);
6245 			}
6246 			older++;
6247 			killed = exec_child(ztest_opts.zo_alt_ztest,
6248 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6249 		}
6250 
6251 		if (killed)
6252 			kills++;
6253 		iters++;
6254 
6255 		if (ztest_opts.zo_verbose >= 1) {
6256 			hrtime_t now = gethrtime();
6257 
6258 			now = MIN(now, zs->zs_proc_stop);
6259 			print_time(zs->zs_proc_stop - now, timebuf);
6260 			nicenum(zs->zs_space, numbuf);
6261 
6262 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6263 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6264 			    iters,
6265 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6266 			    (u_longlong_t)zs->zs_enospc_count,
6267 			    100.0 * zs->zs_alloc / zs->zs_space,
6268 			    numbuf,
6269 			    100.0 * (now - zs->zs_proc_start) /
6270 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6271 		}
6272 
6273 		if (ztest_opts.zo_verbose >= 2) {
6274 			(void) printf("\nWorkload summary:\n\n");
6275 			(void) printf("%7s %9s   %s\n",
6276 			    "Calls", "Time", "Function");
6277 			(void) printf("%7s %9s   %s\n",
6278 			    "-----", "----", "--------");
6279 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6280 				Dl_info dli;
6281 
6282 				zi = &ztest_info[f];
6283 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6284 				print_time(zc->zc_time, timebuf);
6285 				(void) dladdr((void *)zi->zi_func, &dli);
6286 				(void) printf("%7llu %9s   %s\n",
6287 				    (u_longlong_t)zc->zc_count, timebuf,
6288 				    dli.dli_sname);
6289 			}
6290 			(void) printf("\n");
6291 		}
6292 
6293 		/*
6294 		 * It's possible that we killed a child during a rename test,
6295 		 * in which case we'll have a 'ztest_tmp' pool lying around
6296 		 * instead of 'ztest'.  Do a blind rename in case this happened.
6297 		 */
6298 		kernel_init(FREAD);
6299 		if (spa_open(ztest_opts.zo_pool, &spa, FTAG) == 0) {
6300 			spa_close(spa, FTAG);
6301 		} else {
6302 			char tmpname[ZFS_MAX_DATASET_NAME_LEN];
6303 			kernel_fini();
6304 			kernel_init(FREAD | FWRITE);
6305 			(void) snprintf(tmpname, sizeof (tmpname), "%s_tmp",
6306 			    ztest_opts.zo_pool);
6307 			(void) spa_rename(tmpname, ztest_opts.zo_pool);
6308 		}
6309 		kernel_fini();
6310 
6311 		ztest_run_zdb(ztest_opts.zo_pool);
6312 	}
6313 
6314 	if (ztest_opts.zo_verbose >= 1) {
6315 		if (hasalt) {
6316 			(void) printf("%d runs of older ztest: %s\n", older,
6317 			    ztest_opts.zo_alt_ztest);
6318 			(void) printf("%d runs of newer ztest: %s\n", newer,
6319 			    cmd);
6320 		}
6321 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6322 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6323 	}
6324 
6325 	umem_free(cmd, MAXNAMELEN);
6326 
6327 	return (0);
6328 }
6329