xref: /illumos-gate/usr/src/cmd/ztest/ztest.c (revision e0f1c0afa46cc84d4b1e40124032a9a87310386e)
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, 2017 by Delphix. All rights reserved.
24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright (c) 2013 Steven Hartland. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  * Copyright 2017 Joyent, Inc.
28  * Copyright 2017 RackTop Systems.
29  */
30 
31 /*
32  * The objective of this program is to provide a DMU/ZAP/SPA stress test
33  * that runs entirely in userland, is easy to use, and easy to extend.
34  *
35  * The overall design of the ztest program is as follows:
36  *
37  * (1) For each major functional area (e.g. adding vdevs to a pool,
38  *     creating and destroying datasets, reading and writing objects, etc)
39  *     we have a simple routine to test that functionality.  These
40  *     individual routines do not have to do anything "stressful".
41  *
42  * (2) We turn these simple functionality tests into a stress test by
43  *     running them all in parallel, with as many threads as desired,
44  *     and spread across as many datasets, objects, and vdevs as desired.
45  *
46  * (3) While all this is happening, we inject faults into the pool to
47  *     verify that self-healing data really works.
48  *
49  * (4) Every time we open a dataset, we change its checksum and compression
50  *     functions.  Thus even individual objects vary from block to block
51  *     in which checksum they use and whether they're compressed.
52  *
53  * (5) To verify that we never lose on-disk consistency after a crash,
54  *     we run the entire test in a child of the main process.
55  *     At random times, the child self-immolates with a SIGKILL.
56  *     This is the software equivalent of pulling the power cord.
57  *     The parent then runs the test again, using the existing
58  *     storage pool, as many times as desired. If backwards compatibility
59  *     testing is enabled ztest will sometimes run the "older" version
60  *     of ztest after a SIGKILL.
61  *
62  * (6) To verify that we don't have future leaks or temporal incursions,
63  *     many of the functional tests record the transaction group number
64  *     as part of their data.  When reading old data, they verify that
65  *     the transaction group number is less than the current, open txg.
66  *     If you add a new test, please do this if applicable.
67  *
68  * When run with no arguments, ztest runs for about five minutes and
69  * produces no output if successful.  To get a little bit of information,
70  * specify -V.  To get more information, specify -VV, and so on.
71  *
72  * To turn this into an overnight stress test, use -T to specify run time.
73  *
74  * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
75  * to increase the pool capacity, fanout, and overall stress level.
76  *
77  * Use the -k option to set the desired frequency of kills.
78  *
79  * When ztest invokes itself it passes all relevant information through a
80  * temporary file which is mmap-ed in the child process. This allows shared
81  * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
82  * stored at offset 0 of this file and contains information on the size and
83  * number of shared structures in the file. The information stored in this file
84  * must remain backwards compatible with older versions of ztest so that
85  * ztest can invoke them during backwards compatibility testing (-B).
86  */
87 
88 #include <sys/zfs_context.h>
89 #include <sys/spa.h>
90 #include <sys/dmu.h>
91 #include <sys/txg.h>
92 #include <sys/dbuf.h>
93 #include <sys/zap.h>
94 #include <sys/dmu_objset.h>
95 #include <sys/poll.h>
96 #include <sys/stat.h>
97 #include <sys/time.h>
98 #include <sys/wait.h>
99 #include <sys/mman.h>
100 #include <sys/resource.h>
101 #include <sys/zio.h>
102 #include <sys/zil.h>
103 #include <sys/zil_impl.h>
104 #include <sys/vdev_impl.h>
105 #include <sys/vdev_file.h>
106 #include <sys/vdev_initialize.h>
107 #include <sys/spa_impl.h>
108 #include <sys/metaslab_impl.h>
109 #include <sys/dsl_prop.h>
110 #include <sys/dsl_dataset.h>
111 #include <sys/dsl_destroy.h>
112 #include <sys/dsl_scan.h>
113 #include <sys/zio_checksum.h>
114 #include <sys/refcount.h>
115 #include <sys/zfeature.h>
116 #include <sys/dsl_userhold.h>
117 #include <sys/abd.h>
118 #include <stdio.h>
119 #include <stdio_ext.h>
120 #include <stdlib.h>
121 #include <unistd.h>
122 #include <signal.h>
123 #include <umem.h>
124 #include <dlfcn.h>
125 #include <ctype.h>
126 #include <math.h>
127 #include <sys/fs/zfs.h>
128 #include <libnvpair.h>
129 #include <libzfs.h>
130 #include <libcmdutils.h>
131 
132 static int ztest_fd_data = -1;
133 static int ztest_fd_rand = -1;
134 
135 typedef struct ztest_shared_hdr {
136 	uint64_t	zh_hdr_size;
137 	uint64_t	zh_opts_size;
138 	uint64_t	zh_size;
139 	uint64_t	zh_stats_size;
140 	uint64_t	zh_stats_count;
141 	uint64_t	zh_ds_size;
142 	uint64_t	zh_ds_count;
143 } ztest_shared_hdr_t;
144 
145 static ztest_shared_hdr_t *ztest_shared_hdr;
146 
147 typedef struct ztest_shared_opts {
148 	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
149 	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
150 	char zo_alt_ztest[MAXNAMELEN];
151 	char zo_alt_libpath[MAXNAMELEN];
152 	uint64_t zo_vdevs;
153 	uint64_t zo_vdevtime;
154 	size_t zo_vdev_size;
155 	int zo_ashift;
156 	int zo_mirrors;
157 	int zo_raidz;
158 	int zo_raidz_parity;
159 	int zo_datasets;
160 	int zo_threads;
161 	uint64_t zo_passtime;
162 	uint64_t zo_killrate;
163 	int zo_verbose;
164 	int zo_init;
165 	uint64_t zo_time;
166 	uint64_t zo_maxloops;
167 	uint64_t zo_metaslab_force_ganging;
168 	int zo_mmp_test;
169 } ztest_shared_opts_t;
170 
171 static const ztest_shared_opts_t ztest_opts_defaults = {
172 	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
173 	.zo_dir = { '/', 't', 'm', 'p', '\0' },
174 	.zo_alt_ztest = { '\0' },
175 	.zo_alt_libpath = { '\0' },
176 	.zo_vdevs = 5,
177 	.zo_ashift = SPA_MINBLOCKSHIFT,
178 	.zo_mirrors = 2,
179 	.zo_raidz = 4,
180 	.zo_raidz_parity = 1,
181 	.zo_vdev_size = SPA_MINDEVSIZE * 4,	/* 256m default size */
182 	.zo_datasets = 7,
183 	.zo_threads = 23,
184 	.zo_passtime = 60,		/* 60 seconds */
185 	.zo_killrate = 70,		/* 70% kill rate */
186 	.zo_verbose = 0,
187 	.zo_mmp_test = 0,
188 	.zo_init = 1,
189 	.zo_time = 300,			/* 5 minutes */
190 	.zo_maxloops = 50,		/* max loops during spa_freeze() */
191 	.zo_metaslab_force_ganging = 32 << 10
192 };
193 
194 extern uint64_t metaslab_force_ganging;
195 extern uint64_t metaslab_df_alloc_threshold;
196 extern uint64_t zfs_deadman_synctime_ms;
197 extern int metaslab_preload_limit;
198 extern boolean_t zfs_compressed_arc_enabled;
199 extern boolean_t zfs_abd_scatter_enabled;
200 extern int dmu_object_alloc_chunk_shift;
201 extern boolean_t zfs_force_some_double_word_sm_entries;
202 extern unsigned long zfs_reconstruct_indirect_damage_fraction;
203 
204 static ztest_shared_opts_t *ztest_shared_opts;
205 static ztest_shared_opts_t ztest_opts;
206 
207 typedef struct ztest_shared_ds {
208 	uint64_t	zd_seq;
209 } ztest_shared_ds_t;
210 
211 static ztest_shared_ds_t *ztest_shared_ds;
212 #define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
213 
214 #define	BT_MAGIC	0x123456789abcdefULL
215 #define	MAXFAULTS() \
216 	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
217 
218 enum ztest_io_type {
219 	ZTEST_IO_WRITE_TAG,
220 	ZTEST_IO_WRITE_PATTERN,
221 	ZTEST_IO_WRITE_ZEROES,
222 	ZTEST_IO_TRUNCATE,
223 	ZTEST_IO_SETATTR,
224 	ZTEST_IO_REWRITE,
225 	ZTEST_IO_TYPES
226 };
227 
228 typedef struct ztest_block_tag {
229 	uint64_t	bt_magic;
230 	uint64_t	bt_objset;
231 	uint64_t	bt_object;
232 	uint64_t	bt_dnodesize;
233 	uint64_t	bt_offset;
234 	uint64_t	bt_gen;
235 	uint64_t	bt_txg;
236 	uint64_t	bt_crtxg;
237 } ztest_block_tag_t;
238 
239 typedef struct bufwad {
240 	uint64_t	bw_index;
241 	uint64_t	bw_txg;
242 	uint64_t	bw_data;
243 } bufwad_t;
244 
245 /*
246  * It would be better to use a rangelock_t per object.  Unfortunately
247  * the rangelock_t is not a drop-in replacement for rl_t, because we
248  * still need to map from object ID to rangelock_t.
249  */
250 typedef enum {
251 	RL_READER,
252 	RL_WRITER,
253 	RL_APPEND
254 } rl_type_t;
255 
256 typedef struct rll {
257 	void		*rll_writer;
258 	int		rll_readers;
259 	kmutex_t	rll_lock;
260 	kcondvar_t	rll_cv;
261 } rll_t;
262 
263 typedef struct rl {
264 	uint64_t	rl_object;
265 	uint64_t	rl_offset;
266 	uint64_t	rl_size;
267 	rll_t		*rl_lock;
268 } rl_t;
269 
270 #define	ZTEST_RANGE_LOCKS	64
271 #define	ZTEST_OBJECT_LOCKS	64
272 
273 /*
274  * Object descriptor.  Used as a template for object lookup/create/remove.
275  */
276 typedef struct ztest_od {
277 	uint64_t	od_dir;
278 	uint64_t	od_object;
279 	dmu_object_type_t od_type;
280 	dmu_object_type_t od_crtype;
281 	uint64_t	od_blocksize;
282 	uint64_t	od_crblocksize;
283 	uint64_t	od_crdnodesize;
284 	uint64_t	od_gen;
285 	uint64_t	od_crgen;
286 	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
287 } ztest_od_t;
288 
289 /*
290  * Per-dataset state.
291  */
292 typedef struct ztest_ds {
293 	ztest_shared_ds_t *zd_shared;
294 	objset_t	*zd_os;
295 	krwlock_t	zd_zilog_lock;
296 	zilog_t		*zd_zilog;
297 	ztest_od_t	*zd_od;		/* debugging aid */
298 	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
299 	kmutex_t	zd_dirobj_lock;
300 	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
301 	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
302 } ztest_ds_t;
303 
304 /*
305  * Per-iteration state.
306  */
307 typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
308 
309 typedef struct ztest_info {
310 	ztest_func_t	*zi_func;	/* test function */
311 	uint64_t	zi_iters;	/* iterations per execution */
312 	uint64_t	*zi_interval;	/* execute every <interval> seconds */
313 } ztest_info_t;
314 
315 typedef struct ztest_shared_callstate {
316 	uint64_t	zc_count;	/* per-pass count */
317 	uint64_t	zc_time;	/* per-pass time */
318 	uint64_t	zc_next;	/* next time to call this function */
319 } ztest_shared_callstate_t;
320 
321 static ztest_shared_callstate_t *ztest_shared_callstate;
322 #define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
323 
324 /*
325  * Note: these aren't static because we want dladdr() to work.
326  */
327 ztest_func_t ztest_dmu_read_write;
328 ztest_func_t ztest_dmu_write_parallel;
329 ztest_func_t ztest_dmu_object_alloc_free;
330 ztest_func_t ztest_dmu_object_next_chunk;
331 ztest_func_t ztest_dmu_commit_callbacks;
332 ztest_func_t ztest_zap;
333 ztest_func_t ztest_zap_parallel;
334 ztest_func_t ztest_zil_commit;
335 ztest_func_t ztest_zil_remount;
336 ztest_func_t ztest_dmu_read_write_zcopy;
337 ztest_func_t ztest_dmu_objset_create_destroy;
338 ztest_func_t ztest_dmu_prealloc;
339 ztest_func_t ztest_fzap;
340 ztest_func_t ztest_dmu_snapshot_create_destroy;
341 ztest_func_t ztest_dsl_prop_get_set;
342 ztest_func_t ztest_spa_prop_get_set;
343 ztest_func_t ztest_spa_create_destroy;
344 ztest_func_t ztest_fault_inject;
345 ztest_func_t ztest_ddt_repair;
346 ztest_func_t ztest_dmu_snapshot_hold;
347 ztest_func_t ztest_mmp_enable_disable;
348 ztest_func_t ztest_scrub;
349 ztest_func_t ztest_dsl_dataset_promote_busy;
350 ztest_func_t ztest_vdev_attach_detach;
351 ztest_func_t ztest_vdev_LUN_growth;
352 ztest_func_t ztest_vdev_add_remove;
353 ztest_func_t ztest_vdev_aux_add_remove;
354 ztest_func_t ztest_split_pool;
355 ztest_func_t ztest_reguid;
356 ztest_func_t ztest_spa_upgrade;
357 ztest_func_t ztest_device_removal;
358 ztest_func_t ztest_remap_blocks;
359 ztest_func_t ztest_spa_checkpoint_create_discard;
360 ztest_func_t ztest_initialize;
361 ztest_func_t ztest_verify_dnode_bt;
362 
363 uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
364 uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
365 uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
366 uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
367 uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
368 
369 ztest_info_t ztest_info[] = {
370 	{ ztest_dmu_read_write,			1,	&zopt_always	},
371 	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
372 	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
373 	{ ztest_dmu_object_next_chunk,		1,	&zopt_sometimes	},
374 	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
375 	{ ztest_zap,				30,	&zopt_always	},
376 	{ ztest_zap_parallel,			100,	&zopt_always	},
377 	{ ztest_split_pool,			1,	&zopt_always	},
378 	{ ztest_zil_commit,			1,	&zopt_incessant	},
379 	{ ztest_zil_remount,			1,	&zopt_sometimes	},
380 	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
381 	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
382 	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
383 	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
384 #if 0
385 	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
386 #endif
387 	{ ztest_fzap,				1,	&zopt_sometimes	},
388 	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
389 	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
390 	{ ztest_fault_inject,			1,	&zopt_sometimes	},
391 	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
392 	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
393 	{ ztest_mmp_enable_disable,		1,	&zopt_sometimes	},
394 	{ ztest_reguid,				1,	&zopt_rarely	},
395 	{ ztest_scrub,				1,	&zopt_rarely	},
396 	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
397 	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
398 	{ ztest_vdev_attach_detach,		1,	&zopt_sometimes	},
399 	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
400 	{ ztest_vdev_add_remove,		1,
401 	    &ztest_opts.zo_vdevtime				},
402 	{ ztest_vdev_aux_add_remove,		1,
403 	    &ztest_opts.zo_vdevtime				},
404 	{ ztest_device_removal,			1,	&zopt_sometimes	},
405 	{ ztest_remap_blocks,			1,	&zopt_sometimes },
406 	{ ztest_spa_checkpoint_create_discard,	1,	&zopt_rarely	},
407 	{ ztest_initialize,			1,	&zopt_sometimes },
408 	{ ztest_verify_dnode_bt,		1,	&zopt_sometimes }
409 };
410 
411 #define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
412 
413 /*
414  * The following struct is used to hold a list of uncalled commit callbacks.
415  * The callbacks are ordered by txg number.
416  */
417 typedef struct ztest_cb_list {
418 	kmutex_t zcl_callbacks_lock;
419 	list_t	zcl_callbacks;
420 } ztest_cb_list_t;
421 
422 /*
423  * Stuff we need to share writably between parent and child.
424  */
425 typedef struct ztest_shared {
426 	boolean_t	zs_do_init;
427 	hrtime_t	zs_proc_start;
428 	hrtime_t	zs_proc_stop;
429 	hrtime_t	zs_thread_start;
430 	hrtime_t	zs_thread_stop;
431 	hrtime_t	zs_thread_kill;
432 	uint64_t	zs_enospc_count;
433 	uint64_t	zs_vdev_next_leaf;
434 	uint64_t	zs_vdev_aux;
435 	uint64_t	zs_alloc;
436 	uint64_t	zs_space;
437 	uint64_t	zs_splits;
438 	uint64_t	zs_mirrors;
439 	uint64_t	zs_metaslab_sz;
440 	uint64_t	zs_metaslab_df_alloc_threshold;
441 	uint64_t	zs_guid;
442 } ztest_shared_t;
443 
444 #define	ID_PARALLEL	-1ULL
445 
446 static char ztest_dev_template[] = "%s/%s.%llua";
447 static char ztest_aux_template[] = "%s/%s.%s.%llu";
448 ztest_shared_t *ztest_shared;
449 
450 static spa_t *ztest_spa = NULL;
451 static ztest_ds_t *ztest_ds;
452 
453 static kmutex_t ztest_vdev_lock;
454 static boolean_t ztest_device_removal_active = B_FALSE;
455 static kmutex_t ztest_checkpoint_lock;
456 
457 /*
458  * The ztest_name_lock protects the pool and dataset namespace used by
459  * the individual tests. To modify the namespace, consumers must grab
460  * this lock as writer. Grabbing the lock as reader will ensure that the
461  * namespace does not change while the lock is held.
462  */
463 static krwlock_t ztest_name_lock;
464 
465 static boolean_t ztest_dump_core = B_TRUE;
466 static boolean_t ztest_exiting;
467 
468 /* Global commit callback list */
469 static ztest_cb_list_t zcl;
470 
471 enum ztest_object {
472 	ZTEST_META_DNODE = 0,
473 	ZTEST_DIROBJ,
474 	ZTEST_OBJECTS
475 };
476 
477 static void usage(boolean_t) __NORETURN;
478 
479 /*
480  * These libumem hooks provide a reasonable set of defaults for the allocator's
481  * debugging facilities.
482  */
483 const char *
484 _umem_debug_init()
485 {
486 	return ("default,verbose"); /* $UMEM_DEBUG setting */
487 }
488 
489 const char *
490 _umem_logging_init(void)
491 {
492 	return ("fail,contents"); /* $UMEM_LOGGING setting */
493 }
494 
495 #define	FATAL_MSG_SZ	1024
496 
497 char *fatal_msg;
498 
499 static void
500 fatal(int do_perror, char *message, ...)
501 {
502 	va_list args;
503 	int save_errno = errno;
504 	char buf[FATAL_MSG_SZ];
505 
506 	(void) fflush(stdout);
507 
508 	va_start(args, message);
509 	(void) sprintf(buf, "ztest: ");
510 	/* LINTED */
511 	(void) vsprintf(buf + strlen(buf), message, args);
512 	va_end(args);
513 	if (do_perror) {
514 		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
515 		    ": %s", strerror(save_errno));
516 	}
517 	(void) fprintf(stderr, "%s\n", buf);
518 	fatal_msg = buf;			/* to ease debugging */
519 	if (ztest_dump_core)
520 		abort();
521 	exit(3);
522 }
523 
524 static int
525 str2shift(const char *buf)
526 {
527 	const char *ends = "BKMGTPEZ";
528 	int i;
529 
530 	if (buf[0] == '\0')
531 		return (0);
532 	for (i = 0; i < strlen(ends); i++) {
533 		if (toupper(buf[0]) == ends[i])
534 			break;
535 	}
536 	if (i == strlen(ends)) {
537 		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
538 		    buf);
539 		usage(B_FALSE);
540 	}
541 	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
542 		return (10*i);
543 	}
544 	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
545 	usage(B_FALSE);
546 	/* NOTREACHED */
547 }
548 
549 static uint64_t
550 nicenumtoull(const char *buf)
551 {
552 	char *end;
553 	uint64_t val;
554 
555 	val = strtoull(buf, &end, 0);
556 	if (end == buf) {
557 		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
558 		usage(B_FALSE);
559 	} else if (end[0] == '.') {
560 		double fval = strtod(buf, &end);
561 		fval *= pow(2, str2shift(end));
562 		if (fval > UINT64_MAX) {
563 			(void) fprintf(stderr, "ztest: value too large: %s\n",
564 			    buf);
565 			usage(B_FALSE);
566 		}
567 		val = (uint64_t)fval;
568 	} else {
569 		int shift = str2shift(end);
570 		if (shift >= 64 || (val << shift) >> shift != val) {
571 			(void) fprintf(stderr, "ztest: value too large: %s\n",
572 			    buf);
573 			usage(B_FALSE);
574 		}
575 		val <<= shift;
576 	}
577 	return (val);
578 }
579 
580 static void
581 usage(boolean_t requested)
582 {
583 	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
584 
585 	char nice_vdev_size[NN_NUMBUF_SZ];
586 	char nice_force_ganging[NN_NUMBUF_SZ];
587 	FILE *fp = requested ? stdout : stderr;
588 
589 	nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
590 	nicenum(zo->zo_metaslab_force_ganging, nice_force_ganging,
591 	    sizeof (nice_force_ganging));
592 
593 	(void) fprintf(fp, "Usage: %s\n"
594 	    "\t[-v vdevs (default: %llu)]\n"
595 	    "\t[-s size_of_each_vdev (default: %s)]\n"
596 	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
597 	    "\t[-m mirror_copies (default: %d)]\n"
598 	    "\t[-r raidz_disks (default: %d)]\n"
599 	    "\t[-R raidz_parity (default: %d)]\n"
600 	    "\t[-d datasets (default: %d)]\n"
601 	    "\t[-t threads (default: %d)]\n"
602 	    "\t[-g gang_block_threshold (default: %s)]\n"
603 	    "\t[-i init_count (default: %d)] initialize pool i times\n"
604 	    "\t[-k kill_percentage (default: %llu%%)]\n"
605 	    "\t[-p pool_name (default: %s)]\n"
606 	    "\t[-f dir (default: %s)] file directory for vdev files\n"
607 	    "\t[-M] Multi-host simulate pool imported on remote host\n"
608 	    "\t[-V] verbose (use multiple times for ever more blather)\n"
609 	    "\t[-E] use existing pool instead of creating new one\n"
610 	    "\t[-T time (default: %llu sec)] total run time\n"
611 	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
612 	    "\t[-P passtime (default: %llu sec)] time per pass\n"
613 	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
614 	    "\t[-o variable=value] ... set global variable to an unsigned\n"
615 	    "\t    32-bit integer value\n"
616 	    "\t[-h] (print help)\n"
617 	    "",
618 	    zo->zo_pool,
619 	    (u_longlong_t)zo->zo_vdevs,			/* -v */
620 	    nice_vdev_size,				/* -s */
621 	    zo->zo_ashift,				/* -a */
622 	    zo->zo_mirrors,				/* -m */
623 	    zo->zo_raidz,				/* -r */
624 	    zo->zo_raidz_parity,			/* -R */
625 	    zo->zo_datasets,				/* -d */
626 	    zo->zo_threads,				/* -t */
627 	    nice_force_ganging,				/* -g */
628 	    zo->zo_init,				/* -i */
629 	    (u_longlong_t)zo->zo_killrate,		/* -k */
630 	    zo->zo_pool,				/* -p */
631 	    zo->zo_dir,					/* -f */
632 	    (u_longlong_t)zo->zo_time,			/* -T */
633 	    (u_longlong_t)zo->zo_maxloops,		/* -F */
634 	    (u_longlong_t)zo->zo_passtime);
635 	exit(requested ? 0 : 1);
636 }
637 
638 static void
639 process_options(int argc, char **argv)
640 {
641 	char *path;
642 	ztest_shared_opts_t *zo = &ztest_opts;
643 
644 	int opt;
645 	uint64_t value;
646 	char altdir[MAXNAMELEN] = { 0 };
647 
648 	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
649 
650 	while ((opt = getopt(argc, argv,
651 	    "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:o:")) != EOF) {
652 		value = 0;
653 		switch (opt) {
654 		case 'v':
655 		case 's':
656 		case 'a':
657 		case 'm':
658 		case 'r':
659 		case 'R':
660 		case 'd':
661 		case 't':
662 		case 'g':
663 		case 'i':
664 		case 'k':
665 		case 'T':
666 		case 'P':
667 		case 'F':
668 			value = nicenumtoull(optarg);
669 		}
670 		switch (opt) {
671 		case 'v':
672 			zo->zo_vdevs = value;
673 			break;
674 		case 's':
675 			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
676 			break;
677 		case 'a':
678 			zo->zo_ashift = value;
679 			break;
680 		case 'm':
681 			zo->zo_mirrors = value;
682 			break;
683 		case 'r':
684 			zo->zo_raidz = MAX(1, value);
685 			break;
686 		case 'R':
687 			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
688 			break;
689 		case 'd':
690 			zo->zo_datasets = MAX(1, value);
691 			break;
692 		case 't':
693 			zo->zo_threads = MAX(1, value);
694 			break;
695 		case 'g':
696 			zo->zo_metaslab_force_ganging =
697 			    MAX(SPA_MINBLOCKSIZE << 1, value);
698 			break;
699 		case 'i':
700 			zo->zo_init = value;
701 			break;
702 		case 'k':
703 			zo->zo_killrate = value;
704 			break;
705 		case 'p':
706 			(void) strlcpy(zo->zo_pool, optarg,
707 			    sizeof (zo->zo_pool));
708 			break;
709 		case 'f':
710 			path = realpath(optarg, NULL);
711 			if (path == NULL) {
712 				(void) fprintf(stderr, "error: %s: %s\n",
713 				    optarg, strerror(errno));
714 				usage(B_FALSE);
715 			} else {
716 				(void) strlcpy(zo->zo_dir, path,
717 				    sizeof (zo->zo_dir));
718 			}
719 			break;
720 		case 'M':
721 			zo->zo_mmp_test = 1;
722 			break;
723 		case 'V':
724 			zo->zo_verbose++;
725 			break;
726 		case 'E':
727 			zo->zo_init = 0;
728 			break;
729 		case 'T':
730 			zo->zo_time = value;
731 			break;
732 		case 'P':
733 			zo->zo_passtime = MAX(1, value);
734 			break;
735 		case 'F':
736 			zo->zo_maxloops = MAX(1, value);
737 			break;
738 		case 'B':
739 			(void) strlcpy(altdir, optarg, sizeof (altdir));
740 			break;
741 		case 'o':
742 			if (set_global_var(optarg) != 0)
743 				usage(B_FALSE);
744 			break;
745 		case 'h':
746 			usage(B_TRUE);
747 			break;
748 		case '?':
749 		default:
750 			usage(B_FALSE);
751 			break;
752 		}
753 	}
754 
755 	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
756 
757 	zo->zo_vdevtime =
758 	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
759 	    UINT64_MAX >> 2);
760 
761 	if (strlen(altdir) > 0) {
762 		char *cmd;
763 		char *realaltdir;
764 		char *bin;
765 		char *ztest;
766 		char *isa;
767 		int isalen;
768 
769 		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
770 		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
771 
772 		VERIFY(NULL != realpath(getexecname(), cmd));
773 		if (0 != access(altdir, F_OK)) {
774 			ztest_dump_core = B_FALSE;
775 			fatal(B_TRUE, "invalid alternate ztest path: %s",
776 			    altdir);
777 		}
778 		VERIFY(NULL != realpath(altdir, realaltdir));
779 
780 		/*
781 		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
782 		 * We want to extract <isa> to determine if we should use
783 		 * 32 or 64 bit binaries.
784 		 */
785 		bin = strstr(cmd, "/usr/bin/");
786 		ztest = strstr(bin, "/ztest");
787 		isa = bin + 9;
788 		isalen = ztest - isa;
789 		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
790 		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
791 		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
792 		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
793 
794 		if (0 != access(zo->zo_alt_ztest, X_OK)) {
795 			ztest_dump_core = B_FALSE;
796 			fatal(B_TRUE, "invalid alternate ztest: %s",
797 			    zo->zo_alt_ztest);
798 		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
799 			ztest_dump_core = B_FALSE;
800 			fatal(B_TRUE, "invalid alternate lib directory %s",
801 			    zo->zo_alt_libpath);
802 		}
803 
804 		umem_free(cmd, MAXPATHLEN);
805 		umem_free(realaltdir, MAXPATHLEN);
806 	}
807 }
808 
809 static void
810 ztest_kill(ztest_shared_t *zs)
811 {
812 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
813 	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
814 
815 	/*
816 	 * Before we kill off ztest, make sure that the config is updated.
817 	 * See comment above spa_write_cachefile().
818 	 */
819 	mutex_enter(&spa_namespace_lock);
820 	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
821 	mutex_exit(&spa_namespace_lock);
822 
823 	zfs_dbgmsg_print(FTAG);
824 	(void) kill(getpid(), SIGKILL);
825 }
826 
827 static uint64_t
828 ztest_random(uint64_t range)
829 {
830 	uint64_t r;
831 
832 	ASSERT3S(ztest_fd_rand, >=, 0);
833 
834 	if (range == 0)
835 		return (0);
836 
837 	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
838 		fatal(1, "short read from /dev/urandom");
839 
840 	return (r % range);
841 }
842 
843 /* ARGSUSED */
844 static void
845 ztest_record_enospc(const char *s)
846 {
847 	ztest_shared->zs_enospc_count++;
848 }
849 
850 static uint64_t
851 ztest_get_ashift(void)
852 {
853 	if (ztest_opts.zo_ashift == 0)
854 		return (SPA_MINBLOCKSHIFT + ztest_random(5));
855 	return (ztest_opts.zo_ashift);
856 }
857 
858 static nvlist_t *
859 make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
860 {
861 	char pathbuf[MAXPATHLEN];
862 	uint64_t vdev;
863 	nvlist_t *file;
864 
865 	if (ashift == 0)
866 		ashift = ztest_get_ashift();
867 
868 	if (path == NULL) {
869 		path = pathbuf;
870 
871 		if (aux != NULL) {
872 			vdev = ztest_shared->zs_vdev_aux;
873 			(void) snprintf(path, sizeof (pathbuf),
874 			    ztest_aux_template, ztest_opts.zo_dir,
875 			    pool == NULL ? ztest_opts.zo_pool : pool,
876 			    aux, vdev);
877 		} else {
878 			vdev = ztest_shared->zs_vdev_next_leaf++;
879 			(void) snprintf(path, sizeof (pathbuf),
880 			    ztest_dev_template, ztest_opts.zo_dir,
881 			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
882 		}
883 	}
884 
885 	if (size != 0) {
886 		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
887 		if (fd == -1)
888 			fatal(1, "can't open %s", path);
889 		if (ftruncate(fd, size) != 0)
890 			fatal(1, "can't ftruncate %s", path);
891 		(void) close(fd);
892 	}
893 
894 	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
895 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
896 	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
897 	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
898 
899 	return (file);
900 }
901 
902 static nvlist_t *
903 make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
904     uint64_t ashift, int r)
905 {
906 	nvlist_t *raidz, **child;
907 	int c;
908 
909 	if (r < 2)
910 		return (make_vdev_file(path, aux, pool, size, ashift));
911 	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
912 
913 	for (c = 0; c < r; c++)
914 		child[c] = make_vdev_file(path, aux, pool, size, ashift);
915 
916 	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
917 	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
918 	    VDEV_TYPE_RAIDZ) == 0);
919 	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
920 	    ztest_opts.zo_raidz_parity) == 0);
921 	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
922 	    child, r) == 0);
923 
924 	for (c = 0; c < r; c++)
925 		nvlist_free(child[c]);
926 
927 	umem_free(child, r * sizeof (nvlist_t *));
928 
929 	return (raidz);
930 }
931 
932 static nvlist_t *
933 make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
934     uint64_t ashift, int r, int m)
935 {
936 	nvlist_t *mirror, **child;
937 	int c;
938 
939 	if (m < 1)
940 		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
941 
942 	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
943 
944 	for (c = 0; c < m; c++)
945 		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
946 
947 	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
948 	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
949 	    VDEV_TYPE_MIRROR) == 0);
950 	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
951 	    child, m) == 0);
952 
953 	for (c = 0; c < m; c++)
954 		nvlist_free(child[c]);
955 
956 	umem_free(child, m * sizeof (nvlist_t *));
957 
958 	return (mirror);
959 }
960 
961 static nvlist_t *
962 make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
963     int log, int r, int m, int t)
964 {
965 	nvlist_t *root, **child;
966 	int c;
967 
968 	ASSERT(t > 0);
969 
970 	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
971 
972 	for (c = 0; c < t; c++) {
973 		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
974 		    r, m);
975 		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
976 		    log) == 0);
977 	}
978 
979 	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
980 	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
981 	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
982 	    child, t) == 0);
983 
984 	for (c = 0; c < t; c++)
985 		nvlist_free(child[c]);
986 
987 	umem_free(child, t * sizeof (nvlist_t *));
988 
989 	return (root);
990 }
991 
992 /*
993  * Find a random spa version. Returns back a random spa version in the
994  * range [initial_version, SPA_VERSION_FEATURES].
995  */
996 static uint64_t
997 ztest_random_spa_version(uint64_t initial_version)
998 {
999 	uint64_t version = initial_version;
1000 
1001 	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1002 		version = version +
1003 		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1004 	}
1005 
1006 	if (version > SPA_VERSION_BEFORE_FEATURES)
1007 		version = SPA_VERSION_FEATURES;
1008 
1009 	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1010 	return (version);
1011 }
1012 
1013 static int
1014 ztest_random_blocksize(void)
1015 {
1016 	uint64_t block_shift;
1017 	/*
1018 	 * Choose a block size >= the ashift.
1019 	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1020 	 */
1021 	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1022 	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1023 		maxbs = 20;
1024 	block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1025 	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1026 }
1027 
1028 static int
1029 ztest_random_dnodesize(void)
1030 {
1031 	int slots;
1032 	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1033 
1034 	if (max_slots == DNODE_MIN_SLOTS)
1035 		return (DNODE_MIN_SIZE);
1036 
1037 	/*
1038 	 * Weight the random distribution more heavily toward smaller
1039 	 * dnode sizes since that is more likely to reflect real-world
1040 	 * usage.
1041 	 */
1042 	ASSERT3U(max_slots, >, 4);
1043 	switch (ztest_random(10)) {
1044 	case 0:
1045 		slots = 5 + ztest_random(max_slots - 4);
1046 		break;
1047 	case 1 ... 4:
1048 		slots = 2 + ztest_random(3);
1049 		break;
1050 	default:
1051 		slots = 1;
1052 		break;
1053 	}
1054 
1055 	return (slots << DNODE_SHIFT);
1056 }
1057 
1058 static int
1059 ztest_random_ibshift(void)
1060 {
1061 	return (DN_MIN_INDBLKSHIFT +
1062 	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1063 }
1064 
1065 static uint64_t
1066 ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1067 {
1068 	uint64_t top;
1069 	vdev_t *rvd = spa->spa_root_vdev;
1070 	vdev_t *tvd;
1071 
1072 	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1073 
1074 	do {
1075 		top = ztest_random(rvd->vdev_children);
1076 		tvd = rvd->vdev_child[top];
1077 	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1078 	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1079 
1080 	return (top);
1081 }
1082 
1083 static uint64_t
1084 ztest_random_dsl_prop(zfs_prop_t prop)
1085 {
1086 	uint64_t value;
1087 
1088 	do {
1089 		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1090 	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1091 
1092 	return (value);
1093 }
1094 
1095 static int
1096 ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1097     boolean_t inherit)
1098 {
1099 	const char *propname = zfs_prop_to_name(prop);
1100 	const char *valname;
1101 	char setpoint[MAXPATHLEN];
1102 	uint64_t curval;
1103 	int error;
1104 
1105 	error = dsl_prop_set_int(osname, propname,
1106 	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1107 
1108 	if (error == ENOSPC) {
1109 		ztest_record_enospc(FTAG);
1110 		return (error);
1111 	}
1112 	ASSERT0(error);
1113 
1114 	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1115 
1116 	if (ztest_opts.zo_verbose >= 6) {
1117 		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1118 		(void) printf("%s %s = %s at '%s'\n",
1119 		    osname, propname, valname, setpoint);
1120 	}
1121 
1122 	return (error);
1123 }
1124 
1125 static int
1126 ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1127 {
1128 	spa_t *spa = ztest_spa;
1129 	nvlist_t *props = NULL;
1130 	int error;
1131 
1132 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1133 	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1134 
1135 	error = spa_prop_set(spa, props);
1136 
1137 	nvlist_free(props);
1138 
1139 	if (error == ENOSPC) {
1140 		ztest_record_enospc(FTAG);
1141 		return (error);
1142 	}
1143 	ASSERT0(error);
1144 
1145 	return (error);
1146 }
1147 
1148 static void
1149 ztest_rll_init(rll_t *rll)
1150 {
1151 	rll->rll_writer = NULL;
1152 	rll->rll_readers = 0;
1153 	mutex_init(&rll->rll_lock, NULL, USYNC_THREAD, NULL);
1154 	cv_init(&rll->rll_cv, NULL, USYNC_THREAD, NULL);
1155 }
1156 
1157 static void
1158 ztest_rll_destroy(rll_t *rll)
1159 {
1160 	ASSERT(rll->rll_writer == NULL);
1161 	ASSERT(rll->rll_readers == 0);
1162 	mutex_destroy(&rll->rll_lock);
1163 	cv_destroy(&rll->rll_cv);
1164 }
1165 
1166 static void
1167 ztest_rll_lock(rll_t *rll, rl_type_t type)
1168 {
1169 	mutex_enter(&rll->rll_lock);
1170 
1171 	if (type == RL_READER) {
1172 		while (rll->rll_writer != NULL)
1173 			cv_wait(&rll->rll_cv, &rll->rll_lock);
1174 		rll->rll_readers++;
1175 	} else {
1176 		while (rll->rll_writer != NULL || rll->rll_readers)
1177 			cv_wait(&rll->rll_cv, &rll->rll_lock);
1178 		rll->rll_writer = curthread;
1179 	}
1180 
1181 	mutex_exit(&rll->rll_lock);
1182 }
1183 
1184 static void
1185 ztest_rll_unlock(rll_t *rll)
1186 {
1187 	mutex_enter(&rll->rll_lock);
1188 
1189 	if (rll->rll_writer) {
1190 		ASSERT(rll->rll_readers == 0);
1191 		rll->rll_writer = NULL;
1192 	} else {
1193 		ASSERT(rll->rll_readers != 0);
1194 		ASSERT(rll->rll_writer == NULL);
1195 		rll->rll_readers--;
1196 	}
1197 
1198 	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1199 		cv_broadcast(&rll->rll_cv);
1200 
1201 	mutex_exit(&rll->rll_lock);
1202 }
1203 
1204 static void
1205 ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1206 {
1207 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1208 
1209 	ztest_rll_lock(rll, type);
1210 }
1211 
1212 static void
1213 ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1214 {
1215 	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1216 
1217 	ztest_rll_unlock(rll);
1218 }
1219 
1220 static rl_t *
1221 ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1222     uint64_t size, rl_type_t type)
1223 {
1224 	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1225 	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1226 	rl_t *rl;
1227 
1228 	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1229 	rl->rl_object = object;
1230 	rl->rl_offset = offset;
1231 	rl->rl_size = size;
1232 	rl->rl_lock = rll;
1233 
1234 	ztest_rll_lock(rll, type);
1235 
1236 	return (rl);
1237 }
1238 
1239 static void
1240 ztest_range_unlock(rl_t *rl)
1241 {
1242 	rll_t *rll = rl->rl_lock;
1243 
1244 	ztest_rll_unlock(rll);
1245 
1246 	umem_free(rl, sizeof (*rl));
1247 }
1248 
1249 static void
1250 ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1251 {
1252 	zd->zd_os = os;
1253 	zd->zd_zilog = dmu_objset_zil(os);
1254 	zd->zd_shared = szd;
1255 	dmu_objset_name(os, zd->zd_name);
1256 
1257 	if (zd->zd_shared != NULL)
1258 		zd->zd_shared->zd_seq = 0;
1259 
1260 	rw_init(&zd->zd_zilog_lock, NULL, USYNC_THREAD, NULL);
1261 	mutex_init(&zd->zd_dirobj_lock, NULL, USYNC_THREAD, NULL);
1262 
1263 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1264 		ztest_rll_init(&zd->zd_object_lock[l]);
1265 
1266 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1267 		ztest_rll_init(&zd->zd_range_lock[l]);
1268 }
1269 
1270 static void
1271 ztest_zd_fini(ztest_ds_t *zd)
1272 {
1273 	mutex_destroy(&zd->zd_dirobj_lock);
1274 
1275 	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1276 		ztest_rll_destroy(&zd->zd_object_lock[l]);
1277 
1278 	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1279 		ztest_rll_destroy(&zd->zd_range_lock[l]);
1280 }
1281 
1282 #define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1283 
1284 static uint64_t
1285 ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1286 {
1287 	uint64_t txg;
1288 	int error;
1289 
1290 	/*
1291 	 * Attempt to assign tx to some transaction group.
1292 	 */
1293 	error = dmu_tx_assign(tx, txg_how);
1294 	if (error) {
1295 		if (error == ERESTART) {
1296 			ASSERT(txg_how == TXG_NOWAIT);
1297 			dmu_tx_wait(tx);
1298 		} else {
1299 			ASSERT3U(error, ==, ENOSPC);
1300 			ztest_record_enospc(tag);
1301 		}
1302 		dmu_tx_abort(tx);
1303 		return (0);
1304 	}
1305 	txg = dmu_tx_get_txg(tx);
1306 	ASSERT(txg != 0);
1307 	return (txg);
1308 }
1309 
1310 static void
1311 ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1312 {
1313 	uint64_t *ip = buf;
1314 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1315 
1316 	while (ip < ip_end)
1317 		*ip++ = value;
1318 }
1319 
1320 static boolean_t
1321 ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1322 {
1323 	uint64_t *ip = buf;
1324 	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1325 	uint64_t diff = 0;
1326 
1327 	while (ip < ip_end)
1328 		diff |= (value - *ip++);
1329 
1330 	return (diff == 0);
1331 }
1332 
1333 static void
1334 ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1335     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1336     uint64_t crtxg)
1337 {
1338 	bt->bt_magic = BT_MAGIC;
1339 	bt->bt_objset = dmu_objset_id(os);
1340 	bt->bt_object = object;
1341 	bt->bt_dnodesize = dnodesize;
1342 	bt->bt_offset = offset;
1343 	bt->bt_gen = gen;
1344 	bt->bt_txg = txg;
1345 	bt->bt_crtxg = crtxg;
1346 }
1347 
1348 static void
1349 ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1350     uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1351     uint64_t crtxg)
1352 {
1353 	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1354 	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1355 	ASSERT3U(bt->bt_object, ==, object);
1356 	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1357 	ASSERT3U(bt->bt_offset, ==, offset);
1358 	ASSERT3U(bt->bt_gen, <=, gen);
1359 	ASSERT3U(bt->bt_txg, <=, txg);
1360 	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1361 }
1362 
1363 static ztest_block_tag_t *
1364 ztest_bt_bonus(dmu_buf_t *db)
1365 {
1366 	dmu_object_info_t doi;
1367 	ztest_block_tag_t *bt;
1368 
1369 	dmu_object_info_from_db(db, &doi);
1370 	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1371 	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1372 	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1373 
1374 	return (bt);
1375 }
1376 
1377 /*
1378  * Generate a token to fill up unused bonus buffer space.  Try to make
1379  * it unique to the object, generation, and offset to verify that data
1380  * is not getting overwritten by data from other dnodes.
1381  */
1382 #define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset)	\
1383 	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1384 
1385 /*
1386  * Fill up the unused bonus buffer region before the block tag with a
1387  * verifiable pattern. Filling the whole bonus area with non-zero data
1388  * helps ensure that all dnode traversal code properly skips the
1389  * interior regions of large dnodes.
1390  */
1391 void
1392 ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1393     objset_t *os, uint64_t gen)
1394 {
1395 	uint64_t *bonusp;
1396 
1397 	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1398 
1399 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1400 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1401 		    gen, bonusp - (uint64_t *)db->db_data);
1402 		*bonusp = token;
1403 	}
1404 }
1405 
1406 /*
1407  * Verify that the unused area of a bonus buffer is filled with the
1408  * expected tokens.
1409  */
1410 void
1411 ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1412     objset_t *os, uint64_t gen)
1413 {
1414 	uint64_t *bonusp;
1415 
1416 	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1417 		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1418 		    gen, bonusp - (uint64_t *)db->db_data);
1419 		VERIFY3U(*bonusp, ==, token);
1420 	}
1421 }
1422 
1423 /*
1424  * ZIL logging ops
1425  */
1426 
1427 #define	lrz_type	lr_mode
1428 #define	lrz_blocksize	lr_uid
1429 #define	lrz_ibshift	lr_gid
1430 #define	lrz_bonustype	lr_rdev
1431 #define	lrz_dnodesize	lr_crtime[1]
1432 
1433 static void
1434 ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1435 {
1436 	char *name = (void *)(lr + 1);		/* name follows lr */
1437 	size_t namesize = strlen(name) + 1;
1438 	itx_t *itx;
1439 
1440 	if (zil_replaying(zd->zd_zilog, tx))
1441 		return;
1442 
1443 	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1444 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1445 	    sizeof (*lr) + namesize - sizeof (lr_t));
1446 
1447 	zil_itx_assign(zd->zd_zilog, itx, tx);
1448 }
1449 
1450 static void
1451 ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1452 {
1453 	char *name = (void *)(lr + 1);		/* name follows lr */
1454 	size_t namesize = strlen(name) + 1;
1455 	itx_t *itx;
1456 
1457 	if (zil_replaying(zd->zd_zilog, tx))
1458 		return;
1459 
1460 	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1461 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1462 	    sizeof (*lr) + namesize - sizeof (lr_t));
1463 
1464 	itx->itx_oid = object;
1465 	zil_itx_assign(zd->zd_zilog, itx, tx);
1466 }
1467 
1468 static void
1469 ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1470 {
1471 	itx_t *itx;
1472 	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1473 
1474 	if (zil_replaying(zd->zd_zilog, tx))
1475 		return;
1476 
1477 	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1478 		write_state = WR_INDIRECT;
1479 
1480 	itx = zil_itx_create(TX_WRITE,
1481 	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1482 
1483 	if (write_state == WR_COPIED &&
1484 	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1485 	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1486 		zil_itx_destroy(itx);
1487 		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1488 		write_state = WR_NEED_COPY;
1489 	}
1490 	itx->itx_private = zd;
1491 	itx->itx_wr_state = write_state;
1492 	itx->itx_sync = (ztest_random(8) == 0);
1493 
1494 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1495 	    sizeof (*lr) - sizeof (lr_t));
1496 
1497 	zil_itx_assign(zd->zd_zilog, itx, tx);
1498 }
1499 
1500 static void
1501 ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1502 {
1503 	itx_t *itx;
1504 
1505 	if (zil_replaying(zd->zd_zilog, tx))
1506 		return;
1507 
1508 	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1509 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1510 	    sizeof (*lr) - sizeof (lr_t));
1511 
1512 	itx->itx_sync = B_FALSE;
1513 	zil_itx_assign(zd->zd_zilog, itx, tx);
1514 }
1515 
1516 static void
1517 ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1518 {
1519 	itx_t *itx;
1520 
1521 	if (zil_replaying(zd->zd_zilog, tx))
1522 		return;
1523 
1524 	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1525 	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1526 	    sizeof (*lr) - sizeof (lr_t));
1527 
1528 	itx->itx_sync = B_FALSE;
1529 	zil_itx_assign(zd->zd_zilog, itx, tx);
1530 }
1531 
1532 /*
1533  * ZIL replay ops
1534  */
1535 static int
1536 ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1537 {
1538 	ztest_ds_t *zd = arg1;
1539 	lr_create_t *lr = arg2;
1540 	char *name = (void *)(lr + 1);		/* name follows lr */
1541 	objset_t *os = zd->zd_os;
1542 	ztest_block_tag_t *bbt;
1543 	dmu_buf_t *db;
1544 	dmu_tx_t *tx;
1545 	uint64_t txg;
1546 	int error = 0;
1547 	int bonuslen;
1548 
1549 	if (byteswap)
1550 		byteswap_uint64_array(lr, sizeof (*lr));
1551 
1552 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1553 	ASSERT(name[0] != '\0');
1554 
1555 	tx = dmu_tx_create(os);
1556 
1557 	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1558 
1559 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1560 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1561 	} else {
1562 		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1563 	}
1564 
1565 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1566 	if (txg == 0)
1567 		return (ENOSPC);
1568 
1569 	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1570 	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1571 
1572 	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1573 		if (lr->lr_foid == 0) {
1574 			lr->lr_foid = zap_create_dnsize(os,
1575 			    lr->lrz_type, lr->lrz_bonustype,
1576 			    bonuslen, lr->lrz_dnodesize, tx);
1577 		} else {
1578 			error = zap_create_claim_dnsize(os, lr->lr_foid,
1579 			    lr->lrz_type, lr->lrz_bonustype,
1580 			    bonuslen, lr->lrz_dnodesize, tx);
1581 		}
1582 	} else {
1583 		if (lr->lr_foid == 0) {
1584 			lr->lr_foid = dmu_object_alloc_dnsize(os,
1585 			    lr->lrz_type, 0, lr->lrz_bonustype,
1586 			    bonuslen, lr->lrz_dnodesize, tx);
1587 		} else {
1588 			error = dmu_object_claim_dnsize(os, lr->lr_foid,
1589 			    lr->lrz_type, 0, lr->lrz_bonustype,
1590 			    bonuslen, lr->lrz_dnodesize, tx);
1591 		}
1592 	}
1593 
1594 	if (error) {
1595 		ASSERT3U(error, ==, EEXIST);
1596 		ASSERT(zd->zd_zilog->zl_replay);
1597 		dmu_tx_commit(tx);
1598 		return (error);
1599 	}
1600 
1601 	ASSERT(lr->lr_foid != 0);
1602 
1603 	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1604 		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1605 		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1606 
1607 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1608 	bbt = ztest_bt_bonus(db);
1609 	dmu_buf_will_dirty(db, tx);
1610 	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
1611 	    lr->lr_gen, txg, txg);
1612 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
1613 	dmu_buf_rele(db, FTAG);
1614 
1615 	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1616 	    &lr->lr_foid, tx));
1617 
1618 	(void) ztest_log_create(zd, tx, lr);
1619 
1620 	dmu_tx_commit(tx);
1621 
1622 	return (0);
1623 }
1624 
1625 static int
1626 ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1627 {
1628 	ztest_ds_t *zd = arg1;
1629 	lr_remove_t *lr = arg2;
1630 	char *name = (void *)(lr + 1);		/* name follows lr */
1631 	objset_t *os = zd->zd_os;
1632 	dmu_object_info_t doi;
1633 	dmu_tx_t *tx;
1634 	uint64_t object, txg;
1635 
1636 	if (byteswap)
1637 		byteswap_uint64_array(lr, sizeof (*lr));
1638 
1639 	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1640 	ASSERT(name[0] != '\0');
1641 
1642 	VERIFY3U(0, ==,
1643 	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1644 	ASSERT(object != 0);
1645 
1646 	ztest_object_lock(zd, object, RL_WRITER);
1647 
1648 	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1649 
1650 	tx = dmu_tx_create(os);
1651 
1652 	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1653 	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1654 
1655 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1656 	if (txg == 0) {
1657 		ztest_object_unlock(zd, object);
1658 		return (ENOSPC);
1659 	}
1660 
1661 	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1662 		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1663 	} else {
1664 		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1665 	}
1666 
1667 	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1668 
1669 	(void) ztest_log_remove(zd, tx, lr, object);
1670 
1671 	dmu_tx_commit(tx);
1672 
1673 	ztest_object_unlock(zd, object);
1674 
1675 	return (0);
1676 }
1677 
1678 static int
1679 ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1680 {
1681 	ztest_ds_t *zd = arg1;
1682 	lr_write_t *lr = arg2;
1683 	objset_t *os = zd->zd_os;
1684 	void *data = lr + 1;			/* data follows lr */
1685 	uint64_t offset, length;
1686 	ztest_block_tag_t *bt = data;
1687 	ztest_block_tag_t *bbt;
1688 	uint64_t gen, txg, lrtxg, crtxg;
1689 	dmu_object_info_t doi;
1690 	dmu_tx_t *tx;
1691 	dmu_buf_t *db;
1692 	arc_buf_t *abuf = NULL;
1693 	rl_t *rl;
1694 
1695 	if (byteswap)
1696 		byteswap_uint64_array(lr, sizeof (*lr));
1697 
1698 	offset = lr->lr_offset;
1699 	length = lr->lr_length;
1700 
1701 	/* If it's a dmu_sync() block, write the whole block */
1702 	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1703 		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1704 		if (length < blocksize) {
1705 			offset -= offset % blocksize;
1706 			length = blocksize;
1707 		}
1708 	}
1709 
1710 	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1711 		byteswap_uint64_array(bt, sizeof (*bt));
1712 
1713 	if (bt->bt_magic != BT_MAGIC)
1714 		bt = NULL;
1715 
1716 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1717 	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1718 
1719 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1720 
1721 	dmu_object_info_from_db(db, &doi);
1722 
1723 	bbt = ztest_bt_bonus(db);
1724 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1725 	gen = bbt->bt_gen;
1726 	crtxg = bbt->bt_crtxg;
1727 	lrtxg = lr->lr_common.lrc_txg;
1728 
1729 	tx = dmu_tx_create(os);
1730 
1731 	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1732 
1733 	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1734 	    P2PHASE(offset, length) == 0)
1735 		abuf = dmu_request_arcbuf(db, length);
1736 
1737 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1738 	if (txg == 0) {
1739 		if (abuf != NULL)
1740 			dmu_return_arcbuf(abuf);
1741 		dmu_buf_rele(db, FTAG);
1742 		ztest_range_unlock(rl);
1743 		ztest_object_unlock(zd, lr->lr_foid);
1744 		return (ENOSPC);
1745 	}
1746 
1747 	if (bt != NULL) {
1748 		/*
1749 		 * Usually, verify the old data before writing new data --
1750 		 * but not always, because we also want to verify correct
1751 		 * behavior when the data was not recently read into cache.
1752 		 */
1753 		ASSERT(offset % doi.doi_data_block_size == 0);
1754 		if (ztest_random(4) != 0) {
1755 			int prefetch = ztest_random(2) ?
1756 			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1757 			ztest_block_tag_t rbt;
1758 
1759 			VERIFY(dmu_read(os, lr->lr_foid, offset,
1760 			    sizeof (rbt), &rbt, prefetch) == 0);
1761 			if (rbt.bt_magic == BT_MAGIC) {
1762 				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
1763 				    offset, gen, txg, crtxg);
1764 			}
1765 		}
1766 
1767 		/*
1768 		 * Writes can appear to be newer than the bonus buffer because
1769 		 * the ztest_get_data() callback does a dmu_read() of the
1770 		 * open-context data, which may be different than the data
1771 		 * as it was when the write was generated.
1772 		 */
1773 		if (zd->zd_zilog->zl_replay) {
1774 			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
1775 			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1776 			    bt->bt_crtxg);
1777 		}
1778 
1779 		/*
1780 		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1781 		 * so that all of the usual ASSERTs will work.
1782 		 */
1783 		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
1784 		    crtxg);
1785 	}
1786 
1787 	if (abuf == NULL) {
1788 		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1789 	} else {
1790 		bcopy(data, abuf->b_data, length);
1791 		dmu_assign_arcbuf(db, offset, abuf, tx);
1792 	}
1793 
1794 	(void) ztest_log_write(zd, tx, lr);
1795 
1796 	dmu_buf_rele(db, FTAG);
1797 
1798 	dmu_tx_commit(tx);
1799 
1800 	ztest_range_unlock(rl);
1801 	ztest_object_unlock(zd, lr->lr_foid);
1802 
1803 	return (0);
1804 }
1805 
1806 static int
1807 ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
1808 {
1809 	ztest_ds_t *zd = arg1;
1810 	lr_truncate_t *lr = arg2;
1811 	objset_t *os = zd->zd_os;
1812 	dmu_tx_t *tx;
1813 	uint64_t txg;
1814 	rl_t *rl;
1815 
1816 	if (byteswap)
1817 		byteswap_uint64_array(lr, sizeof (*lr));
1818 
1819 	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1820 	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1821 	    RL_WRITER);
1822 
1823 	tx = dmu_tx_create(os);
1824 
1825 	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1826 
1827 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1828 	if (txg == 0) {
1829 		ztest_range_unlock(rl);
1830 		ztest_object_unlock(zd, lr->lr_foid);
1831 		return (ENOSPC);
1832 	}
1833 
1834 	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1835 	    lr->lr_length, tx) == 0);
1836 
1837 	(void) ztest_log_truncate(zd, tx, lr);
1838 
1839 	dmu_tx_commit(tx);
1840 
1841 	ztest_range_unlock(rl);
1842 	ztest_object_unlock(zd, lr->lr_foid);
1843 
1844 	return (0);
1845 }
1846 
1847 static int
1848 ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
1849 {
1850 	ztest_ds_t *zd = arg1;
1851 	lr_setattr_t *lr = arg2;
1852 	objset_t *os = zd->zd_os;
1853 	dmu_tx_t *tx;
1854 	dmu_buf_t *db;
1855 	ztest_block_tag_t *bbt;
1856 	uint64_t txg, lrtxg, crtxg, dnodesize;
1857 
1858 	if (byteswap)
1859 		byteswap_uint64_array(lr, sizeof (*lr));
1860 
1861 	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1862 
1863 	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1864 
1865 	tx = dmu_tx_create(os);
1866 	dmu_tx_hold_bonus(tx, lr->lr_foid);
1867 
1868 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1869 	if (txg == 0) {
1870 		dmu_buf_rele(db, FTAG);
1871 		ztest_object_unlock(zd, lr->lr_foid);
1872 		return (ENOSPC);
1873 	}
1874 
1875 	bbt = ztest_bt_bonus(db);
1876 	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1877 	crtxg = bbt->bt_crtxg;
1878 	lrtxg = lr->lr_common.lrc_txg;
1879 	dnodesize = bbt->bt_dnodesize;
1880 
1881 	if (zd->zd_zilog->zl_replay) {
1882 		ASSERT(lr->lr_size != 0);
1883 		ASSERT(lr->lr_mode != 0);
1884 		ASSERT(lrtxg != 0);
1885 	} else {
1886 		/*
1887 		 * Randomly change the size and increment the generation.
1888 		 */
1889 		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1890 		    sizeof (*bbt);
1891 		lr->lr_mode = bbt->bt_gen + 1;
1892 		ASSERT(lrtxg == 0);
1893 	}
1894 
1895 	/*
1896 	 * Verify that the current bonus buffer is not newer than our txg.
1897 	 */
1898 	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
1899 	    MAX(txg, lrtxg), crtxg);
1900 
1901 	dmu_buf_will_dirty(db, tx);
1902 
1903 	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1904 	ASSERT3U(lr->lr_size, <=, db->db_size);
1905 	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1906 	bbt = ztest_bt_bonus(db);
1907 
1908 	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
1909 	    txg, crtxg);
1910 	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
1911 
1912 	dmu_buf_rele(db, FTAG);
1913 
1914 	(void) ztest_log_setattr(zd, tx, lr);
1915 
1916 	dmu_tx_commit(tx);
1917 
1918 	ztest_object_unlock(zd, lr->lr_foid);
1919 
1920 	return (0);
1921 }
1922 
1923 zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1924 	NULL,			/* 0 no such transaction type */
1925 	ztest_replay_create,	/* TX_CREATE */
1926 	NULL,			/* TX_MKDIR */
1927 	NULL,			/* TX_MKXATTR */
1928 	NULL,			/* TX_SYMLINK */
1929 	ztest_replay_remove,	/* TX_REMOVE */
1930 	NULL,			/* TX_RMDIR */
1931 	NULL,			/* TX_LINK */
1932 	NULL,			/* TX_RENAME */
1933 	ztest_replay_write,	/* TX_WRITE */
1934 	ztest_replay_truncate,	/* TX_TRUNCATE */
1935 	ztest_replay_setattr,	/* TX_SETATTR */
1936 	NULL,			/* TX_ACL */
1937 	NULL,			/* TX_CREATE_ACL */
1938 	NULL,			/* TX_CREATE_ATTR */
1939 	NULL,			/* TX_CREATE_ACL_ATTR */
1940 	NULL,			/* TX_MKDIR_ACL */
1941 	NULL,			/* TX_MKDIR_ATTR */
1942 	NULL,			/* TX_MKDIR_ACL_ATTR */
1943 	NULL,			/* TX_WRITE2 */
1944 };
1945 
1946 /*
1947  * ZIL get_data callbacks
1948  */
1949 
1950 /* ARGSUSED */
1951 static void
1952 ztest_get_done(zgd_t *zgd, int error)
1953 {
1954 	ztest_ds_t *zd = zgd->zgd_private;
1955 	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
1956 
1957 	if (zgd->zgd_db)
1958 		dmu_buf_rele(zgd->zgd_db, zgd);
1959 
1960 	ztest_range_unlock((rl_t *)zgd->zgd_lr);
1961 	ztest_object_unlock(zd, object);
1962 
1963 	umem_free(zgd, sizeof (*zgd));
1964 }
1965 
1966 static int
1967 ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
1968     zio_t *zio)
1969 {
1970 	ztest_ds_t *zd = arg;
1971 	objset_t *os = zd->zd_os;
1972 	uint64_t object = lr->lr_foid;
1973 	uint64_t offset = lr->lr_offset;
1974 	uint64_t size = lr->lr_length;
1975 	uint64_t txg = lr->lr_common.lrc_txg;
1976 	uint64_t crtxg;
1977 	dmu_object_info_t doi;
1978 	dmu_buf_t *db;
1979 	zgd_t *zgd;
1980 	int error;
1981 
1982 	ASSERT3P(lwb, !=, NULL);
1983 	ASSERT3P(zio, !=, NULL);
1984 	ASSERT3U(size, !=, 0);
1985 
1986 	ztest_object_lock(zd, object, RL_READER);
1987 	error = dmu_bonus_hold(os, object, FTAG, &db);
1988 	if (error) {
1989 		ztest_object_unlock(zd, object);
1990 		return (error);
1991 	}
1992 
1993 	crtxg = ztest_bt_bonus(db)->bt_crtxg;
1994 
1995 	if (crtxg == 0 || crtxg > txg) {
1996 		dmu_buf_rele(db, FTAG);
1997 		ztest_object_unlock(zd, object);
1998 		return (ENOENT);
1999 	}
2000 
2001 	dmu_object_info_from_db(db, &doi);
2002 	dmu_buf_rele(db, FTAG);
2003 	db = NULL;
2004 
2005 	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2006 	zgd->zgd_lwb = lwb;
2007 	zgd->zgd_private = zd;
2008 
2009 	if (buf != NULL) {	/* immediate write */
2010 		zgd->zgd_lr = (struct locked_range *)ztest_range_lock(zd,
2011 		    object, offset, size, RL_READER);
2012 
2013 		error = dmu_read(os, object, offset, size, buf,
2014 		    DMU_READ_NO_PREFETCH);
2015 		ASSERT(error == 0);
2016 	} else {
2017 		size = doi.doi_data_block_size;
2018 		if (ISP2(size)) {
2019 			offset = P2ALIGN(offset, size);
2020 		} else {
2021 			ASSERT(offset < size);
2022 			offset = 0;
2023 		}
2024 
2025 		zgd->zgd_lr = (struct locked_range *)ztest_range_lock(zd,
2026 		    object, offset, size, RL_READER);
2027 
2028 		error = dmu_buf_hold(os, object, offset, zgd, &db,
2029 		    DMU_READ_NO_PREFETCH);
2030 
2031 		if (error == 0) {
2032 			blkptr_t *bp = &lr->lr_blkptr;
2033 
2034 			zgd->zgd_db = db;
2035 			zgd->zgd_bp = bp;
2036 
2037 			ASSERT(db->db_offset == offset);
2038 			ASSERT(db->db_size == size);
2039 
2040 			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2041 			    ztest_get_done, zgd);
2042 
2043 			if (error == 0)
2044 				return (0);
2045 		}
2046 	}
2047 
2048 	ztest_get_done(zgd, error);
2049 
2050 	return (error);
2051 }
2052 
2053 static void *
2054 ztest_lr_alloc(size_t lrsize, char *name)
2055 {
2056 	char *lr;
2057 	size_t namesize = name ? strlen(name) + 1 : 0;
2058 
2059 	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2060 
2061 	if (name)
2062 		bcopy(name, lr + lrsize, namesize);
2063 
2064 	return (lr);
2065 }
2066 
2067 void
2068 ztest_lr_free(void *lr, size_t lrsize, char *name)
2069 {
2070 	size_t namesize = name ? strlen(name) + 1 : 0;
2071 
2072 	umem_free(lr, lrsize + namesize);
2073 }
2074 
2075 /*
2076  * Lookup a bunch of objects.  Returns the number of objects not found.
2077  */
2078 static int
2079 ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2080 {
2081 	int missing = 0;
2082 	int error;
2083 
2084 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2085 
2086 	for (int i = 0; i < count; i++, od++) {
2087 		od->od_object = 0;
2088 		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2089 		    sizeof (uint64_t), 1, &od->od_object);
2090 		if (error) {
2091 			ASSERT(error == ENOENT);
2092 			ASSERT(od->od_object == 0);
2093 			missing++;
2094 		} else {
2095 			dmu_buf_t *db;
2096 			ztest_block_tag_t *bbt;
2097 			dmu_object_info_t doi;
2098 
2099 			ASSERT(od->od_object != 0);
2100 			ASSERT(missing == 0);	/* there should be no gaps */
2101 
2102 			ztest_object_lock(zd, od->od_object, RL_READER);
2103 			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2104 			    od->od_object, FTAG, &db));
2105 			dmu_object_info_from_db(db, &doi);
2106 			bbt = ztest_bt_bonus(db);
2107 			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2108 			od->od_type = doi.doi_type;
2109 			od->od_blocksize = doi.doi_data_block_size;
2110 			od->od_gen = bbt->bt_gen;
2111 			dmu_buf_rele(db, FTAG);
2112 			ztest_object_unlock(zd, od->od_object);
2113 		}
2114 	}
2115 
2116 	return (missing);
2117 }
2118 
2119 static int
2120 ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2121 {
2122 	int missing = 0;
2123 
2124 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2125 
2126 	for (int i = 0; i < count; i++, od++) {
2127 		if (missing) {
2128 			od->od_object = 0;
2129 			missing++;
2130 			continue;
2131 		}
2132 
2133 		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2134 
2135 		lr->lr_doid = od->od_dir;
2136 		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2137 		lr->lrz_type = od->od_crtype;
2138 		lr->lrz_blocksize = od->od_crblocksize;
2139 		lr->lrz_ibshift = ztest_random_ibshift();
2140 		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2141 		lr->lrz_dnodesize = od->od_crdnodesize;
2142 		lr->lr_gen = od->od_crgen;
2143 		lr->lr_crtime[0] = time(NULL);
2144 
2145 		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2146 			ASSERT(missing == 0);
2147 			od->od_object = 0;
2148 			missing++;
2149 		} else {
2150 			od->od_object = lr->lr_foid;
2151 			od->od_type = od->od_crtype;
2152 			od->od_blocksize = od->od_crblocksize;
2153 			od->od_gen = od->od_crgen;
2154 			ASSERT(od->od_object != 0);
2155 		}
2156 
2157 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2158 	}
2159 
2160 	return (missing);
2161 }
2162 
2163 static int
2164 ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2165 {
2166 	int missing = 0;
2167 	int error;
2168 
2169 	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2170 
2171 	od += count - 1;
2172 
2173 	for (int i = count - 1; i >= 0; i--, od--) {
2174 		if (missing) {
2175 			missing++;
2176 			continue;
2177 		}
2178 
2179 		/*
2180 		 * No object was found.
2181 		 */
2182 		if (od->od_object == 0)
2183 			continue;
2184 
2185 		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2186 
2187 		lr->lr_doid = od->od_dir;
2188 
2189 		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2190 			ASSERT3U(error, ==, ENOSPC);
2191 			missing++;
2192 		} else {
2193 			od->od_object = 0;
2194 		}
2195 		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2196 	}
2197 
2198 	return (missing);
2199 }
2200 
2201 static int
2202 ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2203     void *data)
2204 {
2205 	lr_write_t *lr;
2206 	int error;
2207 
2208 	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2209 
2210 	lr->lr_foid = object;
2211 	lr->lr_offset = offset;
2212 	lr->lr_length = size;
2213 	lr->lr_blkoff = 0;
2214 	BP_ZERO(&lr->lr_blkptr);
2215 
2216 	bcopy(data, lr + 1, size);
2217 
2218 	error = ztest_replay_write(zd, lr, B_FALSE);
2219 
2220 	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2221 
2222 	return (error);
2223 }
2224 
2225 static int
2226 ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2227 {
2228 	lr_truncate_t *lr;
2229 	int error;
2230 
2231 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2232 
2233 	lr->lr_foid = object;
2234 	lr->lr_offset = offset;
2235 	lr->lr_length = size;
2236 
2237 	error = ztest_replay_truncate(zd, lr, B_FALSE);
2238 
2239 	ztest_lr_free(lr, sizeof (*lr), NULL);
2240 
2241 	return (error);
2242 }
2243 
2244 static int
2245 ztest_setattr(ztest_ds_t *zd, uint64_t object)
2246 {
2247 	lr_setattr_t *lr;
2248 	int error;
2249 
2250 	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2251 
2252 	lr->lr_foid = object;
2253 	lr->lr_size = 0;
2254 	lr->lr_mode = 0;
2255 
2256 	error = ztest_replay_setattr(zd, lr, B_FALSE);
2257 
2258 	ztest_lr_free(lr, sizeof (*lr), NULL);
2259 
2260 	return (error);
2261 }
2262 
2263 static void
2264 ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2265 {
2266 	objset_t *os = zd->zd_os;
2267 	dmu_tx_t *tx;
2268 	uint64_t txg;
2269 	rl_t *rl;
2270 
2271 	txg_wait_synced(dmu_objset_pool(os), 0);
2272 
2273 	ztest_object_lock(zd, object, RL_READER);
2274 	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2275 
2276 	tx = dmu_tx_create(os);
2277 
2278 	dmu_tx_hold_write(tx, object, offset, size);
2279 
2280 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2281 
2282 	if (txg != 0) {
2283 		dmu_prealloc(os, object, offset, size, tx);
2284 		dmu_tx_commit(tx);
2285 		txg_wait_synced(dmu_objset_pool(os), txg);
2286 	} else {
2287 		(void) dmu_free_long_range(os, object, offset, size);
2288 	}
2289 
2290 	ztest_range_unlock(rl);
2291 	ztest_object_unlock(zd, object);
2292 }
2293 
2294 static void
2295 ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2296 {
2297 	int err;
2298 	ztest_block_tag_t wbt;
2299 	dmu_object_info_t doi;
2300 	enum ztest_io_type io_type;
2301 	uint64_t blocksize;
2302 	void *data;
2303 
2304 	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2305 	blocksize = doi.doi_data_block_size;
2306 	data = umem_alloc(blocksize, UMEM_NOFAIL);
2307 
2308 	/*
2309 	 * Pick an i/o type at random, biased toward writing block tags.
2310 	 */
2311 	io_type = ztest_random(ZTEST_IO_TYPES);
2312 	if (ztest_random(2) == 0)
2313 		io_type = ZTEST_IO_WRITE_TAG;
2314 
2315 	rw_enter(&zd->zd_zilog_lock, RW_READER);
2316 
2317 	switch (io_type) {
2318 
2319 	case ZTEST_IO_WRITE_TAG:
2320 		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2321 		    offset, 0, 0, 0);
2322 		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2323 		break;
2324 
2325 	case ZTEST_IO_WRITE_PATTERN:
2326 		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2327 		if (ztest_random(2) == 0) {
2328 			/*
2329 			 * Induce fletcher2 collisions to ensure that
2330 			 * zio_ddt_collision() detects and resolves them
2331 			 * when using fletcher2-verify for deduplication.
2332 			 */
2333 			((uint64_t *)data)[0] ^= 1ULL << 63;
2334 			((uint64_t *)data)[4] ^= 1ULL << 63;
2335 		}
2336 		(void) ztest_write(zd, object, offset, blocksize, data);
2337 		break;
2338 
2339 	case ZTEST_IO_WRITE_ZEROES:
2340 		bzero(data, blocksize);
2341 		(void) ztest_write(zd, object, offset, blocksize, data);
2342 		break;
2343 
2344 	case ZTEST_IO_TRUNCATE:
2345 		(void) ztest_truncate(zd, object, offset, blocksize);
2346 		break;
2347 
2348 	case ZTEST_IO_SETATTR:
2349 		(void) ztest_setattr(zd, object);
2350 		break;
2351 
2352 	case ZTEST_IO_REWRITE:
2353 		rw_enter(&ztest_name_lock, RW_READER);
2354 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2355 		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2356 		    B_FALSE);
2357 		VERIFY(err == 0 || err == ENOSPC);
2358 		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2359 		    ZFS_PROP_COMPRESSION,
2360 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2361 		    B_FALSE);
2362 		VERIFY(err == 0 || err == ENOSPC);
2363 		rw_exit(&ztest_name_lock);
2364 
2365 		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2366 		    DMU_READ_NO_PREFETCH));
2367 
2368 		(void) ztest_write(zd, object, offset, blocksize, data);
2369 		break;
2370 	}
2371 
2372 	rw_exit(&zd->zd_zilog_lock);
2373 
2374 	umem_free(data, blocksize);
2375 }
2376 
2377 /*
2378  * Initialize an object description template.
2379  */
2380 static void
2381 ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2382     dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2383     uint64_t gen)
2384 {
2385 	od->od_dir = ZTEST_DIROBJ;
2386 	od->od_object = 0;
2387 
2388 	od->od_crtype = type;
2389 	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2390 	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2391 	od->od_crgen = gen;
2392 
2393 	od->od_type = DMU_OT_NONE;
2394 	od->od_blocksize = 0;
2395 	od->od_gen = 0;
2396 
2397 	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2398 	    tag, (int64_t)id, index);
2399 }
2400 
2401 /*
2402  * Lookup or create the objects for a test using the od template.
2403  * If the objects do not all exist, or if 'remove' is specified,
2404  * remove any existing objects and create new ones.  Otherwise,
2405  * use the existing objects.
2406  */
2407 static int
2408 ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2409 {
2410 	int count = size / sizeof (*od);
2411 	int rv = 0;
2412 
2413 	mutex_enter(&zd->zd_dirobj_lock);
2414 	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2415 	    (ztest_remove(zd, od, count) != 0 ||
2416 	    ztest_create(zd, od, count) != 0))
2417 		rv = -1;
2418 	zd->zd_od = od;
2419 	mutex_exit(&zd->zd_dirobj_lock);
2420 
2421 	return (rv);
2422 }
2423 
2424 /* ARGSUSED */
2425 void
2426 ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2427 {
2428 	zilog_t *zilog = zd->zd_zilog;
2429 
2430 	rw_enter(&zd->zd_zilog_lock, RW_READER);
2431 
2432 	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2433 
2434 	/*
2435 	 * Remember the committed values in zd, which is in parent/child
2436 	 * shared memory.  If we die, the next iteration of ztest_run()
2437 	 * will verify that the log really does contain this record.
2438 	 */
2439 	mutex_enter(&zilog->zl_lock);
2440 	ASSERT(zd->zd_shared != NULL);
2441 	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2442 	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2443 	mutex_exit(&zilog->zl_lock);
2444 
2445 	rw_exit(&zd->zd_zilog_lock);
2446 }
2447 
2448 /*
2449  * This function is designed to simulate the operations that occur during a
2450  * mount/unmount operation.  We hold the dataset across these operations in an
2451  * attempt to expose any implicit assumptions about ZIL management.
2452  */
2453 /* ARGSUSED */
2454 void
2455 ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2456 {
2457 	objset_t *os = zd->zd_os;
2458 
2459 	/*
2460 	 * We grab the zd_dirobj_lock to ensure that no other thread is
2461 	 * updating the zil (i.e. adding in-memory log records) and the
2462 	 * zd_zilog_lock to block any I/O.
2463 	 */
2464 	mutex_enter(&zd->zd_dirobj_lock);
2465 	rw_enter(&zd->zd_zilog_lock, RW_WRITER);
2466 
2467 	/* zfsvfs_teardown() */
2468 	zil_close(zd->zd_zilog);
2469 
2470 	/* zfsvfs_setup() */
2471 	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2472 	zil_replay(os, zd, ztest_replay_vector);
2473 
2474 	rw_exit(&zd->zd_zilog_lock);
2475 	mutex_exit(&zd->zd_dirobj_lock);
2476 }
2477 
2478 /*
2479  * Verify that we can't destroy an active pool, create an existing pool,
2480  * or create a pool with a bad vdev spec.
2481  */
2482 /* ARGSUSED */
2483 void
2484 ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2485 {
2486 	ztest_shared_opts_t *zo = &ztest_opts;
2487 	spa_t *spa;
2488 	nvlist_t *nvroot;
2489 
2490 	if (zo->zo_mmp_test)
2491 		return;
2492 
2493 	/*
2494 	 * Attempt to create using a bad file.
2495 	 */
2496 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2497 	VERIFY3U(ENOENT, ==,
2498 	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2499 	nvlist_free(nvroot);
2500 
2501 	/*
2502 	 * Attempt to create using a bad mirror.
2503 	 */
2504 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 2, 1);
2505 	VERIFY3U(ENOENT, ==,
2506 	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2507 	nvlist_free(nvroot);
2508 
2509 	/*
2510 	 * Attempt to create an existing pool.  It shouldn't matter
2511 	 * what's in the nvroot; we should fail with EEXIST.
2512 	 */
2513 	rw_enter(&ztest_name_lock, RW_READER);
2514 	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, 0, 0, 0, 1);
2515 	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2516 	nvlist_free(nvroot);
2517 	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2518 	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2519 	spa_close(spa, FTAG);
2520 
2521 	rw_exit(&ztest_name_lock);
2522 }
2523 
2524 /*
2525  * Start and then stop the MMP threads to ensure the startup and shutdown code
2526  * works properly.  Actual protection and property-related code tested via ZTS.
2527  */
2528 /* ARGSUSED */
2529 void
2530 ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2531 {
2532 	ztest_shared_opts_t *zo = &ztest_opts;
2533 	spa_t *spa = ztest_spa;
2534 
2535 	if (zo->zo_mmp_test)
2536 		return;
2537 
2538 	/*
2539 	 * Since enabling MMP involves setting a property, it could not be done
2540 	 * while the pool is suspended.
2541 	 */
2542 	if (spa_suspended(spa))
2543 		return;
2544 
2545 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2546 	mutex_enter(&spa->spa_props_lock);
2547 
2548 	zfs_multihost_fail_intervals = 0;
2549 
2550 	if (!spa_multihost(spa)) {
2551 		spa->spa_multihost = B_TRUE;
2552 		mmp_thread_start(spa);
2553 	}
2554 
2555 	mutex_exit(&spa->spa_props_lock);
2556 	spa_config_exit(spa, SCL_CONFIG, FTAG);
2557 
2558 	txg_wait_synced(spa_get_dsl(spa), 0);
2559 	mmp_signal_all_threads();
2560 	txg_wait_synced(spa_get_dsl(spa), 0);
2561 
2562 	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2563 	mutex_enter(&spa->spa_props_lock);
2564 
2565 	if (spa_multihost(spa)) {
2566 		mmp_thread_stop(spa);
2567 		spa->spa_multihost = B_FALSE;
2568 	}
2569 
2570 	mutex_exit(&spa->spa_props_lock);
2571 	spa_config_exit(spa, SCL_CONFIG, FTAG);
2572 }
2573 
2574 /* ARGSUSED */
2575 void
2576 ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2577 {
2578 	spa_t *spa;
2579 	uint64_t initial_version = SPA_VERSION_INITIAL;
2580 	uint64_t version, newversion;
2581 	nvlist_t *nvroot, *props;
2582 	char *name;
2583 
2584 	if (ztest_opts.zo_mmp_test)
2585 		return;
2586 
2587 	mutex_enter(&ztest_vdev_lock);
2588 	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2589 
2590 	/*
2591 	 * Clean up from previous runs.
2592 	 */
2593 	(void) spa_destroy(name);
2594 
2595 	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2596 	    0, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2597 
2598 	/*
2599 	 * If we're configuring a RAIDZ device then make sure that the
2600 	 * the initial version is capable of supporting that feature.
2601 	 */
2602 	switch (ztest_opts.zo_raidz_parity) {
2603 	case 0:
2604 	case 1:
2605 		initial_version = SPA_VERSION_INITIAL;
2606 		break;
2607 	case 2:
2608 		initial_version = SPA_VERSION_RAIDZ2;
2609 		break;
2610 	case 3:
2611 		initial_version = SPA_VERSION_RAIDZ3;
2612 		break;
2613 	}
2614 
2615 	/*
2616 	 * Create a pool with a spa version that can be upgraded. Pick
2617 	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2618 	 */
2619 	do {
2620 		version = ztest_random_spa_version(initial_version);
2621 	} while (version > SPA_VERSION_BEFORE_FEATURES);
2622 
2623 	props = fnvlist_alloc();
2624 	fnvlist_add_uint64(props,
2625 	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2626 	VERIFY0(spa_create(name, nvroot, props, NULL));
2627 	fnvlist_free(nvroot);
2628 	fnvlist_free(props);
2629 
2630 	VERIFY0(spa_open(name, &spa, FTAG));
2631 	VERIFY3U(spa_version(spa), ==, version);
2632 	newversion = ztest_random_spa_version(version + 1);
2633 
2634 	if (ztest_opts.zo_verbose >= 4) {
2635 		(void) printf("upgrading spa version from %llu to %llu\n",
2636 		    (u_longlong_t)version, (u_longlong_t)newversion);
2637 	}
2638 
2639 	spa_upgrade(spa, newversion);
2640 	VERIFY3U(spa_version(spa), >, version);
2641 	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2642 	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2643 	spa_close(spa, FTAG);
2644 
2645 	strfree(name);
2646 	mutex_exit(&ztest_vdev_lock);
2647 }
2648 
2649 static void
2650 ztest_spa_checkpoint(spa_t *spa)
2651 {
2652 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2653 
2654 	int error = spa_checkpoint(spa->spa_name);
2655 
2656 	switch (error) {
2657 	case 0:
2658 	case ZFS_ERR_DEVRM_IN_PROGRESS:
2659 	case ZFS_ERR_DISCARDING_CHECKPOINT:
2660 	case ZFS_ERR_CHECKPOINT_EXISTS:
2661 		break;
2662 	case ENOSPC:
2663 		ztest_record_enospc(FTAG);
2664 		break;
2665 	default:
2666 		fatal(0, "spa_checkpoint(%s) = %d", spa->spa_name, error);
2667 	}
2668 }
2669 
2670 static void
2671 ztest_spa_discard_checkpoint(spa_t *spa)
2672 {
2673 	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2674 
2675 	int error = spa_checkpoint_discard(spa->spa_name);
2676 
2677 	switch (error) {
2678 	case 0:
2679 	case ZFS_ERR_DISCARDING_CHECKPOINT:
2680 	case ZFS_ERR_NO_CHECKPOINT:
2681 		break;
2682 	default:
2683 		fatal(0, "spa_discard_checkpoint(%s) = %d",
2684 		    spa->spa_name, error);
2685 	}
2686 
2687 }
2688 
2689 /* ARGSUSED */
2690 void
2691 ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
2692 {
2693 	spa_t *spa = ztest_spa;
2694 
2695 	mutex_enter(&ztest_checkpoint_lock);
2696 	if (ztest_random(2) == 0) {
2697 		ztest_spa_checkpoint(spa);
2698 	} else {
2699 		ztest_spa_discard_checkpoint(spa);
2700 	}
2701 	mutex_exit(&ztest_checkpoint_lock);
2702 }
2703 
2704 
2705 static vdev_t *
2706 vdev_lookup_by_path(vdev_t *vd, const char *path)
2707 {
2708 	vdev_t *mvd;
2709 
2710 	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2711 		return (vd);
2712 
2713 	for (int c = 0; c < vd->vdev_children; c++)
2714 		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2715 		    NULL)
2716 			return (mvd);
2717 
2718 	return (NULL);
2719 }
2720 
2721 /*
2722  * Find the first available hole which can be used as a top-level.
2723  */
2724 int
2725 find_vdev_hole(spa_t *spa)
2726 {
2727 	vdev_t *rvd = spa->spa_root_vdev;
2728 	int c;
2729 
2730 	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2731 
2732 	for (c = 0; c < rvd->vdev_children; c++) {
2733 		vdev_t *cvd = rvd->vdev_child[c];
2734 
2735 		if (cvd->vdev_ishole)
2736 			break;
2737 	}
2738 	return (c);
2739 }
2740 
2741 /*
2742  * Verify that vdev_add() works as expected.
2743  */
2744 /* ARGSUSED */
2745 void
2746 ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2747 {
2748 	ztest_shared_t *zs = ztest_shared;
2749 	spa_t *spa = ztest_spa;
2750 	uint64_t leaves;
2751 	uint64_t guid;
2752 	nvlist_t *nvroot;
2753 	int error;
2754 
2755 	if (ztest_opts.zo_mmp_test)
2756 		return;
2757 
2758 	mutex_enter(&ztest_vdev_lock);
2759 	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2760 
2761 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2762 
2763 	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2764 
2765 	/*
2766 	 * If we have slogs then remove them 1/4 of the time.
2767 	 */
2768 	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2769 		/*
2770 		 * Grab the guid from the head of the log class rotor.
2771 		 */
2772 		guid = spa_log_class(spa)->mc_rotor->mg_vd->vdev_guid;
2773 
2774 		spa_config_exit(spa, SCL_VDEV, FTAG);
2775 
2776 		/*
2777 		 * We have to grab the zs_name_lock as writer to
2778 		 * prevent a race between removing a slog (dmu_objset_find)
2779 		 * and destroying a dataset. Removing the slog will
2780 		 * grab a reference on the dataset which may cause
2781 		 * dmu_objset_destroy() to fail with EBUSY thus
2782 		 * leaving the dataset in an inconsistent state.
2783 		 */
2784 		rw_enter(&ztest_name_lock, RW_WRITER);
2785 		error = spa_vdev_remove(spa, guid, B_FALSE);
2786 		rw_exit(&ztest_name_lock);
2787 
2788 		switch (error) {
2789 		case 0:
2790 		case EEXIST:
2791 		case ZFS_ERR_CHECKPOINT_EXISTS:
2792 		case ZFS_ERR_DISCARDING_CHECKPOINT:
2793 			break;
2794 		default:
2795 			fatal(0, "spa_vdev_remove() = %d", error);
2796 		}
2797 	} else {
2798 		spa_config_exit(spa, SCL_VDEV, FTAG);
2799 
2800 		/*
2801 		 * Make 1/4 of the devices be log devices.
2802 		 */
2803 		nvroot = make_vdev_root(NULL, NULL, NULL,
2804 		    ztest_opts.zo_vdev_size, 0,
2805 		    ztest_random(4) == 0, ztest_opts.zo_raidz,
2806 		    zs->zs_mirrors, 1);
2807 
2808 		error = spa_vdev_add(spa, nvroot);
2809 		nvlist_free(nvroot);
2810 
2811 		switch (error) {
2812 		case 0:
2813 			break;
2814 		case ENOSPC:
2815 			ztest_record_enospc("spa_vdev_add");
2816 			break;
2817 		default:
2818 			fatal(0, "spa_vdev_add() = %d", error);
2819 		}
2820 	}
2821 
2822 	mutex_exit(&ztest_vdev_lock);
2823 }
2824 
2825 /*
2826  * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2827  */
2828 /* ARGSUSED */
2829 void
2830 ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2831 {
2832 	ztest_shared_t *zs = ztest_shared;
2833 	spa_t *spa = ztest_spa;
2834 	vdev_t *rvd = spa->spa_root_vdev;
2835 	spa_aux_vdev_t *sav;
2836 	char *aux;
2837 	uint64_t guid = 0;
2838 	int error;
2839 
2840 	if (ztest_opts.zo_mmp_test)
2841 		return;
2842 
2843 	if (ztest_random(2) == 0) {
2844 		sav = &spa->spa_spares;
2845 		aux = ZPOOL_CONFIG_SPARES;
2846 	} else {
2847 		sav = &spa->spa_l2cache;
2848 		aux = ZPOOL_CONFIG_L2CACHE;
2849 	}
2850 
2851 	mutex_enter(&ztest_vdev_lock);
2852 
2853 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2854 
2855 	if (sav->sav_count != 0 && ztest_random(4) == 0) {
2856 		/*
2857 		 * Pick a random device to remove.
2858 		 */
2859 		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
2860 	} else {
2861 		/*
2862 		 * Find an unused device we can add.
2863 		 */
2864 		zs->zs_vdev_aux = 0;
2865 		for (;;) {
2866 			char path[MAXPATHLEN];
2867 			int c;
2868 			(void) snprintf(path, sizeof (path), ztest_aux_template,
2869 			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
2870 			    zs->zs_vdev_aux);
2871 			for (c = 0; c < sav->sav_count; c++)
2872 				if (strcmp(sav->sav_vdevs[c]->vdev_path,
2873 				    path) == 0)
2874 					break;
2875 			if (c == sav->sav_count &&
2876 			    vdev_lookup_by_path(rvd, path) == NULL)
2877 				break;
2878 			zs->zs_vdev_aux++;
2879 		}
2880 	}
2881 
2882 	spa_config_exit(spa, SCL_VDEV, FTAG);
2883 
2884 	if (guid == 0) {
2885 		/*
2886 		 * Add a new device.
2887 		 */
2888 		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
2889 		    (ztest_opts.zo_vdev_size * 5) / 4, 0, 0, 0, 0, 1);
2890 		error = spa_vdev_add(spa, nvroot);
2891 
2892 		switch (error) {
2893 		case 0:
2894 			break;
2895 		default:
2896 			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
2897 		}
2898 		nvlist_free(nvroot);
2899 	} else {
2900 		/*
2901 		 * Remove an existing device.  Sometimes, dirty its
2902 		 * vdev state first to make sure we handle removal
2903 		 * of devices that have pending state changes.
2904 		 */
2905 		if (ztest_random(2) == 0)
2906 			(void) vdev_online(spa, guid, 0, NULL);
2907 
2908 		error = spa_vdev_remove(spa, guid, B_FALSE);
2909 
2910 		switch (error) {
2911 		case 0:
2912 		case EBUSY:
2913 		case ZFS_ERR_CHECKPOINT_EXISTS:
2914 		case ZFS_ERR_DISCARDING_CHECKPOINT:
2915 			break;
2916 		default:
2917 			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
2918 		}
2919 	}
2920 
2921 	mutex_exit(&ztest_vdev_lock);
2922 }
2923 
2924 /*
2925  * split a pool if it has mirror tlvdevs
2926  */
2927 /* ARGSUSED */
2928 void
2929 ztest_split_pool(ztest_ds_t *zd, uint64_t id)
2930 {
2931 	ztest_shared_t *zs = ztest_shared;
2932 	spa_t *spa = ztest_spa;
2933 	vdev_t *rvd = spa->spa_root_vdev;
2934 	nvlist_t *tree, **child, *config, *split, **schild;
2935 	uint_t c, children, schildren = 0, lastlogid = 0;
2936 	int error = 0;
2937 
2938 	if (ztest_opts.zo_mmp_test)
2939 		return;
2940 
2941 	mutex_enter(&ztest_vdev_lock);
2942 
2943 	/* ensure we have a useable config; mirrors of raidz aren't supported */
2944 	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
2945 		mutex_exit(&ztest_vdev_lock);
2946 		return;
2947 	}
2948 
2949 	/* clean up the old pool, if any */
2950 	(void) spa_destroy("splitp");
2951 
2952 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2953 
2954 	/* generate a config from the existing config */
2955 	mutex_enter(&spa->spa_props_lock);
2956 	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
2957 	    &tree) == 0);
2958 	mutex_exit(&spa->spa_props_lock);
2959 
2960 	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2961 	    &children) == 0);
2962 
2963 	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
2964 	for (c = 0; c < children; c++) {
2965 		vdev_t *tvd = rvd->vdev_child[c];
2966 		nvlist_t **mchild;
2967 		uint_t mchildren;
2968 
2969 		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
2970 			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
2971 			    0) == 0);
2972 			VERIFY(nvlist_add_string(schild[schildren],
2973 			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
2974 			VERIFY(nvlist_add_uint64(schild[schildren],
2975 			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
2976 			if (lastlogid == 0)
2977 				lastlogid = schildren;
2978 			++schildren;
2979 			continue;
2980 		}
2981 		lastlogid = 0;
2982 		VERIFY(nvlist_lookup_nvlist_array(child[c],
2983 		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2984 		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
2985 	}
2986 
2987 	/* OK, create a config that can be used to split */
2988 	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
2989 	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
2990 	    VDEV_TYPE_ROOT) == 0);
2991 	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
2992 	    lastlogid != 0 ? lastlogid : schildren) == 0);
2993 
2994 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
2995 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
2996 
2997 	for (c = 0; c < schildren; c++)
2998 		nvlist_free(schild[c]);
2999 	free(schild);
3000 	nvlist_free(split);
3001 
3002 	spa_config_exit(spa, SCL_VDEV, FTAG);
3003 
3004 	rw_enter(&ztest_name_lock, RW_WRITER);
3005 	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3006 	rw_exit(&ztest_name_lock);
3007 
3008 	nvlist_free(config);
3009 
3010 	if (error == 0) {
3011 		(void) printf("successful split - results:\n");
3012 		mutex_enter(&spa_namespace_lock);
3013 		show_pool_stats(spa);
3014 		show_pool_stats(spa_lookup("splitp"));
3015 		mutex_exit(&spa_namespace_lock);
3016 		++zs->zs_splits;
3017 		--zs->zs_mirrors;
3018 	}
3019 	mutex_exit(&ztest_vdev_lock);
3020 }
3021 
3022 /*
3023  * Verify that we can attach and detach devices.
3024  */
3025 /* ARGSUSED */
3026 void
3027 ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3028 {
3029 	ztest_shared_t *zs = ztest_shared;
3030 	spa_t *spa = ztest_spa;
3031 	spa_aux_vdev_t *sav = &spa->spa_spares;
3032 	vdev_t *rvd = spa->spa_root_vdev;
3033 	vdev_t *oldvd, *newvd, *pvd;
3034 	nvlist_t *root;
3035 	uint64_t leaves;
3036 	uint64_t leaf, top;
3037 	uint64_t ashift = ztest_get_ashift();
3038 	uint64_t oldguid, pguid;
3039 	uint64_t oldsize, newsize;
3040 	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
3041 	int replacing;
3042 	int oldvd_has_siblings = B_FALSE;
3043 	int newvd_is_spare = B_FALSE;
3044 	int oldvd_is_log;
3045 	int error, expected_error;
3046 
3047 	if (ztest_opts.zo_mmp_test)
3048 		return;
3049 
3050 	mutex_enter(&ztest_vdev_lock);
3051 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
3052 
3053 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3054 
3055 	/*
3056 	 * If a vdev is in the process of being removed, its removal may
3057 	 * finish while we are in progress, leading to an unexpected error
3058 	 * value.  Don't bother trying to attach while we are in the middle
3059 	 * of removal.
3060 	 */
3061 	if (ztest_device_removal_active) {
3062 		spa_config_exit(spa, SCL_ALL, FTAG);
3063 		mutex_exit(&ztest_vdev_lock);
3064 		return;
3065 	}
3066 
3067 	/*
3068 	 * Decide whether to do an attach or a replace.
3069 	 */
3070 	replacing = ztest_random(2);
3071 
3072 	/*
3073 	 * Pick a random top-level vdev.
3074 	 */
3075 	top = ztest_random_vdev_top(spa, B_TRUE);
3076 
3077 	/*
3078 	 * Pick a random leaf within it.
3079 	 */
3080 	leaf = ztest_random(leaves);
3081 
3082 	/*
3083 	 * Locate this vdev.
3084 	 */
3085 	oldvd = rvd->vdev_child[top];
3086 	if (zs->zs_mirrors >= 1) {
3087 		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
3088 		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
3089 		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
3090 	}
3091 	if (ztest_opts.zo_raidz > 1) {
3092 		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
3093 		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
3094 		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
3095 	}
3096 
3097 	/*
3098 	 * If we're already doing an attach or replace, oldvd may be a
3099 	 * mirror vdev -- in which case, pick a random child.
3100 	 */
3101 	while (oldvd->vdev_children != 0) {
3102 		oldvd_has_siblings = B_TRUE;
3103 		ASSERT(oldvd->vdev_children >= 2);
3104 		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3105 	}
3106 
3107 	oldguid = oldvd->vdev_guid;
3108 	oldsize = vdev_get_min_asize(oldvd);
3109 	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3110 	(void) strcpy(oldpath, oldvd->vdev_path);
3111 	pvd = oldvd->vdev_parent;
3112 	pguid = pvd->vdev_guid;
3113 
3114 	/*
3115 	 * If oldvd has siblings, then half of the time, detach it.
3116 	 */
3117 	if (oldvd_has_siblings && ztest_random(2) == 0) {
3118 		spa_config_exit(spa, SCL_ALL, FTAG);
3119 		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3120 		if (error != 0 && error != ENODEV && error != EBUSY &&
3121 		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3122 		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3123 			fatal(0, "detach (%s) returned %d", oldpath, error);
3124 		mutex_exit(&ztest_vdev_lock);
3125 		return;
3126 	}
3127 
3128 	/*
3129 	 * For the new vdev, choose with equal probability between the two
3130 	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3131 	 */
3132 	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3133 		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3134 		newvd_is_spare = B_TRUE;
3135 		(void) strcpy(newpath, newvd->vdev_path);
3136 	} else {
3137 		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
3138 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3139 		    top * leaves + leaf);
3140 		if (ztest_random(2) == 0)
3141 			newpath[strlen(newpath) - 1] = 'b';
3142 		newvd = vdev_lookup_by_path(rvd, newpath);
3143 	}
3144 
3145 	if (newvd) {
3146 		/*
3147 		 * Reopen to ensure the vdev's asize field isn't stale.
3148 		 */
3149 		vdev_reopen(newvd);
3150 		newsize = vdev_get_min_asize(newvd);
3151 	} else {
3152 		/*
3153 		 * Make newsize a little bigger or smaller than oldsize.
3154 		 * If it's smaller, the attach should fail.
3155 		 * If it's larger, and we're doing a replace,
3156 		 * we should get dynamic LUN growth when we're done.
3157 		 */
3158 		newsize = 10 * oldsize / (9 + ztest_random(3));
3159 	}
3160 
3161 	/*
3162 	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3163 	 * unless it's a replace; in that case any non-replacing parent is OK.
3164 	 *
3165 	 * If newvd is already part of the pool, it should fail with EBUSY.
3166 	 *
3167 	 * If newvd is too small, it should fail with EOVERFLOW.
3168 	 */
3169 	if (pvd->vdev_ops != &vdev_mirror_ops &&
3170 	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3171 	    pvd->vdev_ops == &vdev_replacing_ops ||
3172 	    pvd->vdev_ops == &vdev_spare_ops))
3173 		expected_error = ENOTSUP;
3174 	else if (newvd_is_spare && (!replacing || oldvd_is_log))
3175 		expected_error = ENOTSUP;
3176 	else if (newvd == oldvd)
3177 		expected_error = replacing ? 0 : EBUSY;
3178 	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3179 		expected_error = EBUSY;
3180 	else if (newsize < oldsize)
3181 		expected_error = EOVERFLOW;
3182 	else if (ashift > oldvd->vdev_top->vdev_ashift)
3183 		expected_error = EDOM;
3184 	else
3185 		expected_error = 0;
3186 
3187 	spa_config_exit(spa, SCL_ALL, FTAG);
3188 
3189 	/*
3190 	 * Build the nvlist describing newpath.
3191 	 */
3192 	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3193 	    ashift, 0, 0, 0, 1);
3194 
3195 	error = spa_vdev_attach(spa, oldguid, root, replacing);
3196 
3197 	nvlist_free(root);
3198 
3199 	/*
3200 	 * If our parent was the replacing vdev, but the replace completed,
3201 	 * then instead of failing with ENOTSUP we may either succeed,
3202 	 * fail with ENODEV, or fail with EOVERFLOW.
3203 	 */
3204 	if (expected_error == ENOTSUP &&
3205 	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3206 		expected_error = error;
3207 
3208 	/*
3209 	 * If someone grew the LUN, the replacement may be too small.
3210 	 */
3211 	if (error == EOVERFLOW || error == EBUSY)
3212 		expected_error = error;
3213 
3214 	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3215 	    error == ZFS_ERR_DISCARDING_CHECKPOINT)
3216 		expected_error = error;
3217 
3218 	/* XXX workaround 6690467 */
3219 	if (error != expected_error && expected_error != EBUSY) {
3220 		fatal(0, "attach (%s %llu, %s %llu, %d) "
3221 		    "returned %d, expected %d",
3222 		    oldpath, oldsize, newpath,
3223 		    newsize, replacing, error, expected_error);
3224 	}
3225 
3226 	mutex_exit(&ztest_vdev_lock);
3227 }
3228 
3229 /* ARGSUSED */
3230 void
3231 ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3232 {
3233 	spa_t *spa = ztest_spa;
3234 	vdev_t *vd;
3235 	uint64_t guid;
3236 	int error;
3237 
3238 	mutex_enter(&ztest_vdev_lock);
3239 
3240 	if (ztest_device_removal_active) {
3241 		mutex_exit(&ztest_vdev_lock);
3242 		return;
3243 	}
3244 
3245 	/*
3246 	 * Remove a random top-level vdev and wait for removal to finish.
3247 	 */
3248 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3249 	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3250 	guid = vd->vdev_guid;
3251 	spa_config_exit(spa, SCL_VDEV, FTAG);
3252 
3253 	error = spa_vdev_remove(spa, guid, B_FALSE);
3254 	if (error == 0) {
3255 		ztest_device_removal_active = B_TRUE;
3256 		mutex_exit(&ztest_vdev_lock);
3257 
3258 		while (spa->spa_vdev_removal != NULL)
3259 			txg_wait_synced(spa_get_dsl(spa), 0);
3260 	} else {
3261 		mutex_exit(&ztest_vdev_lock);
3262 		return;
3263 	}
3264 
3265 	/*
3266 	 * The pool needs to be scrubbed after completing device removal.
3267 	 * Failure to do so may result in checksum errors due to the
3268 	 * strategy employed by ztest_fault_inject() when selecting which
3269 	 * offset are redundant and can be damaged.
3270 	 */
3271 	error = spa_scan(spa, POOL_SCAN_SCRUB);
3272 	if (error == 0) {
3273 		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3274 			txg_wait_synced(spa_get_dsl(spa), 0);
3275 	}
3276 
3277 	mutex_enter(&ztest_vdev_lock);
3278 	ztest_device_removal_active = B_FALSE;
3279 	mutex_exit(&ztest_vdev_lock);
3280 }
3281 
3282 /*
3283  * Callback function which expands the physical size of the vdev.
3284  */
3285 vdev_t *
3286 grow_vdev(vdev_t *vd, void *arg)
3287 {
3288 	spa_t *spa = vd->vdev_spa;
3289 	size_t *newsize = arg;
3290 	size_t fsize;
3291 	int fd;
3292 
3293 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3294 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3295 
3296 	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3297 		return (vd);
3298 
3299 	fsize = lseek(fd, 0, SEEK_END);
3300 	(void) ftruncate(fd, *newsize);
3301 
3302 	if (ztest_opts.zo_verbose >= 6) {
3303 		(void) printf("%s grew from %lu to %lu bytes\n",
3304 		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3305 	}
3306 	(void) close(fd);
3307 	return (NULL);
3308 }
3309 
3310 /*
3311  * Callback function which expands a given vdev by calling vdev_online().
3312  */
3313 /* ARGSUSED */
3314 vdev_t *
3315 online_vdev(vdev_t *vd, void *arg)
3316 {
3317 	spa_t *spa = vd->vdev_spa;
3318 	vdev_t *tvd = vd->vdev_top;
3319 	uint64_t guid = vd->vdev_guid;
3320 	uint64_t generation = spa->spa_config_generation + 1;
3321 	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3322 	int error;
3323 
3324 	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3325 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3326 
3327 	/* Calling vdev_online will initialize the new metaslabs */
3328 	spa_config_exit(spa, SCL_STATE, spa);
3329 	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3330 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3331 
3332 	/*
3333 	 * If vdev_online returned an error or the underlying vdev_open
3334 	 * failed then we abort the expand. The only way to know that
3335 	 * vdev_open fails is by checking the returned newstate.
3336 	 */
3337 	if (error || newstate != VDEV_STATE_HEALTHY) {
3338 		if (ztest_opts.zo_verbose >= 5) {
3339 			(void) printf("Unable to expand vdev, state %llu, "
3340 			    "error %d\n", (u_longlong_t)newstate, error);
3341 		}
3342 		return (vd);
3343 	}
3344 	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3345 
3346 	/*
3347 	 * Since we dropped the lock we need to ensure that we're
3348 	 * still talking to the original vdev. It's possible this
3349 	 * vdev may have been detached/replaced while we were
3350 	 * trying to online it.
3351 	 */
3352 	if (generation != spa->spa_config_generation) {
3353 		if (ztest_opts.zo_verbose >= 5) {
3354 			(void) printf("vdev configuration has changed, "
3355 			    "guid %llu, state %llu, expected gen %llu, "
3356 			    "got gen %llu\n",
3357 			    (u_longlong_t)guid,
3358 			    (u_longlong_t)tvd->vdev_state,
3359 			    (u_longlong_t)generation,
3360 			    (u_longlong_t)spa->spa_config_generation);
3361 		}
3362 		return (vd);
3363 	}
3364 	return (NULL);
3365 }
3366 
3367 /*
3368  * Traverse the vdev tree calling the supplied function.
3369  * We continue to walk the tree until we either have walked all
3370  * children or we receive a non-NULL return from the callback.
3371  * If a NULL callback is passed, then we just return back the first
3372  * leaf vdev we encounter.
3373  */
3374 vdev_t *
3375 vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3376 {
3377 	if (vd->vdev_ops->vdev_op_leaf) {
3378 		if (func == NULL)
3379 			return (vd);
3380 		else
3381 			return (func(vd, arg));
3382 	}
3383 
3384 	for (uint_t c = 0; c < vd->vdev_children; c++) {
3385 		vdev_t *cvd = vd->vdev_child[c];
3386 		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3387 			return (cvd);
3388 	}
3389 	return (NULL);
3390 }
3391 
3392 /*
3393  * Verify that dynamic LUN growth works as expected.
3394  */
3395 /* ARGSUSED */
3396 void
3397 ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3398 {
3399 	spa_t *spa = ztest_spa;
3400 	vdev_t *vd, *tvd;
3401 	metaslab_class_t *mc;
3402 	metaslab_group_t *mg;
3403 	size_t psize, newsize;
3404 	uint64_t top;
3405 	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3406 
3407 	mutex_enter(&ztest_checkpoint_lock);
3408 	mutex_enter(&ztest_vdev_lock);
3409 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3410 
3411 	/*
3412 	 * If there is a vdev removal in progress, it could complete while
3413 	 * we are running, in which case we would not be able to verify
3414 	 * that the metaslab_class space increased (because it decreases
3415 	 * when the device removal completes).
3416 	 */
3417 	if (ztest_device_removal_active) {
3418 		spa_config_exit(spa, SCL_STATE, spa);
3419 		mutex_exit(&ztest_vdev_lock);
3420 		mutex_exit(&ztest_checkpoint_lock);
3421 		return;
3422 	}
3423 
3424 	top = ztest_random_vdev_top(spa, B_TRUE);
3425 
3426 	tvd = spa->spa_root_vdev->vdev_child[top];
3427 	mg = tvd->vdev_mg;
3428 	mc = mg->mg_class;
3429 	old_ms_count = tvd->vdev_ms_count;
3430 	old_class_space = metaslab_class_get_space(mc);
3431 
3432 	/*
3433 	 * Determine the size of the first leaf vdev associated with
3434 	 * our top-level device.
3435 	 */
3436 	vd = vdev_walk_tree(tvd, NULL, NULL);
3437 	ASSERT3P(vd, !=, NULL);
3438 	ASSERT(vd->vdev_ops->vdev_op_leaf);
3439 
3440 	psize = vd->vdev_psize;
3441 
3442 	/*
3443 	 * We only try to expand the vdev if it's healthy, less than 4x its
3444 	 * original size, and it has a valid psize.
3445 	 */
3446 	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3447 	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3448 		spa_config_exit(spa, SCL_STATE, spa);
3449 		mutex_exit(&ztest_vdev_lock);
3450 		mutex_exit(&ztest_checkpoint_lock);
3451 		return;
3452 	}
3453 	ASSERT(psize > 0);
3454 	newsize = psize + psize / 8;
3455 	ASSERT3U(newsize, >, psize);
3456 
3457 	if (ztest_opts.zo_verbose >= 6) {
3458 		(void) printf("Expanding LUN %s from %lu to %lu\n",
3459 		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3460 	}
3461 
3462 	/*
3463 	 * Growing the vdev is a two step process:
3464 	 *	1). expand the physical size (i.e. relabel)
3465 	 *	2). online the vdev to create the new metaslabs
3466 	 */
3467 	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3468 	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3469 	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3470 		if (ztest_opts.zo_verbose >= 5) {
3471 			(void) printf("Could not expand LUN because "
3472 			    "the vdev configuration changed.\n");
3473 		}
3474 		spa_config_exit(spa, SCL_STATE, spa);
3475 		mutex_exit(&ztest_vdev_lock);
3476 		mutex_exit(&ztest_checkpoint_lock);
3477 		return;
3478 	}
3479 
3480 	spa_config_exit(spa, SCL_STATE, spa);
3481 
3482 	/*
3483 	 * Expanding the LUN will update the config asynchronously,
3484 	 * thus we must wait for the async thread to complete any
3485 	 * pending tasks before proceeding.
3486 	 */
3487 	for (;;) {
3488 		boolean_t done;
3489 		mutex_enter(&spa->spa_async_lock);
3490 		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3491 		mutex_exit(&spa->spa_async_lock);
3492 		if (done)
3493 			break;
3494 		txg_wait_synced(spa_get_dsl(spa), 0);
3495 		(void) poll(NULL, 0, 100);
3496 	}
3497 
3498 	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3499 
3500 	tvd = spa->spa_root_vdev->vdev_child[top];
3501 	new_ms_count = tvd->vdev_ms_count;
3502 	new_class_space = metaslab_class_get_space(mc);
3503 
3504 	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3505 		if (ztest_opts.zo_verbose >= 5) {
3506 			(void) printf("Could not verify LUN expansion due to "
3507 			    "intervening vdev offline or remove.\n");
3508 		}
3509 		spa_config_exit(spa, SCL_STATE, spa);
3510 		mutex_exit(&ztest_vdev_lock);
3511 		mutex_exit(&ztest_checkpoint_lock);
3512 		return;
3513 	}
3514 
3515 	/*
3516 	 * Make sure we were able to grow the vdev.
3517 	 */
3518 	if (new_ms_count <= old_ms_count) {
3519 		fatal(0, "LUN expansion failed: ms_count %llu < %llu\n",
3520 		    old_ms_count, new_ms_count);
3521 	}
3522 
3523 	/*
3524 	 * Make sure we were able to grow the pool.
3525 	 */
3526 	if (new_class_space <= old_class_space) {
3527 		fatal(0, "LUN expansion failed: class_space %llu < %llu\n",
3528 		    old_class_space, new_class_space);
3529 	}
3530 
3531 	if (ztest_opts.zo_verbose >= 5) {
3532 		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3533 
3534 		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3535 		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3536 		(void) printf("%s grew from %s to %s\n",
3537 		    spa->spa_name, oldnumbuf, newnumbuf);
3538 	}
3539 
3540 	spa_config_exit(spa, SCL_STATE, spa);
3541 	mutex_exit(&ztest_vdev_lock);
3542 	mutex_exit(&ztest_checkpoint_lock);
3543 }
3544 
3545 /*
3546  * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3547  */
3548 /* ARGSUSED */
3549 static void
3550 ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3551 {
3552 	/*
3553 	 * Create the objects common to all ztest datasets.
3554 	 */
3555 	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3556 	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3557 }
3558 
3559 static int
3560 ztest_dataset_create(char *dsname)
3561 {
3562 	uint64_t zilset = ztest_random(100);
3563 	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3564 	    ztest_objset_create_cb, NULL);
3565 
3566 	if (err || zilset < 80)
3567 		return (err);
3568 
3569 	if (ztest_opts.zo_verbose >= 6)
3570 		(void) printf("Setting dataset %s to sync always\n", dsname);
3571 	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3572 	    ZFS_SYNC_ALWAYS, B_FALSE));
3573 }
3574 
3575 /* ARGSUSED */
3576 static int
3577 ztest_objset_destroy_cb(const char *name, void *arg)
3578 {
3579 	objset_t *os;
3580 	dmu_object_info_t doi;
3581 	int error;
3582 
3583 	/*
3584 	 * Verify that the dataset contains a directory object.
3585 	 */
3586 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3587 	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3588 	if (error != ENOENT) {
3589 		/* We could have crashed in the middle of destroying it */
3590 		ASSERT0(error);
3591 		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3592 		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3593 	}
3594 	dmu_objset_disown(os, FTAG);
3595 
3596 	/*
3597 	 * Destroy the dataset.
3598 	 */
3599 	if (strchr(name, '@') != NULL) {
3600 		VERIFY0(dsl_destroy_snapshot(name, B_TRUE));
3601 	} else {
3602 		error = dsl_destroy_head(name);
3603 		/* There could be a hold on this dataset */
3604 		if (error != EBUSY)
3605 			ASSERT0(error);
3606 	}
3607 	return (0);
3608 }
3609 
3610 static boolean_t
3611 ztest_snapshot_create(char *osname, uint64_t id)
3612 {
3613 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3614 	int error;
3615 
3616 	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3617 
3618 	error = dmu_objset_snapshot_one(osname, snapname);
3619 	if (error == ENOSPC) {
3620 		ztest_record_enospc(FTAG);
3621 		return (B_FALSE);
3622 	}
3623 	if (error != 0 && error != EEXIST) {
3624 		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3625 		    snapname, error);
3626 	}
3627 	return (B_TRUE);
3628 }
3629 
3630 static boolean_t
3631 ztest_snapshot_destroy(char *osname, uint64_t id)
3632 {
3633 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3634 	int error;
3635 
3636 	(void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3637 	    (u_longlong_t)id);
3638 
3639 	error = dsl_destroy_snapshot(snapname, B_FALSE);
3640 	if (error != 0 && error != ENOENT)
3641 		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3642 	return (B_TRUE);
3643 }
3644 
3645 /* ARGSUSED */
3646 void
3647 ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3648 {
3649 	ztest_ds_t zdtmp;
3650 	int iters;
3651 	int error;
3652 	objset_t *os, *os2;
3653 	char name[ZFS_MAX_DATASET_NAME_LEN];
3654 	zilog_t *zilog;
3655 
3656 	rw_enter(&ztest_name_lock, RW_READER);
3657 
3658 	(void) snprintf(name, sizeof (name), "%s/temp_%llu",
3659 	    ztest_opts.zo_pool, (u_longlong_t)id);
3660 
3661 	/*
3662 	 * If this dataset exists from a previous run, process its replay log
3663 	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3664 	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3665 	 */
3666 	if (ztest_random(2) == 0 &&
3667 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3668 		ztest_zd_init(&zdtmp, NULL, os);
3669 		zil_replay(os, &zdtmp, ztest_replay_vector);
3670 		ztest_zd_fini(&zdtmp);
3671 		dmu_objset_disown(os, FTAG);
3672 	}
3673 
3674 	/*
3675 	 * There may be an old instance of the dataset we're about to
3676 	 * create lying around from a previous run.  If so, destroy it
3677 	 * and all of its snapshots.
3678 	 */
3679 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3680 	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3681 
3682 	/*
3683 	 * Verify that the destroyed dataset is no longer in the namespace.
3684 	 */
3685 	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3686 	    FTAG, &os));
3687 
3688 	/*
3689 	 * Verify that we can create a new dataset.
3690 	 */
3691 	error = ztest_dataset_create(name);
3692 	if (error) {
3693 		if (error == ENOSPC) {
3694 			ztest_record_enospc(FTAG);
3695 			rw_exit(&ztest_name_lock);
3696 			return;
3697 		}
3698 		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3699 	}
3700 
3701 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3702 
3703 	ztest_zd_init(&zdtmp, NULL, os);
3704 
3705 	/*
3706 	 * Open the intent log for it.
3707 	 */
3708 	zilog = zil_open(os, ztest_get_data);
3709 
3710 	/*
3711 	 * Put some objects in there, do a little I/O to them,
3712 	 * and randomly take a couple of snapshots along the way.
3713 	 */
3714 	iters = ztest_random(5);
3715 	for (int i = 0; i < iters; i++) {
3716 		ztest_dmu_object_alloc_free(&zdtmp, id);
3717 		if (ztest_random(iters) == 0)
3718 			(void) ztest_snapshot_create(name, i);
3719 	}
3720 
3721 	/*
3722 	 * Verify that we cannot create an existing dataset.
3723 	 */
3724 	VERIFY3U(EEXIST, ==,
3725 	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3726 
3727 	/*
3728 	 * Verify that we can hold an objset that is also owned.
3729 	 */
3730 	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3731 	dmu_objset_rele(os2, FTAG);
3732 
3733 	/*
3734 	 * Verify that we cannot own an objset that is already owned.
3735 	 */
3736 	VERIFY3U(EBUSY, ==,
3737 	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3738 
3739 	zil_close(zilog);
3740 	dmu_objset_disown(os, FTAG);
3741 	ztest_zd_fini(&zdtmp);
3742 
3743 	rw_exit(&ztest_name_lock);
3744 }
3745 
3746 /*
3747  * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3748  */
3749 void
3750 ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3751 {
3752 	rw_enter(&ztest_name_lock, RW_READER);
3753 	(void) ztest_snapshot_destroy(zd->zd_name, id);
3754 	(void) ztest_snapshot_create(zd->zd_name, id);
3755 	rw_exit(&ztest_name_lock);
3756 }
3757 
3758 /*
3759  * Cleanup non-standard snapshots and clones.
3760  */
3761 void
3762 ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3763 {
3764 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3765 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3766 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3767 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3768 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3769 	int error;
3770 
3771 	(void) snprintf(snap1name, sizeof (snap1name),
3772 	    "%s@s1_%llu", osname, id);
3773 	(void) snprintf(clone1name, sizeof (clone1name),
3774 	    "%s/c1_%llu", osname, id);
3775 	(void) snprintf(snap2name, sizeof (snap2name),
3776 	    "%s@s2_%llu", clone1name, id);
3777 	(void) snprintf(clone2name, sizeof (clone2name),
3778 	    "%s/c2_%llu", osname, id);
3779 	(void) snprintf(snap3name, sizeof (snap3name),
3780 	    "%s@s3_%llu", clone1name, id);
3781 
3782 	error = dsl_destroy_head(clone2name);
3783 	if (error && error != ENOENT)
3784 		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3785 	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3786 	if (error && error != ENOENT)
3787 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3788 	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3789 	if (error && error != ENOENT)
3790 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3791 	error = dsl_destroy_head(clone1name);
3792 	if (error && error != ENOENT)
3793 		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3794 	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3795 	if (error && error != ENOENT)
3796 		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3797 }
3798 
3799 /*
3800  * Verify dsl_dataset_promote handles EBUSY
3801  */
3802 void
3803 ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3804 {
3805 	objset_t *os;
3806 	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3807 	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3808 	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3809 	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3810 	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3811 	char *osname = zd->zd_name;
3812 	int error;
3813 
3814 	rw_enter(&ztest_name_lock, RW_READER);
3815 
3816 	ztest_dsl_dataset_cleanup(osname, id);
3817 
3818 	(void) snprintf(snap1name, sizeof (snap1name),
3819 	    "%s@s1_%llu", osname, id);
3820 	(void) snprintf(clone1name, sizeof (clone1name),
3821 	    "%s/c1_%llu", osname, id);
3822 	(void) snprintf(snap2name, sizeof (snap2name),
3823 	    "%s@s2_%llu", clone1name, id);
3824 	(void) snprintf(clone2name, sizeof (clone2name),
3825 	    "%s/c2_%llu", osname, id);
3826 	(void) snprintf(snap3name, sizeof (snap3name),
3827 	    "%s@s3_%llu", clone1name, id);
3828 
3829 	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3830 	if (error && error != EEXIST) {
3831 		if (error == ENOSPC) {
3832 			ztest_record_enospc(FTAG);
3833 			goto out;
3834 		}
3835 		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3836 	}
3837 
3838 	error = dmu_objset_clone(clone1name, snap1name);
3839 	if (error) {
3840 		if (error == ENOSPC) {
3841 			ztest_record_enospc(FTAG);
3842 			goto out;
3843 		}
3844 		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3845 	}
3846 
3847 	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
3848 	if (error && error != EEXIST) {
3849 		if (error == ENOSPC) {
3850 			ztest_record_enospc(FTAG);
3851 			goto out;
3852 		}
3853 		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
3854 	}
3855 
3856 	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
3857 	if (error && error != EEXIST) {
3858 		if (error == ENOSPC) {
3859 			ztest_record_enospc(FTAG);
3860 			goto out;
3861 		}
3862 		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
3863 	}
3864 
3865 	error = dmu_objset_clone(clone2name, snap3name);
3866 	if (error) {
3867 		if (error == ENOSPC) {
3868 			ztest_record_enospc(FTAG);
3869 			goto out;
3870 		}
3871 		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
3872 	}
3873 
3874 	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
3875 	if (error)
3876 		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
3877 	error = dsl_dataset_promote(clone2name, NULL);
3878 	if (error == ENOSPC) {
3879 		dmu_objset_disown(os, FTAG);
3880 		ztest_record_enospc(FTAG);
3881 		goto out;
3882 	}
3883 	if (error != EBUSY)
3884 		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
3885 		    error);
3886 	dmu_objset_disown(os, FTAG);
3887 
3888 out:
3889 	ztest_dsl_dataset_cleanup(osname, id);
3890 
3891 	rw_exit(&ztest_name_lock);
3892 }
3893 
3894 /*
3895  * Verify that dmu_object_{alloc,free} work as expected.
3896  */
3897 void
3898 ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
3899 {
3900 	ztest_od_t od[4];
3901 	int batchsize = sizeof (od) / sizeof (od[0]);
3902 
3903 	for (int b = 0; b < batchsize; b++) {
3904 		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER,
3905 		    0, 0, 0);
3906 	}
3907 
3908 	/*
3909 	 * Destroy the previous batch of objects, create a new batch,
3910 	 * and do some I/O on the new objects.
3911 	 */
3912 	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
3913 		return;
3914 
3915 	while (ztest_random(4 * batchsize) != 0)
3916 		ztest_io(zd, od[ztest_random(batchsize)].od_object,
3917 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
3918 }
3919 
3920 /*
3921  * Rewind the global allocator to verify object allocation backfilling.
3922  */
3923 void
3924 ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
3925 {
3926 	objset_t *os = zd->zd_os;
3927 	int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
3928 	uint64_t object;
3929 
3930 	/*
3931 	 * Rewind the global allocator randomly back to a lower object number
3932 	 * to force backfilling and reclamation of recently freed dnodes.
3933 	 */
3934 	mutex_enter(&os->os_obj_lock);
3935 	object = ztest_random(os->os_obj_next_chunk);
3936 	os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
3937 	mutex_exit(&os->os_obj_lock);
3938 }
3939 
3940 /*
3941  * Verify that dmu_{read,write} work as expected.
3942  */
3943 void
3944 ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
3945 {
3946 	objset_t *os = zd->zd_os;
3947 	ztest_od_t od[2];
3948 	dmu_tx_t *tx;
3949 	int i, freeit, error;
3950 	uint64_t n, s, txg;
3951 	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
3952 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
3953 	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
3954 	uint64_t regions = 997;
3955 	uint64_t stride = 123456789ULL;
3956 	uint64_t width = 40;
3957 	int free_percent = 5;
3958 
3959 	/*
3960 	 * This test uses two objects, packobj and bigobj, that are always
3961 	 * updated together (i.e. in the same tx) so that their contents are
3962 	 * in sync and can be compared.  Their contents relate to each other
3963 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
3964 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
3965 	 * for any index n, there are three bufwads that should be identical:
3966 	 *
3967 	 *	packobj, at offset n * sizeof (bufwad_t)
3968 	 *	bigobj, at the head of the nth chunk
3969 	 *	bigobj, at the tail of the nth chunk
3970 	 *
3971 	 * The chunk size is arbitrary. It doesn't have to be a power of two,
3972 	 * and it doesn't have any relation to the object blocksize.
3973 	 * The only requirement is that it can hold at least two bufwads.
3974 	 *
3975 	 * Normally, we write the bufwad to each of these locations.
3976 	 * However, free_percent of the time we instead write zeroes to
3977 	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
3978 	 * bigobj to packobj, we can verify that the DMU is correctly
3979 	 * tracking which parts of an object are allocated and free,
3980 	 * and that the contents of the allocated blocks are correct.
3981 	 */
3982 
3983 	/*
3984 	 * Read the directory info.  If it's the first time, set things up.
3985 	 */
3986 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0,
3987 	    chunksize);
3988 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
3989 	    chunksize);
3990 
3991 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
3992 		return;
3993 
3994 	bigobj = od[0].od_object;
3995 	packobj = od[1].od_object;
3996 	chunksize = od[0].od_gen;
3997 	ASSERT(chunksize == od[1].od_gen);
3998 
3999 	/*
4000 	 * Prefetch a random chunk of the big object.
4001 	 * Our aim here is to get some async reads in flight
4002 	 * for blocks that we may free below; the DMU should
4003 	 * handle this race correctly.
4004 	 */
4005 	n = ztest_random(regions) * stride + ztest_random(width);
4006 	s = 1 + ztest_random(2 * width - 1);
4007 	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4008 	    ZIO_PRIORITY_SYNC_READ);
4009 
4010 	/*
4011 	 * Pick a random index and compute the offsets into packobj and bigobj.
4012 	 */
4013 	n = ztest_random(regions) * stride + ztest_random(width);
4014 	s = 1 + ztest_random(width - 1);
4015 
4016 	packoff = n * sizeof (bufwad_t);
4017 	packsize = s * sizeof (bufwad_t);
4018 
4019 	bigoff = n * chunksize;
4020 	bigsize = s * chunksize;
4021 
4022 	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4023 	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4024 
4025 	/*
4026 	 * free_percent of the time, free a range of bigobj rather than
4027 	 * overwriting it.
4028 	 */
4029 	freeit = (ztest_random(100) < free_percent);
4030 
4031 	/*
4032 	 * Read the current contents of our objects.
4033 	 */
4034 	error = dmu_read(os, packobj, packoff, packsize, packbuf,
4035 	    DMU_READ_PREFETCH);
4036 	ASSERT0(error);
4037 	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4038 	    DMU_READ_PREFETCH);
4039 	ASSERT0(error);
4040 
4041 	/*
4042 	 * Get a tx for the mods to both packobj and bigobj.
4043 	 */
4044 	tx = dmu_tx_create(os);
4045 
4046 	dmu_tx_hold_write(tx, packobj, packoff, packsize);
4047 
4048 	if (freeit)
4049 		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4050 	else
4051 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4052 
4053 	/* This accounts for setting the checksum/compression. */
4054 	dmu_tx_hold_bonus(tx, bigobj);
4055 
4056 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4057 	if (txg == 0) {
4058 		umem_free(packbuf, packsize);
4059 		umem_free(bigbuf, bigsize);
4060 		return;
4061 	}
4062 
4063 	enum zio_checksum cksum;
4064 	do {
4065 		cksum = (enum zio_checksum)
4066 		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4067 	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4068 	dmu_object_set_checksum(os, bigobj, cksum, tx);
4069 
4070 	enum zio_compress comp;
4071 	do {
4072 		comp = (enum zio_compress)
4073 		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4074 	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4075 	dmu_object_set_compress(os, bigobj, comp, tx);
4076 
4077 	/*
4078 	 * For each index from n to n + s, verify that the existing bufwad
4079 	 * in packobj matches the bufwads at the head and tail of the
4080 	 * corresponding chunk in bigobj.  Then update all three bufwads
4081 	 * with the new values we want to write out.
4082 	 */
4083 	for (i = 0; i < s; i++) {
4084 		/* LINTED */
4085 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4086 		/* LINTED */
4087 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4088 		/* LINTED */
4089 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4090 
4091 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4092 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4093 
4094 		if (pack->bw_txg > txg)
4095 			fatal(0, "future leak: got %llx, open txg is %llx",
4096 			    pack->bw_txg, txg);
4097 
4098 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4099 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4100 			    pack->bw_index, n, i);
4101 
4102 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4103 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4104 
4105 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4106 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4107 
4108 		if (freeit) {
4109 			bzero(pack, sizeof (bufwad_t));
4110 		} else {
4111 			pack->bw_index = n + i;
4112 			pack->bw_txg = txg;
4113 			pack->bw_data = 1 + ztest_random(-2ULL);
4114 		}
4115 		*bigH = *pack;
4116 		*bigT = *pack;
4117 	}
4118 
4119 	/*
4120 	 * We've verified all the old bufwads, and made new ones.
4121 	 * Now write them out.
4122 	 */
4123 	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4124 
4125 	if (freeit) {
4126 		if (ztest_opts.zo_verbose >= 7) {
4127 			(void) printf("freeing offset %llx size %llx"
4128 			    " txg %llx\n",
4129 			    (u_longlong_t)bigoff,
4130 			    (u_longlong_t)bigsize,
4131 			    (u_longlong_t)txg);
4132 		}
4133 		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4134 	} else {
4135 		if (ztest_opts.zo_verbose >= 7) {
4136 			(void) printf("writing offset %llx size %llx"
4137 			    " txg %llx\n",
4138 			    (u_longlong_t)bigoff,
4139 			    (u_longlong_t)bigsize,
4140 			    (u_longlong_t)txg);
4141 		}
4142 		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4143 	}
4144 
4145 	dmu_tx_commit(tx);
4146 
4147 	/*
4148 	 * Sanity check the stuff we just wrote.
4149 	 */
4150 	{
4151 		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4152 		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4153 
4154 		VERIFY(0 == dmu_read(os, packobj, packoff,
4155 		    packsize, packcheck, DMU_READ_PREFETCH));
4156 		VERIFY(0 == dmu_read(os, bigobj, bigoff,
4157 		    bigsize, bigcheck, DMU_READ_PREFETCH));
4158 
4159 		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4160 		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4161 
4162 		umem_free(packcheck, packsize);
4163 		umem_free(bigcheck, bigsize);
4164 	}
4165 
4166 	umem_free(packbuf, packsize);
4167 	umem_free(bigbuf, bigsize);
4168 }
4169 
4170 void
4171 compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4172     uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4173 {
4174 	uint64_t i;
4175 	bufwad_t *pack;
4176 	bufwad_t *bigH;
4177 	bufwad_t *bigT;
4178 
4179 	/*
4180 	 * For each index from n to n + s, verify that the existing bufwad
4181 	 * in packobj matches the bufwads at the head and tail of the
4182 	 * corresponding chunk in bigobj.  Then update all three bufwads
4183 	 * with the new values we want to write out.
4184 	 */
4185 	for (i = 0; i < s; i++) {
4186 		/* LINTED */
4187 		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4188 		/* LINTED */
4189 		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4190 		/* LINTED */
4191 		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4192 
4193 		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4194 		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4195 
4196 		if (pack->bw_txg > txg)
4197 			fatal(0, "future leak: got %llx, open txg is %llx",
4198 			    pack->bw_txg, txg);
4199 
4200 		if (pack->bw_data != 0 && pack->bw_index != n + i)
4201 			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4202 			    pack->bw_index, n, i);
4203 
4204 		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4205 			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4206 
4207 		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4208 			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4209 
4210 		pack->bw_index = n + i;
4211 		pack->bw_txg = txg;
4212 		pack->bw_data = 1 + ztest_random(-2ULL);
4213 
4214 		*bigH = *pack;
4215 		*bigT = *pack;
4216 	}
4217 }
4218 
4219 void
4220 ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4221 {
4222 	objset_t *os = zd->zd_os;
4223 	ztest_od_t od[2];
4224 	dmu_tx_t *tx;
4225 	uint64_t i;
4226 	int error;
4227 	uint64_t n, s, txg;
4228 	bufwad_t *packbuf, *bigbuf;
4229 	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4230 	uint64_t blocksize = ztest_random_blocksize();
4231 	uint64_t chunksize = blocksize;
4232 	uint64_t regions = 997;
4233 	uint64_t stride = 123456789ULL;
4234 	uint64_t width = 9;
4235 	dmu_buf_t *bonus_db;
4236 	arc_buf_t **bigbuf_arcbufs;
4237 	dmu_object_info_t doi;
4238 
4239 	/*
4240 	 * This test uses two objects, packobj and bigobj, that are always
4241 	 * updated together (i.e. in the same tx) so that their contents are
4242 	 * in sync and can be compared.  Their contents relate to each other
4243 	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4244 	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4245 	 * for any index n, there are three bufwads that should be identical:
4246 	 *
4247 	 *	packobj, at offset n * sizeof (bufwad_t)
4248 	 *	bigobj, at the head of the nth chunk
4249 	 *	bigobj, at the tail of the nth chunk
4250 	 *
4251 	 * The chunk size is set equal to bigobj block size so that
4252 	 * dmu_assign_arcbuf() can be tested for object updates.
4253 	 */
4254 
4255 	/*
4256 	 * Read the directory info.  If it's the first time, set things up.
4257 	 */
4258 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
4259 	    0, 0);
4260 	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4261 	    chunksize);
4262 
4263 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4264 		return;
4265 
4266 	bigobj = od[0].od_object;
4267 	packobj = od[1].od_object;
4268 	blocksize = od[0].od_blocksize;
4269 	chunksize = blocksize;
4270 	ASSERT(chunksize == od[1].od_gen);
4271 
4272 	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4273 	VERIFY(ISP2(doi.doi_data_block_size));
4274 	VERIFY(chunksize == doi.doi_data_block_size);
4275 	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4276 
4277 	/*
4278 	 * Pick a random index and compute the offsets into packobj and bigobj.
4279 	 */
4280 	n = ztest_random(regions) * stride + ztest_random(width);
4281 	s = 1 + ztest_random(width - 1);
4282 
4283 	packoff = n * sizeof (bufwad_t);
4284 	packsize = s * sizeof (bufwad_t);
4285 
4286 	bigoff = n * chunksize;
4287 	bigsize = s * chunksize;
4288 
4289 	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4290 	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4291 
4292 	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4293 
4294 	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4295 
4296 	/*
4297 	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4298 	 * Iteration 1 test zcopy to already referenced dbufs.
4299 	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4300 	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4301 	 * Iteration 4 test zcopy when dbuf is no longer dirty.
4302 	 * Iteration 5 test zcopy when it can't be done.
4303 	 * Iteration 6 one more zcopy write.
4304 	 */
4305 	for (i = 0; i < 7; i++) {
4306 		uint64_t j;
4307 		uint64_t off;
4308 
4309 		/*
4310 		 * In iteration 5 (i == 5) use arcbufs
4311 		 * that don't match bigobj blksz to test
4312 		 * dmu_assign_arcbuf() when it can't directly
4313 		 * assign an arcbuf to a dbuf.
4314 		 */
4315 		for (j = 0; j < s; j++) {
4316 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4317 				bigbuf_arcbufs[j] =
4318 				    dmu_request_arcbuf(bonus_db, chunksize);
4319 			} else {
4320 				bigbuf_arcbufs[2 * j] =
4321 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
4322 				bigbuf_arcbufs[2 * j + 1] =
4323 				    dmu_request_arcbuf(bonus_db, chunksize / 2);
4324 			}
4325 		}
4326 
4327 		/*
4328 		 * Get a tx for the mods to both packobj and bigobj.
4329 		 */
4330 		tx = dmu_tx_create(os);
4331 
4332 		dmu_tx_hold_write(tx, packobj, packoff, packsize);
4333 		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4334 
4335 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4336 		if (txg == 0) {
4337 			umem_free(packbuf, packsize);
4338 			umem_free(bigbuf, bigsize);
4339 			for (j = 0; j < s; j++) {
4340 				if (i != 5 ||
4341 				    chunksize < (SPA_MINBLOCKSIZE * 2)) {
4342 					dmu_return_arcbuf(bigbuf_arcbufs[j]);
4343 				} else {
4344 					dmu_return_arcbuf(
4345 					    bigbuf_arcbufs[2 * j]);
4346 					dmu_return_arcbuf(
4347 					    bigbuf_arcbufs[2 * j + 1]);
4348 				}
4349 			}
4350 			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4351 			dmu_buf_rele(bonus_db, FTAG);
4352 			return;
4353 		}
4354 
4355 		/*
4356 		 * 50% of the time don't read objects in the 1st iteration to
4357 		 * test dmu_assign_arcbuf() for the case when there're no
4358 		 * existing dbufs for the specified offsets.
4359 		 */
4360 		if (i != 0 || ztest_random(2) != 0) {
4361 			error = dmu_read(os, packobj, packoff,
4362 			    packsize, packbuf, DMU_READ_PREFETCH);
4363 			ASSERT0(error);
4364 			error = dmu_read(os, bigobj, bigoff, bigsize,
4365 			    bigbuf, DMU_READ_PREFETCH);
4366 			ASSERT0(error);
4367 		}
4368 		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4369 		    n, chunksize, txg);
4370 
4371 		/*
4372 		 * We've verified all the old bufwads, and made new ones.
4373 		 * Now write them out.
4374 		 */
4375 		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4376 		if (ztest_opts.zo_verbose >= 7) {
4377 			(void) printf("writing offset %llx size %llx"
4378 			    " txg %llx\n",
4379 			    (u_longlong_t)bigoff,
4380 			    (u_longlong_t)bigsize,
4381 			    (u_longlong_t)txg);
4382 		}
4383 		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4384 			dmu_buf_t *dbt;
4385 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4386 				bcopy((caddr_t)bigbuf + (off - bigoff),
4387 				    bigbuf_arcbufs[j]->b_data, chunksize);
4388 			} else {
4389 				bcopy((caddr_t)bigbuf + (off - bigoff),
4390 				    bigbuf_arcbufs[2 * j]->b_data,
4391 				    chunksize / 2);
4392 				bcopy((caddr_t)bigbuf + (off - bigoff) +
4393 				    chunksize / 2,
4394 				    bigbuf_arcbufs[2 * j + 1]->b_data,
4395 				    chunksize / 2);
4396 			}
4397 
4398 			if (i == 1) {
4399 				VERIFY(dmu_buf_hold(os, bigobj, off,
4400 				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4401 			}
4402 			if (i != 5 || chunksize < (SPA_MINBLOCKSIZE * 2)) {
4403 				dmu_assign_arcbuf(bonus_db, off,
4404 				    bigbuf_arcbufs[j], tx);
4405 			} else {
4406 				dmu_assign_arcbuf(bonus_db, off,
4407 				    bigbuf_arcbufs[2 * j], tx);
4408 				dmu_assign_arcbuf(bonus_db,
4409 				    off + chunksize / 2,
4410 				    bigbuf_arcbufs[2 * j + 1], tx);
4411 			}
4412 			if (i == 1) {
4413 				dmu_buf_rele(dbt, FTAG);
4414 			}
4415 		}
4416 		dmu_tx_commit(tx);
4417 
4418 		/*
4419 		 * Sanity check the stuff we just wrote.
4420 		 */
4421 		{
4422 			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4423 			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4424 
4425 			VERIFY(0 == dmu_read(os, packobj, packoff,
4426 			    packsize, packcheck, DMU_READ_PREFETCH));
4427 			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4428 			    bigsize, bigcheck, DMU_READ_PREFETCH));
4429 
4430 			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4431 			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4432 
4433 			umem_free(packcheck, packsize);
4434 			umem_free(bigcheck, bigsize);
4435 		}
4436 		if (i == 2) {
4437 			txg_wait_open(dmu_objset_pool(os), 0);
4438 		} else if (i == 3) {
4439 			txg_wait_synced(dmu_objset_pool(os), 0);
4440 		}
4441 	}
4442 
4443 	dmu_buf_rele(bonus_db, FTAG);
4444 	umem_free(packbuf, packsize);
4445 	umem_free(bigbuf, bigsize);
4446 	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4447 }
4448 
4449 /* ARGSUSED */
4450 void
4451 ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4452 {
4453 	ztest_od_t od[1];
4454 	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4455 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4456 
4457 	/*
4458 	 * Have multiple threads write to large offsets in an object
4459 	 * to verify that parallel writes to an object -- even to the
4460 	 * same blocks within the object -- doesn't cause any trouble.
4461 	 */
4462 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER,
4463 	    0, 0, 0);
4464 
4465 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4466 		return;
4467 
4468 	while (ztest_random(10) != 0)
4469 		ztest_io(zd, od[0].od_object, offset);
4470 }
4471 
4472 void
4473 ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4474 {
4475 	ztest_od_t od[1];
4476 	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4477 	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4478 	uint64_t count = ztest_random(20) + 1;
4479 	uint64_t blocksize = ztest_random_blocksize();
4480 	void *data;
4481 
4482 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
4483 	    0, 0);
4484 
4485 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4486 		return;
4487 
4488 	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4489 		return;
4490 
4491 	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4492 
4493 	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4494 
4495 	while (ztest_random(count) != 0) {
4496 		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4497 		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4498 		    data) != 0)
4499 			break;
4500 		while (ztest_random(4) != 0)
4501 			ztest_io(zd, od[0].od_object, randoff);
4502 	}
4503 
4504 	umem_free(data, blocksize);
4505 }
4506 
4507 /*
4508  * Verify that zap_{create,destroy,add,remove,update} work as expected.
4509  */
4510 #define	ZTEST_ZAP_MIN_INTS	1
4511 #define	ZTEST_ZAP_MAX_INTS	4
4512 #define	ZTEST_ZAP_MAX_PROPS	1000
4513 
4514 void
4515 ztest_zap(ztest_ds_t *zd, uint64_t id)
4516 {
4517 	objset_t *os = zd->zd_os;
4518 	ztest_od_t od[1];
4519 	uint64_t object;
4520 	uint64_t txg, last_txg;
4521 	uint64_t value[ZTEST_ZAP_MAX_INTS];
4522 	uint64_t zl_ints, zl_intsize, prop;
4523 	int i, ints;
4524 	dmu_tx_t *tx;
4525 	char propname[100], txgname[100];
4526 	int error;
4527 	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4528 
4529 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4530 
4531 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4532 		return;
4533 
4534 	object = od[0].od_object;
4535 
4536 	/*
4537 	 * Generate a known hash collision, and verify that
4538 	 * we can lookup and remove both entries.
4539 	 */
4540 	tx = dmu_tx_create(os);
4541 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4542 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4543 	if (txg == 0)
4544 		return;
4545 	for (i = 0; i < 2; i++) {
4546 		value[i] = i;
4547 		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4548 		    1, &value[i], tx));
4549 	}
4550 	for (i = 0; i < 2; i++) {
4551 		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4552 		    sizeof (uint64_t), 1, &value[i], tx));
4553 		VERIFY3U(0, ==,
4554 		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4555 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4556 		ASSERT3U(zl_ints, ==, 1);
4557 	}
4558 	for (i = 0; i < 2; i++) {
4559 		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4560 	}
4561 	dmu_tx_commit(tx);
4562 
4563 	/*
4564 	 * Generate a buch of random entries.
4565 	 */
4566 	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4567 
4568 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4569 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4570 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4571 	bzero(value, sizeof (value));
4572 	last_txg = 0;
4573 
4574 	/*
4575 	 * If these zap entries already exist, validate their contents.
4576 	 */
4577 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4578 	if (error == 0) {
4579 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4580 		ASSERT3U(zl_ints, ==, 1);
4581 
4582 		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4583 		    zl_ints, &last_txg) == 0);
4584 
4585 		VERIFY(zap_length(os, object, propname, &zl_intsize,
4586 		    &zl_ints) == 0);
4587 
4588 		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4589 		ASSERT3U(zl_ints, ==, ints);
4590 
4591 		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4592 		    zl_ints, value) == 0);
4593 
4594 		for (i = 0; i < ints; i++) {
4595 			ASSERT3U(value[i], ==, last_txg + object + i);
4596 		}
4597 	} else {
4598 		ASSERT3U(error, ==, ENOENT);
4599 	}
4600 
4601 	/*
4602 	 * Atomically update two entries in our zap object.
4603 	 * The first is named txg_%llu, and contains the txg
4604 	 * in which the property was last updated.  The second
4605 	 * is named prop_%llu, and the nth element of its value
4606 	 * should be txg + object + n.
4607 	 */
4608 	tx = dmu_tx_create(os);
4609 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4610 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4611 	if (txg == 0)
4612 		return;
4613 
4614 	if (last_txg > txg)
4615 		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4616 
4617 	for (i = 0; i < ints; i++)
4618 		value[i] = txg + object + i;
4619 
4620 	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4621 	    1, &txg, tx));
4622 	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4623 	    ints, value, tx));
4624 
4625 	dmu_tx_commit(tx);
4626 
4627 	/*
4628 	 * Remove a random pair of entries.
4629 	 */
4630 	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4631 	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4632 	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4633 
4634 	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4635 
4636 	if (error == ENOENT)
4637 		return;
4638 
4639 	ASSERT0(error);
4640 
4641 	tx = dmu_tx_create(os);
4642 	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4643 	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4644 	if (txg == 0)
4645 		return;
4646 	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4647 	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4648 	dmu_tx_commit(tx);
4649 }
4650 
4651 /*
4652  * Testcase to test the upgrading of a microzap to fatzap.
4653  */
4654 void
4655 ztest_fzap(ztest_ds_t *zd, uint64_t id)
4656 {
4657 	objset_t *os = zd->zd_os;
4658 	ztest_od_t od[1];
4659 	uint64_t object, txg;
4660 
4661 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4662 
4663 	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4664 		return;
4665 
4666 	object = od[0].od_object;
4667 
4668 	/*
4669 	 * Add entries to this ZAP and make sure it spills over
4670 	 * and gets upgraded to a fatzap. Also, since we are adding
4671 	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4672 	 */
4673 	for (int i = 0; i < 2050; i++) {
4674 		char name[ZFS_MAX_DATASET_NAME_LEN];
4675 		uint64_t value = i;
4676 		dmu_tx_t *tx;
4677 		int error;
4678 
4679 		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4680 		    id, value);
4681 
4682 		tx = dmu_tx_create(os);
4683 		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4684 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4685 		if (txg == 0)
4686 			return;
4687 		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4688 		    &value, tx);
4689 		ASSERT(error == 0 || error == EEXIST);
4690 		dmu_tx_commit(tx);
4691 	}
4692 }
4693 
4694 /* ARGSUSED */
4695 void
4696 ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4697 {
4698 	objset_t *os = zd->zd_os;
4699 	ztest_od_t od[1];
4700 	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4701 	dmu_tx_t *tx;
4702 	int i, namelen, error;
4703 	int micro = ztest_random(2);
4704 	char name[20], string_value[20];
4705 	void *data;
4706 
4707 	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER,
4708 	    0, 0, 0);
4709 
4710 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4711 		return;
4712 
4713 	object = od[0].od_object;
4714 
4715 	/*
4716 	 * Generate a random name of the form 'xxx.....' where each
4717 	 * x is a random printable character and the dots are dots.
4718 	 * There are 94 such characters, and the name length goes from
4719 	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4720 	 */
4721 	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4722 
4723 	for (i = 0; i < 3; i++)
4724 		name[i] = '!' + ztest_random('~' - '!' + 1);
4725 	for (; i < namelen - 1; i++)
4726 		name[i] = '.';
4727 	name[i] = '\0';
4728 
4729 	if ((namelen & 1) || micro) {
4730 		wsize = sizeof (txg);
4731 		wc = 1;
4732 		data = &txg;
4733 	} else {
4734 		wsize = 1;
4735 		wc = namelen;
4736 		data = string_value;
4737 	}
4738 
4739 	count = -1ULL;
4740 	VERIFY0(zap_count(os, object, &count));
4741 	ASSERT(count != -1ULL);
4742 
4743 	/*
4744 	 * Select an operation: length, lookup, add, update, remove.
4745 	 */
4746 	i = ztest_random(5);
4747 
4748 	if (i >= 2) {
4749 		tx = dmu_tx_create(os);
4750 		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4751 		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4752 		if (txg == 0)
4753 			return;
4754 		bcopy(name, string_value, namelen);
4755 	} else {
4756 		tx = NULL;
4757 		txg = 0;
4758 		bzero(string_value, namelen);
4759 	}
4760 
4761 	switch (i) {
4762 
4763 	case 0:
4764 		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4765 		if (error == 0) {
4766 			ASSERT3U(wsize, ==, zl_wsize);
4767 			ASSERT3U(wc, ==, zl_wc);
4768 		} else {
4769 			ASSERT3U(error, ==, ENOENT);
4770 		}
4771 		break;
4772 
4773 	case 1:
4774 		error = zap_lookup(os, object, name, wsize, wc, data);
4775 		if (error == 0) {
4776 			if (data == string_value &&
4777 			    bcmp(name, data, namelen) != 0)
4778 				fatal(0, "name '%s' != val '%s' len %d",
4779 				    name, data, namelen);
4780 		} else {
4781 			ASSERT3U(error, ==, ENOENT);
4782 		}
4783 		break;
4784 
4785 	case 2:
4786 		error = zap_add(os, object, name, wsize, wc, data, tx);
4787 		ASSERT(error == 0 || error == EEXIST);
4788 		break;
4789 
4790 	case 3:
4791 		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4792 		break;
4793 
4794 	case 4:
4795 		error = zap_remove(os, object, name, tx);
4796 		ASSERT(error == 0 || error == ENOENT);
4797 		break;
4798 	}
4799 
4800 	if (tx != NULL)
4801 		dmu_tx_commit(tx);
4802 }
4803 
4804 /*
4805  * Commit callback data.
4806  */
4807 typedef struct ztest_cb_data {
4808 	list_node_t		zcd_node;
4809 	uint64_t		zcd_txg;
4810 	int			zcd_expected_err;
4811 	boolean_t		zcd_added;
4812 	boolean_t		zcd_called;
4813 	spa_t			*zcd_spa;
4814 } ztest_cb_data_t;
4815 
4816 /* This is the actual commit callback function */
4817 static void
4818 ztest_commit_callback(void *arg, int error)
4819 {
4820 	ztest_cb_data_t *data = arg;
4821 	uint64_t synced_txg;
4822 
4823 	VERIFY(data != NULL);
4824 	VERIFY3S(data->zcd_expected_err, ==, error);
4825 	VERIFY(!data->zcd_called);
4826 
4827 	synced_txg = spa_last_synced_txg(data->zcd_spa);
4828 	if (data->zcd_txg > synced_txg)
4829 		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4830 		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4831 		    synced_txg);
4832 
4833 	data->zcd_called = B_TRUE;
4834 
4835 	if (error == ECANCELED) {
4836 		ASSERT0(data->zcd_txg);
4837 		ASSERT(!data->zcd_added);
4838 
4839 		/*
4840 		 * The private callback data should be destroyed here, but
4841 		 * since we are going to check the zcd_called field after
4842 		 * dmu_tx_abort(), we will destroy it there.
4843 		 */
4844 		return;
4845 	}
4846 
4847 	/* Was this callback added to the global callback list? */
4848 	if (!data->zcd_added)
4849 		goto out;
4850 
4851 	ASSERT3U(data->zcd_txg, !=, 0);
4852 
4853 	/* Remove our callback from the list */
4854 	mutex_enter(&zcl.zcl_callbacks_lock);
4855 	list_remove(&zcl.zcl_callbacks, data);
4856 	mutex_exit(&zcl.zcl_callbacks_lock);
4857 
4858 out:
4859 	umem_free(data, sizeof (ztest_cb_data_t));
4860 }
4861 
4862 /* Allocate and initialize callback data structure */
4863 static ztest_cb_data_t *
4864 ztest_create_cb_data(objset_t *os, uint64_t txg)
4865 {
4866 	ztest_cb_data_t *cb_data;
4867 
4868 	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
4869 
4870 	cb_data->zcd_txg = txg;
4871 	cb_data->zcd_spa = dmu_objset_spa(os);
4872 
4873 	return (cb_data);
4874 }
4875 
4876 /*
4877  * If a number of txgs equal to this threshold have been created after a commit
4878  * callback has been registered but not called, then we assume there is an
4879  * implementation bug.
4880  */
4881 #define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
4882 
4883 /*
4884  * Commit callback test.
4885  */
4886 void
4887 ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
4888 {
4889 	objset_t *os = zd->zd_os;
4890 	ztest_od_t od[1];
4891 	dmu_tx_t *tx;
4892 	ztest_cb_data_t *cb_data[3], *tmp_cb;
4893 	uint64_t old_txg, txg;
4894 	int i, error;
4895 
4896 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
4897 
4898 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4899 		return;
4900 
4901 	tx = dmu_tx_create(os);
4902 
4903 	cb_data[0] = ztest_create_cb_data(os, 0);
4904 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
4905 
4906 	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
4907 
4908 	/* Every once in a while, abort the transaction on purpose */
4909 	if (ztest_random(100) == 0)
4910 		error = -1;
4911 
4912 	if (!error)
4913 		error = dmu_tx_assign(tx, TXG_NOWAIT);
4914 
4915 	txg = error ? 0 : dmu_tx_get_txg(tx);
4916 
4917 	cb_data[0]->zcd_txg = txg;
4918 	cb_data[1] = ztest_create_cb_data(os, txg);
4919 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
4920 
4921 	if (error) {
4922 		/*
4923 		 * It's not a strict requirement to call the registered
4924 		 * callbacks from inside dmu_tx_abort(), but that's what
4925 		 * it's supposed to happen in the current implementation
4926 		 * so we will check for that.
4927 		 */
4928 		for (i = 0; i < 2; i++) {
4929 			cb_data[i]->zcd_expected_err = ECANCELED;
4930 			VERIFY(!cb_data[i]->zcd_called);
4931 		}
4932 
4933 		dmu_tx_abort(tx);
4934 
4935 		for (i = 0; i < 2; i++) {
4936 			VERIFY(cb_data[i]->zcd_called);
4937 			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
4938 		}
4939 
4940 		return;
4941 	}
4942 
4943 	cb_data[2] = ztest_create_cb_data(os, txg);
4944 	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
4945 
4946 	/*
4947 	 * Read existing data to make sure there isn't a future leak.
4948 	 */
4949 	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
4950 	    &old_txg, DMU_READ_PREFETCH));
4951 
4952 	if (old_txg > txg)
4953 		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
4954 		    old_txg, txg);
4955 
4956 	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
4957 
4958 	mutex_enter(&zcl.zcl_callbacks_lock);
4959 
4960 	/*
4961 	 * Since commit callbacks don't have any ordering requirement and since
4962 	 * it is theoretically possible for a commit callback to be called
4963 	 * after an arbitrary amount of time has elapsed since its txg has been
4964 	 * synced, it is difficult to reliably determine whether a commit
4965 	 * callback hasn't been called due to high load or due to a flawed
4966 	 * implementation.
4967 	 *
4968 	 * In practice, we will assume that if after a certain number of txgs a
4969 	 * commit callback hasn't been called, then most likely there's an
4970 	 * implementation bug..
4971 	 */
4972 	tmp_cb = list_head(&zcl.zcl_callbacks);
4973 	if (tmp_cb != NULL &&
4974 	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
4975 		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
4976 		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
4977 	}
4978 
4979 	/*
4980 	 * Let's find the place to insert our callbacks.
4981 	 *
4982 	 * Even though the list is ordered by txg, it is possible for the
4983 	 * insertion point to not be the end because our txg may already be
4984 	 * quiescing at this point and other callbacks in the open txg
4985 	 * (from other objsets) may have sneaked in.
4986 	 */
4987 	tmp_cb = list_tail(&zcl.zcl_callbacks);
4988 	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
4989 		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
4990 
4991 	/* Add the 3 callbacks to the list */
4992 	for (i = 0; i < 3; i++) {
4993 		if (tmp_cb == NULL)
4994 			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
4995 		else
4996 			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
4997 			    cb_data[i]);
4998 
4999 		cb_data[i]->zcd_added = B_TRUE;
5000 		VERIFY(!cb_data[i]->zcd_called);
5001 
5002 		tmp_cb = cb_data[i];
5003 	}
5004 
5005 	mutex_exit(&zcl.zcl_callbacks_lock);
5006 
5007 	dmu_tx_commit(tx);
5008 }
5009 
5010 /*
5011  * Visit each object in the dataset. Verify that its properties
5012  * are consistent what was stored in the block tag when it was created,
5013  * and that its unused bonus buffer space has not been overwritten.
5014  */
5015 void
5016 ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5017 {
5018 	objset_t *os = zd->zd_os;
5019 	uint64_t obj;
5020 	int err = 0;
5021 
5022 	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5023 		ztest_block_tag_t *bt = NULL;
5024 		dmu_object_info_t doi;
5025 		dmu_buf_t *db;
5026 
5027 		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0)
5028 			continue;
5029 
5030 		dmu_object_info_from_db(db, &doi);
5031 		if (doi.doi_bonus_size >= sizeof (*bt))
5032 			bt = ztest_bt_bonus(db);
5033 
5034 		if (bt && bt->bt_magic == BT_MAGIC) {
5035 			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5036 			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
5037 			    bt->bt_crtxg);
5038 			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5039 		}
5040 
5041 		dmu_buf_rele(db, FTAG);
5042 	}
5043 }
5044 
5045 /* ARGSUSED */
5046 void
5047 ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5048 {
5049 	zfs_prop_t proplist[] = {
5050 		ZFS_PROP_CHECKSUM,
5051 		ZFS_PROP_COMPRESSION,
5052 		ZFS_PROP_COPIES,
5053 		ZFS_PROP_DEDUP
5054 	};
5055 
5056 	rw_enter(&ztest_name_lock, RW_READER);
5057 
5058 	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5059 		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5060 		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5061 
5062 	rw_exit(&ztest_name_lock);
5063 }
5064 
5065 /* ARGSUSED */
5066 void
5067 ztest_remap_blocks(ztest_ds_t *zd, uint64_t id)
5068 {
5069 	rw_enter(&ztest_name_lock, RW_READER);
5070 
5071 	int error = dmu_objset_remap_indirects(zd->zd_name);
5072 	if (error == ENOSPC)
5073 		error = 0;
5074 	ASSERT0(error);
5075 
5076 	rw_exit(&ztest_name_lock);
5077 }
5078 
5079 /* ARGSUSED */
5080 void
5081 ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5082 {
5083 	nvlist_t *props = NULL;
5084 
5085 	rw_enter(&ztest_name_lock, RW_READER);
5086 
5087 	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
5088 	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
5089 
5090 	VERIFY0(spa_prop_get(ztest_spa, &props));
5091 
5092 	if (ztest_opts.zo_verbose >= 6)
5093 		dump_nvlist(props, 4);
5094 
5095 	nvlist_free(props);
5096 
5097 	rw_exit(&ztest_name_lock);
5098 }
5099 
5100 static int
5101 user_release_one(const char *snapname, const char *holdname)
5102 {
5103 	nvlist_t *snaps, *holds;
5104 	int error;
5105 
5106 	snaps = fnvlist_alloc();
5107 	holds = fnvlist_alloc();
5108 	fnvlist_add_boolean(holds, holdname);
5109 	fnvlist_add_nvlist(snaps, snapname, holds);
5110 	fnvlist_free(holds);
5111 	error = dsl_dataset_user_release(snaps, NULL);
5112 	fnvlist_free(snaps);
5113 	return (error);
5114 }
5115 
5116 /*
5117  * Test snapshot hold/release and deferred destroy.
5118  */
5119 void
5120 ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5121 {
5122 	int error;
5123 	objset_t *os = zd->zd_os;
5124 	objset_t *origin;
5125 	char snapname[100];
5126 	char fullname[100];
5127 	char clonename[100];
5128 	char tag[100];
5129 	char osname[ZFS_MAX_DATASET_NAME_LEN];
5130 	nvlist_t *holds;
5131 
5132 	rw_enter(&ztest_name_lock, RW_READER);
5133 
5134 	dmu_objset_name(os, osname);
5135 
5136 	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
5137 	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5138 	(void) snprintf(clonename, sizeof (clonename),
5139 	    "%s/ch1_%llu", osname, id);
5140 	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
5141 
5142 	/*
5143 	 * Clean up from any previous run.
5144 	 */
5145 	error = dsl_destroy_head(clonename);
5146 	if (error != ENOENT)
5147 		ASSERT0(error);
5148 	error = user_release_one(fullname, tag);
5149 	if (error != ESRCH && error != ENOENT)
5150 		ASSERT0(error);
5151 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5152 	if (error != ENOENT)
5153 		ASSERT0(error);
5154 
5155 	/*
5156 	 * Create snapshot, clone it, mark snap for deferred destroy,
5157 	 * destroy clone, verify snap was also destroyed.
5158 	 */
5159 	error = dmu_objset_snapshot_one(osname, snapname);
5160 	if (error) {
5161 		if (error == ENOSPC) {
5162 			ztest_record_enospc("dmu_objset_snapshot");
5163 			goto out;
5164 		}
5165 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5166 	}
5167 
5168 	error = dmu_objset_clone(clonename, fullname);
5169 	if (error) {
5170 		if (error == ENOSPC) {
5171 			ztest_record_enospc("dmu_objset_clone");
5172 			goto out;
5173 		}
5174 		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
5175 	}
5176 
5177 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5178 	if (error) {
5179 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5180 		    fullname, error);
5181 	}
5182 
5183 	error = dsl_destroy_head(clonename);
5184 	if (error)
5185 		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
5186 
5187 	error = dmu_objset_hold(fullname, FTAG, &origin);
5188 	if (error != ENOENT)
5189 		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
5190 
5191 	/*
5192 	 * Create snapshot, add temporary hold, verify that we can't
5193 	 * destroy a held snapshot, mark for deferred destroy,
5194 	 * release hold, verify snapshot was destroyed.
5195 	 */
5196 	error = dmu_objset_snapshot_one(osname, snapname);
5197 	if (error) {
5198 		if (error == ENOSPC) {
5199 			ztest_record_enospc("dmu_objset_snapshot");
5200 			goto out;
5201 		}
5202 		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5203 	}
5204 
5205 	holds = fnvlist_alloc();
5206 	fnvlist_add_string(holds, fullname, tag);
5207 	error = dsl_dataset_user_hold(holds, 0, NULL);
5208 	fnvlist_free(holds);
5209 
5210 	if (error == ENOSPC) {
5211 		ztest_record_enospc("dsl_dataset_user_hold");
5212 		goto out;
5213 	} else if (error) {
5214 		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5215 		    fullname, tag, error);
5216 	}
5217 
5218 	error = dsl_destroy_snapshot(fullname, B_FALSE);
5219 	if (error != EBUSY) {
5220 		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5221 		    fullname, error);
5222 	}
5223 
5224 	error = dsl_destroy_snapshot(fullname, B_TRUE);
5225 	if (error) {
5226 		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5227 		    fullname, error);
5228 	}
5229 
5230 	error = user_release_one(fullname, tag);
5231 	if (error)
5232 		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
5233 
5234 	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5235 
5236 out:
5237 	rw_exit(&ztest_name_lock);
5238 }
5239 
5240 /*
5241  * Inject random faults into the on-disk data.
5242  */
5243 /* ARGSUSED */
5244 void
5245 ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5246 {
5247 	ztest_shared_t *zs = ztest_shared;
5248 	spa_t *spa = ztest_spa;
5249 	int fd;
5250 	uint64_t offset;
5251 	uint64_t leaves;
5252 	uint64_t bad = 0x1990c0ffeedecade;
5253 	uint64_t top, leaf;
5254 	char path0[MAXPATHLEN];
5255 	char pathrand[MAXPATHLEN];
5256 	size_t fsize;
5257 	int bshift = SPA_MAXBLOCKSHIFT + 2;
5258 	int iters = 1000;
5259 	int maxfaults;
5260 	int mirror_save;
5261 	vdev_t *vd0 = NULL;
5262 	uint64_t guid0 = 0;
5263 	boolean_t islog = B_FALSE;
5264 
5265 	mutex_enter(&ztest_vdev_lock);
5266 
5267 	/*
5268 	 * Device removal is in progress, fault injection must be disabled
5269 	 * until it completes and the pool is scrubbed.  The fault injection
5270 	 * strategy for damaging blocks does not take in to account evacuated
5271 	 * blocks which may have already been damaged.
5272 	 */
5273 	if (ztest_device_removal_active) {
5274 		mutex_exit(&ztest_vdev_lock);
5275 		return;
5276 	}
5277 
5278 	maxfaults = MAXFAULTS();
5279 	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5280 	mirror_save = zs->zs_mirrors;
5281 	mutex_exit(&ztest_vdev_lock);
5282 
5283 	ASSERT(leaves >= 1);
5284 
5285 	/*
5286 	 * Grab the name lock as reader. There are some operations
5287 	 * which don't like to have their vdevs changed while
5288 	 * they are in progress (i.e. spa_change_guid). Those
5289 	 * operations will have grabbed the name lock as writer.
5290 	 */
5291 	rw_enter(&ztest_name_lock, RW_READER);
5292 
5293 	/*
5294 	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5295 	 */
5296 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5297 
5298 	if (ztest_random(2) == 0) {
5299 		/*
5300 		 * Inject errors on a normal data device or slog device.
5301 		 */
5302 		top = ztest_random_vdev_top(spa, B_TRUE);
5303 		leaf = ztest_random(leaves) + zs->zs_splits;
5304 
5305 		/*
5306 		 * Generate paths to the first leaf in this top-level vdev,
5307 		 * and to the random leaf we selected.  We'll induce transient
5308 		 * write failures and random online/offline activity on leaf 0,
5309 		 * and we'll write random garbage to the randomly chosen leaf.
5310 		 */
5311 		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
5312 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
5313 		    top * leaves + zs->zs_splits);
5314 		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
5315 		    ztest_opts.zo_dir, ztest_opts.zo_pool,
5316 		    top * leaves + leaf);
5317 
5318 		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5319 		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5320 			islog = B_TRUE;
5321 
5322 		/*
5323 		 * If the top-level vdev needs to be resilvered
5324 		 * then we only allow faults on the device that is
5325 		 * resilvering.
5326 		 */
5327 		if (vd0 != NULL && maxfaults != 1 &&
5328 		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5329 		    vd0->vdev_resilver_txg != 0)) {
5330 			/*
5331 			 * Make vd0 explicitly claim to be unreadable,
5332 			 * or unwriteable, or reach behind its back
5333 			 * and close the underlying fd.  We can do this if
5334 			 * maxfaults == 0 because we'll fail and reexecute,
5335 			 * and we can do it if maxfaults >= 2 because we'll
5336 			 * have enough redundancy.  If maxfaults == 1, the
5337 			 * combination of this with injection of random data
5338 			 * corruption below exceeds the pool's fault tolerance.
5339 			 */
5340 			vdev_file_t *vf = vd0->vdev_tsd;
5341 
5342 			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
5343 			    (long long)vd0->vdev_id, (int)maxfaults);
5344 
5345 			if (vf != NULL && ztest_random(3) == 0) {
5346 				(void) close(vf->vf_vnode->v_fd);
5347 				vf->vf_vnode->v_fd = -1;
5348 			} else if (ztest_random(2) == 0) {
5349 				vd0->vdev_cant_read = B_TRUE;
5350 			} else {
5351 				vd0->vdev_cant_write = B_TRUE;
5352 			}
5353 			guid0 = vd0->vdev_guid;
5354 		}
5355 	} else {
5356 		/*
5357 		 * Inject errors on an l2cache device.
5358 		 */
5359 		spa_aux_vdev_t *sav = &spa->spa_l2cache;
5360 
5361 		if (sav->sav_count == 0) {
5362 			spa_config_exit(spa, SCL_STATE, FTAG);
5363 			rw_exit(&ztest_name_lock);
5364 			return;
5365 		}
5366 		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5367 		guid0 = vd0->vdev_guid;
5368 		(void) strcpy(path0, vd0->vdev_path);
5369 		(void) strcpy(pathrand, vd0->vdev_path);
5370 
5371 		leaf = 0;
5372 		leaves = 1;
5373 		maxfaults = INT_MAX;	/* no limit on cache devices */
5374 	}
5375 
5376 	spa_config_exit(spa, SCL_STATE, FTAG);
5377 	rw_exit(&ztest_name_lock);
5378 
5379 	/*
5380 	 * If we can tolerate two or more faults, or we're dealing
5381 	 * with a slog, randomly online/offline vd0.
5382 	 */
5383 	if ((maxfaults >= 2 || islog) && guid0 != 0) {
5384 		if (ztest_random(10) < 6) {
5385 			int flags = (ztest_random(2) == 0 ?
5386 			    ZFS_OFFLINE_TEMPORARY : 0);
5387 
5388 			/*
5389 			 * We have to grab the zs_name_lock as writer to
5390 			 * prevent a race between offlining a slog and
5391 			 * destroying a dataset. Offlining the slog will
5392 			 * grab a reference on the dataset which may cause
5393 			 * dmu_objset_destroy() to fail with EBUSY thus
5394 			 * leaving the dataset in an inconsistent state.
5395 			 */
5396 			if (islog)
5397 				rw_enter(&ztest_name_lock, RW_WRITER);
5398 
5399 			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5400 
5401 			if (islog)
5402 				rw_exit(&ztest_name_lock);
5403 		} else {
5404 			/*
5405 			 * Ideally we would like to be able to randomly
5406 			 * call vdev_[on|off]line without holding locks
5407 			 * to force unpredictable failures but the side
5408 			 * effects of vdev_[on|off]line prevent us from
5409 			 * doing so. We grab the ztest_vdev_lock here to
5410 			 * prevent a race between injection testing and
5411 			 * aux_vdev removal.
5412 			 */
5413 			mutex_enter(&ztest_vdev_lock);
5414 			(void) vdev_online(spa, guid0, 0, NULL);
5415 			mutex_exit(&ztest_vdev_lock);
5416 		}
5417 	}
5418 
5419 	if (maxfaults == 0)
5420 		return;
5421 
5422 	/*
5423 	 * We have at least single-fault tolerance, so inject data corruption.
5424 	 */
5425 	fd = open(pathrand, O_RDWR);
5426 
5427 	if (fd == -1) /* we hit a gap in the device namespace */
5428 		return;
5429 
5430 	fsize = lseek(fd, 0, SEEK_END);
5431 
5432 	while (--iters != 0) {
5433 		/*
5434 		 * The offset must be chosen carefully to ensure that
5435 		 * we do not inject a given logical block with errors
5436 		 * on two different leaf devices, because ZFS can not
5437 		 * tolerate that (if maxfaults==1).
5438 		 *
5439 		 * We divide each leaf into chunks of size
5440 		 * (# leaves * SPA_MAXBLOCKSIZE * 4).  Within each chunk
5441 		 * there is a series of ranges to which we can inject errors.
5442 		 * Each range can accept errors on only a single leaf vdev.
5443 		 * The error injection ranges are separated by ranges
5444 		 * which we will not inject errors on any device (DMZs).
5445 		 * Each DMZ must be large enough such that a single block
5446 		 * can not straddle it, so that a single block can not be
5447 		 * a target in two different injection ranges (on different
5448 		 * leaf vdevs).
5449 		 *
5450 		 * For example, with 3 leaves, each chunk looks like:
5451 		 *    0 to  32M: injection range for leaf 0
5452 		 *  32M to  64M: DMZ - no injection allowed
5453 		 *  64M to  96M: injection range for leaf 1
5454 		 *  96M to 128M: DMZ - no injection allowed
5455 		 * 128M to 160M: injection range for leaf 2
5456 		 * 160M to 192M: DMZ - no injection allowed
5457 		 */
5458 		offset = ztest_random(fsize / (leaves << bshift)) *
5459 		    (leaves << bshift) + (leaf << bshift) +
5460 		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5461 
5462 		/*
5463 		 * Only allow damage to the labels at one end of the vdev.
5464 		 *
5465 		 * If all labels are damaged, the device will be totally
5466 		 * inaccessible, which will result in loss of data,
5467 		 * because we also damage (parts of) the other side of
5468 		 * the mirror/raidz.
5469 		 *
5470 		 * Additionally, we will always have both an even and an
5471 		 * odd label, so that we can handle crashes in the
5472 		 * middle of vdev_config_sync().
5473 		 */
5474 		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5475 			continue;
5476 
5477 		/*
5478 		 * The two end labels are stored at the "end" of the disk, but
5479 		 * the end of the disk (vdev_psize) is aligned to
5480 		 * sizeof (vdev_label_t).
5481 		 */
5482 		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5483 		if ((leaf & 1) == 1 &&
5484 		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5485 			continue;
5486 
5487 		mutex_enter(&ztest_vdev_lock);
5488 		if (mirror_save != zs->zs_mirrors) {
5489 			mutex_exit(&ztest_vdev_lock);
5490 			(void) close(fd);
5491 			return;
5492 		}
5493 
5494 		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5495 			fatal(1, "can't inject bad word at 0x%llx in %s",
5496 			    offset, pathrand);
5497 
5498 		mutex_exit(&ztest_vdev_lock);
5499 
5500 		if (ztest_opts.zo_verbose >= 7)
5501 			(void) printf("injected bad word into %s,"
5502 			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5503 	}
5504 
5505 	(void) close(fd);
5506 }
5507 
5508 /*
5509  * Verify that DDT repair works as expected.
5510  */
5511 void
5512 ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5513 {
5514 	ztest_shared_t *zs = ztest_shared;
5515 	spa_t *spa = ztest_spa;
5516 	objset_t *os = zd->zd_os;
5517 	ztest_od_t od[1];
5518 	uint64_t object, blocksize, txg, pattern, psize;
5519 	enum zio_checksum checksum = spa_dedup_checksum(spa);
5520 	dmu_buf_t *db;
5521 	dmu_tx_t *tx;
5522 	abd_t *abd;
5523 	blkptr_t blk;
5524 	int copies = 2 * ZIO_DEDUPDITTO_MIN;
5525 
5526 	blocksize = ztest_random_blocksize();
5527 	blocksize = MIN(blocksize, 2048);	/* because we write so many */
5528 
5529 	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
5530 	    0, 0);
5531 
5532 	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5533 		return;
5534 
5535 	/*
5536 	 * Take the name lock as writer to prevent anyone else from changing
5537 	 * the pool and dataset properies we need to maintain during this test.
5538 	 */
5539 	rw_enter(&ztest_name_lock, RW_WRITER);
5540 
5541 	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5542 	    B_FALSE) != 0 ||
5543 	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5544 	    B_FALSE) != 0) {
5545 		rw_exit(&ztest_name_lock);
5546 		return;
5547 	}
5548 
5549 	dmu_objset_stats_t dds;
5550 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5551 	dmu_objset_fast_stat(os, &dds);
5552 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5553 
5554 	object = od[0].od_object;
5555 	blocksize = od[0].od_blocksize;
5556 	pattern = zs->zs_guid ^ dds.dds_guid;
5557 
5558 	ASSERT(object != 0);
5559 
5560 	tx = dmu_tx_create(os);
5561 	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5562 	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5563 	if (txg == 0) {
5564 		rw_exit(&ztest_name_lock);
5565 		return;
5566 	}
5567 
5568 	/*
5569 	 * Write all the copies of our block.
5570 	 */
5571 	for (int i = 0; i < copies; i++) {
5572 		uint64_t offset = i * blocksize;
5573 		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5574 		    DMU_READ_NO_PREFETCH);
5575 		if (error != 0) {
5576 			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5577 			    os, (long long)object, (long long) offset, error);
5578 		}
5579 		ASSERT(db->db_offset == offset);
5580 		ASSERT(db->db_size == blocksize);
5581 		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5582 		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5583 		dmu_buf_will_fill(db, tx);
5584 		ztest_pattern_set(db->db_data, db->db_size, pattern);
5585 		dmu_buf_rele(db, FTAG);
5586 	}
5587 
5588 	dmu_tx_commit(tx);
5589 	txg_wait_synced(spa_get_dsl(spa), txg);
5590 
5591 	/*
5592 	 * Find out what block we got.
5593 	 */
5594 	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5595 	    DMU_READ_NO_PREFETCH));
5596 	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5597 	dmu_buf_rele(db, FTAG);
5598 
5599 	/*
5600 	 * Damage the block.  Dedup-ditto will save us when we read it later.
5601 	 */
5602 	psize = BP_GET_PSIZE(&blk);
5603 	abd = abd_alloc_linear(psize, B_TRUE);
5604 	ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5605 
5606 	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5607 	    abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5608 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5609 
5610 	abd_free(abd);
5611 
5612 	rw_exit(&ztest_name_lock);
5613 }
5614 
5615 /*
5616  * Scrub the pool.
5617  */
5618 /* ARGSUSED */
5619 void
5620 ztest_scrub(ztest_ds_t *zd, uint64_t id)
5621 {
5622 	spa_t *spa = ztest_spa;
5623 
5624 	/*
5625 	 * Scrub in progress by device removal.
5626 	 */
5627 	if (ztest_device_removal_active)
5628 		return;
5629 
5630 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5631 	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5632 	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5633 }
5634 
5635 /*
5636  * Change the guid for the pool.
5637  */
5638 /* ARGSUSED */
5639 void
5640 ztest_reguid(ztest_ds_t *zd, uint64_t id)
5641 {
5642 	spa_t *spa = ztest_spa;
5643 	uint64_t orig, load;
5644 	int error;
5645 
5646 	if (ztest_opts.zo_mmp_test)
5647 		return;
5648 
5649 	orig = spa_guid(spa);
5650 	load = spa_load_guid(spa);
5651 
5652 	rw_enter(&ztest_name_lock, RW_WRITER);
5653 	error = spa_change_guid(spa);
5654 	rw_exit(&ztest_name_lock);
5655 
5656 	if (error != 0)
5657 		return;
5658 
5659 	if (ztest_opts.zo_verbose >= 4) {
5660 		(void) printf("Changed guid old %llu -> %llu\n",
5661 		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5662 	}
5663 
5664 	VERIFY3U(orig, !=, spa_guid(spa));
5665 	VERIFY3U(load, ==, spa_load_guid(spa));
5666 }
5667 
5668 static vdev_t *
5669 ztest_random_concrete_vdev_leaf(vdev_t *vd)
5670 {
5671 	if (vd == NULL)
5672 		return (NULL);
5673 
5674 	if (vd->vdev_children == 0)
5675 		return (vd);
5676 
5677 	vdev_t *eligible[vd->vdev_children];
5678 	int eligible_idx = 0, i;
5679 	for (i = 0; i < vd->vdev_children; i++) {
5680 		vdev_t *cvd = vd->vdev_child[i];
5681 		if (cvd->vdev_top->vdev_removing)
5682 			continue;
5683 		if (cvd->vdev_children > 0 ||
5684 		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
5685 			eligible[eligible_idx++] = cvd;
5686 		}
5687 	}
5688 	VERIFY(eligible_idx > 0);
5689 
5690 	uint64_t child_no = ztest_random(eligible_idx);
5691 	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
5692 }
5693 
5694 /* ARGSUSED */
5695 void
5696 ztest_initialize(ztest_ds_t *zd, uint64_t id)
5697 {
5698 	spa_t *spa = ztest_spa;
5699 	int error = 0;
5700 
5701 	mutex_enter(&ztest_vdev_lock);
5702 
5703 	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
5704 
5705 	/* Random leaf vdev */
5706 	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
5707 	if (rand_vd == NULL) {
5708 		spa_config_exit(spa, SCL_VDEV, FTAG);
5709 		mutex_exit(&ztest_vdev_lock);
5710 		return;
5711 	}
5712 
5713 	/*
5714 	 * The random vdev we've selected may change as soon as we
5715 	 * drop the spa_config_lock. We create local copies of things
5716 	 * we're interested in.
5717 	 */
5718 	uint64_t guid = rand_vd->vdev_guid;
5719 	char *path = strdup(rand_vd->vdev_path);
5720 	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
5721 
5722 	zfs_dbgmsg("vd %p, guid %llu", rand_vd, guid);
5723 	spa_config_exit(spa, SCL_VDEV, FTAG);
5724 
5725 	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
5726 	error = spa_vdev_initialize(spa, guid, cmd);
5727 	switch (cmd) {
5728 	case POOL_INITIALIZE_CANCEL:
5729 		if (ztest_opts.zo_verbose >= 4) {
5730 			(void) printf("Cancel initialize %s", path);
5731 			if (!active)
5732 				(void) printf(" failed (no initialize active)");
5733 			(void) printf("\n");
5734 		}
5735 		break;
5736 	case POOL_INITIALIZE_DO:
5737 		if (ztest_opts.zo_verbose >= 4) {
5738 			(void) printf("Start initialize %s", path);
5739 			if (active && error == 0)
5740 				(void) printf(" failed (already active)");
5741 			else if (error != 0)
5742 				(void) printf(" failed (error %d)", error);
5743 			(void) printf("\n");
5744 		}
5745 		break;
5746 	case POOL_INITIALIZE_SUSPEND:
5747 		if (ztest_opts.zo_verbose >= 4) {
5748 			(void) printf("Suspend initialize %s", path);
5749 			if (!active)
5750 				(void) printf(" failed (no initialize active)");
5751 			(void) printf("\n");
5752 		}
5753 		break;
5754 	}
5755 	free(path);
5756 	mutex_exit(&ztest_vdev_lock);
5757 }
5758 
5759 /*
5760  * Verify pool integrity by running zdb.
5761  */
5762 static void
5763 ztest_run_zdb(char *pool)
5764 {
5765 	int status;
5766 	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5767 	char zbuf[1024];
5768 	char *bin;
5769 	char *ztest;
5770 	char *isa;
5771 	int isalen;
5772 	FILE *fp;
5773 
5774 	(void) realpath(getexecname(), zdb);
5775 
5776 	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5777 	bin = strstr(zdb, "/usr/bin/");
5778 	ztest = strstr(bin, "/ztest");
5779 	isa = bin + 8;
5780 	isalen = ztest - isa;
5781 	isa = strdup(isa);
5782 	/* LINTED */
5783 	(void) sprintf(bin,
5784 	    "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s "
5785 	    "-o zfs_reconstruct_indirect_combinations_max=65536 %s",
5786 	    isalen,
5787 	    isa,
5788 	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5789 	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5790 	    spa_config_path,
5791 	    pool);
5792 	free(isa);
5793 
5794 	if (ztest_opts.zo_verbose >= 5)
5795 		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5796 
5797 	fp = popen(zdb, "r");
5798 
5799 	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5800 		if (ztest_opts.zo_verbose >= 3)
5801 			(void) printf("%s", zbuf);
5802 
5803 	status = pclose(fp);
5804 
5805 	if (status == 0)
5806 		return;
5807 
5808 	ztest_dump_core = 0;
5809 	if (WIFEXITED(status))
5810 		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5811 	else
5812 		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5813 }
5814 
5815 static void
5816 ztest_walk_pool_directory(char *header)
5817 {
5818 	spa_t *spa = NULL;
5819 
5820 	if (ztest_opts.zo_verbose >= 6)
5821 		(void) printf("%s\n", header);
5822 
5823 	mutex_enter(&spa_namespace_lock);
5824 	while ((spa = spa_next(spa)) != NULL)
5825 		if (ztest_opts.zo_verbose >= 6)
5826 			(void) printf("\t%s\n", spa_name(spa));
5827 	mutex_exit(&spa_namespace_lock);
5828 }
5829 
5830 static void
5831 ztest_spa_import_export(char *oldname, char *newname)
5832 {
5833 	nvlist_t *config, *newconfig;
5834 	uint64_t pool_guid;
5835 	spa_t *spa;
5836 	int error;
5837 
5838 	if (ztest_opts.zo_verbose >= 4) {
5839 		(void) printf("import/export: old = %s, new = %s\n",
5840 		    oldname, newname);
5841 	}
5842 
5843 	/*
5844 	 * Clean up from previous runs.
5845 	 */
5846 	(void) spa_destroy(newname);
5847 
5848 	/*
5849 	 * Get the pool's configuration and guid.
5850 	 */
5851 	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
5852 
5853 	/*
5854 	 * Kick off a scrub to tickle scrub/export races.
5855 	 */
5856 	if (ztest_random(2) == 0)
5857 		(void) spa_scan(spa, POOL_SCAN_SCRUB);
5858 
5859 	pool_guid = spa_guid(spa);
5860 	spa_close(spa, FTAG);
5861 
5862 	ztest_walk_pool_directory("pools before export");
5863 
5864 	/*
5865 	 * Export it.
5866 	 */
5867 	VERIFY3U(0, ==, spa_export(oldname, &config, B_FALSE, B_FALSE));
5868 
5869 	ztest_walk_pool_directory("pools after export");
5870 
5871 	/*
5872 	 * Try to import it.
5873 	 */
5874 	newconfig = spa_tryimport(config);
5875 	ASSERT(newconfig != NULL);
5876 	nvlist_free(newconfig);
5877 
5878 	/*
5879 	 * Import it under the new name.
5880 	 */
5881 	error = spa_import(newname, config, NULL, 0);
5882 	if (error != 0) {
5883 		dump_nvlist(config, 0);
5884 		fatal(B_FALSE, "couldn't import pool %s as %s: error %u",
5885 		    oldname, newname, error);
5886 	}
5887 
5888 	ztest_walk_pool_directory("pools after import");
5889 
5890 	/*
5891 	 * Try to import it again -- should fail with EEXIST.
5892 	 */
5893 	VERIFY3U(EEXIST, ==, spa_import(newname, config, NULL, 0));
5894 
5895 	/*
5896 	 * Try to import it under a different name -- should fail with EEXIST.
5897 	 */
5898 	VERIFY3U(EEXIST, ==, spa_import(oldname, config, NULL, 0));
5899 
5900 	/*
5901 	 * Verify that the pool is no longer visible under the old name.
5902 	 */
5903 	VERIFY3U(ENOENT, ==, spa_open(oldname, &spa, FTAG));
5904 
5905 	/*
5906 	 * Verify that we can open and close the pool using the new name.
5907 	 */
5908 	VERIFY3U(0, ==, spa_open(newname, &spa, FTAG));
5909 	ASSERT(pool_guid == spa_guid(spa));
5910 	spa_close(spa, FTAG);
5911 
5912 	nvlist_free(config);
5913 }
5914 
5915 static void
5916 ztest_resume(spa_t *spa)
5917 {
5918 	if (spa_suspended(spa) && ztest_opts.zo_verbose >= 6)
5919 		(void) printf("resuming from suspended state\n");
5920 	spa_vdev_state_enter(spa, SCL_NONE);
5921 	vdev_clear(spa, NULL);
5922 	(void) spa_vdev_state_exit(spa, NULL, 0);
5923 	(void) zio_resume(spa);
5924 }
5925 
5926 static void *
5927 ztest_resume_thread(void *arg)
5928 {
5929 	spa_t *spa = arg;
5930 
5931 	while (!ztest_exiting) {
5932 		if (spa_suspended(spa))
5933 			ztest_resume(spa);
5934 		(void) poll(NULL, 0, 100);
5935 
5936 		/*
5937 		 * Periodically change the zfs_compressed_arc_enabled setting.
5938 		 */
5939 		if (ztest_random(10) == 0)
5940 			zfs_compressed_arc_enabled = ztest_random(2);
5941 
5942 		/*
5943 		 * Periodically change the zfs_abd_scatter_enabled setting.
5944 		 */
5945 		if (ztest_random(10) == 0)
5946 			zfs_abd_scatter_enabled = ztest_random(2);
5947 	}
5948 	return (NULL);
5949 }
5950 
5951 static void *
5952 ztest_deadman_thread(void *arg)
5953 {
5954 	ztest_shared_t *zs = arg;
5955 	spa_t *spa = ztest_spa;
5956 	hrtime_t delta, total = 0;
5957 
5958 	for (;;) {
5959 		delta = zs->zs_thread_stop - zs->zs_thread_start +
5960 		    MSEC2NSEC(zfs_deadman_synctime_ms);
5961 
5962 		(void) poll(NULL, 0, (int)NSEC2MSEC(delta));
5963 
5964 		/*
5965 		 * If the pool is suspended then fail immediately. Otherwise,
5966 		 * check to see if the pool is making any progress. If
5967 		 * vdev_deadman() discovers that there hasn't been any recent
5968 		 * I/Os then it will end up aborting the tests.
5969 		 */
5970 		if (spa_suspended(spa) || spa->spa_root_vdev == NULL) {
5971 			fatal(0, "aborting test after %llu seconds because "
5972 			    "pool has transitioned to a suspended state.",
5973 			    zfs_deadman_synctime_ms / 1000);
5974 			return (NULL);
5975 		}
5976 		vdev_deadman(spa->spa_root_vdev);
5977 
5978 		total += zfs_deadman_synctime_ms/1000;
5979 		(void) printf("ztest has been running for %lld seconds\n",
5980 		    total);
5981 	}
5982 }
5983 
5984 static void
5985 ztest_execute(int test, ztest_info_t *zi, uint64_t id)
5986 {
5987 	ztest_ds_t *zd = &ztest_ds[id % ztest_opts.zo_datasets];
5988 	ztest_shared_callstate_t *zc = ZTEST_GET_SHARED_CALLSTATE(test);
5989 	hrtime_t functime = gethrtime();
5990 
5991 	for (int i = 0; i < zi->zi_iters; i++)
5992 		zi->zi_func(zd, id);
5993 
5994 	functime = gethrtime() - functime;
5995 
5996 	atomic_add_64(&zc->zc_count, 1);
5997 	atomic_add_64(&zc->zc_time, functime);
5998 
5999 	if (ztest_opts.zo_verbose >= 4) {
6000 		Dl_info dli;
6001 		(void) dladdr((void *)zi->zi_func, &dli);
6002 		(void) printf("%6.2f sec in %s\n",
6003 		    (double)functime / NANOSEC, dli.dli_sname);
6004 	}
6005 }
6006 
6007 static void *
6008 ztest_thread(void *arg)
6009 {
6010 	int rand;
6011 	uint64_t id = (uintptr_t)arg;
6012 	ztest_shared_t *zs = ztest_shared;
6013 	uint64_t call_next;
6014 	hrtime_t now;
6015 	ztest_info_t *zi;
6016 	ztest_shared_callstate_t *zc;
6017 
6018 	while ((now = gethrtime()) < zs->zs_thread_stop) {
6019 		/*
6020 		 * See if it's time to force a crash.
6021 		 */
6022 		if (now > zs->zs_thread_kill)
6023 			ztest_kill(zs);
6024 
6025 		/*
6026 		 * If we're getting ENOSPC with some regularity, stop.
6027 		 */
6028 		if (zs->zs_enospc_count > 10)
6029 			break;
6030 
6031 		/*
6032 		 * Pick a random function to execute.
6033 		 */
6034 		rand = ztest_random(ZTEST_FUNCS);
6035 		zi = &ztest_info[rand];
6036 		zc = ZTEST_GET_SHARED_CALLSTATE(rand);
6037 		call_next = zc->zc_next;
6038 
6039 		if (now >= call_next &&
6040 		    atomic_cas_64(&zc->zc_next, call_next, call_next +
6041 		    ztest_random(2 * zi->zi_interval[0] + 1)) == call_next) {
6042 			ztest_execute(rand, zi, id);
6043 		}
6044 	}
6045 
6046 	return (NULL);
6047 }
6048 
6049 static void
6050 ztest_dataset_name(char *dsname, char *pool, int d)
6051 {
6052 	(void) snprintf(dsname, ZFS_MAX_DATASET_NAME_LEN, "%s/ds_%d", pool, d);
6053 }
6054 
6055 static void
6056 ztest_dataset_destroy(int d)
6057 {
6058 	char name[ZFS_MAX_DATASET_NAME_LEN];
6059 
6060 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
6061 
6062 	if (ztest_opts.zo_verbose >= 3)
6063 		(void) printf("Destroying %s to free up space\n", name);
6064 
6065 	/*
6066 	 * Cleanup any non-standard clones and snapshots.  In general,
6067 	 * ztest thread t operates on dataset (t % zopt_datasets),
6068 	 * so there may be more than one thing to clean up.
6069 	 */
6070 	for (int t = d; t < ztest_opts.zo_threads;
6071 	    t += ztest_opts.zo_datasets) {
6072 		ztest_dsl_dataset_cleanup(name, t);
6073 	}
6074 
6075 	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
6076 	    DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
6077 }
6078 
6079 static void
6080 ztest_dataset_dirobj_verify(ztest_ds_t *zd)
6081 {
6082 	uint64_t usedobjs, dirobjs, scratch;
6083 
6084 	/*
6085 	 * ZTEST_DIROBJ is the object directory for the entire dataset.
6086 	 * Therefore, the number of objects in use should equal the
6087 	 * number of ZTEST_DIROBJ entries, +1 for ZTEST_DIROBJ itself.
6088 	 * If not, we have an object leak.
6089 	 *
6090 	 * Note that we can only check this in ztest_dataset_open(),
6091 	 * when the open-context and syncing-context values agree.
6092 	 * That's because zap_count() returns the open-context value,
6093 	 * while dmu_objset_space() returns the rootbp fill count.
6094 	 */
6095 	VERIFY3U(0, ==, zap_count(zd->zd_os, ZTEST_DIROBJ, &dirobjs));
6096 	dmu_objset_space(zd->zd_os, &scratch, &scratch, &usedobjs, &scratch);
6097 	ASSERT3U(dirobjs + 1, ==, usedobjs);
6098 }
6099 
6100 static int
6101 ztest_dataset_open(int d)
6102 {
6103 	ztest_ds_t *zd = &ztest_ds[d];
6104 	uint64_t committed_seq = ZTEST_GET_SHARED_DS(d)->zd_seq;
6105 	objset_t *os;
6106 	zilog_t *zilog;
6107 	char name[ZFS_MAX_DATASET_NAME_LEN];
6108 	int error;
6109 
6110 	ztest_dataset_name(name, ztest_opts.zo_pool, d);
6111 
6112 	rw_enter(&ztest_name_lock, RW_READER);
6113 
6114 	error = ztest_dataset_create(name);
6115 	if (error == ENOSPC) {
6116 		rw_exit(&ztest_name_lock);
6117 		ztest_record_enospc(FTAG);
6118 		return (error);
6119 	}
6120 	ASSERT(error == 0 || error == EEXIST);
6121 
6122 	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, zd, &os));
6123 	rw_exit(&ztest_name_lock);
6124 
6125 	ztest_zd_init(zd, ZTEST_GET_SHARED_DS(d), os);
6126 
6127 	zilog = zd->zd_zilog;
6128 
6129 	if (zilog->zl_header->zh_claim_lr_seq != 0 &&
6130 	    zilog->zl_header->zh_claim_lr_seq < committed_seq)
6131 		fatal(0, "missing log records: claimed %llu < committed %llu",
6132 		    zilog->zl_header->zh_claim_lr_seq, committed_seq);
6133 
6134 	ztest_dataset_dirobj_verify(zd);
6135 
6136 	zil_replay(os, zd, ztest_replay_vector);
6137 
6138 	ztest_dataset_dirobj_verify(zd);
6139 
6140 	if (ztest_opts.zo_verbose >= 6)
6141 		(void) printf("%s replay %llu blocks, %llu records, seq %llu\n",
6142 		    zd->zd_name,
6143 		    (u_longlong_t)zilog->zl_parse_blk_count,
6144 		    (u_longlong_t)zilog->zl_parse_lr_count,
6145 		    (u_longlong_t)zilog->zl_replaying_seq);
6146 
6147 	zilog = zil_open(os, ztest_get_data);
6148 
6149 	if (zilog->zl_replaying_seq != 0 &&
6150 	    zilog->zl_replaying_seq < committed_seq)
6151 		fatal(0, "missing log records: replayed %llu < committed %llu",
6152 		    zilog->zl_replaying_seq, committed_seq);
6153 
6154 	return (0);
6155 }
6156 
6157 static void
6158 ztest_dataset_close(int d)
6159 {
6160 	ztest_ds_t *zd = &ztest_ds[d];
6161 
6162 	zil_close(zd->zd_zilog);
6163 	dmu_objset_disown(zd->zd_os, zd);
6164 
6165 	ztest_zd_fini(zd);
6166 }
6167 
6168 /*
6169  * Kick off threads to run tests on all datasets in parallel.
6170  */
6171 static void
6172 ztest_run(ztest_shared_t *zs)
6173 {
6174 	thread_t *tid;
6175 	spa_t *spa;
6176 	objset_t *os;
6177 	thread_t resume_tid;
6178 	int error;
6179 
6180 	ztest_exiting = B_FALSE;
6181 
6182 	/*
6183 	 * Initialize parent/child shared state.
6184 	 */
6185 	mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);
6186 	mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
6187 	rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
6188 
6189 	zs->zs_thread_start = gethrtime();
6190 	zs->zs_thread_stop =
6191 	    zs->zs_thread_start + ztest_opts.zo_passtime * NANOSEC;
6192 	zs->zs_thread_stop = MIN(zs->zs_thread_stop, zs->zs_proc_stop);
6193 	zs->zs_thread_kill = zs->zs_thread_stop;
6194 	if (ztest_random(100) < ztest_opts.zo_killrate) {
6195 		zs->zs_thread_kill -=
6196 		    ztest_random(ztest_opts.zo_passtime * NANOSEC);
6197 	}
6198 
6199 	mutex_init(&zcl.zcl_callbacks_lock, NULL, USYNC_THREAD, NULL);
6200 
6201 	list_create(&zcl.zcl_callbacks, sizeof (ztest_cb_data_t),
6202 	    offsetof(ztest_cb_data_t, zcd_node));
6203 
6204 	/*
6205 	 * Open our pool.
6206 	 */
6207 	kernel_init(FREAD | FWRITE);
6208 	VERIFY0(spa_open(ztest_opts.zo_pool, &spa, FTAG));
6209 	metaslab_preload_limit = ztest_random(20) + 1;
6210 	ztest_spa = spa;
6211 
6212 	dmu_objset_stats_t dds;
6213 	VERIFY0(dmu_objset_own(ztest_opts.zo_pool,
6214 	    DMU_OST_ANY, B_TRUE, FTAG, &os));
6215 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
6216 	dmu_objset_fast_stat(os, &dds);
6217 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
6218 	zs->zs_guid = dds.dds_guid;
6219 	dmu_objset_disown(os, FTAG);
6220 
6221 	spa->spa_dedup_ditto = 2 * ZIO_DEDUPDITTO_MIN;
6222 
6223 	/*
6224 	 * We don't expect the pool to suspend unless maxfaults == 0,
6225 	 * in which case ztest_fault_inject() temporarily takes away
6226 	 * the only valid replica.
6227 	 */
6228 	if (MAXFAULTS() == 0)
6229 		spa->spa_failmode = ZIO_FAILURE_MODE_WAIT;
6230 	else
6231 		spa->spa_failmode = ZIO_FAILURE_MODE_PANIC;
6232 
6233 	/*
6234 	 * Create a thread to periodically resume suspended I/O.
6235 	 */
6236 	VERIFY(thr_create(0, 0, ztest_resume_thread, spa, THR_BOUND,
6237 	    &resume_tid) == 0);
6238 
6239 	/*
6240 	 * Create a deadman thread to abort() if we hang.
6241 	 */
6242 	VERIFY(thr_create(0, 0, ztest_deadman_thread, zs, THR_BOUND,
6243 	    NULL) == 0);
6244 
6245 	/*
6246 	 * Verify that we can safely inquire about any object,
6247 	 * whether it's allocated or not.  To make it interesting,
6248 	 * we probe a 5-wide window around each power of two.
6249 	 * This hits all edge cases, including zero and the max.
6250 	 */
6251 	for (int t = 0; t < 64; t++) {
6252 		for (int d = -5; d <= 5; d++) {
6253 			error = dmu_object_info(spa->spa_meta_objset,
6254 			    (1ULL << t) + d, NULL);
6255 			ASSERT(error == 0 || error == ENOENT ||
6256 			    error == EINVAL);
6257 		}
6258 	}
6259 
6260 	/*
6261 	 * If we got any ENOSPC errors on the previous run, destroy something.
6262 	 */
6263 	if (zs->zs_enospc_count != 0) {
6264 		int d = ztest_random(ztest_opts.zo_datasets);
6265 		ztest_dataset_destroy(d);
6266 	}
6267 	zs->zs_enospc_count = 0;
6268 
6269 	tid = umem_zalloc(ztest_opts.zo_threads * sizeof (thread_t),
6270 	    UMEM_NOFAIL);
6271 
6272 	if (ztest_opts.zo_verbose >= 4)
6273 		(void) printf("starting main threads...\n");
6274 
6275 	/*
6276 	 * Kick off all the tests that run in parallel.
6277 	 */
6278 	for (int t = 0; t < ztest_opts.zo_threads; t++) {
6279 		if (t < ztest_opts.zo_datasets &&
6280 		    ztest_dataset_open(t) != 0)
6281 			return;
6282 		VERIFY(thr_create(0, 0, ztest_thread, (void *)(uintptr_t)t,
6283 		    THR_BOUND, &tid[t]) == 0);
6284 	}
6285 
6286 	/*
6287 	 * Wait for all of the tests to complete.  We go in reverse order
6288 	 * so we don't close datasets while threads are still using them.
6289 	 */
6290 	for (int t = ztest_opts.zo_threads - 1; t >= 0; t--) {
6291 		VERIFY(thr_join(tid[t], NULL, NULL) == 0);
6292 		if (t < ztest_opts.zo_datasets)
6293 			ztest_dataset_close(t);
6294 	}
6295 
6296 	txg_wait_synced(spa_get_dsl(spa), 0);
6297 
6298 	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
6299 	zs->zs_space = metaslab_class_get_space(spa_normal_class(spa));
6300 	zfs_dbgmsg_print(FTAG);
6301 
6302 	umem_free(tid, ztest_opts.zo_threads * sizeof (thread_t));
6303 
6304 	/* Kill the resume thread */
6305 	ztest_exiting = B_TRUE;
6306 	VERIFY(thr_join(resume_tid, NULL, NULL) == 0);
6307 	ztest_resume(spa);
6308 
6309 	/*
6310 	 * Right before closing the pool, kick off a bunch of async I/O;
6311 	 * spa_close() should wait for it to complete.
6312 	 */
6313 	for (uint64_t object = 1; object < 50; object++) {
6314 		dmu_prefetch(spa->spa_meta_objset, object, 0, 0, 1ULL << 20,
6315 		    ZIO_PRIORITY_SYNC_READ);
6316 	}
6317 
6318 	spa_close(spa, FTAG);
6319 
6320 	/*
6321 	 * Verify that we can loop over all pools.
6322 	 */
6323 	mutex_enter(&spa_namespace_lock);
6324 	for (spa = spa_next(NULL); spa != NULL; spa = spa_next(spa))
6325 		if (ztest_opts.zo_verbose > 3)
6326 			(void) printf("spa_next: found %s\n", spa_name(spa));
6327 	mutex_exit(&spa_namespace_lock);
6328 
6329 	/*
6330 	 * Verify that we can export the pool and reimport it under a
6331 	 * different name.
6332 	 */
6333 	if ((ztest_random(2) == 0) && !ztest_opts.zo_mmp_test) {
6334 		char name[ZFS_MAX_DATASET_NAME_LEN];
6335 		(void) snprintf(name, sizeof (name), "%s_import",
6336 		    ztest_opts.zo_pool);
6337 		ztest_spa_import_export(ztest_opts.zo_pool, name);
6338 		ztest_spa_import_export(name, ztest_opts.zo_pool);
6339 	}
6340 
6341 	kernel_fini();
6342 
6343 	list_destroy(&zcl.zcl_callbacks);
6344 
6345 	mutex_destroy(&zcl.zcl_callbacks_lock);
6346 
6347 	rw_destroy(&ztest_name_lock);
6348 	mutex_destroy(&ztest_vdev_lock);
6349 	mutex_destroy(&ztest_checkpoint_lock);
6350 }
6351 
6352 static void
6353 ztest_freeze(void)
6354 {
6355 	ztest_ds_t *zd = &ztest_ds[0];
6356 	spa_t *spa;
6357 	int numloops = 0;
6358 
6359 	if (ztest_opts.zo_verbose >= 3)
6360 		(void) printf("testing spa_freeze()...\n");
6361 
6362 	kernel_init(FREAD | FWRITE);
6363 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6364 	VERIFY3U(0, ==, ztest_dataset_open(0));
6365 	ztest_spa = spa;
6366 
6367 	/*
6368 	 * Force the first log block to be transactionally allocated.
6369 	 * We have to do this before we freeze the pool -- otherwise
6370 	 * the log chain won't be anchored.
6371 	 */
6372 	while (BP_IS_HOLE(&zd->zd_zilog->zl_header->zh_log)) {
6373 		ztest_dmu_object_alloc_free(zd, 0);
6374 		zil_commit(zd->zd_zilog, 0);
6375 	}
6376 
6377 	txg_wait_synced(spa_get_dsl(spa), 0);
6378 
6379 	/*
6380 	 * Freeze the pool.  This stops spa_sync() from doing anything,
6381 	 * so that the only way to record changes from now on is the ZIL.
6382 	 */
6383 	spa_freeze(spa);
6384 
6385 	/*
6386 	 * Because it is hard to predict how much space a write will actually
6387 	 * require beforehand, we leave ourselves some fudge space to write over
6388 	 * capacity.
6389 	 */
6390 	uint64_t capacity = metaslab_class_get_space(spa_normal_class(spa)) / 2;
6391 
6392 	/*
6393 	 * Run tests that generate log records but don't alter the pool config
6394 	 * or depend on DSL sync tasks (snapshots, objset create/destroy, etc).
6395 	 * We do a txg_wait_synced() after each iteration to force the txg
6396 	 * to increase well beyond the last synced value in the uberblock.
6397 	 * The ZIL should be OK with that.
6398 	 *
6399 	 * Run a random number of times less than zo_maxloops and ensure we do
6400 	 * not run out of space on the pool.
6401 	 */
6402 	while (ztest_random(10) != 0 &&
6403 	    numloops++ < ztest_opts.zo_maxloops &&
6404 	    metaslab_class_get_alloc(spa_normal_class(spa)) < capacity) {
6405 		ztest_od_t od;
6406 		ztest_od_init(&od, 0, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
6407 		VERIFY0(ztest_object_init(zd, &od, sizeof (od), B_FALSE));
6408 		ztest_io(zd, od.od_object,
6409 		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
6410 		txg_wait_synced(spa_get_dsl(spa), 0);
6411 	}
6412 
6413 	/*
6414 	 * Commit all of the changes we just generated.
6415 	 */
6416 	zil_commit(zd->zd_zilog, 0);
6417 	txg_wait_synced(spa_get_dsl(spa), 0);
6418 
6419 	/*
6420 	 * Close our dataset and close the pool.
6421 	 */
6422 	ztest_dataset_close(0);
6423 	spa_close(spa, FTAG);
6424 	kernel_fini();
6425 
6426 	/*
6427 	 * Open and close the pool and dataset to induce log replay.
6428 	 */
6429 	kernel_init(FREAD | FWRITE);
6430 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6431 	ASSERT(spa_freeze_txg(spa) == UINT64_MAX);
6432 	VERIFY3U(0, ==, ztest_dataset_open(0));
6433 	ztest_dataset_close(0);
6434 
6435 	ztest_spa = spa;
6436 	txg_wait_synced(spa_get_dsl(spa), 0);
6437 	ztest_reguid(NULL, 0);
6438 
6439 	spa_close(spa, FTAG);
6440 	kernel_fini();
6441 }
6442 
6443 void
6444 print_time(hrtime_t t, char *timebuf)
6445 {
6446 	hrtime_t s = t / NANOSEC;
6447 	hrtime_t m = s / 60;
6448 	hrtime_t h = m / 60;
6449 	hrtime_t d = h / 24;
6450 
6451 	s -= m * 60;
6452 	m -= h * 60;
6453 	h -= d * 24;
6454 
6455 	timebuf[0] = '\0';
6456 
6457 	if (d)
6458 		(void) sprintf(timebuf,
6459 		    "%llud%02lluh%02llum%02llus", d, h, m, s);
6460 	else if (h)
6461 		(void) sprintf(timebuf, "%lluh%02llum%02llus", h, m, s);
6462 	else if (m)
6463 		(void) sprintf(timebuf, "%llum%02llus", m, s);
6464 	else
6465 		(void) sprintf(timebuf, "%llus", s);
6466 }
6467 
6468 static nvlist_t *
6469 make_random_props()
6470 {
6471 	nvlist_t *props;
6472 
6473 	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
6474 	if (ztest_random(2) == 0)
6475 		return (props);
6476 	VERIFY(nvlist_add_uint64(props, "autoreplace", 1) == 0);
6477 
6478 	return (props);
6479 }
6480 
6481 /*
6482  * Import a storage pool with the given name.
6483  */
6484 static void
6485 ztest_import(ztest_shared_t *zs)
6486 {
6487 	libzfs_handle_t *hdl;
6488 	importargs_t args = { 0 };
6489 	spa_t *spa;
6490 	nvlist_t *cfg = NULL;
6491 	int nsearch = 1;
6492 	char *searchdirs[nsearch];
6493 	char *name = ztest_opts.zo_pool;
6494 	int flags = ZFS_IMPORT_MISSING_LOG;
6495 	int error;
6496 
6497 	mutex_init(&ztest_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
6498 	rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
6499 
6500 	kernel_init(FREAD | FWRITE);
6501 	hdl = libzfs_init();
6502 
6503 	searchdirs[0] = ztest_opts.zo_dir;
6504 	args.paths = nsearch;
6505 	args.path = searchdirs;
6506 	args.can_be_active = B_FALSE;
6507 
6508 	error = zpool_tryimport(hdl, name, &cfg, &args);
6509 	if (error)
6510 		(void) fatal(0, "No pools found\n");
6511 
6512 	VERIFY0(spa_import(name, cfg, NULL, flags));
6513 	VERIFY0(spa_open(name, &spa, FTAG));
6514 	zs->zs_metaslab_sz =
6515 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6516 	spa_close(spa, FTAG);
6517 
6518 	libzfs_fini(hdl);
6519 	kernel_fini();
6520 
6521 	if (!ztest_opts.zo_mmp_test) {
6522 		ztest_run_zdb(ztest_opts.zo_pool);
6523 		ztest_freeze();
6524 		ztest_run_zdb(ztest_opts.zo_pool);
6525 	}
6526 
6527 	rw_destroy(&ztest_name_lock);
6528 	mutex_destroy(&ztest_vdev_lock);
6529 }
6530 
6531 /*
6532  * Create a storage pool with the given name and initial vdev size.
6533  * Then test spa_freeze() functionality.
6534  */
6535 static void
6536 ztest_init(ztest_shared_t *zs)
6537 {
6538 	spa_t *spa;
6539 	nvlist_t *nvroot, *props;
6540 
6541 	mutex_init(&ztest_vdev_lock, NULL, USYNC_THREAD, NULL);
6542 	mutex_init(&ztest_checkpoint_lock, NULL, USYNC_THREAD, NULL);
6543 	rw_init(&ztest_name_lock, NULL, USYNC_THREAD, NULL);
6544 
6545 	kernel_init(FREAD | FWRITE);
6546 
6547 	/*
6548 	 * Create the storage pool.
6549 	 */
6550 	(void) spa_destroy(ztest_opts.zo_pool);
6551 	ztest_shared->zs_vdev_next_leaf = 0;
6552 	zs->zs_splits = 0;
6553 	zs->zs_mirrors = ztest_opts.zo_mirrors;
6554 	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
6555 	    0, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
6556 	props = make_random_props();
6557 	for (int i = 0; i < SPA_FEATURES; i++) {
6558 		char buf[1024];
6559 		(void) snprintf(buf, sizeof (buf), "feature@%s",
6560 		    spa_feature_table[i].fi_uname);
6561 		VERIFY3U(0, ==, nvlist_add_uint64(props, buf, 0));
6562 	}
6563 	VERIFY3U(0, ==, spa_create(ztest_opts.zo_pool, nvroot, props, NULL));
6564 	nvlist_free(nvroot);
6565 	nvlist_free(props);
6566 
6567 	VERIFY3U(0, ==, spa_open(ztest_opts.zo_pool, &spa, FTAG));
6568 	zs->zs_metaslab_sz =
6569 	    1ULL << spa->spa_root_vdev->vdev_child[0]->vdev_ms_shift;
6570 
6571 	spa_close(spa, FTAG);
6572 
6573 	kernel_fini();
6574 
6575 	if (!ztest_opts.zo_mmp_test) {
6576 		ztest_run_zdb(ztest_opts.zo_pool);
6577 		ztest_freeze();
6578 		ztest_run_zdb(ztest_opts.zo_pool);
6579 	}
6580 
6581 	rw_destroy(&ztest_name_lock);
6582 	mutex_destroy(&ztest_vdev_lock);
6583 	mutex_destroy(&ztest_checkpoint_lock);
6584 }
6585 
6586 static void
6587 setup_data_fd(void)
6588 {
6589 	static char ztest_name_data[] = "/tmp/ztest.data.XXXXXX";
6590 
6591 	ztest_fd_data = mkstemp(ztest_name_data);
6592 	ASSERT3S(ztest_fd_data, >=, 0);
6593 	(void) unlink(ztest_name_data);
6594 }
6595 
6596 
6597 static int
6598 shared_data_size(ztest_shared_hdr_t *hdr)
6599 {
6600 	int size;
6601 
6602 	size = hdr->zh_hdr_size;
6603 	size += hdr->zh_opts_size;
6604 	size += hdr->zh_size;
6605 	size += hdr->zh_stats_size * hdr->zh_stats_count;
6606 	size += hdr->zh_ds_size * hdr->zh_ds_count;
6607 
6608 	return (size);
6609 }
6610 
6611 static void
6612 setup_hdr(void)
6613 {
6614 	int size;
6615 	ztest_shared_hdr_t *hdr;
6616 
6617 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6618 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6619 	ASSERT(hdr != MAP_FAILED);
6620 
6621 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, sizeof (ztest_shared_hdr_t)));
6622 
6623 	hdr->zh_hdr_size = sizeof (ztest_shared_hdr_t);
6624 	hdr->zh_opts_size = sizeof (ztest_shared_opts_t);
6625 	hdr->zh_size = sizeof (ztest_shared_t);
6626 	hdr->zh_stats_size = sizeof (ztest_shared_callstate_t);
6627 	hdr->zh_stats_count = ZTEST_FUNCS;
6628 	hdr->zh_ds_size = sizeof (ztest_shared_ds_t);
6629 	hdr->zh_ds_count = ztest_opts.zo_datasets;
6630 
6631 	size = shared_data_size(hdr);
6632 	VERIFY3U(0, ==, ftruncate(ztest_fd_data, size));
6633 
6634 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6635 }
6636 
6637 static void
6638 setup_data(void)
6639 {
6640 	int size, offset;
6641 	ztest_shared_hdr_t *hdr;
6642 	uint8_t *buf;
6643 
6644 	hdr = (void *)mmap(0, P2ROUNDUP(sizeof (*hdr), getpagesize()),
6645 	    PROT_READ, MAP_SHARED, ztest_fd_data, 0);
6646 	ASSERT(hdr != MAP_FAILED);
6647 
6648 	size = shared_data_size(hdr);
6649 
6650 	(void) munmap((caddr_t)hdr, P2ROUNDUP(sizeof (*hdr), getpagesize()));
6651 	hdr = ztest_shared_hdr = (void *)mmap(0, P2ROUNDUP(size, getpagesize()),
6652 	    PROT_READ | PROT_WRITE, MAP_SHARED, ztest_fd_data, 0);
6653 	ASSERT(hdr != MAP_FAILED);
6654 	buf = (uint8_t *)hdr;
6655 
6656 	offset = hdr->zh_hdr_size;
6657 	ztest_shared_opts = (void *)&buf[offset];
6658 	offset += hdr->zh_opts_size;
6659 	ztest_shared = (void *)&buf[offset];
6660 	offset += hdr->zh_size;
6661 	ztest_shared_callstate = (void *)&buf[offset];
6662 	offset += hdr->zh_stats_size * hdr->zh_stats_count;
6663 	ztest_shared_ds = (void *)&buf[offset];
6664 }
6665 
6666 static boolean_t
6667 exec_child(char *cmd, char *libpath, boolean_t ignorekill, int *statusp)
6668 {
6669 	pid_t pid;
6670 	int status;
6671 	char *cmdbuf = NULL;
6672 
6673 	pid = fork();
6674 
6675 	if (cmd == NULL) {
6676 		cmdbuf = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
6677 		(void) strlcpy(cmdbuf, getexecname(), MAXPATHLEN);
6678 		cmd = cmdbuf;
6679 	}
6680 
6681 	if (pid == -1)
6682 		fatal(1, "fork failed");
6683 
6684 	if (pid == 0) {	/* child */
6685 		char *emptyargv[2] = { cmd, NULL };
6686 		char fd_data_str[12];
6687 
6688 		struct rlimit rl = { 1024, 1024 };
6689 		(void) setrlimit(RLIMIT_NOFILE, &rl);
6690 
6691 		(void) close(ztest_fd_rand);
6692 		VERIFY3U(11, >=,
6693 		    snprintf(fd_data_str, 12, "%d", ztest_fd_data));
6694 		VERIFY0(setenv("ZTEST_FD_DATA", fd_data_str, 1));
6695 
6696 		(void) enable_extended_FILE_stdio(-1, -1);
6697 		if (libpath != NULL)
6698 			VERIFY(0 == setenv("LD_LIBRARY_PATH", libpath, 1));
6699 		(void) execv(cmd, emptyargv);
6700 		ztest_dump_core = B_FALSE;
6701 		fatal(B_TRUE, "exec failed: %s", cmd);
6702 	}
6703 
6704 	if (cmdbuf != NULL) {
6705 		umem_free(cmdbuf, MAXPATHLEN);
6706 		cmd = NULL;
6707 	}
6708 
6709 	while (waitpid(pid, &status, 0) != pid)
6710 		continue;
6711 	if (statusp != NULL)
6712 		*statusp = status;
6713 
6714 	if (WIFEXITED(status)) {
6715 		if (WEXITSTATUS(status) != 0) {
6716 			(void) fprintf(stderr, "child exited with code %d\n",
6717 			    WEXITSTATUS(status));
6718 			exit(2);
6719 		}
6720 		return (B_FALSE);
6721 	} else if (WIFSIGNALED(status)) {
6722 		if (!ignorekill || WTERMSIG(status) != SIGKILL) {
6723 			(void) fprintf(stderr, "child died with signal %d\n",
6724 			    WTERMSIG(status));
6725 			exit(3);
6726 		}
6727 		return (B_TRUE);
6728 	} else {
6729 		(void) fprintf(stderr, "something strange happened to child\n");
6730 		exit(4);
6731 		/* NOTREACHED */
6732 	}
6733 }
6734 
6735 static void
6736 ztest_run_init(void)
6737 {
6738 	ztest_shared_t *zs = ztest_shared;
6739 
6740 	/*
6741 	 * Blow away any existing copy of zpool.cache
6742 	 */
6743 	(void) remove(spa_config_path);
6744 
6745 	if (ztest_opts.zo_init == 0) {
6746 		if (ztest_opts.zo_verbose >= 1)
6747 			(void) printf("Importing pool %s\n",
6748 			    ztest_opts.zo_pool);
6749 		ztest_import(zs);
6750 		return;
6751 	}
6752 
6753 	/*
6754 	 * Create and initialize our storage pool.
6755 	 */
6756 	for (int i = 1; i <= ztest_opts.zo_init; i++) {
6757 		bzero(zs, sizeof (ztest_shared_t));
6758 		if (ztest_opts.zo_verbose >= 3 &&
6759 		    ztest_opts.zo_init != 1) {
6760 			(void) printf("ztest_init(), pass %d\n", i);
6761 		}
6762 		ztest_init(zs);
6763 	}
6764 }
6765 
6766 int
6767 main(int argc, char **argv)
6768 {
6769 	int kills = 0;
6770 	int iters = 0;
6771 	int older = 0;
6772 	int newer = 0;
6773 	ztest_shared_t *zs;
6774 	ztest_info_t *zi;
6775 	ztest_shared_callstate_t *zc;
6776 	char timebuf[100];
6777 	char numbuf[NN_NUMBUF_SZ];
6778 	char *cmd;
6779 	boolean_t hasalt;
6780 	char *fd_data_str = getenv("ZTEST_FD_DATA");
6781 
6782 	(void) setvbuf(stdout, NULL, _IOLBF, 0);
6783 
6784 	dprintf_setup(&argc, argv);
6785 	zfs_deadman_synctime_ms = 300000;
6786 	/*
6787 	 * As two-word space map entries may not come up often (especially
6788 	 * if pool and vdev sizes are small) we want to force at least some
6789 	 * of them so the feature get tested.
6790 	 */
6791 	zfs_force_some_double_word_sm_entries = B_TRUE;
6792 
6793 	/*
6794 	 * Verify that even extensively damaged split blocks with many
6795 	 * segments can be reconstructed in a reasonable amount of time
6796 	 * when reconstruction is known to be possible.
6797 	 */
6798 	zfs_reconstruct_indirect_damage_fraction = 4;
6799 
6800 	ztest_fd_rand = open("/dev/urandom", O_RDONLY);
6801 	ASSERT3S(ztest_fd_rand, >=, 0);
6802 
6803 	if (!fd_data_str) {
6804 		process_options(argc, argv);
6805 
6806 		setup_data_fd();
6807 		setup_hdr();
6808 		setup_data();
6809 		bcopy(&ztest_opts, ztest_shared_opts,
6810 		    sizeof (*ztest_shared_opts));
6811 	} else {
6812 		ztest_fd_data = atoi(fd_data_str);
6813 		setup_data();
6814 		bcopy(ztest_shared_opts, &ztest_opts, sizeof (ztest_opts));
6815 	}
6816 	ASSERT3U(ztest_opts.zo_datasets, ==, ztest_shared_hdr->zh_ds_count);
6817 
6818 	/* Override location of zpool.cache */
6819 	VERIFY3U(asprintf((char **)&spa_config_path, "%s/zpool.cache",
6820 	    ztest_opts.zo_dir), !=, -1);
6821 
6822 	ztest_ds = umem_alloc(ztest_opts.zo_datasets * sizeof (ztest_ds_t),
6823 	    UMEM_NOFAIL);
6824 	zs = ztest_shared;
6825 
6826 	if (fd_data_str) {
6827 		metaslab_force_ganging = ztest_opts.zo_metaslab_force_ganging;
6828 		metaslab_df_alloc_threshold =
6829 		    zs->zs_metaslab_df_alloc_threshold;
6830 
6831 		if (zs->zs_do_init)
6832 			ztest_run_init();
6833 		else
6834 			ztest_run(zs);
6835 		exit(0);
6836 	}
6837 
6838 	hasalt = (strlen(ztest_opts.zo_alt_ztest) != 0);
6839 
6840 	if (ztest_opts.zo_verbose >= 1) {
6841 		(void) printf("%llu vdevs, %d datasets, %d threads,"
6842 		    " %llu seconds...\n",
6843 		    (u_longlong_t)ztest_opts.zo_vdevs,
6844 		    ztest_opts.zo_datasets,
6845 		    ztest_opts.zo_threads,
6846 		    (u_longlong_t)ztest_opts.zo_time);
6847 	}
6848 
6849 	cmd = umem_alloc(MAXNAMELEN, UMEM_NOFAIL);
6850 	(void) strlcpy(cmd, getexecname(), MAXNAMELEN);
6851 
6852 	zs->zs_do_init = B_TRUE;
6853 	if (strlen(ztest_opts.zo_alt_ztest) != 0) {
6854 		if (ztest_opts.zo_verbose >= 1) {
6855 			(void) printf("Executing older ztest for "
6856 			    "initialization: %s\n", ztest_opts.zo_alt_ztest);
6857 		}
6858 		VERIFY(!exec_child(ztest_opts.zo_alt_ztest,
6859 		    ztest_opts.zo_alt_libpath, B_FALSE, NULL));
6860 	} else {
6861 		VERIFY(!exec_child(NULL, NULL, B_FALSE, NULL));
6862 	}
6863 	zs->zs_do_init = B_FALSE;
6864 
6865 	zs->zs_proc_start = gethrtime();
6866 	zs->zs_proc_stop = zs->zs_proc_start + ztest_opts.zo_time * NANOSEC;
6867 
6868 	for (int f = 0; f < ZTEST_FUNCS; f++) {
6869 		zi = &ztest_info[f];
6870 		zc = ZTEST_GET_SHARED_CALLSTATE(f);
6871 		if (zs->zs_proc_start + zi->zi_interval[0] > zs->zs_proc_stop)
6872 			zc->zc_next = UINT64_MAX;
6873 		else
6874 			zc->zc_next = zs->zs_proc_start +
6875 			    ztest_random(2 * zi->zi_interval[0] + 1);
6876 	}
6877 
6878 	/*
6879 	 * Run the tests in a loop.  These tests include fault injection
6880 	 * to verify that self-healing data works, and forced crashes
6881 	 * to verify that we never lose on-disk consistency.
6882 	 */
6883 	while (gethrtime() < zs->zs_proc_stop) {
6884 		int status;
6885 		boolean_t killed;
6886 
6887 		/*
6888 		 * Initialize the workload counters for each function.
6889 		 */
6890 		for (int f = 0; f < ZTEST_FUNCS; f++) {
6891 			zc = ZTEST_GET_SHARED_CALLSTATE(f);
6892 			zc->zc_count = 0;
6893 			zc->zc_time = 0;
6894 		}
6895 
6896 		/* Set the allocation switch size */
6897 		zs->zs_metaslab_df_alloc_threshold =
6898 		    ztest_random(zs->zs_metaslab_sz / 4) + 1;
6899 
6900 		if (!hasalt || ztest_random(2) == 0) {
6901 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6902 				(void) printf("Executing newer ztest: %s\n",
6903 				    cmd);
6904 			}
6905 			newer++;
6906 			killed = exec_child(cmd, NULL, B_TRUE, &status);
6907 		} else {
6908 			if (hasalt && ztest_opts.zo_verbose >= 1) {
6909 				(void) printf("Executing older ztest: %s\n",
6910 				    ztest_opts.zo_alt_ztest);
6911 			}
6912 			older++;
6913 			killed = exec_child(ztest_opts.zo_alt_ztest,
6914 			    ztest_opts.zo_alt_libpath, B_TRUE, &status);
6915 		}
6916 
6917 		if (killed)
6918 			kills++;
6919 		iters++;
6920 
6921 		if (ztest_opts.zo_verbose >= 1) {
6922 			hrtime_t now = gethrtime();
6923 
6924 			now = MIN(now, zs->zs_proc_stop);
6925 			print_time(zs->zs_proc_stop - now, timebuf);
6926 			nicenum(zs->zs_space, numbuf, sizeof (numbuf));
6927 
6928 			(void) printf("Pass %3d, %8s, %3llu ENOSPC, "
6929 			    "%4.1f%% of %5s used, %3.0f%% done, %8s to go\n",
6930 			    iters,
6931 			    WIFEXITED(status) ? "Complete" : "SIGKILL",
6932 			    (u_longlong_t)zs->zs_enospc_count,
6933 			    100.0 * zs->zs_alloc / zs->zs_space,
6934 			    numbuf,
6935 			    100.0 * (now - zs->zs_proc_start) /
6936 			    (ztest_opts.zo_time * NANOSEC), timebuf);
6937 		}
6938 
6939 		if (ztest_opts.zo_verbose >= 2) {
6940 			(void) printf("\nWorkload summary:\n\n");
6941 			(void) printf("%7s %9s   %s\n",
6942 			    "Calls", "Time", "Function");
6943 			(void) printf("%7s %9s   %s\n",
6944 			    "-----", "----", "--------");
6945 			for (int f = 0; f < ZTEST_FUNCS; f++) {
6946 				Dl_info dli;
6947 
6948 				zi = &ztest_info[f];
6949 				zc = ZTEST_GET_SHARED_CALLSTATE(f);
6950 				print_time(zc->zc_time, timebuf);
6951 				(void) dladdr((void *)zi->zi_func, &dli);
6952 				(void) printf("%7llu %9s   %s\n",
6953 				    (u_longlong_t)zc->zc_count, timebuf,
6954 				    dli.dli_sname);
6955 			}
6956 			(void) printf("\n");
6957 		}
6958 
6959 		if (!ztest_opts.zo_mmp_test)
6960 			ztest_run_zdb(ztest_opts.zo_pool);
6961 	}
6962 
6963 	if (ztest_opts.zo_verbose >= 1) {
6964 		if (hasalt) {
6965 			(void) printf("%d runs of older ztest: %s\n", older,
6966 			    ztest_opts.zo_alt_ztest);
6967 			(void) printf("%d runs of newer ztest: %s\n", newer,
6968 			    cmd);
6969 		}
6970 		(void) printf("%d killed, %d completed, %.0f%% kill rate\n",
6971 		    kills, iters - kills, (100.0 * kills) / MAX(1, iters));
6972 	}
6973 
6974 	umem_free(cmd, MAXNAMELEN);
6975 
6976 	return (0);
6977 }
6978