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, 2018 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25 * Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>.  All rights reserved.
26 * Copyright (c) 2013 Steven Hartland. All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 * Copyright 2017 Joyent, Inc.
29 * Copyright (c) 2017, Intel Corporation.
30 * Copyright 2017 RackTop Systems.
31 */
32
33/*
34 * The objective of this program is to provide a DMU/ZAP/SPA stress test
35 * that runs entirely in userland, is easy to use, and easy to extend.
36 *
37 * The overall design of the ztest program is as follows:
38 *
39 * (1) For each major functional area (e.g. adding vdevs to a pool,
40 *     creating and destroying datasets, reading and writing objects, etc)
41 *     we have a simple routine to test that functionality.  These
42 *     individual routines do not have to do anything "stressful".
43 *
44 * (2) We turn these simple functionality tests into a stress test by
45 *     running them all in parallel, with as many threads as desired,
46 *     and spread across as many datasets, objects, and vdevs as desired.
47 *
48 * (3) While all this is happening, we inject faults into the pool to
49 *     verify that self-healing data really works.
50 *
51 * (4) Every time we open a dataset, we change its checksum and compression
52 *     functions.  Thus even individual objects vary from block to block
53 *     in which checksum they use and whether they're compressed.
54 *
55 * (5) To verify that we never lose on-disk consistency after a crash,
56 *     we run the entire test in a child of the main process.
57 *     At random times, the child self-immolates with a SIGKILL.
58 *     This is the software equivalent of pulling the power cord.
59 *     The parent then runs the test again, using the existing
60 *     storage pool, as many times as desired. If backwards compatibility
61 *     testing is enabled ztest will sometimes run the "older" version
62 *     of ztest after a SIGKILL.
63 *
64 * (6) To verify that we don't have future leaks or temporal incursions,
65 *     many of the functional tests record the transaction group number
66 *     as part of their data.  When reading old data, they verify that
67 *     the transaction group number is less than the current, open txg.
68 *     If you add a new test, please do this if applicable.
69 *
70 * When run with no arguments, ztest runs for about five minutes and
71 * produces no output if successful.  To get a little bit of information,
72 * specify -V.  To get more information, specify -VV, and so on.
73 *
74 * To turn this into an overnight stress test, use -T to specify run time.
75 *
76 * You can ask more more vdevs [-v], datasets [-d], or threads [-t]
77 * to increase the pool capacity, fanout, and overall stress level.
78 *
79 * Use the -k option to set the desired frequency of kills.
80 *
81 * When ztest invokes itself it passes all relevant information through a
82 * temporary file which is mmap-ed in the child process. This allows shared
83 * memory to survive the exec syscall. The ztest_shared_hdr_t struct is always
84 * stored at offset 0 of this file and contains information on the size and
85 * number of shared structures in the file. The information stored in this file
86 * must remain backwards compatible with older versions of ztest so that
87 * ztest can invoke them during backwards compatibility testing (-B).
88 */
89
90#include <sys/zfs_context.h>
91#include <sys/spa.h>
92#include <sys/dmu.h>
93#include <sys/txg.h>
94#include <sys/dbuf.h>
95#include <sys/zap.h>
96#include <sys/dmu_objset.h>
97#include <sys/poll.h>
98#include <sys/stat.h>
99#include <sys/time.h>
100#include <sys/wait.h>
101#include <sys/mman.h>
102#include <sys/resource.h>
103#include <sys/zio.h>
104#include <sys/zil.h>
105#include <sys/zil_impl.h>
106#include <sys/vdev_impl.h>
107#include <sys/vdev_file.h>
108#include <sys/vdev_initialize.h>
109#include <sys/spa_impl.h>
110#include <sys/metaslab_impl.h>
111#include <sys/dsl_prop.h>
112#include <sys/dsl_dataset.h>
113#include <sys/dsl_destroy.h>
114#include <sys/dsl_scan.h>
115#include <sys/zio_checksum.h>
116#include <sys/refcount.h>
117#include <sys/zfeature.h>
118#include <sys/dsl_userhold.h>
119#include <sys/abd.h>
120#include <stdio.h>
121#include <stdio_ext.h>
122#include <stdlib.h>
123#include <unistd.h>
124#include <signal.h>
125#include <umem.h>
126#include <dlfcn.h>
127#include <ctype.h>
128#include <math.h>
129#include <errno.h>
130#include <sys/fs/zfs.h>
131#include <libnvpair.h>
132#include <libzfs.h>
133#include <libcmdutils.h>
134
135static int ztest_fd_data = -1;
136static int ztest_fd_rand = -1;
137
138typedef struct ztest_shared_hdr {
139	uint64_t	zh_hdr_size;
140	uint64_t	zh_opts_size;
141	uint64_t	zh_size;
142	uint64_t	zh_stats_size;
143	uint64_t	zh_stats_count;
144	uint64_t	zh_ds_size;
145	uint64_t	zh_ds_count;
146} ztest_shared_hdr_t;
147
148static ztest_shared_hdr_t *ztest_shared_hdr;
149
150enum ztest_class_state {
151	ZTEST_VDEV_CLASS_OFF,
152	ZTEST_VDEV_CLASS_ON,
153	ZTEST_VDEV_CLASS_RND
154};
155
156typedef struct ztest_shared_opts {
157	char zo_pool[ZFS_MAX_DATASET_NAME_LEN];
158	char zo_dir[ZFS_MAX_DATASET_NAME_LEN];
159	char zo_alt_ztest[MAXNAMELEN];
160	char zo_alt_libpath[MAXNAMELEN];
161	uint64_t zo_vdevs;
162	uint64_t zo_vdevtime;
163	size_t zo_vdev_size;
164	int zo_ashift;
165	int zo_mirrors;
166	int zo_raidz;
167	int zo_raidz_parity;
168	int zo_datasets;
169	int zo_threads;
170	uint64_t zo_passtime;
171	uint64_t zo_killrate;
172	int zo_verbose;
173	int zo_init;
174	uint64_t zo_time;
175	uint64_t zo_maxloops;
176	uint64_t zo_metaslab_force_ganging;
177	int zo_mmp_test;
178	int zo_special_vdevs;
179} ztest_shared_opts_t;
180
181static const ztest_shared_opts_t ztest_opts_defaults = {
182	.zo_pool = { 'z', 't', 'e', 's', 't', '\0' },
183	.zo_dir = { '/', 't', 'm', 'p', '\0' },
184	.zo_alt_ztest = { '\0' },
185	.zo_alt_libpath = { '\0' },
186	.zo_vdevs = 5,
187	.zo_ashift = SPA_MINBLOCKSHIFT,
188	.zo_mirrors = 2,
189	.zo_raidz = 4,
190	.zo_raidz_parity = 1,
191	.zo_vdev_size = SPA_MINDEVSIZE * 4,	/* 256m default size */
192	.zo_datasets = 7,
193	.zo_threads = 23,
194	.zo_passtime = 60,		/* 60 seconds */
195	.zo_killrate = 70,		/* 70% kill rate */
196	.zo_verbose = 0,
197	.zo_mmp_test = 0,
198	.zo_init = 1,
199	.zo_time = 300,			/* 5 minutes */
200	.zo_maxloops = 50,		/* max loops during spa_freeze() */
201	.zo_metaslab_force_ganging = 32 << 10,
202	.zo_special_vdevs = ZTEST_VDEV_CLASS_RND,
203};
204
205extern uint64_t metaslab_force_ganging;
206extern uint64_t metaslab_df_alloc_threshold;
207extern uint64_t zfs_deadman_synctime_ms;
208extern int metaslab_preload_limit;
209extern boolean_t zfs_compressed_arc_enabled;
210extern boolean_t zfs_abd_scatter_enabled;
211extern int dmu_object_alloc_chunk_shift;
212extern boolean_t zfs_force_some_double_word_sm_entries;
213extern unsigned long zfs_reconstruct_indirect_damage_fraction;
214
215static ztest_shared_opts_t *ztest_shared_opts;
216static ztest_shared_opts_t ztest_opts;
217
218typedef struct ztest_shared_ds {
219	uint64_t	zd_seq;
220} ztest_shared_ds_t;
221
222static ztest_shared_ds_t *ztest_shared_ds;
223#define	ZTEST_GET_SHARED_DS(d) (&ztest_shared_ds[d])
224
225#define	BT_MAGIC	0x123456789abcdefULL
226#define	MAXFAULTS() \
227	(MAX(zs->zs_mirrors, 1) * (ztest_opts.zo_raidz_parity + 1) - 1)
228
229enum ztest_io_type {
230	ZTEST_IO_WRITE_TAG,
231	ZTEST_IO_WRITE_PATTERN,
232	ZTEST_IO_WRITE_ZEROES,
233	ZTEST_IO_TRUNCATE,
234	ZTEST_IO_SETATTR,
235	ZTEST_IO_REWRITE,
236	ZTEST_IO_TYPES
237};
238
239typedef struct ztest_block_tag {
240	uint64_t	bt_magic;
241	uint64_t	bt_objset;
242	uint64_t	bt_object;
243	uint64_t	bt_dnodesize;
244	uint64_t	bt_offset;
245	uint64_t	bt_gen;
246	uint64_t	bt_txg;
247	uint64_t	bt_crtxg;
248} ztest_block_tag_t;
249
250typedef struct bufwad {
251	uint64_t	bw_index;
252	uint64_t	bw_txg;
253	uint64_t	bw_data;
254} bufwad_t;
255
256/*
257 * It would be better to use a rangelock_t per object.  Unfortunately
258 * the rangelock_t is not a drop-in replacement for rl_t, because we
259 * still need to map from object ID to rangelock_t.
260 */
261typedef enum {
262	RL_READER,
263	RL_WRITER,
264	RL_APPEND
265} rl_type_t;
266
267typedef struct rll {
268	void		*rll_writer;
269	int		rll_readers;
270	kmutex_t	rll_lock;
271	kcondvar_t	rll_cv;
272} rll_t;
273
274typedef struct rl {
275	uint64_t	rl_object;
276	uint64_t	rl_offset;
277	uint64_t	rl_size;
278	rll_t		*rl_lock;
279} rl_t;
280
281#define	ZTEST_RANGE_LOCKS	64
282#define	ZTEST_OBJECT_LOCKS	64
283
284/*
285 * Object descriptor.  Used as a template for object lookup/create/remove.
286 */
287typedef struct ztest_od {
288	uint64_t	od_dir;
289	uint64_t	od_object;
290	dmu_object_type_t od_type;
291	dmu_object_type_t od_crtype;
292	uint64_t	od_blocksize;
293	uint64_t	od_crblocksize;
294	uint64_t	od_crdnodesize;
295	uint64_t	od_gen;
296	uint64_t	od_crgen;
297	char		od_name[ZFS_MAX_DATASET_NAME_LEN];
298} ztest_od_t;
299
300/*
301 * Per-dataset state.
302 */
303typedef struct ztest_ds {
304	ztest_shared_ds_t *zd_shared;
305	objset_t	*zd_os;
306	krwlock_t	zd_zilog_lock;
307	zilog_t		*zd_zilog;
308	ztest_od_t	*zd_od;		/* debugging aid */
309	char		zd_name[ZFS_MAX_DATASET_NAME_LEN];
310	kmutex_t	zd_dirobj_lock;
311	rll_t		zd_object_lock[ZTEST_OBJECT_LOCKS];
312	rll_t		zd_range_lock[ZTEST_RANGE_LOCKS];
313} ztest_ds_t;
314
315/*
316 * Per-iteration state.
317 */
318typedef void ztest_func_t(ztest_ds_t *zd, uint64_t id);
319
320typedef struct ztest_info {
321	ztest_func_t	*zi_func;	/* test function */
322	uint64_t	zi_iters;	/* iterations per execution */
323	uint64_t	*zi_interval;	/* execute every <interval> seconds */
324} ztest_info_t;
325
326typedef struct ztest_shared_callstate {
327	uint64_t	zc_count;	/* per-pass count */
328	uint64_t	zc_time;	/* per-pass time */
329	uint64_t	zc_next;	/* next time to call this function */
330} ztest_shared_callstate_t;
331
332static ztest_shared_callstate_t *ztest_shared_callstate;
333#define	ZTEST_GET_SHARED_CALLSTATE(c) (&ztest_shared_callstate[c])
334
335/*
336 * Note: these aren't static because we want dladdr() to work.
337 */
338ztest_func_t ztest_dmu_read_write;
339ztest_func_t ztest_dmu_write_parallel;
340ztest_func_t ztest_dmu_object_alloc_free;
341ztest_func_t ztest_dmu_object_next_chunk;
342ztest_func_t ztest_dmu_commit_callbacks;
343ztest_func_t ztest_zap;
344ztest_func_t ztest_zap_parallel;
345ztest_func_t ztest_zil_commit;
346ztest_func_t ztest_zil_remount;
347ztest_func_t ztest_dmu_read_write_zcopy;
348ztest_func_t ztest_dmu_objset_create_destroy;
349ztest_func_t ztest_dmu_prealloc;
350ztest_func_t ztest_fzap;
351ztest_func_t ztest_dmu_snapshot_create_destroy;
352ztest_func_t ztest_dsl_prop_get_set;
353ztest_func_t ztest_spa_prop_get_set;
354ztest_func_t ztest_spa_create_destroy;
355ztest_func_t ztest_fault_inject;
356ztest_func_t ztest_ddt_repair;
357ztest_func_t ztest_dmu_snapshot_hold;
358ztest_func_t ztest_mmp_enable_disable;
359ztest_func_t ztest_scrub;
360ztest_func_t ztest_dsl_dataset_promote_busy;
361ztest_func_t ztest_vdev_attach_detach;
362ztest_func_t ztest_vdev_LUN_growth;
363ztest_func_t ztest_vdev_add_remove;
364ztest_func_t ztest_vdev_class_add;
365ztest_func_t ztest_vdev_aux_add_remove;
366ztest_func_t ztest_split_pool;
367ztest_func_t ztest_reguid;
368ztest_func_t ztest_spa_upgrade;
369ztest_func_t ztest_device_removal;
370ztest_func_t ztest_remap_blocks;
371ztest_func_t ztest_spa_checkpoint_create_discard;
372ztest_func_t ztest_initialize;
373ztest_func_t ztest_verify_dnode_bt;
374
375uint64_t zopt_always = 0ULL * NANOSEC;		/* all the time */
376uint64_t zopt_incessant = 1ULL * NANOSEC / 10;	/* every 1/10 second */
377uint64_t zopt_often = 1ULL * NANOSEC;		/* every second */
378uint64_t zopt_sometimes = 10ULL * NANOSEC;	/* every 10 seconds */
379uint64_t zopt_rarely = 60ULL * NANOSEC;		/* every 60 seconds */
380
381ztest_info_t ztest_info[] = {
382	{ ztest_dmu_read_write,			1,	&zopt_always	},
383	{ ztest_dmu_write_parallel,		10,	&zopt_always	},
384	{ ztest_dmu_object_alloc_free,		1,	&zopt_always	},
385	{ ztest_dmu_object_next_chunk,		1,	&zopt_sometimes	},
386	{ ztest_dmu_commit_callbacks,		1,	&zopt_always	},
387	{ ztest_zap,				30,	&zopt_always	},
388	{ ztest_zap_parallel,			100,	&zopt_always	},
389	{ ztest_split_pool,			1,	&zopt_always	},
390	{ ztest_zil_commit,			1,	&zopt_incessant	},
391	{ ztest_zil_remount,			1,	&zopt_sometimes	},
392	{ ztest_dmu_read_write_zcopy,		1,	&zopt_often	},
393	{ ztest_dmu_objset_create_destroy,	1,	&zopt_often	},
394	{ ztest_dsl_prop_get_set,		1,	&zopt_often	},
395	{ ztest_spa_prop_get_set,		1,	&zopt_sometimes	},
396#if 0
397	{ ztest_dmu_prealloc,			1,	&zopt_sometimes	},
398#endif
399	{ ztest_fzap,				1,	&zopt_sometimes	},
400	{ ztest_dmu_snapshot_create_destroy,	1,	&zopt_sometimes	},
401	{ ztest_spa_create_destroy,		1,	&zopt_sometimes	},
402	{ ztest_fault_inject,			1,	&zopt_incessant	},
403	{ ztest_ddt_repair,			1,	&zopt_sometimes	},
404	{ ztest_dmu_snapshot_hold,		1,	&zopt_sometimes	},
405	{ ztest_mmp_enable_disable,		1,	&zopt_sometimes	},
406	{ ztest_reguid,				1,	&zopt_rarely	},
407	{ ztest_scrub,				1,	&zopt_often	},
408	{ ztest_spa_upgrade,			1,	&zopt_rarely	},
409	{ ztest_dsl_dataset_promote_busy,	1,	&zopt_rarely	},
410	{ ztest_vdev_attach_detach,		1,	&zopt_incessant	},
411	{ ztest_vdev_LUN_growth,		1,	&zopt_rarely	},
412	{ ztest_vdev_add_remove,		1,
413	    &ztest_opts.zo_vdevtime				},
414	{ ztest_vdev_class_add,			1,
415	    &ztest_opts.zo_vdevtime				},
416	{ ztest_vdev_aux_add_remove,		1,
417	    &ztest_opts.zo_vdevtime				},
418	{ ztest_device_removal,			1,	&zopt_sometimes	},
419	{ ztest_remap_blocks,			1,	&zopt_sometimes },
420	{ ztest_spa_checkpoint_create_discard,	1,	&zopt_rarely	},
421	{ ztest_initialize,			1,	&zopt_sometimes },
422	{ ztest_verify_dnode_bt,		1,	&zopt_sometimes }
423};
424
425#define	ZTEST_FUNCS	(sizeof (ztest_info) / sizeof (ztest_info_t))
426
427/*
428 * The following struct is used to hold a list of uncalled commit callbacks.
429 * The callbacks are ordered by txg number.
430 */
431typedef struct ztest_cb_list {
432	kmutex_t zcl_callbacks_lock;
433	list_t	zcl_callbacks;
434} ztest_cb_list_t;
435
436/*
437 * Stuff we need to share writably between parent and child.
438 */
439typedef struct ztest_shared {
440	boolean_t	zs_do_init;
441	hrtime_t	zs_proc_start;
442	hrtime_t	zs_proc_stop;
443	hrtime_t	zs_thread_start;
444	hrtime_t	zs_thread_stop;
445	hrtime_t	zs_thread_kill;
446	uint64_t	zs_enospc_count;
447	uint64_t	zs_vdev_next_leaf;
448	uint64_t	zs_vdev_aux;
449	uint64_t	zs_alloc;
450	uint64_t	zs_space;
451	uint64_t	zs_splits;
452	uint64_t	zs_mirrors;
453	uint64_t	zs_metaslab_sz;
454	uint64_t	zs_metaslab_df_alloc_threshold;
455	uint64_t	zs_guid;
456} ztest_shared_t;
457
458#define	ID_PARALLEL	-1ULL
459
460static char ztest_dev_template[] = "%s/%s.%llua";
461static char ztest_aux_template[] = "%s/%s.%s.%llu";
462ztest_shared_t *ztest_shared;
463
464static spa_t *ztest_spa = NULL;
465static ztest_ds_t *ztest_ds;
466
467static kmutex_t ztest_vdev_lock;
468static boolean_t ztest_device_removal_active = B_FALSE;
469static kmutex_t ztest_checkpoint_lock;
470
471/*
472 * The ztest_name_lock protects the pool and dataset namespace used by
473 * the individual tests. To modify the namespace, consumers must grab
474 * this lock as writer. Grabbing the lock as reader will ensure that the
475 * namespace does not change while the lock is held.
476 */
477static krwlock_t ztest_name_lock;
478
479static boolean_t ztest_dump_core = B_TRUE;
480static boolean_t ztest_exiting;
481
482/* Global commit callback list */
483static ztest_cb_list_t zcl;
484
485enum ztest_object {
486	ZTEST_META_DNODE = 0,
487	ZTEST_DIROBJ,
488	ZTEST_OBJECTS
489};
490
491static void usage(boolean_t) __NORETURN;
492
493/*
494 * These libumem hooks provide a reasonable set of defaults for the allocator's
495 * debugging facilities.
496 */
497const char *
498_umem_debug_init()
499{
500	return ("default,verbose"); /* $UMEM_DEBUG setting */
501}
502
503const char *
504_umem_logging_init(void)
505{
506	return ("fail,contents"); /* $UMEM_LOGGING setting */
507}
508
509#define	FATAL_MSG_SZ	1024
510
511char *fatal_msg;
512
513static void
514fatal(int do_perror, char *message, ...)
515{
516	va_list args;
517	int save_errno = errno;
518	char buf[FATAL_MSG_SZ];
519
520	(void) fflush(stdout);
521
522	va_start(args, message);
523	(void) sprintf(buf, "ztest: ");
524	/* LINTED */
525	(void) vsprintf(buf + strlen(buf), message, args);
526	va_end(args);
527	if (do_perror) {
528		(void) snprintf(buf + strlen(buf), FATAL_MSG_SZ - strlen(buf),
529		    ": %s", strerror(save_errno));
530	}
531	(void) fprintf(stderr, "%s\n", buf);
532	fatal_msg = buf;			/* to ease debugging */
533	if (ztest_dump_core)
534		abort();
535	exit(3);
536}
537
538static int
539str2shift(const char *buf)
540{
541	const char *ends = "BKMGTPEZ";
542	int i;
543
544	if (buf[0] == '\0')
545		return (0);
546	for (i = 0; i < strlen(ends); i++) {
547		if (toupper(buf[0]) == ends[i])
548			break;
549	}
550	if (i == strlen(ends)) {
551		(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n",
552		    buf);
553		usage(B_FALSE);
554	}
555	if (buf[1] == '\0' || (toupper(buf[1]) == 'B' && buf[2] == '\0')) {
556		return (10*i);
557	}
558	(void) fprintf(stderr, "ztest: invalid bytes suffix: %s\n", buf);
559	usage(B_FALSE);
560	/* NOTREACHED */
561}
562
563static uint64_t
564nicenumtoull(const char *buf)
565{
566	char *end;
567	uint64_t val;
568
569	val = strtoull(buf, &end, 0);
570	if (end == buf) {
571		(void) fprintf(stderr, "ztest: bad numeric value: %s\n", buf);
572		usage(B_FALSE);
573	} else if (end[0] == '.') {
574		double fval = strtod(buf, &end);
575		fval *= pow(2, str2shift(end));
576		if (fval > UINT64_MAX) {
577			(void) fprintf(stderr, "ztest: value too large: %s\n",
578			    buf);
579			usage(B_FALSE);
580		}
581		val = (uint64_t)fval;
582	} else {
583		int shift = str2shift(end);
584		if (shift >= 64 || (val << shift) >> shift != val) {
585			(void) fprintf(stderr, "ztest: value too large: %s\n",
586			    buf);
587			usage(B_FALSE);
588		}
589		val <<= shift;
590	}
591	return (val);
592}
593
594static void
595usage(boolean_t requested)
596{
597	const ztest_shared_opts_t *zo = &ztest_opts_defaults;
598
599	char nice_vdev_size[NN_NUMBUF_SZ];
600	char nice_force_ganging[NN_NUMBUF_SZ];
601	FILE *fp = requested ? stdout : stderr;
602
603	nicenum(zo->zo_vdev_size, nice_vdev_size, sizeof (nice_vdev_size));
604	nicenum(zo->zo_metaslab_force_ganging, nice_force_ganging,
605	    sizeof (nice_force_ganging));
606
607	(void) fprintf(fp, "Usage: %s\n"
608	    "\t[-v vdevs (default: %llu)]\n"
609	    "\t[-s size_of_each_vdev (default: %s)]\n"
610	    "\t[-a alignment_shift (default: %d)] use 0 for random\n"
611	    "\t[-m mirror_copies (default: %d)]\n"
612	    "\t[-r raidz_disks (default: %d)]\n"
613	    "\t[-R raidz_parity (default: %d)]\n"
614	    "\t[-d datasets (default: %d)]\n"
615	    "\t[-t threads (default: %d)]\n"
616	    "\t[-g gang_block_threshold (default: %s)]\n"
617	    "\t[-i init_count (default: %d)] initialize pool i times\n"
618	    "\t[-k kill_percentage (default: %llu%%)]\n"
619	    "\t[-p pool_name (default: %s)]\n"
620	    "\t[-f dir (default: %s)] file directory for vdev files\n"
621	    "\t[-M] Multi-host simulate pool imported on remote host\n"
622	    "\t[-V] verbose (use multiple times for ever more blather)\n"
623	    "\t[-E] use existing pool instead of creating new one\n"
624	    "\t[-T time (default: %llu sec)] total run time\n"
625	    "\t[-F freezeloops (default: %llu)] max loops in spa_freeze()\n"
626	    "\t[-P passtime (default: %llu sec)] time per pass\n"
627	    "\t[-B alt_ztest (default: <none>)] alternate ztest path\n"
628	    "\t[-C vdev class state (default: random)] special=on|off|random\n"
629	    "\t[-o variable=value] ... set global variable to an unsigned\n"
630	    "\t    32-bit integer value\n"
631	    "\t[-h] (print help)\n"
632	    "",
633	    zo->zo_pool,
634	    (u_longlong_t)zo->zo_vdevs,			/* -v */
635	    nice_vdev_size,				/* -s */
636	    zo->zo_ashift,				/* -a */
637	    zo->zo_mirrors,				/* -m */
638	    zo->zo_raidz,				/* -r */
639	    zo->zo_raidz_parity,			/* -R */
640	    zo->zo_datasets,				/* -d */
641	    zo->zo_threads,				/* -t */
642	    nice_force_ganging,				/* -g */
643	    zo->zo_init,				/* -i */
644	    (u_longlong_t)zo->zo_killrate,		/* -k */
645	    zo->zo_pool,				/* -p */
646	    zo->zo_dir,					/* -f */
647	    (u_longlong_t)zo->zo_time,			/* -T */
648	    (u_longlong_t)zo->zo_maxloops,		/* -F */
649	    (u_longlong_t)zo->zo_passtime);
650	exit(requested ? 0 : 1);
651}
652
653
654static void
655ztest_parse_name_value(const char *input, ztest_shared_opts_t *zo)
656{
657	char name[32];
658	char *value;
659	int state = ZTEST_VDEV_CLASS_RND;
660
661	(void) strlcpy(name, input, sizeof (name));
662
663	value = strchr(name, '=');
664	if (value == NULL) {
665		(void) fprintf(stderr, "missing value in property=value "
666		    "'-C' argument (%s)\n", input);
667		usage(B_FALSE);
668	}
669	*(value) = '\0';
670	value++;
671
672	if (strcmp(value, "on") == 0) {
673		state = ZTEST_VDEV_CLASS_ON;
674	} else if (strcmp(value, "off") == 0) {
675		state = ZTEST_VDEV_CLASS_OFF;
676	} else if (strcmp(value, "random") == 0) {
677		state = ZTEST_VDEV_CLASS_RND;
678	} else {
679		(void) fprintf(stderr, "invalid property value '%s'\n", value);
680		usage(B_FALSE);
681	}
682
683	if (strcmp(name, "special") == 0) {
684		zo->zo_special_vdevs = state;
685	} else {
686		(void) fprintf(stderr, "invalid property name '%s'\n", name);
687		usage(B_FALSE);
688	}
689	if (zo->zo_verbose >= 3)
690		(void) printf("%s vdev state is '%s'\n", name, value);
691}
692
693static void
694process_options(int argc, char **argv)
695{
696	char *path;
697	ztest_shared_opts_t *zo = &ztest_opts;
698
699	int opt;
700	uint64_t value;
701	char altdir[MAXNAMELEN] = { 0 };
702
703	bcopy(&ztest_opts_defaults, zo, sizeof (*zo));
704
705	while ((opt = getopt(argc, argv,
706	    "v:s:a:m:r:R:d:t:g:i:k:p:f:MVET:P:hF:B:C:o:")) != EOF) {
707		value = 0;
708		switch (opt) {
709		case 'v':
710		case 's':
711		case 'a':
712		case 'm':
713		case 'r':
714		case 'R':
715		case 'd':
716		case 't':
717		case 'g':
718		case 'i':
719		case 'k':
720		case 'T':
721		case 'P':
722		case 'F':
723			value = nicenumtoull(optarg);
724		}
725		switch (opt) {
726		case 'v':
727			zo->zo_vdevs = value;
728			break;
729		case 's':
730			zo->zo_vdev_size = MAX(SPA_MINDEVSIZE, value);
731			break;
732		case 'a':
733			zo->zo_ashift = value;
734			break;
735		case 'm':
736			zo->zo_mirrors = value;
737			break;
738		case 'r':
739			zo->zo_raidz = MAX(1, value);
740			break;
741		case 'R':
742			zo->zo_raidz_parity = MIN(MAX(value, 1), 3);
743			break;
744		case 'd':
745			zo->zo_datasets = MAX(1, value);
746			break;
747		case 't':
748			zo->zo_threads = MAX(1, value);
749			break;
750		case 'g':
751			zo->zo_metaslab_force_ganging =
752			    MAX(SPA_MINBLOCKSIZE << 1, value);
753			break;
754		case 'i':
755			zo->zo_init = value;
756			break;
757		case 'k':
758			zo->zo_killrate = value;
759			break;
760		case 'p':
761			(void) strlcpy(zo->zo_pool, optarg,
762			    sizeof (zo->zo_pool));
763			break;
764		case 'f':
765			path = realpath(optarg, NULL);
766			if (path == NULL) {
767				(void) fprintf(stderr, "error: %s: %s\n",
768				    optarg, strerror(errno));
769				usage(B_FALSE);
770			} else {
771				(void) strlcpy(zo->zo_dir, path,
772				    sizeof (zo->zo_dir));
773			}
774			break;
775		case 'M':
776			zo->zo_mmp_test = 1;
777			break;
778		case 'V':
779			zo->zo_verbose++;
780			break;
781		case 'E':
782			zo->zo_init = 0;
783			break;
784		case 'T':
785			zo->zo_time = value;
786			break;
787		case 'P':
788			zo->zo_passtime = MAX(1, value);
789			break;
790		case 'F':
791			zo->zo_maxloops = MAX(1, value);
792			break;
793		case 'B':
794			(void) strlcpy(altdir, optarg, sizeof (altdir));
795			break;
796		case 'C':
797			ztest_parse_name_value(optarg, zo);
798			break;
799		case 'o':
800			if (set_global_var(optarg) != 0)
801				usage(B_FALSE);
802			break;
803		case 'h':
804			usage(B_TRUE);
805			break;
806		case '?':
807		default:
808			usage(B_FALSE);
809			break;
810		}
811	}
812
813	zo->zo_raidz_parity = MIN(zo->zo_raidz_parity, zo->zo_raidz - 1);
814
815	zo->zo_vdevtime =
816	    (zo->zo_vdevs > 0 ? zo->zo_time * NANOSEC / zo->zo_vdevs :
817	    UINT64_MAX >> 2);
818
819	if (strlen(altdir) > 0) {
820		char *cmd;
821		char *realaltdir;
822		char *bin;
823		char *ztest;
824		char *isa;
825		int isalen;
826
827		cmd = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
828		realaltdir = umem_alloc(MAXPATHLEN, UMEM_NOFAIL);
829
830		VERIFY(NULL != realpath(getexecname(), cmd));
831		if (0 != access(altdir, F_OK)) {
832			ztest_dump_core = B_FALSE;
833			fatal(B_TRUE, "invalid alternate ztest path: %s",
834			    altdir);
835		}
836		VERIFY(NULL != realpath(altdir, realaltdir));
837
838		/*
839		 * 'cmd' should be of the form "<anything>/usr/bin/<isa>/ztest".
840		 * We want to extract <isa> to determine if we should use
841		 * 32 or 64 bit binaries.
842		 */
843		bin = strstr(cmd, "/usr/bin/");
844		ztest = strstr(bin, "/ztest");
845		isa = bin + 9;
846		isalen = ztest - isa;
847		(void) snprintf(zo->zo_alt_ztest, sizeof (zo->zo_alt_ztest),
848		    "%s/usr/bin/%.*s/ztest", realaltdir, isalen, isa);
849		(void) snprintf(zo->zo_alt_libpath, sizeof (zo->zo_alt_libpath),
850		    "%s/usr/lib/%.*s", realaltdir, isalen, isa);
851
852		if (0 != access(zo->zo_alt_ztest, X_OK)) {
853			ztest_dump_core = B_FALSE;
854			fatal(B_TRUE, "invalid alternate ztest: %s",
855			    zo->zo_alt_ztest);
856		} else if (0 != access(zo->zo_alt_libpath, X_OK)) {
857			ztest_dump_core = B_FALSE;
858			fatal(B_TRUE, "invalid alternate lib directory %s",
859			    zo->zo_alt_libpath);
860		}
861
862		umem_free(cmd, MAXPATHLEN);
863		umem_free(realaltdir, MAXPATHLEN);
864	}
865}
866
867static void
868ztest_kill(ztest_shared_t *zs)
869{
870	zs->zs_alloc = metaslab_class_get_alloc(spa_normal_class(ztest_spa));
871	zs->zs_space = metaslab_class_get_space(spa_normal_class(ztest_spa));
872
873	/*
874	 * Before we kill off ztest, make sure that the config is updated.
875	 * See comment above spa_write_cachefile().
876	 */
877	mutex_enter(&spa_namespace_lock);
878	spa_write_cachefile(ztest_spa, B_FALSE, B_FALSE);
879	mutex_exit(&spa_namespace_lock);
880
881	zfs_dbgmsg_print(FTAG);
882	(void) kill(getpid(), SIGKILL);
883}
884
885static uint64_t
886ztest_random(uint64_t range)
887{
888	uint64_t r;
889
890	ASSERT3S(ztest_fd_rand, >=, 0);
891
892	if (range == 0)
893		return (0);
894
895	if (read(ztest_fd_rand, &r, sizeof (r)) != sizeof (r))
896		fatal(1, "short read from /dev/urandom");
897
898	return (r % range);
899}
900
901/* ARGSUSED */
902static void
903ztest_record_enospc(const char *s)
904{
905	ztest_shared->zs_enospc_count++;
906}
907
908static uint64_t
909ztest_get_ashift(void)
910{
911	if (ztest_opts.zo_ashift == 0)
912		return (SPA_MINBLOCKSHIFT + ztest_random(5));
913	return (ztest_opts.zo_ashift);
914}
915
916static nvlist_t *
917make_vdev_file(char *path, char *aux, char *pool, size_t size, uint64_t ashift)
918{
919	char pathbuf[MAXPATHLEN];
920	uint64_t vdev;
921	nvlist_t *file;
922
923	if (ashift == 0)
924		ashift = ztest_get_ashift();
925
926	if (path == NULL) {
927		path = pathbuf;
928
929		if (aux != NULL) {
930			vdev = ztest_shared->zs_vdev_aux;
931			(void) snprintf(path, sizeof (pathbuf),
932			    ztest_aux_template, ztest_opts.zo_dir,
933			    pool == NULL ? ztest_opts.zo_pool : pool,
934			    aux, vdev);
935		} else {
936			vdev = ztest_shared->zs_vdev_next_leaf++;
937			(void) snprintf(path, sizeof (pathbuf),
938			    ztest_dev_template, ztest_opts.zo_dir,
939			    pool == NULL ? ztest_opts.zo_pool : pool, vdev);
940		}
941	}
942
943	if (size != 0) {
944		int fd = open(path, O_RDWR | O_CREAT | O_TRUNC, 0666);
945		if (fd == -1)
946			fatal(1, "can't open %s", path);
947		if (ftruncate(fd, size) != 0)
948			fatal(1, "can't ftruncate %s", path);
949		(void) close(fd);
950	}
951
952	VERIFY(nvlist_alloc(&file, NV_UNIQUE_NAME, 0) == 0);
953	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_TYPE, VDEV_TYPE_FILE) == 0);
954	VERIFY(nvlist_add_string(file, ZPOOL_CONFIG_PATH, path) == 0);
955	VERIFY(nvlist_add_uint64(file, ZPOOL_CONFIG_ASHIFT, ashift) == 0);
956
957	return (file);
958}
959
960static nvlist_t *
961make_vdev_raidz(char *path, char *aux, char *pool, size_t size,
962    uint64_t ashift, int r)
963{
964	nvlist_t *raidz, **child;
965	int c;
966
967	if (r < 2)
968		return (make_vdev_file(path, aux, pool, size, ashift));
969	child = umem_alloc(r * sizeof (nvlist_t *), UMEM_NOFAIL);
970
971	for (c = 0; c < r; c++)
972		child[c] = make_vdev_file(path, aux, pool, size, ashift);
973
974	VERIFY(nvlist_alloc(&raidz, NV_UNIQUE_NAME, 0) == 0);
975	VERIFY(nvlist_add_string(raidz, ZPOOL_CONFIG_TYPE,
976	    VDEV_TYPE_RAIDZ) == 0);
977	VERIFY(nvlist_add_uint64(raidz, ZPOOL_CONFIG_NPARITY,
978	    ztest_opts.zo_raidz_parity) == 0);
979	VERIFY(nvlist_add_nvlist_array(raidz, ZPOOL_CONFIG_CHILDREN,
980	    child, r) == 0);
981
982	for (c = 0; c < r; c++)
983		nvlist_free(child[c]);
984
985	umem_free(child, r * sizeof (nvlist_t *));
986
987	return (raidz);
988}
989
990static nvlist_t *
991make_vdev_mirror(char *path, char *aux, char *pool, size_t size,
992    uint64_t ashift, int r, int m)
993{
994	nvlist_t *mirror, **child;
995	int c;
996
997	if (m < 1)
998		return (make_vdev_raidz(path, aux, pool, size, ashift, r));
999
1000	child = umem_alloc(m * sizeof (nvlist_t *), UMEM_NOFAIL);
1001
1002	for (c = 0; c < m; c++)
1003		child[c] = make_vdev_raidz(path, aux, pool, size, ashift, r);
1004
1005	VERIFY(nvlist_alloc(&mirror, NV_UNIQUE_NAME, 0) == 0);
1006	VERIFY(nvlist_add_string(mirror, ZPOOL_CONFIG_TYPE,
1007	    VDEV_TYPE_MIRROR) == 0);
1008	VERIFY(nvlist_add_nvlist_array(mirror, ZPOOL_CONFIG_CHILDREN,
1009	    child, m) == 0);
1010
1011	for (c = 0; c < m; c++)
1012		nvlist_free(child[c]);
1013
1014	umem_free(child, m * sizeof (nvlist_t *));
1015
1016	return (mirror);
1017}
1018
1019static nvlist_t *
1020make_vdev_root(char *path, char *aux, char *pool, size_t size, uint64_t ashift,
1021    const char *class, int r, int m, int t)
1022{
1023	nvlist_t *root, **child;
1024	int c;
1025	boolean_t log;
1026
1027	ASSERT(t > 0);
1028
1029	log = (class != NULL && strcmp(class, "log") == 0);
1030
1031	child = umem_alloc(t * sizeof (nvlist_t *), UMEM_NOFAIL);
1032
1033	for (c = 0; c < t; c++) {
1034		child[c] = make_vdev_mirror(path, aux, pool, size, ashift,
1035		    r, m);
1036		VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
1037		    log) == 0);
1038
1039		if (class != NULL && class[0] != '\0') {
1040			ASSERT(m > 1 || log);   /* expecting a mirror */
1041			VERIFY(nvlist_add_string(child[c],
1042			    ZPOOL_CONFIG_ALLOCATION_BIAS, class) == 0);
1043		}
1044	}
1045
1046	VERIFY(nvlist_alloc(&root, NV_UNIQUE_NAME, 0) == 0);
1047	VERIFY(nvlist_add_string(root, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT) == 0);
1048	VERIFY(nvlist_add_nvlist_array(root, aux ? aux : ZPOOL_CONFIG_CHILDREN,
1049	    child, t) == 0);
1050
1051	for (c = 0; c < t; c++)
1052		nvlist_free(child[c]);
1053
1054	umem_free(child, t * sizeof (nvlist_t *));
1055
1056	return (root);
1057}
1058
1059/*
1060 * Find a random spa version. Returns back a random spa version in the
1061 * range [initial_version, SPA_VERSION_FEATURES].
1062 */
1063static uint64_t
1064ztest_random_spa_version(uint64_t initial_version)
1065{
1066	uint64_t version = initial_version;
1067
1068	if (version <= SPA_VERSION_BEFORE_FEATURES) {
1069		version = version +
1070		    ztest_random(SPA_VERSION_BEFORE_FEATURES - version + 1);
1071	}
1072
1073	if (version > SPA_VERSION_BEFORE_FEATURES)
1074		version = SPA_VERSION_FEATURES;
1075
1076	ASSERT(SPA_VERSION_IS_SUPPORTED(version));
1077	return (version);
1078}
1079
1080static int
1081ztest_random_blocksize(void)
1082{
1083	uint64_t block_shift;
1084
1085	ASSERT(ztest_spa->spa_max_ashift != 0);
1086
1087	/*
1088	 * Choose a block size >= the ashift.
1089	 * If the SPA supports new MAXBLOCKSIZE, test up to 1MB blocks.
1090	 */
1091	int maxbs = SPA_OLD_MAXBLOCKSHIFT;
1092	if (spa_maxblocksize(ztest_spa) == SPA_MAXBLOCKSIZE)
1093		maxbs = 20;
1094	block_shift = ztest_random(maxbs - ztest_spa->spa_max_ashift + 1);
1095	return (1 << (SPA_MINBLOCKSHIFT + block_shift));
1096}
1097
1098static int
1099ztest_random_dnodesize(void)
1100{
1101	int slots;
1102	int max_slots = spa_maxdnodesize(ztest_spa) >> DNODE_SHIFT;
1103
1104	if (max_slots == DNODE_MIN_SLOTS)
1105		return (DNODE_MIN_SIZE);
1106
1107	/*
1108	 * Weight the random distribution more heavily toward smaller
1109	 * dnode sizes since that is more likely to reflect real-world
1110	 * usage.
1111	 */
1112	ASSERT3U(max_slots, >, 4);
1113	switch (ztest_random(10)) {
1114	case 0:
1115		slots = 5 + ztest_random(max_slots - 4);
1116		break;
1117	case 1 ... 4:
1118		slots = 2 + ztest_random(3);
1119		break;
1120	default:
1121		slots = 1;
1122		break;
1123	}
1124
1125	return (slots << DNODE_SHIFT);
1126}
1127
1128static int
1129ztest_random_ibshift(void)
1130{
1131	return (DN_MIN_INDBLKSHIFT +
1132	    ztest_random(DN_MAX_INDBLKSHIFT - DN_MIN_INDBLKSHIFT + 1));
1133}
1134
1135static uint64_t
1136ztest_random_vdev_top(spa_t *spa, boolean_t log_ok)
1137{
1138	uint64_t top;
1139	vdev_t *rvd = spa->spa_root_vdev;
1140	vdev_t *tvd;
1141
1142	ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1143
1144	do {
1145		top = ztest_random(rvd->vdev_children);
1146		tvd = rvd->vdev_child[top];
1147	} while (!vdev_is_concrete(tvd) || (tvd->vdev_islog && !log_ok) ||
1148	    tvd->vdev_mg == NULL || tvd->vdev_mg->mg_class == NULL);
1149
1150	return (top);
1151}
1152
1153static uint64_t
1154ztest_random_dsl_prop(zfs_prop_t prop)
1155{
1156	uint64_t value;
1157
1158	do {
1159		value = zfs_prop_random_value(prop, ztest_random(-1ULL));
1160	} while (prop == ZFS_PROP_CHECKSUM && value == ZIO_CHECKSUM_OFF);
1161
1162	return (value);
1163}
1164
1165static int
1166ztest_dsl_prop_set_uint64(char *osname, zfs_prop_t prop, uint64_t value,
1167    boolean_t inherit)
1168{
1169	const char *propname = zfs_prop_to_name(prop);
1170	const char *valname;
1171	char setpoint[MAXPATHLEN];
1172	uint64_t curval;
1173	int error;
1174
1175	error = dsl_prop_set_int(osname, propname,
1176	    (inherit ? ZPROP_SRC_NONE : ZPROP_SRC_LOCAL), value);
1177
1178	if (error == ENOSPC) {
1179		ztest_record_enospc(FTAG);
1180		return (error);
1181	}
1182	ASSERT0(error);
1183
1184	VERIFY0(dsl_prop_get_integer(osname, propname, &curval, setpoint));
1185
1186	if (ztest_opts.zo_verbose >= 6) {
1187		VERIFY(zfs_prop_index_to_string(prop, curval, &valname) == 0);
1188		(void) printf("%s %s = %s at '%s'\n",
1189		    osname, propname, valname, setpoint);
1190	}
1191
1192	return (error);
1193}
1194
1195static int
1196ztest_spa_prop_set_uint64(zpool_prop_t prop, uint64_t value)
1197{
1198	spa_t *spa = ztest_spa;
1199	nvlist_t *props = NULL;
1200	int error;
1201
1202	VERIFY(nvlist_alloc(&props, NV_UNIQUE_NAME, 0) == 0);
1203	VERIFY(nvlist_add_uint64(props, zpool_prop_to_name(prop), value) == 0);
1204
1205	error = spa_prop_set(spa, props);
1206
1207	nvlist_free(props);
1208
1209	if (error == ENOSPC) {
1210		ztest_record_enospc(FTAG);
1211		return (error);
1212	}
1213	ASSERT0(error);
1214
1215	return (error);
1216}
1217
1218static void
1219ztest_rll_init(rll_t *rll)
1220{
1221	rll->rll_writer = NULL;
1222	rll->rll_readers = 0;
1223	mutex_init(&rll->rll_lock, NULL, USYNC_THREAD, NULL);
1224	cv_init(&rll->rll_cv, NULL, USYNC_THREAD, NULL);
1225}
1226
1227static void
1228ztest_rll_destroy(rll_t *rll)
1229{
1230	ASSERT(rll->rll_writer == NULL);
1231	ASSERT(rll->rll_readers == 0);
1232	mutex_destroy(&rll->rll_lock);
1233	cv_destroy(&rll->rll_cv);
1234}
1235
1236static void
1237ztest_rll_lock(rll_t *rll, rl_type_t type)
1238{
1239	mutex_enter(&rll->rll_lock);
1240
1241	if (type == RL_READER) {
1242		while (rll->rll_writer != NULL)
1243			cv_wait(&rll->rll_cv, &rll->rll_lock);
1244		rll->rll_readers++;
1245	} else {
1246		while (rll->rll_writer != NULL || rll->rll_readers)
1247			cv_wait(&rll->rll_cv, &rll->rll_lock);
1248		rll->rll_writer = curthread;
1249	}
1250
1251	mutex_exit(&rll->rll_lock);
1252}
1253
1254static void
1255ztest_rll_unlock(rll_t *rll)
1256{
1257	mutex_enter(&rll->rll_lock);
1258
1259	if (rll->rll_writer) {
1260		ASSERT(rll->rll_readers == 0);
1261		rll->rll_writer = NULL;
1262	} else {
1263		ASSERT(rll->rll_readers != 0);
1264		ASSERT(rll->rll_writer == NULL);
1265		rll->rll_readers--;
1266	}
1267
1268	if (rll->rll_writer == NULL && rll->rll_readers == 0)
1269		cv_broadcast(&rll->rll_cv);
1270
1271	mutex_exit(&rll->rll_lock);
1272}
1273
1274static void
1275ztest_object_lock(ztest_ds_t *zd, uint64_t object, rl_type_t type)
1276{
1277	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1278
1279	ztest_rll_lock(rll, type);
1280}
1281
1282static void
1283ztest_object_unlock(ztest_ds_t *zd, uint64_t object)
1284{
1285	rll_t *rll = &zd->zd_object_lock[object & (ZTEST_OBJECT_LOCKS - 1)];
1286
1287	ztest_rll_unlock(rll);
1288}
1289
1290static rl_t *
1291ztest_range_lock(ztest_ds_t *zd, uint64_t object, uint64_t offset,
1292    uint64_t size, rl_type_t type)
1293{
1294	uint64_t hash = object ^ (offset % (ZTEST_RANGE_LOCKS + 1));
1295	rll_t *rll = &zd->zd_range_lock[hash & (ZTEST_RANGE_LOCKS - 1)];
1296	rl_t *rl;
1297
1298	rl = umem_alloc(sizeof (*rl), UMEM_NOFAIL);
1299	rl->rl_object = object;
1300	rl->rl_offset = offset;
1301	rl->rl_size = size;
1302	rl->rl_lock = rll;
1303
1304	ztest_rll_lock(rll, type);
1305
1306	return (rl);
1307}
1308
1309static void
1310ztest_range_unlock(rl_t *rl)
1311{
1312	rll_t *rll = rl->rl_lock;
1313
1314	ztest_rll_unlock(rll);
1315
1316	umem_free(rl, sizeof (*rl));
1317}
1318
1319static void
1320ztest_zd_init(ztest_ds_t *zd, ztest_shared_ds_t *szd, objset_t *os)
1321{
1322	zd->zd_os = os;
1323	zd->zd_zilog = dmu_objset_zil(os);
1324	zd->zd_shared = szd;
1325	dmu_objset_name(os, zd->zd_name);
1326
1327	if (zd->zd_shared != NULL)
1328		zd->zd_shared->zd_seq = 0;
1329
1330	rw_init(&zd->zd_zilog_lock, NULL, USYNC_THREAD, NULL);
1331	mutex_init(&zd->zd_dirobj_lock, NULL, USYNC_THREAD, NULL);
1332
1333	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1334		ztest_rll_init(&zd->zd_object_lock[l]);
1335
1336	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1337		ztest_rll_init(&zd->zd_range_lock[l]);
1338}
1339
1340static void
1341ztest_zd_fini(ztest_ds_t *zd)
1342{
1343	mutex_destroy(&zd->zd_dirobj_lock);
1344
1345	for (int l = 0; l < ZTEST_OBJECT_LOCKS; l++)
1346		ztest_rll_destroy(&zd->zd_object_lock[l]);
1347
1348	for (int l = 0; l < ZTEST_RANGE_LOCKS; l++)
1349		ztest_rll_destroy(&zd->zd_range_lock[l]);
1350}
1351
1352#define	TXG_MIGHTWAIT	(ztest_random(10) == 0 ? TXG_NOWAIT : TXG_WAIT)
1353
1354static uint64_t
1355ztest_tx_assign(dmu_tx_t *tx, uint64_t txg_how, const char *tag)
1356{
1357	uint64_t txg;
1358	int error;
1359
1360	/*
1361	 * Attempt to assign tx to some transaction group.
1362	 */
1363	error = dmu_tx_assign(tx, txg_how);
1364	if (error) {
1365		if (error == ERESTART) {
1366			ASSERT(txg_how == TXG_NOWAIT);
1367			dmu_tx_wait(tx);
1368		} else {
1369			ASSERT3U(error, ==, ENOSPC);
1370			ztest_record_enospc(tag);
1371		}
1372		dmu_tx_abort(tx);
1373		return (0);
1374	}
1375	txg = dmu_tx_get_txg(tx);
1376	ASSERT(txg != 0);
1377	return (txg);
1378}
1379
1380static void
1381ztest_pattern_set(void *buf, uint64_t size, uint64_t value)
1382{
1383	uint64_t *ip = buf;
1384	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1385
1386	while (ip < ip_end)
1387		*ip++ = value;
1388}
1389
1390static boolean_t
1391ztest_pattern_match(void *buf, uint64_t size, uint64_t value)
1392{
1393	uint64_t *ip = buf;
1394	uint64_t *ip_end = (uint64_t *)((uintptr_t)buf + (uintptr_t)size);
1395	uint64_t diff = 0;
1396
1397	while (ip < ip_end)
1398		diff |= (value - *ip++);
1399
1400	return (diff == 0);
1401}
1402
1403static void
1404ztest_bt_generate(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1405    uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1406    uint64_t crtxg)
1407{
1408	bt->bt_magic = BT_MAGIC;
1409	bt->bt_objset = dmu_objset_id(os);
1410	bt->bt_object = object;
1411	bt->bt_dnodesize = dnodesize;
1412	bt->bt_offset = offset;
1413	bt->bt_gen = gen;
1414	bt->bt_txg = txg;
1415	bt->bt_crtxg = crtxg;
1416}
1417
1418static void
1419ztest_bt_verify(ztest_block_tag_t *bt, objset_t *os, uint64_t object,
1420    uint64_t dnodesize, uint64_t offset, uint64_t gen, uint64_t txg,
1421    uint64_t crtxg)
1422{
1423	ASSERT3U(bt->bt_magic, ==, BT_MAGIC);
1424	ASSERT3U(bt->bt_objset, ==, dmu_objset_id(os));
1425	ASSERT3U(bt->bt_object, ==, object);
1426	ASSERT3U(bt->bt_dnodesize, ==, dnodesize);
1427	ASSERT3U(bt->bt_offset, ==, offset);
1428	ASSERT3U(bt->bt_gen, <=, gen);
1429	ASSERT3U(bt->bt_txg, <=, txg);
1430	ASSERT3U(bt->bt_crtxg, ==, crtxg);
1431}
1432
1433static ztest_block_tag_t *
1434ztest_bt_bonus(dmu_buf_t *db)
1435{
1436	dmu_object_info_t doi;
1437	ztest_block_tag_t *bt;
1438
1439	dmu_object_info_from_db(db, &doi);
1440	ASSERT3U(doi.doi_bonus_size, <=, db->db_size);
1441	ASSERT3U(doi.doi_bonus_size, >=, sizeof (*bt));
1442	bt = (void *)((char *)db->db_data + doi.doi_bonus_size - sizeof (*bt));
1443
1444	return (bt);
1445}
1446
1447/*
1448 * Generate a token to fill up unused bonus buffer space.  Try to make
1449 * it unique to the object, generation, and offset to verify that data
1450 * is not getting overwritten by data from other dnodes.
1451 */
1452#define	ZTEST_BONUS_FILL_TOKEN(obj, ds, gen, offset)	\
1453	(((ds) << 48) | ((gen) << 32) | ((obj) << 8) | (offset))
1454
1455/*
1456 * Fill up the unused bonus buffer region before the block tag with a
1457 * verifiable pattern. Filling the whole bonus area with non-zero data
1458 * helps ensure that all dnode traversal code properly skips the
1459 * interior regions of large dnodes.
1460 */
1461void
1462ztest_fill_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1463    objset_t *os, uint64_t gen)
1464{
1465	uint64_t *bonusp;
1466
1467	ASSERT(IS_P2ALIGNED((char *)end - (char *)db->db_data, 8));
1468
1469	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1470		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1471		    gen, bonusp - (uint64_t *)db->db_data);
1472		*bonusp = token;
1473	}
1474}
1475
1476/*
1477 * Verify that the unused area of a bonus buffer is filled with the
1478 * expected tokens.
1479 */
1480void
1481ztest_verify_unused_bonus(dmu_buf_t *db, void *end, uint64_t obj,
1482    objset_t *os, uint64_t gen)
1483{
1484	uint64_t *bonusp;
1485
1486	for (bonusp = db->db_data; bonusp < (uint64_t *)end; bonusp++) {
1487		uint64_t token = ZTEST_BONUS_FILL_TOKEN(obj, dmu_objset_id(os),
1488		    gen, bonusp - (uint64_t *)db->db_data);
1489		VERIFY3U(*bonusp, ==, token);
1490	}
1491}
1492
1493/*
1494 * ZIL logging ops
1495 */
1496
1497#define	lrz_type	lr_mode
1498#define	lrz_blocksize	lr_uid
1499#define	lrz_ibshift	lr_gid
1500#define	lrz_bonustype	lr_rdev
1501#define	lrz_dnodesize	lr_crtime[1]
1502
1503static void
1504ztest_log_create(ztest_ds_t *zd, dmu_tx_t *tx, lr_create_t *lr)
1505{
1506	char *name = (void *)(lr + 1);		/* name follows lr */
1507	size_t namesize = strlen(name) + 1;
1508	itx_t *itx;
1509
1510	if (zil_replaying(zd->zd_zilog, tx))
1511		return;
1512
1513	itx = zil_itx_create(TX_CREATE, sizeof (*lr) + namesize);
1514	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1515	    sizeof (*lr) + namesize - sizeof (lr_t));
1516
1517	zil_itx_assign(zd->zd_zilog, itx, tx);
1518}
1519
1520static void
1521ztest_log_remove(ztest_ds_t *zd, dmu_tx_t *tx, lr_remove_t *lr, uint64_t object)
1522{
1523	char *name = (void *)(lr + 1);		/* name follows lr */
1524	size_t namesize = strlen(name) + 1;
1525	itx_t *itx;
1526
1527	if (zil_replaying(zd->zd_zilog, tx))
1528		return;
1529
1530	itx = zil_itx_create(TX_REMOVE, sizeof (*lr) + namesize);
1531	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1532	    sizeof (*lr) + namesize - sizeof (lr_t));
1533
1534	itx->itx_oid = object;
1535	zil_itx_assign(zd->zd_zilog, itx, tx);
1536}
1537
1538static void
1539ztest_log_write(ztest_ds_t *zd, dmu_tx_t *tx, lr_write_t *lr)
1540{
1541	itx_t *itx;
1542	itx_wr_state_t write_state = ztest_random(WR_NUM_STATES);
1543
1544	if (zil_replaying(zd->zd_zilog, tx))
1545		return;
1546
1547	if (lr->lr_length > ZIL_MAX_LOG_DATA)
1548		write_state = WR_INDIRECT;
1549
1550	itx = zil_itx_create(TX_WRITE,
1551	    sizeof (*lr) + (write_state == WR_COPIED ? lr->lr_length : 0));
1552
1553	if (write_state == WR_COPIED &&
1554	    dmu_read(zd->zd_os, lr->lr_foid, lr->lr_offset, lr->lr_length,
1555	    ((lr_write_t *)&itx->itx_lr) + 1, DMU_READ_NO_PREFETCH) != 0) {
1556		zil_itx_destroy(itx);
1557		itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1558		write_state = WR_NEED_COPY;
1559	}
1560	itx->itx_private = zd;
1561	itx->itx_wr_state = write_state;
1562	itx->itx_sync = (ztest_random(8) == 0);
1563
1564	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1565	    sizeof (*lr) - sizeof (lr_t));
1566
1567	zil_itx_assign(zd->zd_zilog, itx, tx);
1568}
1569
1570static void
1571ztest_log_truncate(ztest_ds_t *zd, dmu_tx_t *tx, lr_truncate_t *lr)
1572{
1573	itx_t *itx;
1574
1575	if (zil_replaying(zd->zd_zilog, tx))
1576		return;
1577
1578	itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1579	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1580	    sizeof (*lr) - sizeof (lr_t));
1581
1582	itx->itx_sync = B_FALSE;
1583	zil_itx_assign(zd->zd_zilog, itx, tx);
1584}
1585
1586static void
1587ztest_log_setattr(ztest_ds_t *zd, dmu_tx_t *tx, lr_setattr_t *lr)
1588{
1589	itx_t *itx;
1590
1591	if (zil_replaying(zd->zd_zilog, tx))
1592		return;
1593
1594	itx = zil_itx_create(TX_SETATTR, sizeof (*lr));
1595	bcopy(&lr->lr_common + 1, &itx->itx_lr + 1,
1596	    sizeof (*lr) - sizeof (lr_t));
1597
1598	itx->itx_sync = B_FALSE;
1599	zil_itx_assign(zd->zd_zilog, itx, tx);
1600}
1601
1602/*
1603 * ZIL replay ops
1604 */
1605static int
1606ztest_replay_create(void *arg1, void *arg2, boolean_t byteswap)
1607{
1608	ztest_ds_t *zd = arg1;
1609	lr_create_t *lr = arg2;
1610	char *name = (void *)(lr + 1);		/* name follows lr */
1611	objset_t *os = zd->zd_os;
1612	ztest_block_tag_t *bbt;
1613	dmu_buf_t *db;
1614	dmu_tx_t *tx;
1615	uint64_t txg;
1616	int error = 0;
1617	int bonuslen;
1618
1619	if (byteswap)
1620		byteswap_uint64_array(lr, sizeof (*lr));
1621
1622	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1623	ASSERT(name[0] != '\0');
1624
1625	tx = dmu_tx_create(os);
1626
1627	dmu_tx_hold_zap(tx, lr->lr_doid, B_TRUE, name);
1628
1629	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1630		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, B_TRUE, NULL);
1631	} else {
1632		dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1633	}
1634
1635	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1636	if (txg == 0)
1637		return (ENOSPC);
1638
1639	ASSERT(dmu_objset_zil(os)->zl_replay == !!lr->lr_foid);
1640	bonuslen = DN_BONUS_SIZE(lr->lrz_dnodesize);
1641
1642	if (lr->lrz_type == DMU_OT_ZAP_OTHER) {
1643		if (lr->lr_foid == 0) {
1644			lr->lr_foid = zap_create_dnsize(os,
1645			    lr->lrz_type, lr->lrz_bonustype,
1646			    bonuslen, lr->lrz_dnodesize, tx);
1647		} else {
1648			error = zap_create_claim_dnsize(os, lr->lr_foid,
1649			    lr->lrz_type, lr->lrz_bonustype,
1650			    bonuslen, lr->lrz_dnodesize, tx);
1651		}
1652	} else {
1653		if (lr->lr_foid == 0) {
1654			lr->lr_foid = dmu_object_alloc_dnsize(os,
1655			    lr->lrz_type, 0, lr->lrz_bonustype,
1656			    bonuslen, lr->lrz_dnodesize, tx);
1657		} else {
1658			error = dmu_object_claim_dnsize(os, lr->lr_foid,
1659			    lr->lrz_type, 0, lr->lrz_bonustype,
1660			    bonuslen, lr->lrz_dnodesize, tx);
1661		}
1662	}
1663
1664	if (error) {
1665		ASSERT3U(error, ==, EEXIST);
1666		ASSERT(zd->zd_zilog->zl_replay);
1667		dmu_tx_commit(tx);
1668		return (error);
1669	}
1670
1671	ASSERT(lr->lr_foid != 0);
1672
1673	if (lr->lrz_type != DMU_OT_ZAP_OTHER)
1674		VERIFY3U(0, ==, dmu_object_set_blocksize(os, lr->lr_foid,
1675		    lr->lrz_blocksize, lr->lrz_ibshift, tx));
1676
1677	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1678	bbt = ztest_bt_bonus(db);
1679	dmu_buf_will_dirty(db, tx);
1680	ztest_bt_generate(bbt, os, lr->lr_foid, lr->lrz_dnodesize, -1ULL,
1681	    lr->lr_gen, txg, txg);
1682	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, lr->lr_gen);
1683	dmu_buf_rele(db, FTAG);
1684
1685	VERIFY3U(0, ==, zap_add(os, lr->lr_doid, name, sizeof (uint64_t), 1,
1686	    &lr->lr_foid, tx));
1687
1688	(void) ztest_log_create(zd, tx, lr);
1689
1690	dmu_tx_commit(tx);
1691
1692	return (0);
1693}
1694
1695static int
1696ztest_replay_remove(void *arg1, void *arg2, boolean_t byteswap)
1697{
1698	ztest_ds_t *zd = arg1;
1699	lr_remove_t *lr = arg2;
1700	char *name = (void *)(lr + 1);		/* name follows lr */
1701	objset_t *os = zd->zd_os;
1702	dmu_object_info_t doi;
1703	dmu_tx_t *tx;
1704	uint64_t object, txg;
1705
1706	if (byteswap)
1707		byteswap_uint64_array(lr, sizeof (*lr));
1708
1709	ASSERT(lr->lr_doid == ZTEST_DIROBJ);
1710	ASSERT(name[0] != '\0');
1711
1712	VERIFY3U(0, ==,
1713	    zap_lookup(os, lr->lr_doid, name, sizeof (object), 1, &object));
1714	ASSERT(object != 0);
1715
1716	ztest_object_lock(zd, object, RL_WRITER);
1717
1718	VERIFY3U(0, ==, dmu_object_info(os, object, &doi));
1719
1720	tx = dmu_tx_create(os);
1721
1722	dmu_tx_hold_zap(tx, lr->lr_doid, B_FALSE, name);
1723	dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END);
1724
1725	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1726	if (txg == 0) {
1727		ztest_object_unlock(zd, object);
1728		return (ENOSPC);
1729	}
1730
1731	if (doi.doi_type == DMU_OT_ZAP_OTHER) {
1732		VERIFY3U(0, ==, zap_destroy(os, object, tx));
1733	} else {
1734		VERIFY3U(0, ==, dmu_object_free(os, object, tx));
1735	}
1736
1737	VERIFY3U(0, ==, zap_remove(os, lr->lr_doid, name, tx));
1738
1739	(void) ztest_log_remove(zd, tx, lr, object);
1740
1741	dmu_tx_commit(tx);
1742
1743	ztest_object_unlock(zd, object);
1744
1745	return (0);
1746}
1747
1748static int
1749ztest_replay_write(void *arg1, void *arg2, boolean_t byteswap)
1750{
1751	ztest_ds_t *zd = arg1;
1752	lr_write_t *lr = arg2;
1753	objset_t *os = zd->zd_os;
1754	void *data = lr + 1;			/* data follows lr */
1755	uint64_t offset, length;
1756	ztest_block_tag_t *bt = data;
1757	ztest_block_tag_t *bbt;
1758	uint64_t gen, txg, lrtxg, crtxg;
1759	dmu_object_info_t doi;
1760	dmu_tx_t *tx;
1761	dmu_buf_t *db;
1762	arc_buf_t *abuf = NULL;
1763	rl_t *rl;
1764
1765	if (byteswap)
1766		byteswap_uint64_array(lr, sizeof (*lr));
1767
1768	offset = lr->lr_offset;
1769	length = lr->lr_length;
1770
1771	/* If it's a dmu_sync() block, write the whole block */
1772	if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
1773		uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
1774		if (length < blocksize) {
1775			offset -= offset % blocksize;
1776			length = blocksize;
1777		}
1778	}
1779
1780	if (bt->bt_magic == BSWAP_64(BT_MAGIC))
1781		byteswap_uint64_array(bt, sizeof (*bt));
1782
1783	if (bt->bt_magic != BT_MAGIC)
1784		bt = NULL;
1785
1786	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1787	rl = ztest_range_lock(zd, lr->lr_foid, offset, length, RL_WRITER);
1788
1789	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1790
1791	dmu_object_info_from_db(db, &doi);
1792
1793	bbt = ztest_bt_bonus(db);
1794	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1795	gen = bbt->bt_gen;
1796	crtxg = bbt->bt_crtxg;
1797	lrtxg = lr->lr_common.lrc_txg;
1798
1799	tx = dmu_tx_create(os);
1800
1801	dmu_tx_hold_write(tx, lr->lr_foid, offset, length);
1802
1803	if (ztest_random(8) == 0 && length == doi.doi_data_block_size &&
1804	    P2PHASE(offset, length) == 0)
1805		abuf = dmu_request_arcbuf(db, length);
1806
1807	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1808	if (txg == 0) {
1809		if (abuf != NULL)
1810			dmu_return_arcbuf(abuf);
1811		dmu_buf_rele(db, FTAG);
1812		ztest_range_unlock(rl);
1813		ztest_object_unlock(zd, lr->lr_foid);
1814		return (ENOSPC);
1815	}
1816
1817	if (bt != NULL) {
1818		/*
1819		 * Usually, verify the old data before writing new data --
1820		 * but not always, because we also want to verify correct
1821		 * behavior when the data was not recently read into cache.
1822		 */
1823		ASSERT(offset % doi.doi_data_block_size == 0);
1824		if (ztest_random(4) != 0) {
1825			int prefetch = ztest_random(2) ?
1826			    DMU_READ_PREFETCH : DMU_READ_NO_PREFETCH;
1827			ztest_block_tag_t rbt;
1828
1829			VERIFY(dmu_read(os, lr->lr_foid, offset,
1830			    sizeof (rbt), &rbt, prefetch) == 0);
1831			if (rbt.bt_magic == BT_MAGIC) {
1832				ztest_bt_verify(&rbt, os, lr->lr_foid, 0,
1833				    offset, gen, txg, crtxg);
1834			}
1835		}
1836
1837		/*
1838		 * Writes can appear to be newer than the bonus buffer because
1839		 * the ztest_get_data() callback does a dmu_read() of the
1840		 * open-context data, which may be different than the data
1841		 * as it was when the write was generated.
1842		 */
1843		if (zd->zd_zilog->zl_replay) {
1844			ztest_bt_verify(bt, os, lr->lr_foid, 0, offset,
1845			    MAX(gen, bt->bt_gen), MAX(txg, lrtxg),
1846			    bt->bt_crtxg);
1847		}
1848
1849		/*
1850		 * Set the bt's gen/txg to the bonus buffer's gen/txg
1851		 * so that all of the usual ASSERTs will work.
1852		 */
1853		ztest_bt_generate(bt, os, lr->lr_foid, 0, offset, gen, txg,
1854		    crtxg);
1855	}
1856
1857	if (abuf == NULL) {
1858		dmu_write(os, lr->lr_foid, offset, length, data, tx);
1859	} else {
1860		bcopy(data, abuf->b_data, length);
1861		dmu_assign_arcbuf(db, offset, abuf, tx);
1862	}
1863
1864	(void) ztest_log_write(zd, tx, lr);
1865
1866	dmu_buf_rele(db, FTAG);
1867
1868	dmu_tx_commit(tx);
1869
1870	ztest_range_unlock(rl);
1871	ztest_object_unlock(zd, lr->lr_foid);
1872
1873	return (0);
1874}
1875
1876static int
1877ztest_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
1878{
1879	ztest_ds_t *zd = arg1;
1880	lr_truncate_t *lr = arg2;
1881	objset_t *os = zd->zd_os;
1882	dmu_tx_t *tx;
1883	uint64_t txg;
1884	rl_t *rl;
1885
1886	if (byteswap)
1887		byteswap_uint64_array(lr, sizeof (*lr));
1888
1889	ztest_object_lock(zd, lr->lr_foid, RL_READER);
1890	rl = ztest_range_lock(zd, lr->lr_foid, lr->lr_offset, lr->lr_length,
1891	    RL_WRITER);
1892
1893	tx = dmu_tx_create(os);
1894
1895	dmu_tx_hold_free(tx, lr->lr_foid, lr->lr_offset, lr->lr_length);
1896
1897	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1898	if (txg == 0) {
1899		ztest_range_unlock(rl);
1900		ztest_object_unlock(zd, lr->lr_foid);
1901		return (ENOSPC);
1902	}
1903
1904	VERIFY(dmu_free_range(os, lr->lr_foid, lr->lr_offset,
1905	    lr->lr_length, tx) == 0);
1906
1907	(void) ztest_log_truncate(zd, tx, lr);
1908
1909	dmu_tx_commit(tx);
1910
1911	ztest_range_unlock(rl);
1912	ztest_object_unlock(zd, lr->lr_foid);
1913
1914	return (0);
1915}
1916
1917static int
1918ztest_replay_setattr(void *arg1, void *arg2, boolean_t byteswap)
1919{
1920	ztest_ds_t *zd = arg1;
1921	lr_setattr_t *lr = arg2;
1922	objset_t *os = zd->zd_os;
1923	dmu_tx_t *tx;
1924	dmu_buf_t *db;
1925	ztest_block_tag_t *bbt;
1926	uint64_t txg, lrtxg, crtxg, dnodesize;
1927
1928	if (byteswap)
1929		byteswap_uint64_array(lr, sizeof (*lr));
1930
1931	ztest_object_lock(zd, lr->lr_foid, RL_WRITER);
1932
1933	VERIFY3U(0, ==, dmu_bonus_hold(os, lr->lr_foid, FTAG, &db));
1934
1935	tx = dmu_tx_create(os);
1936	dmu_tx_hold_bonus(tx, lr->lr_foid);
1937
1938	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
1939	if (txg == 0) {
1940		dmu_buf_rele(db, FTAG);
1941		ztest_object_unlock(zd, lr->lr_foid);
1942		return (ENOSPC);
1943	}
1944
1945	bbt = ztest_bt_bonus(db);
1946	ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
1947	crtxg = bbt->bt_crtxg;
1948	lrtxg = lr->lr_common.lrc_txg;
1949	dnodesize = bbt->bt_dnodesize;
1950
1951	if (zd->zd_zilog->zl_replay) {
1952		ASSERT(lr->lr_size != 0);
1953		ASSERT(lr->lr_mode != 0);
1954		ASSERT(lrtxg != 0);
1955	} else {
1956		/*
1957		 * Randomly change the size and increment the generation.
1958		 */
1959		lr->lr_size = (ztest_random(db->db_size / sizeof (*bbt)) + 1) *
1960		    sizeof (*bbt);
1961		lr->lr_mode = bbt->bt_gen + 1;
1962		ASSERT(lrtxg == 0);
1963	}
1964
1965	/*
1966	 * Verify that the current bonus buffer is not newer than our txg.
1967	 */
1968	ztest_bt_verify(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
1969	    MAX(txg, lrtxg), crtxg);
1970
1971	dmu_buf_will_dirty(db, tx);
1972
1973	ASSERT3U(lr->lr_size, >=, sizeof (*bbt));
1974	ASSERT3U(lr->lr_size, <=, db->db_size);
1975	VERIFY0(dmu_set_bonus(db, lr->lr_size, tx));
1976	bbt = ztest_bt_bonus(db);
1977
1978	ztest_bt_generate(bbt, os, lr->lr_foid, dnodesize, -1ULL, lr->lr_mode,
1979	    txg, crtxg);
1980	ztest_fill_unused_bonus(db, bbt, lr->lr_foid, os, bbt->bt_gen);
1981
1982	dmu_buf_rele(db, FTAG);
1983
1984	(void) ztest_log_setattr(zd, tx, lr);
1985
1986	dmu_tx_commit(tx);
1987
1988	ztest_object_unlock(zd, lr->lr_foid);
1989
1990	return (0);
1991}
1992
1993zil_replay_func_t *ztest_replay_vector[TX_MAX_TYPE] = {
1994	NULL,			/* 0 no such transaction type */
1995	ztest_replay_create,	/* TX_CREATE */
1996	NULL,			/* TX_MKDIR */
1997	NULL,			/* TX_MKXATTR */
1998	NULL,			/* TX_SYMLINK */
1999	ztest_replay_remove,	/* TX_REMOVE */
2000	NULL,			/* TX_RMDIR */
2001	NULL,			/* TX_LINK */
2002	NULL,			/* TX_RENAME */
2003	ztest_replay_write,	/* TX_WRITE */
2004	ztest_replay_truncate,	/* TX_TRUNCATE */
2005	ztest_replay_setattr,	/* TX_SETATTR */
2006	NULL,			/* TX_ACL */
2007	NULL,			/* TX_CREATE_ACL */
2008	NULL,			/* TX_CREATE_ATTR */
2009	NULL,			/* TX_CREATE_ACL_ATTR */
2010	NULL,			/* TX_MKDIR_ACL */
2011	NULL,			/* TX_MKDIR_ATTR */
2012	NULL,			/* TX_MKDIR_ACL_ATTR */
2013	NULL,			/* TX_WRITE2 */
2014};
2015
2016/*
2017 * ZIL get_data callbacks
2018 */
2019
2020/* ARGSUSED */
2021static void
2022ztest_get_done(zgd_t *zgd, int error)
2023{
2024	ztest_ds_t *zd = zgd->zgd_private;
2025	uint64_t object = ((rl_t *)zgd->zgd_lr)->rl_object;
2026
2027	if (zgd->zgd_db)
2028		dmu_buf_rele(zgd->zgd_db, zgd);
2029
2030	ztest_range_unlock((rl_t *)zgd->zgd_lr);
2031	ztest_object_unlock(zd, object);
2032
2033	umem_free(zgd, sizeof (*zgd));
2034}
2035
2036static int
2037ztest_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb,
2038    zio_t *zio)
2039{
2040	ztest_ds_t *zd = arg;
2041	objset_t *os = zd->zd_os;
2042	uint64_t object = lr->lr_foid;
2043	uint64_t offset = lr->lr_offset;
2044	uint64_t size = lr->lr_length;
2045	uint64_t txg = lr->lr_common.lrc_txg;
2046	uint64_t crtxg;
2047	dmu_object_info_t doi;
2048	dmu_buf_t *db;
2049	zgd_t *zgd;
2050	int error;
2051
2052	ASSERT3P(lwb, !=, NULL);
2053	ASSERT3P(zio, !=, NULL);
2054	ASSERT3U(size, !=, 0);
2055
2056	ztest_object_lock(zd, object, RL_READER);
2057	error = dmu_bonus_hold(os, object, FTAG, &db);
2058	if (error) {
2059		ztest_object_unlock(zd, object);
2060		return (error);
2061	}
2062
2063	crtxg = ztest_bt_bonus(db)->bt_crtxg;
2064
2065	if (crtxg == 0 || crtxg > txg) {
2066		dmu_buf_rele(db, FTAG);
2067		ztest_object_unlock(zd, object);
2068		return (ENOENT);
2069	}
2070
2071	dmu_object_info_from_db(db, &doi);
2072	dmu_buf_rele(db, FTAG);
2073	db = NULL;
2074
2075	zgd = umem_zalloc(sizeof (*zgd), UMEM_NOFAIL);
2076	zgd->zgd_lwb = lwb;
2077	zgd->zgd_private = zd;
2078
2079	if (buf != NULL) {	/* immediate write */
2080		zgd->zgd_lr = (struct locked_range *)ztest_range_lock(zd,
2081		    object, offset, size, RL_READER);
2082
2083		error = dmu_read(os, object, offset, size, buf,
2084		    DMU_READ_NO_PREFETCH);
2085		ASSERT(error == 0);
2086	} else {
2087		size = doi.doi_data_block_size;
2088		if (ISP2(size)) {
2089			offset = P2ALIGN(offset, size);
2090		} else {
2091			ASSERT(offset < size);
2092			offset = 0;
2093		}
2094
2095		zgd->zgd_lr = (struct locked_range *)ztest_range_lock(zd,
2096		    object, offset, size, RL_READER);
2097
2098		error = dmu_buf_hold(os, object, offset, zgd, &db,
2099		    DMU_READ_NO_PREFETCH);
2100
2101		if (error == 0) {
2102			blkptr_t *bp = &lr->lr_blkptr;
2103
2104			zgd->zgd_db = db;
2105			zgd->zgd_bp = bp;
2106
2107			ASSERT(db->db_offset == offset);
2108			ASSERT(db->db_size == size);
2109
2110			error = dmu_sync(zio, lr->lr_common.lrc_txg,
2111			    ztest_get_done, zgd);
2112
2113			if (error == 0)
2114				return (0);
2115		}
2116	}
2117
2118	ztest_get_done(zgd, error);
2119
2120	return (error);
2121}
2122
2123static void *
2124ztest_lr_alloc(size_t lrsize, char *name)
2125{
2126	char *lr;
2127	size_t namesize = name ? strlen(name) + 1 : 0;
2128
2129	lr = umem_zalloc(lrsize + namesize, UMEM_NOFAIL);
2130
2131	if (name)
2132		bcopy(name, lr + lrsize, namesize);
2133
2134	return (lr);
2135}
2136
2137void
2138ztest_lr_free(void *lr, size_t lrsize, char *name)
2139{
2140	size_t namesize = name ? strlen(name) + 1 : 0;
2141
2142	umem_free(lr, lrsize + namesize);
2143}
2144
2145/*
2146 * Lookup a bunch of objects.  Returns the number of objects not found.
2147 */
2148static int
2149ztest_lookup(ztest_ds_t *zd, ztest_od_t *od, int count)
2150{
2151	int missing = 0;
2152	int error;
2153
2154	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2155
2156	for (int i = 0; i < count; i++, od++) {
2157		od->od_object = 0;
2158		error = zap_lookup(zd->zd_os, od->od_dir, od->od_name,
2159		    sizeof (uint64_t), 1, &od->od_object);
2160		if (error) {
2161			ASSERT(error == ENOENT);
2162			ASSERT(od->od_object == 0);
2163			missing++;
2164		} else {
2165			dmu_buf_t *db;
2166			ztest_block_tag_t *bbt;
2167			dmu_object_info_t doi;
2168
2169			ASSERT(od->od_object != 0);
2170			ASSERT(missing == 0);	/* there should be no gaps */
2171
2172			ztest_object_lock(zd, od->od_object, RL_READER);
2173			VERIFY3U(0, ==, dmu_bonus_hold(zd->zd_os,
2174			    od->od_object, FTAG, &db));
2175			dmu_object_info_from_db(db, &doi);
2176			bbt = ztest_bt_bonus(db);
2177			ASSERT3U(bbt->bt_magic, ==, BT_MAGIC);
2178			od->od_type = doi.doi_type;
2179			od->od_blocksize = doi.doi_data_block_size;
2180			od->od_gen = bbt->bt_gen;
2181			dmu_buf_rele(db, FTAG);
2182			ztest_object_unlock(zd, od->od_object);
2183		}
2184	}
2185
2186	return (missing);
2187}
2188
2189static int
2190ztest_create(ztest_ds_t *zd, ztest_od_t *od, int count)
2191{
2192	int missing = 0;
2193
2194	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2195
2196	for (int i = 0; i < count; i++, od++) {
2197		if (missing) {
2198			od->od_object = 0;
2199			missing++;
2200			continue;
2201		}
2202
2203		lr_create_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2204
2205		lr->lr_doid = od->od_dir;
2206		lr->lr_foid = 0;	/* 0 to allocate, > 0 to claim */
2207		lr->lrz_type = od->od_crtype;
2208		lr->lrz_blocksize = od->od_crblocksize;
2209		lr->lrz_ibshift = ztest_random_ibshift();
2210		lr->lrz_bonustype = DMU_OT_UINT64_OTHER;
2211		lr->lrz_dnodesize = od->od_crdnodesize;
2212		lr->lr_gen = od->od_crgen;
2213		lr->lr_crtime[0] = time(NULL);
2214
2215		if (ztest_replay_create(zd, lr, B_FALSE) != 0) {
2216			ASSERT(missing == 0);
2217			od->od_object = 0;
2218			missing++;
2219		} else {
2220			od->od_object = lr->lr_foid;
2221			od->od_type = od->od_crtype;
2222			od->od_blocksize = od->od_crblocksize;
2223			od->od_gen = od->od_crgen;
2224			ASSERT(od->od_object != 0);
2225		}
2226
2227		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2228	}
2229
2230	return (missing);
2231}
2232
2233static int
2234ztest_remove(ztest_ds_t *zd, ztest_od_t *od, int count)
2235{
2236	int missing = 0;
2237	int error;
2238
2239	ASSERT(MUTEX_HELD(&zd->zd_dirobj_lock));
2240
2241	od += count - 1;
2242
2243	for (int i = count - 1; i >= 0; i--, od--) {
2244		if (missing) {
2245			missing++;
2246			continue;
2247		}
2248
2249		/*
2250		 * No object was found.
2251		 */
2252		if (od->od_object == 0)
2253			continue;
2254
2255		lr_remove_t *lr = ztest_lr_alloc(sizeof (*lr), od->od_name);
2256
2257		lr->lr_doid = od->od_dir;
2258
2259		if ((error = ztest_replay_remove(zd, lr, B_FALSE)) != 0) {
2260			ASSERT3U(error, ==, ENOSPC);
2261			missing++;
2262		} else {
2263			od->od_object = 0;
2264		}
2265		ztest_lr_free(lr, sizeof (*lr), od->od_name);
2266	}
2267
2268	return (missing);
2269}
2270
2271static int
2272ztest_write(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size,
2273    void *data)
2274{
2275	lr_write_t *lr;
2276	int error;
2277
2278	lr = ztest_lr_alloc(sizeof (*lr) + size, NULL);
2279
2280	lr->lr_foid = object;
2281	lr->lr_offset = offset;
2282	lr->lr_length = size;
2283	lr->lr_blkoff = 0;
2284	BP_ZERO(&lr->lr_blkptr);
2285
2286	bcopy(data, lr + 1, size);
2287
2288	error = ztest_replay_write(zd, lr, B_FALSE);
2289
2290	ztest_lr_free(lr, sizeof (*lr) + size, NULL);
2291
2292	return (error);
2293}
2294
2295static int
2296ztest_truncate(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2297{
2298	lr_truncate_t *lr;
2299	int error;
2300
2301	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2302
2303	lr->lr_foid = object;
2304	lr->lr_offset = offset;
2305	lr->lr_length = size;
2306
2307	error = ztest_replay_truncate(zd, lr, B_FALSE);
2308
2309	ztest_lr_free(lr, sizeof (*lr), NULL);
2310
2311	return (error);
2312}
2313
2314static int
2315ztest_setattr(ztest_ds_t *zd, uint64_t object)
2316{
2317	lr_setattr_t *lr;
2318	int error;
2319
2320	lr = ztest_lr_alloc(sizeof (*lr), NULL);
2321
2322	lr->lr_foid = object;
2323	lr->lr_size = 0;
2324	lr->lr_mode = 0;
2325
2326	error = ztest_replay_setattr(zd, lr, B_FALSE);
2327
2328	ztest_lr_free(lr, sizeof (*lr), NULL);
2329
2330	return (error);
2331}
2332
2333static void
2334ztest_prealloc(ztest_ds_t *zd, uint64_t object, uint64_t offset, uint64_t size)
2335{
2336	objset_t *os = zd->zd_os;
2337	dmu_tx_t *tx;
2338	uint64_t txg;
2339	rl_t *rl;
2340
2341	txg_wait_synced(dmu_objset_pool(os), 0);
2342
2343	ztest_object_lock(zd, object, RL_READER);
2344	rl = ztest_range_lock(zd, object, offset, size, RL_WRITER);
2345
2346	tx = dmu_tx_create(os);
2347
2348	dmu_tx_hold_write(tx, object, offset, size);
2349
2350	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
2351
2352	if (txg != 0) {
2353		dmu_prealloc(os, object, offset, size, tx);
2354		dmu_tx_commit(tx);
2355		txg_wait_synced(dmu_objset_pool(os), txg);
2356	} else {
2357		(void) dmu_free_long_range(os, object, offset, size);
2358	}
2359
2360	ztest_range_unlock(rl);
2361	ztest_object_unlock(zd, object);
2362}
2363
2364static void
2365ztest_io(ztest_ds_t *zd, uint64_t object, uint64_t offset)
2366{
2367	int err;
2368	ztest_block_tag_t wbt;
2369	dmu_object_info_t doi;
2370	enum ztest_io_type io_type;
2371	uint64_t blocksize;
2372	void *data;
2373
2374	VERIFY(dmu_object_info(zd->zd_os, object, &doi) == 0);
2375	blocksize = doi.doi_data_block_size;
2376	data = umem_alloc(blocksize, UMEM_NOFAIL);
2377
2378	/*
2379	 * Pick an i/o type at random, biased toward writing block tags.
2380	 */
2381	io_type = ztest_random(ZTEST_IO_TYPES);
2382	if (ztest_random(2) == 0)
2383		io_type = ZTEST_IO_WRITE_TAG;
2384
2385	rw_enter(&zd->zd_zilog_lock, RW_READER);
2386
2387	switch (io_type) {
2388
2389	case ZTEST_IO_WRITE_TAG:
2390		ztest_bt_generate(&wbt, zd->zd_os, object, doi.doi_dnodesize,
2391		    offset, 0, 0, 0);
2392		(void) ztest_write(zd, object, offset, sizeof (wbt), &wbt);
2393		break;
2394
2395	case ZTEST_IO_WRITE_PATTERN:
2396		(void) memset(data, 'a' + (object + offset) % 5, blocksize);
2397		if (ztest_random(2) == 0) {
2398			/*
2399			 * Induce fletcher2 collisions to ensure that
2400			 * zio_ddt_collision() detects and resolves them
2401			 * when using fletcher2-verify for deduplication.
2402			 */
2403			((uint64_t *)data)[0] ^= 1ULL << 63;
2404			((uint64_t *)data)[4] ^= 1ULL << 63;
2405		}
2406		(void) ztest_write(zd, object, offset, blocksize, data);
2407		break;
2408
2409	case ZTEST_IO_WRITE_ZEROES:
2410		bzero(data, blocksize);
2411		(void) ztest_write(zd, object, offset, blocksize, data);
2412		break;
2413
2414	case ZTEST_IO_TRUNCATE:
2415		(void) ztest_truncate(zd, object, offset, blocksize);
2416		break;
2417
2418	case ZTEST_IO_SETATTR:
2419		(void) ztest_setattr(zd, object);
2420		break;
2421
2422	case ZTEST_IO_REWRITE:
2423		rw_enter(&ztest_name_lock, RW_READER);
2424		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2425		    ZFS_PROP_CHECKSUM, spa_dedup_checksum(ztest_spa),
2426		    B_FALSE);
2427		VERIFY(err == 0 || err == ENOSPC);
2428		err = ztest_dsl_prop_set_uint64(zd->zd_name,
2429		    ZFS_PROP_COMPRESSION,
2430		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION),
2431		    B_FALSE);
2432		VERIFY(err == 0 || err == ENOSPC);
2433		rw_exit(&ztest_name_lock);
2434
2435		VERIFY0(dmu_read(zd->zd_os, object, offset, blocksize, data,
2436		    DMU_READ_NO_PREFETCH));
2437
2438		(void) ztest_write(zd, object, offset, blocksize, data);
2439		break;
2440	}
2441
2442	rw_exit(&zd->zd_zilog_lock);
2443
2444	umem_free(data, blocksize);
2445}
2446
2447/*
2448 * Initialize an object description template.
2449 */
2450static void
2451ztest_od_init(ztest_od_t *od, uint64_t id, char *tag, uint64_t index,
2452    dmu_object_type_t type, uint64_t blocksize, uint64_t dnodesize,
2453    uint64_t gen)
2454{
2455	od->od_dir = ZTEST_DIROBJ;
2456	od->od_object = 0;
2457
2458	od->od_crtype = type;
2459	od->od_crblocksize = blocksize ? blocksize : ztest_random_blocksize();
2460	od->od_crdnodesize = dnodesize ? dnodesize : ztest_random_dnodesize();
2461	od->od_crgen = gen;
2462
2463	od->od_type = DMU_OT_NONE;
2464	od->od_blocksize = 0;
2465	od->od_gen = 0;
2466
2467	(void) snprintf(od->od_name, sizeof (od->od_name), "%s(%lld)[%llu]",
2468	    tag, (int64_t)id, index);
2469}
2470
2471/*
2472 * Lookup or create the objects for a test using the od template.
2473 * If the objects do not all exist, or if 'remove' is specified,
2474 * remove any existing objects and create new ones.  Otherwise,
2475 * use the existing objects.
2476 */
2477static int
2478ztest_object_init(ztest_ds_t *zd, ztest_od_t *od, size_t size, boolean_t remove)
2479{
2480	int count = size / sizeof (*od);
2481	int rv = 0;
2482
2483	mutex_enter(&zd->zd_dirobj_lock);
2484	if ((ztest_lookup(zd, od, count) != 0 || remove) &&
2485	    (ztest_remove(zd, od, count) != 0 ||
2486	    ztest_create(zd, od, count) != 0))
2487		rv = -1;
2488	zd->zd_od = od;
2489	mutex_exit(&zd->zd_dirobj_lock);
2490
2491	return (rv);
2492}
2493
2494/* ARGSUSED */
2495void
2496ztest_zil_commit(ztest_ds_t *zd, uint64_t id)
2497{
2498	zilog_t *zilog = zd->zd_zilog;
2499
2500	rw_enter(&zd->zd_zilog_lock, RW_READER);
2501
2502	zil_commit(zilog, ztest_random(ZTEST_OBJECTS));
2503
2504	/*
2505	 * Remember the committed values in zd, which is in parent/child
2506	 * shared memory.  If we die, the next iteration of ztest_run()
2507	 * will verify that the log really does contain this record.
2508	 */
2509	mutex_enter(&zilog->zl_lock);
2510	ASSERT(zd->zd_shared != NULL);
2511	ASSERT3U(zd->zd_shared->zd_seq, <=, zilog->zl_commit_lr_seq);
2512	zd->zd_shared->zd_seq = zilog->zl_commit_lr_seq;
2513	mutex_exit(&zilog->zl_lock);
2514
2515	rw_exit(&zd->zd_zilog_lock);
2516}
2517
2518/*
2519 * This function is designed to simulate the operations that occur during a
2520 * mount/unmount operation.  We hold the dataset across these operations in an
2521 * attempt to expose any implicit assumptions about ZIL management.
2522 */
2523/* ARGSUSED */
2524void
2525ztest_zil_remount(ztest_ds_t *zd, uint64_t id)
2526{
2527	objset_t *os = zd->zd_os;
2528
2529	/*
2530	 * We grab the zd_dirobj_lock to ensure that no other thread is
2531	 * updating the zil (i.e. adding in-memory log records) and the
2532	 * zd_zilog_lock to block any I/O.
2533	 */
2534	mutex_enter(&zd->zd_dirobj_lock);
2535	rw_enter(&zd->zd_zilog_lock, RW_WRITER);
2536
2537	/* zfsvfs_teardown() */
2538	zil_close(zd->zd_zilog);
2539
2540	/* zfsvfs_setup() */
2541	VERIFY(zil_open(os, ztest_get_data) == zd->zd_zilog);
2542	zil_replay(os, zd, ztest_replay_vector);
2543
2544	rw_exit(&zd->zd_zilog_lock);
2545	mutex_exit(&zd->zd_dirobj_lock);
2546}
2547
2548/*
2549 * Verify that we can't destroy an active pool, create an existing pool,
2550 * or create a pool with a bad vdev spec.
2551 */
2552/* ARGSUSED */
2553void
2554ztest_spa_create_destroy(ztest_ds_t *zd, uint64_t id)
2555{
2556	ztest_shared_opts_t *zo = &ztest_opts;
2557	spa_t *spa;
2558	nvlist_t *nvroot;
2559
2560	if (zo->zo_mmp_test)
2561		return;
2562
2563	/*
2564	 * Attempt to create using a bad file.
2565	 */
2566	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2567	VERIFY3U(ENOENT, ==,
2568	    spa_create("ztest_bad_file", nvroot, NULL, NULL));
2569	nvlist_free(nvroot);
2570
2571	/*
2572	 * Attempt to create using a bad mirror.
2573	 */
2574	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 2, 1);
2575	VERIFY3U(ENOENT, ==,
2576	    spa_create("ztest_bad_mirror", nvroot, NULL, NULL));
2577	nvlist_free(nvroot);
2578
2579	/*
2580	 * Attempt to create an existing pool.  It shouldn't matter
2581	 * what's in the nvroot; we should fail with EEXIST.
2582	 */
2583	rw_enter(&ztest_name_lock, RW_READER);
2584	nvroot = make_vdev_root("/dev/bogus", NULL, NULL, 0, 0, NULL, 0, 0, 1);
2585	VERIFY3U(EEXIST, ==, spa_create(zo->zo_pool, nvroot, NULL, NULL));
2586	nvlist_free(nvroot);
2587	VERIFY3U(0, ==, spa_open(zo->zo_pool, &spa, FTAG));
2588	VERIFY3U(EBUSY, ==, spa_destroy(zo->zo_pool));
2589	spa_close(spa, FTAG);
2590
2591	rw_exit(&ztest_name_lock);
2592}
2593
2594/*
2595 * Start and then stop the MMP threads to ensure the startup and shutdown code
2596 * works properly.  Actual protection and property-related code tested via ZTS.
2597 */
2598/* ARGSUSED */
2599void
2600ztest_mmp_enable_disable(ztest_ds_t *zd, uint64_t id)
2601{
2602	ztest_shared_opts_t *zo = &ztest_opts;
2603	spa_t *spa = ztest_spa;
2604
2605	if (zo->zo_mmp_test)
2606		return;
2607
2608	/*
2609	 * Since enabling MMP involves setting a property, it could not be done
2610	 * while the pool is suspended.
2611	 */
2612	if (spa_suspended(spa))
2613		return;
2614
2615	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2616	mutex_enter(&spa->spa_props_lock);
2617
2618	zfs_multihost_fail_intervals = 0;
2619
2620	if (!spa_multihost(spa)) {
2621		spa->spa_multihost = B_TRUE;
2622		mmp_thread_start(spa);
2623	}
2624
2625	mutex_exit(&spa->spa_props_lock);
2626	spa_config_exit(spa, SCL_CONFIG, FTAG);
2627
2628	txg_wait_synced(spa_get_dsl(spa), 0);
2629	mmp_signal_all_threads();
2630	txg_wait_synced(spa_get_dsl(spa), 0);
2631
2632	spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2633	mutex_enter(&spa->spa_props_lock);
2634
2635	if (spa_multihost(spa)) {
2636		mmp_thread_stop(spa);
2637		spa->spa_multihost = B_FALSE;
2638	}
2639
2640	mutex_exit(&spa->spa_props_lock);
2641	spa_config_exit(spa, SCL_CONFIG, FTAG);
2642}
2643
2644/* ARGSUSED */
2645void
2646ztest_spa_upgrade(ztest_ds_t *zd, uint64_t id)
2647{
2648	spa_t *spa;
2649	uint64_t initial_version = SPA_VERSION_INITIAL;
2650	uint64_t version, newversion;
2651	nvlist_t *nvroot, *props;
2652	char *name;
2653
2654	if (ztest_opts.zo_mmp_test)
2655		return;
2656
2657	mutex_enter(&ztest_vdev_lock);
2658	name = kmem_asprintf("%s_upgrade", ztest_opts.zo_pool);
2659
2660	/*
2661	 * Clean up from previous runs.
2662	 */
2663	(void) spa_destroy(name);
2664
2665	nvroot = make_vdev_root(NULL, NULL, name, ztest_opts.zo_vdev_size, 0,
2666	    NULL, ztest_opts.zo_raidz, ztest_opts.zo_mirrors, 1);
2667
2668	/*
2669	 * If we're configuring a RAIDZ device then make sure that the
2670	 * the initial version is capable of supporting that feature.
2671	 */
2672	switch (ztest_opts.zo_raidz_parity) {
2673	case 0:
2674	case 1:
2675		initial_version = SPA_VERSION_INITIAL;
2676		break;
2677	case 2:
2678		initial_version = SPA_VERSION_RAIDZ2;
2679		break;
2680	case 3:
2681		initial_version = SPA_VERSION_RAIDZ3;
2682		break;
2683	}
2684
2685	/*
2686	 * Create a pool with a spa version that can be upgraded. Pick
2687	 * a value between initial_version and SPA_VERSION_BEFORE_FEATURES.
2688	 */
2689	do {
2690		version = ztest_random_spa_version(initial_version);
2691	} while (version > SPA_VERSION_BEFORE_FEATURES);
2692
2693	props = fnvlist_alloc();
2694	fnvlist_add_uint64(props,
2695	    zpool_prop_to_name(ZPOOL_PROP_VERSION), version);
2696	VERIFY0(spa_create(name, nvroot, props, NULL));
2697	fnvlist_free(nvroot);
2698	fnvlist_free(props);
2699
2700	VERIFY0(spa_open(name, &spa, FTAG));
2701	VERIFY3U(spa_version(spa), ==, version);
2702	newversion = ztest_random_spa_version(version + 1);
2703
2704	if (ztest_opts.zo_verbose >= 4) {
2705		(void) printf("upgrading spa version from %llu to %llu\n",
2706		    (u_longlong_t)version, (u_longlong_t)newversion);
2707	}
2708
2709	spa_upgrade(spa, newversion);
2710	VERIFY3U(spa_version(spa), >, version);
2711	VERIFY3U(spa_version(spa), ==, fnvlist_lookup_uint64(spa->spa_config,
2712	    zpool_prop_to_name(ZPOOL_PROP_VERSION)));
2713	spa_close(spa, FTAG);
2714
2715	strfree(name);
2716	mutex_exit(&ztest_vdev_lock);
2717}
2718
2719static void
2720ztest_spa_checkpoint(spa_t *spa)
2721{
2722	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2723
2724	int error = spa_checkpoint(spa->spa_name);
2725
2726	switch (error) {
2727	case 0:
2728	case ZFS_ERR_DEVRM_IN_PROGRESS:
2729	case ZFS_ERR_DISCARDING_CHECKPOINT:
2730	case ZFS_ERR_CHECKPOINT_EXISTS:
2731		break;
2732	case ENOSPC:
2733		ztest_record_enospc(FTAG);
2734		break;
2735	default:
2736		fatal(0, "spa_checkpoint(%s) = %d", spa->spa_name, error);
2737	}
2738}
2739
2740static void
2741ztest_spa_discard_checkpoint(spa_t *spa)
2742{
2743	ASSERT(MUTEX_HELD(&ztest_checkpoint_lock));
2744
2745	int error = spa_checkpoint_discard(spa->spa_name);
2746
2747	switch (error) {
2748	case 0:
2749	case ZFS_ERR_DISCARDING_CHECKPOINT:
2750	case ZFS_ERR_NO_CHECKPOINT:
2751		break;
2752	default:
2753		fatal(0, "spa_discard_checkpoint(%s) = %d",
2754		    spa->spa_name, error);
2755	}
2756
2757}
2758
2759/* ARGSUSED */
2760void
2761ztest_spa_checkpoint_create_discard(ztest_ds_t *zd, uint64_t id)
2762{
2763	spa_t *spa = ztest_spa;
2764
2765	mutex_enter(&ztest_checkpoint_lock);
2766	if (ztest_random(2) == 0) {
2767		ztest_spa_checkpoint(spa);
2768	} else {
2769		ztest_spa_discard_checkpoint(spa);
2770	}
2771	mutex_exit(&ztest_checkpoint_lock);
2772}
2773
2774
2775static vdev_t *
2776vdev_lookup_by_path(vdev_t *vd, const char *path)
2777{
2778	vdev_t *mvd;
2779
2780	if (vd->vdev_path != NULL && strcmp(path, vd->vdev_path) == 0)
2781		return (vd);
2782
2783	for (int c = 0; c < vd->vdev_children; c++)
2784		if ((mvd = vdev_lookup_by_path(vd->vdev_child[c], path)) !=
2785		    NULL)
2786			return (mvd);
2787
2788	return (NULL);
2789}
2790
2791/*
2792 * Find the first available hole which can be used as a top-level.
2793 */
2794int
2795find_vdev_hole(spa_t *spa)
2796{
2797	vdev_t *rvd = spa->spa_root_vdev;
2798	int c;
2799
2800	ASSERT(spa_config_held(spa, SCL_VDEV, RW_READER) == SCL_VDEV);
2801
2802	for (c = 0; c < rvd->vdev_children; c++) {
2803		vdev_t *cvd = rvd->vdev_child[c];
2804
2805		if (cvd->vdev_ishole)
2806			break;
2807	}
2808	return (c);
2809}
2810
2811/*
2812 * Verify that vdev_add() works as expected.
2813 */
2814/* ARGSUSED */
2815void
2816ztest_vdev_add_remove(ztest_ds_t *zd, uint64_t id)
2817{
2818	ztest_shared_t *zs = ztest_shared;
2819	spa_t *spa = ztest_spa;
2820	uint64_t leaves;
2821	uint64_t guid;
2822	nvlist_t *nvroot;
2823	int error;
2824
2825	if (ztest_opts.zo_mmp_test)
2826		return;
2827
2828	mutex_enter(&ztest_vdev_lock);
2829	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2830
2831	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2832
2833	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2834
2835	/*
2836	 * If we have slogs then remove them 1/4 of the time.
2837	 */
2838	if (spa_has_slogs(spa) && ztest_random(4) == 0) {
2839		metaslab_group_t *mg;
2840
2841		/*
2842		 * find the first real slog in log allocation class
2843		 */
2844		mg =  spa_log_class(spa)->mc_rotor;
2845		while (!mg->mg_vd->vdev_islog)
2846			mg = mg->mg_next;
2847
2848		guid = mg->mg_vd->vdev_guid;
2849
2850		spa_config_exit(spa, SCL_VDEV, FTAG);
2851
2852		/*
2853		 * We have to grab the zs_name_lock as writer to
2854		 * prevent a race between removing a slog (dmu_objset_find)
2855		 * and destroying a dataset. Removing the slog will
2856		 * grab a reference on the dataset which may cause
2857		 * dmu_objset_destroy() to fail with EBUSY thus
2858		 * leaving the dataset in an inconsistent state.
2859		 */
2860		rw_enter(&ztest_name_lock, RW_WRITER);
2861		error = spa_vdev_remove(spa, guid, B_FALSE);
2862		rw_exit(&ztest_name_lock);
2863
2864		switch (error) {
2865		case 0:
2866		case EEXIST:
2867		case ZFS_ERR_CHECKPOINT_EXISTS:
2868		case ZFS_ERR_DISCARDING_CHECKPOINT:
2869			break;
2870		default:
2871			fatal(0, "spa_vdev_remove() = %d", error);
2872		}
2873	} else {
2874		spa_config_exit(spa, SCL_VDEV, FTAG);
2875
2876		/*
2877		 * Make 1/4 of the devices be log devices
2878		 */
2879		nvroot = make_vdev_root(NULL, NULL, NULL,
2880		    ztest_opts.zo_vdev_size, 0, (ztest_random(4) == 0) ?
2881		    "log" : NULL, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
2882
2883		error = spa_vdev_add(spa, nvroot);
2884		nvlist_free(nvroot);
2885
2886		switch (error) {
2887		case 0:
2888			break;
2889		case ENOSPC:
2890			ztest_record_enospc("spa_vdev_add");
2891			break;
2892		default:
2893			fatal(0, "spa_vdev_add() = %d", error);
2894		}
2895	}
2896
2897	mutex_exit(&ztest_vdev_lock);
2898}
2899
2900/* ARGSUSED */
2901void
2902ztest_vdev_class_add(ztest_ds_t *zd, uint64_t id)
2903{
2904	ztest_shared_t *zs = ztest_shared;
2905	spa_t *spa = ztest_spa;
2906	uint64_t leaves;
2907	nvlist_t *nvroot;
2908	const char *class = (ztest_random(2) == 0) ?
2909	    VDEV_ALLOC_BIAS_SPECIAL : VDEV_ALLOC_BIAS_DEDUP;
2910	int error;
2911
2912	/*
2913	 * By default add a special vdev 50% of the time
2914	 */
2915	if ((ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_OFF) ||
2916	    (ztest_opts.zo_special_vdevs == ZTEST_VDEV_CLASS_RND &&
2917	    ztest_random(2) == 0)) {
2918		return;
2919	}
2920
2921	mutex_enter(&ztest_vdev_lock);
2922
2923	/* Only test with mirrors */
2924	if (zs->zs_mirrors < 2) {
2925		mutex_exit(&ztest_vdev_lock);
2926		return;
2927	}
2928
2929	/* requires feature@allocation_classes */
2930	if (!spa_feature_is_enabled(spa, SPA_FEATURE_ALLOCATION_CLASSES)) {
2931		mutex_exit(&ztest_vdev_lock);
2932		return;
2933	}
2934
2935	leaves = MAX(zs->zs_mirrors + zs->zs_splits, 1) * ztest_opts.zo_raidz;
2936
2937	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
2938	ztest_shared->zs_vdev_next_leaf = find_vdev_hole(spa) * leaves;
2939	spa_config_exit(spa, SCL_VDEV, FTAG);
2940
2941	nvroot = make_vdev_root(NULL, NULL, NULL, ztest_opts.zo_vdev_size, 0,
2942	    class, ztest_opts.zo_raidz, zs->zs_mirrors, 1);
2943
2944	error = spa_vdev_add(spa, nvroot);
2945	nvlist_free(nvroot);
2946
2947	if (error == ENOSPC)
2948		ztest_record_enospc("spa_vdev_add");
2949	else if (error != 0)
2950		fatal(0, "spa_vdev_add() = %d", error);
2951
2952	/*
2953	 * 50% of the time allow small blocks in the special class
2954	 */
2955	if (error == 0 &&
2956	    spa_special_class(spa)->mc_groups == 1 && ztest_random(2) == 0) {
2957		if (ztest_opts.zo_verbose >= 3)
2958			(void) printf("Enabling special VDEV small blocks\n");
2959		(void) ztest_dsl_prop_set_uint64(zd->zd_name,
2960		    ZFS_PROP_SPECIAL_SMALL_BLOCKS, 32768, B_FALSE);
2961	}
2962
2963	mutex_exit(&ztest_vdev_lock);
2964
2965	if (ztest_opts.zo_verbose >= 3) {
2966		metaslab_class_t *mc;
2967
2968		if (strcmp(class, VDEV_ALLOC_BIAS_SPECIAL) == 0)
2969			mc = spa_special_class(spa);
2970		else
2971			mc = spa_dedup_class(spa);
2972		(void) printf("Added a %s mirrored vdev (of %d)\n",
2973		    class, (int)mc->mc_groups);
2974	}
2975}
2976
2977/*
2978 * Verify that adding/removing aux devices (l2arc, hot spare) works as expected.
2979 */
2980/* ARGSUSED */
2981void
2982ztest_vdev_aux_add_remove(ztest_ds_t *zd, uint64_t id)
2983{
2984	ztest_shared_t *zs = ztest_shared;
2985	spa_t *spa = ztest_spa;
2986	vdev_t *rvd = spa->spa_root_vdev;
2987	spa_aux_vdev_t *sav;
2988	char *aux;
2989	uint64_t guid = 0;
2990	int error;
2991
2992	if (ztest_opts.zo_mmp_test)
2993		return;
2994
2995	if (ztest_random(2) == 0) {
2996		sav = &spa->spa_spares;
2997		aux = ZPOOL_CONFIG_SPARES;
2998	} else {
2999		sav = &spa->spa_l2cache;
3000		aux = ZPOOL_CONFIG_L2CACHE;
3001	}
3002
3003	mutex_enter(&ztest_vdev_lock);
3004
3005	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3006
3007	if (sav->sav_count != 0 && ztest_random(4) == 0) {
3008		/*
3009		 * Pick a random device to remove.
3010		 */
3011		guid = sav->sav_vdevs[ztest_random(sav->sav_count)]->vdev_guid;
3012	} else {
3013		/*
3014		 * Find an unused device we can add.
3015		 */
3016		zs->zs_vdev_aux = 0;
3017		for (;;) {
3018			char path[MAXPATHLEN];
3019			int c;
3020			(void) snprintf(path, sizeof (path), ztest_aux_template,
3021			    ztest_opts.zo_dir, ztest_opts.zo_pool, aux,
3022			    zs->zs_vdev_aux);
3023			for (c = 0; c < sav->sav_count; c++)
3024				if (strcmp(sav->sav_vdevs[c]->vdev_path,
3025				    path) == 0)
3026					break;
3027			if (c == sav->sav_count &&
3028			    vdev_lookup_by_path(rvd, path) == NULL)
3029				break;
3030			zs->zs_vdev_aux++;
3031		}
3032	}
3033
3034	spa_config_exit(spa, SCL_VDEV, FTAG);
3035
3036	if (guid == 0) {
3037		/*
3038		 * Add a new device.
3039		 */
3040		nvlist_t *nvroot = make_vdev_root(NULL, aux, NULL,
3041		    (ztest_opts.zo_vdev_size * 5) / 4, 0, NULL, 0, 0, 1);
3042		error = spa_vdev_add(spa, nvroot);
3043
3044		switch (error) {
3045		case 0:
3046			break;
3047		default:
3048			fatal(0, "spa_vdev_add(%p) = %d", nvroot, error);
3049		}
3050		nvlist_free(nvroot);
3051	} else {
3052		/*
3053		 * Remove an existing device.  Sometimes, dirty its
3054		 * vdev state first to make sure we handle removal
3055		 * of devices that have pending state changes.
3056		 */
3057		if (ztest_random(2) == 0)
3058			(void) vdev_online(spa, guid, 0, NULL);
3059
3060		error = spa_vdev_remove(spa, guid, B_FALSE);
3061
3062		switch (error) {
3063		case 0:
3064		case EBUSY:
3065		case ZFS_ERR_CHECKPOINT_EXISTS:
3066		case ZFS_ERR_DISCARDING_CHECKPOINT:
3067			break;
3068		default:
3069			fatal(0, "spa_vdev_remove(%llu) = %d", guid, error);
3070		}
3071	}
3072
3073	mutex_exit(&ztest_vdev_lock);
3074}
3075
3076/*
3077 * split a pool if it has mirror tlvdevs
3078 */
3079/* ARGSUSED */
3080void
3081ztest_split_pool(ztest_ds_t *zd, uint64_t id)
3082{
3083	ztest_shared_t *zs = ztest_shared;
3084	spa_t *spa = ztest_spa;
3085	vdev_t *rvd = spa->spa_root_vdev;
3086	nvlist_t *tree, **child, *config, *split, **schild;
3087	uint_t c, children, schildren = 0, lastlogid = 0;
3088	int error = 0;
3089
3090	if (ztest_opts.zo_mmp_test)
3091		return;
3092
3093	mutex_enter(&ztest_vdev_lock);
3094
3095	/* ensure we have a useable config; mirrors of raidz aren't supported */
3096	if (zs->zs_mirrors < 3 || ztest_opts.zo_raidz > 1) {
3097		mutex_exit(&ztest_vdev_lock);
3098		return;
3099	}
3100
3101	/* clean up the old pool, if any */
3102	(void) spa_destroy("splitp");
3103
3104	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3105
3106	/* generate a config from the existing config */
3107	mutex_enter(&spa->spa_props_lock);
3108	VERIFY(nvlist_lookup_nvlist(spa->spa_config, ZPOOL_CONFIG_VDEV_TREE,
3109	    &tree) == 0);
3110	mutex_exit(&spa->spa_props_lock);
3111
3112	VERIFY(nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3113	    &children) == 0);
3114
3115	schild = malloc(rvd->vdev_children * sizeof (nvlist_t *));
3116	for (c = 0; c < children; c++) {
3117		vdev_t *tvd = rvd->vdev_child[c];
3118		nvlist_t **mchild;
3119		uint_t mchildren;
3120
3121		if (tvd->vdev_islog || tvd->vdev_ops == &vdev_hole_ops) {
3122			VERIFY(nvlist_alloc(&schild[schildren], NV_UNIQUE_NAME,
3123			    0) == 0);
3124			VERIFY(nvlist_add_string(schild[schildren],
3125			    ZPOOL_CONFIG_TYPE, VDEV_TYPE_HOLE) == 0);
3126			VERIFY(nvlist_add_uint64(schild[schildren],
3127			    ZPOOL_CONFIG_IS_HOLE, 1) == 0);
3128			if (lastlogid == 0)
3129				lastlogid = schildren;
3130			++schildren;
3131			continue;
3132		}
3133		lastlogid = 0;
3134		VERIFY(nvlist_lookup_nvlist_array(child[c],
3135		    ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3136		VERIFY(nvlist_dup(mchild[0], &schild[schildren++], 0) == 0);
3137	}
3138
3139	/* OK, create a config that can be used to split */
3140	VERIFY(nvlist_alloc(&split, NV_UNIQUE_NAME, 0) == 0);
3141	VERIFY(nvlist_add_string(split, ZPOOL_CONFIG_TYPE,
3142	    VDEV_TYPE_ROOT) == 0);
3143	VERIFY(nvlist_add_nvlist_array(split, ZPOOL_CONFIG_CHILDREN, schild,
3144	    lastlogid != 0 ? lastlogid : schildren) == 0);
3145
3146	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, 0) == 0);
3147	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, split) == 0);
3148
3149	for (c = 0; c < schildren; c++)
3150		nvlist_free(schild[c]);
3151	free(schild);
3152	nvlist_free(split);
3153
3154	spa_config_exit(spa, SCL_VDEV, FTAG);
3155
3156	rw_enter(&ztest_name_lock, RW_WRITER);
3157	error = spa_vdev_split_mirror(spa, "splitp", config, NULL, B_FALSE);
3158	rw_exit(&ztest_name_lock);
3159
3160	nvlist_free(config);
3161
3162	if (error == 0) {
3163		(void) printf("successful split - results:\n");
3164		mutex_enter(&spa_namespace_lock);
3165		show_pool_stats(spa);
3166		show_pool_stats(spa_lookup("splitp"));
3167		mutex_exit(&spa_namespace_lock);
3168		++zs->zs_splits;
3169		--zs->zs_mirrors;
3170	}
3171	mutex_exit(&ztest_vdev_lock);
3172}
3173
3174/*
3175 * Verify that we can attach and detach devices.
3176 */
3177/* ARGSUSED */
3178void
3179ztest_vdev_attach_detach(ztest_ds_t *zd, uint64_t id)
3180{
3181	ztest_shared_t *zs = ztest_shared;
3182	spa_t *spa = ztest_spa;
3183	spa_aux_vdev_t *sav = &spa->spa_spares;
3184	vdev_t *rvd = spa->spa_root_vdev;
3185	vdev_t *oldvd, *newvd, *pvd;
3186	nvlist_t *root;
3187	uint64_t leaves;
3188	uint64_t leaf, top;
3189	uint64_t ashift = ztest_get_ashift();
3190	uint64_t oldguid, pguid;
3191	uint64_t oldsize, newsize;
3192	char oldpath[MAXPATHLEN], newpath[MAXPATHLEN];
3193	int replacing;
3194	int oldvd_has_siblings = B_FALSE;
3195	int newvd_is_spare = B_FALSE;
3196	int oldvd_is_log;
3197	int error, expected_error;
3198
3199	if (ztest_opts.zo_mmp_test)
3200		return;
3201
3202	mutex_enter(&ztest_vdev_lock);
3203	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
3204
3205	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3206
3207	/*
3208	 * If a vdev is in the process of being removed, its removal may
3209	 * finish while we are in progress, leading to an unexpected error
3210	 * value.  Don't bother trying to attach while we are in the middle
3211	 * of removal.
3212	 */
3213	if (ztest_device_removal_active) {
3214		spa_config_exit(spa, SCL_ALL, FTAG);
3215		mutex_exit(&ztest_vdev_lock);
3216		return;
3217	}
3218
3219	/*
3220	 * Decide whether to do an attach or a replace.
3221	 */
3222	replacing = ztest_random(2);
3223
3224	/*
3225	 * Pick a random top-level vdev.
3226	 */
3227	top = ztest_random_vdev_top(spa, B_TRUE);
3228
3229	/*
3230	 * Pick a random leaf within it.
3231	 */
3232	leaf = ztest_random(leaves);
3233
3234	/*
3235	 * Locate this vdev.
3236	 */
3237	oldvd = rvd->vdev_child[top];
3238
3239	/* pick a child from the mirror */
3240	if (zs->zs_mirrors >= 1) {
3241		ASSERT(oldvd->vdev_ops == &vdev_mirror_ops);
3242		ASSERT(oldvd->vdev_children >= zs->zs_mirrors);
3243		oldvd = oldvd->vdev_child[leaf / ztest_opts.zo_raidz];
3244	}
3245
3246	/* pick a child out of the raidz group */
3247	if (ztest_opts.zo_raidz > 1) {
3248		ASSERT(oldvd->vdev_ops == &vdev_raidz_ops);
3249		ASSERT(oldvd->vdev_children == ztest_opts.zo_raidz);
3250		oldvd = oldvd->vdev_child[leaf % ztest_opts.zo_raidz];
3251	}
3252
3253	/*
3254	 * If we're already doing an attach or replace, oldvd may be a
3255	 * mirror vdev -- in which case, pick a random child.
3256	 */
3257	while (oldvd->vdev_children != 0) {
3258		oldvd_has_siblings = B_TRUE;
3259		ASSERT(oldvd->vdev_children >= 2);
3260		oldvd = oldvd->vdev_child[ztest_random(oldvd->vdev_children)];
3261	}
3262
3263	oldguid = oldvd->vdev_guid;
3264	oldsize = vdev_get_min_asize(oldvd);
3265	oldvd_is_log = oldvd->vdev_top->vdev_islog;
3266	(void) strcpy(oldpath, oldvd->vdev_path);
3267	pvd = oldvd->vdev_parent;
3268	pguid = pvd->vdev_guid;
3269
3270	/*
3271	 * If oldvd has siblings, then half of the time, detach it.
3272	 */
3273	if (oldvd_has_siblings && ztest_random(2) == 0) {
3274		spa_config_exit(spa, SCL_ALL, FTAG);
3275		error = spa_vdev_detach(spa, oldguid, pguid, B_FALSE);
3276		if (error != 0 && error != ENODEV && error != EBUSY &&
3277		    error != ENOTSUP && error != ZFS_ERR_CHECKPOINT_EXISTS &&
3278		    error != ZFS_ERR_DISCARDING_CHECKPOINT)
3279			fatal(0, "detach (%s) returned %d", oldpath, error);
3280		mutex_exit(&ztest_vdev_lock);
3281		return;
3282	}
3283
3284	/*
3285	 * For the new vdev, choose with equal probability between the two
3286	 * standard paths (ending in either 'a' or 'b') or a random hot spare.
3287	 */
3288	if (sav->sav_count != 0 && ztest_random(3) == 0) {
3289		newvd = sav->sav_vdevs[ztest_random(sav->sav_count)];
3290		newvd_is_spare = B_TRUE;
3291		(void) strcpy(newpath, newvd->vdev_path);
3292	} else {
3293		(void) snprintf(newpath, sizeof (newpath), ztest_dev_template,
3294		    ztest_opts.zo_dir, ztest_opts.zo_pool,
3295		    top * leaves + leaf);
3296		if (ztest_random(2) == 0)
3297			newpath[strlen(newpath) - 1] = 'b';
3298		newvd = vdev_lookup_by_path(rvd, newpath);
3299	}
3300
3301	if (newvd) {
3302		/*
3303		 * Reopen to ensure the vdev's asize field isn't stale.
3304		 */
3305		vdev_reopen(newvd);
3306		newsize = vdev_get_min_asize(newvd);
3307	} else {
3308		/*
3309		 * Make newsize a little bigger or smaller than oldsize.
3310		 * If it's smaller, the attach should fail.
3311		 * If it's larger, and we're doing a replace,
3312		 * we should get dynamic LUN growth when we're done.
3313		 */
3314		newsize = 10 * oldsize / (9 + ztest_random(3));
3315	}
3316
3317	/*
3318	 * If pvd is not a mirror or root, the attach should fail with ENOTSUP,
3319	 * unless it's a replace; in that case any non-replacing parent is OK.
3320	 *
3321	 * If newvd is already part of the pool, it should fail with EBUSY.
3322	 *
3323	 * If newvd is too small, it should fail with EOVERFLOW.
3324	 */
3325	if (pvd->vdev_ops != &vdev_mirror_ops &&
3326	    pvd->vdev_ops != &vdev_root_ops && (!replacing ||
3327	    pvd->vdev_ops == &vdev_replacing_ops ||
3328	    pvd->vdev_ops == &vdev_spare_ops))
3329		expected_error = ENOTSUP;
3330	else if (newvd_is_spare && (!replacing || oldvd_is_log))
3331		expected_error = ENOTSUP;
3332	else if (newvd == oldvd)
3333		expected_error = replacing ? 0 : EBUSY;
3334	else if (vdev_lookup_by_path(rvd, newpath) != NULL)
3335		expected_error = EBUSY;
3336	else if (newsize < oldsize)
3337		expected_error = EOVERFLOW;
3338	else if (ashift > oldvd->vdev_top->vdev_ashift)
3339		expected_error = EDOM;
3340	else
3341		expected_error = 0;
3342
3343	spa_config_exit(spa, SCL_ALL, FTAG);
3344
3345	/*
3346	 * Build the nvlist describing newpath.
3347	 */
3348	root = make_vdev_root(newpath, NULL, NULL, newvd == NULL ? newsize : 0,
3349	    ashift, NULL, 0, 0, 1);
3350
3351	error = spa_vdev_attach(spa, oldguid, root, replacing);
3352
3353	nvlist_free(root);
3354
3355	/*
3356	 * If our parent was the replacing vdev, but the replace completed,
3357	 * then instead of failing with ENOTSUP we may either succeed,
3358	 * fail with ENODEV, or fail with EOVERFLOW.
3359	 */
3360	if (expected_error == ENOTSUP &&
3361	    (error == 0 || error == ENODEV || error == EOVERFLOW))
3362		expected_error = error;
3363
3364	/*
3365	 * If someone grew the LUN, the replacement may be too small.
3366	 */
3367	if (error == EOVERFLOW || error == EBUSY)
3368		expected_error = error;
3369
3370	if (error == ZFS_ERR_CHECKPOINT_EXISTS ||
3371	    error == ZFS_ERR_DISCARDING_CHECKPOINT)
3372		expected_error = error;
3373
3374	/* XXX workaround 6690467 */
3375	if (error != expected_error && expected_error != EBUSY) {
3376		fatal(0, "attach (%s %llu, %s %llu, %d) "
3377		    "returned %d, expected %d",
3378		    oldpath, oldsize, newpath,
3379		    newsize, replacing, error, expected_error);
3380	}
3381
3382	mutex_exit(&ztest_vdev_lock);
3383}
3384
3385/* ARGSUSED */
3386void
3387ztest_device_removal(ztest_ds_t *zd, uint64_t id)
3388{
3389	spa_t *spa = ztest_spa;
3390	vdev_t *vd;
3391	uint64_t guid;
3392	int error;
3393
3394	mutex_enter(&ztest_vdev_lock);
3395
3396	if (ztest_device_removal_active) {
3397		mutex_exit(&ztest_vdev_lock);
3398		return;
3399	}
3400
3401	/*
3402	 * Remove a random top-level vdev and wait for removal to finish.
3403	 */
3404	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3405	vd = vdev_lookup_top(spa, ztest_random_vdev_top(spa, B_FALSE));
3406	guid = vd->vdev_guid;
3407	spa_config_exit(spa, SCL_VDEV, FTAG);
3408
3409	error = spa_vdev_remove(spa, guid, B_FALSE);
3410	if (error == 0) {
3411		ztest_device_removal_active = B_TRUE;
3412		mutex_exit(&ztest_vdev_lock);
3413
3414		while (spa->spa_vdev_removal != NULL)
3415			txg_wait_synced(spa_get_dsl(spa), 0);
3416	} else {
3417		mutex_exit(&ztest_vdev_lock);
3418		return;
3419	}
3420
3421	/*
3422	 * The pool needs to be scrubbed after completing device removal.
3423	 * Failure to do so may result in checksum errors due to the
3424	 * strategy employed by ztest_fault_inject() when selecting which
3425	 * offset are redundant and can be damaged.
3426	 */
3427	error = spa_scan(spa, POOL_SCAN_SCRUB);
3428	if (error == 0) {
3429		while (dsl_scan_scrubbing(spa_get_dsl(spa)))
3430			txg_wait_synced(spa_get_dsl(spa), 0);
3431	}
3432
3433	mutex_enter(&ztest_vdev_lock);
3434	ztest_device_removal_active = B_FALSE;
3435	mutex_exit(&ztest_vdev_lock);
3436}
3437
3438/*
3439 * Callback function which expands the physical size of the vdev.
3440 */
3441vdev_t *
3442grow_vdev(vdev_t *vd, void *arg)
3443{
3444	spa_t *spa = vd->vdev_spa;
3445	size_t *newsize = arg;
3446	size_t fsize;
3447	int fd;
3448
3449	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3450	ASSERT(vd->vdev_ops->vdev_op_leaf);
3451
3452	if ((fd = open(vd->vdev_path, O_RDWR)) == -1)
3453		return (vd);
3454
3455	fsize = lseek(fd, 0, SEEK_END);
3456	(void) ftruncate(fd, *newsize);
3457
3458	if (ztest_opts.zo_verbose >= 6) {
3459		(void) printf("%s grew from %lu to %lu bytes\n",
3460		    vd->vdev_path, (ulong_t)fsize, (ulong_t)*newsize);
3461	}
3462	(void) close(fd);
3463	return (NULL);
3464}
3465
3466/*
3467 * Callback function which expands a given vdev by calling vdev_online().
3468 */
3469/* ARGSUSED */
3470vdev_t *
3471online_vdev(vdev_t *vd, void *arg)
3472{
3473	spa_t *spa = vd->vdev_spa;
3474	vdev_t *tvd = vd->vdev_top;
3475	uint64_t guid = vd->vdev_guid;
3476	uint64_t generation = spa->spa_config_generation + 1;
3477	vdev_state_t newstate = VDEV_STATE_UNKNOWN;
3478	int error;
3479
3480	ASSERT(spa_config_held(spa, SCL_STATE, RW_READER) == SCL_STATE);
3481	ASSERT(vd->vdev_ops->vdev_op_leaf);
3482
3483	/* Calling vdev_online will initialize the new metaslabs */
3484	spa_config_exit(spa, SCL_STATE, spa);
3485	error = vdev_online(spa, guid, ZFS_ONLINE_EXPAND, &newstate);
3486	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3487
3488	/*
3489	 * If vdev_online returned an error or the underlying vdev_open
3490	 * failed then we abort the expand. The only way to know that
3491	 * vdev_open fails is by checking the returned newstate.
3492	 */
3493	if (error || newstate != VDEV_STATE_HEALTHY) {
3494		if (ztest_opts.zo_verbose >= 5) {
3495			(void) printf("Unable to expand vdev, state %llu, "
3496			    "error %d\n", (u_longlong_t)newstate, error);
3497		}
3498		return (vd);
3499	}
3500	ASSERT3U(newstate, ==, VDEV_STATE_HEALTHY);
3501
3502	/*
3503	 * Since we dropped the lock we need to ensure that we're
3504	 * still talking to the original vdev. It's possible this
3505	 * vdev may have been detached/replaced while we were
3506	 * trying to online it.
3507	 */
3508	if (generation != spa->spa_config_generation) {
3509		if (ztest_opts.zo_verbose >= 5) {
3510			(void) printf("vdev configuration has changed, "
3511			    "guid %llu, state %llu, expected gen %llu, "
3512			    "got gen %llu\n",
3513			    (u_longlong_t)guid,
3514			    (u_longlong_t)tvd->vdev_state,
3515			    (u_longlong_t)generation,
3516			    (u_longlong_t)spa->spa_config_generation);
3517		}
3518		return (vd);
3519	}
3520	return (NULL);
3521}
3522
3523/*
3524 * Traverse the vdev tree calling the supplied function.
3525 * We continue to walk the tree until we either have walked all
3526 * children or we receive a non-NULL return from the callback.
3527 * If a NULL callback is passed, then we just return back the first
3528 * leaf vdev we encounter.
3529 */
3530vdev_t *
3531vdev_walk_tree(vdev_t *vd, vdev_t *(*func)(vdev_t *, void *), void *arg)
3532{
3533	if (vd->vdev_ops->vdev_op_leaf) {
3534		if (func == NULL)
3535			return (vd);
3536		else
3537			return (func(vd, arg));
3538	}
3539
3540	for (uint_t c = 0; c < vd->vdev_children; c++) {
3541		vdev_t *cvd = vd->vdev_child[c];
3542		if ((cvd = vdev_walk_tree(cvd, func, arg)) != NULL)
3543			return (cvd);
3544	}
3545	return (NULL);
3546}
3547
3548/*
3549 * Verify that dynamic LUN growth works as expected.
3550 */
3551/* ARGSUSED */
3552void
3553ztest_vdev_LUN_growth(ztest_ds_t *zd, uint64_t id)
3554{
3555	spa_t *spa = ztest_spa;
3556	vdev_t *vd, *tvd;
3557	metaslab_class_t *mc;
3558	metaslab_group_t *mg;
3559	size_t psize, newsize;
3560	uint64_t top;
3561	uint64_t old_class_space, new_class_space, old_ms_count, new_ms_count;
3562
3563	mutex_enter(&ztest_checkpoint_lock);
3564	mutex_enter(&ztest_vdev_lock);
3565	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3566
3567	/*
3568	 * If there is a vdev removal in progress, it could complete while
3569	 * we are running, in which case we would not be able to verify
3570	 * that the metaslab_class space increased (because it decreases
3571	 * when the device removal completes).
3572	 */
3573	if (ztest_device_removal_active) {
3574		spa_config_exit(spa, SCL_STATE, spa);
3575		mutex_exit(&ztest_vdev_lock);
3576		mutex_exit(&ztest_checkpoint_lock);
3577		return;
3578	}
3579
3580	top = ztest_random_vdev_top(spa, B_TRUE);
3581
3582	tvd = spa->spa_root_vdev->vdev_child[top];
3583	mg = tvd->vdev_mg;
3584	mc = mg->mg_class;
3585	old_ms_count = tvd->vdev_ms_count;
3586	old_class_space = metaslab_class_get_space(mc);
3587
3588	/*
3589	 * Determine the size of the first leaf vdev associated with
3590	 * our top-level device.
3591	 */
3592	vd = vdev_walk_tree(tvd, NULL, NULL);
3593	ASSERT3P(vd, !=, NULL);
3594	ASSERT(vd->vdev_ops->vdev_op_leaf);
3595
3596	psize = vd->vdev_psize;
3597
3598	/*
3599	 * We only try to expand the vdev if it's healthy, less than 4x its
3600	 * original size, and it has a valid psize.
3601	 */
3602	if (tvd->vdev_state != VDEV_STATE_HEALTHY ||
3603	    psize == 0 || psize >= 4 * ztest_opts.zo_vdev_size) {
3604		spa_config_exit(spa, SCL_STATE, spa);
3605		mutex_exit(&ztest_vdev_lock);
3606		mutex_exit(&ztest_checkpoint_lock);
3607		return;
3608	}
3609	ASSERT(psize > 0);
3610	newsize = psize + MAX(psize / 8, SPA_MAXBLOCKSIZE);
3611	ASSERT3U(newsize, >, psize);
3612
3613	if (ztest_opts.zo_verbose >= 6) {
3614		(void) printf("Expanding LUN %s from %lu to %lu\n",
3615		    vd->vdev_path, (ulong_t)psize, (ulong_t)newsize);
3616	}
3617
3618	/*
3619	 * Growing the vdev is a two step process:
3620	 *	1). expand the physical size (i.e. relabel)
3621	 *	2). online the vdev to create the new metaslabs
3622	 */
3623	if (vdev_walk_tree(tvd, grow_vdev, &newsize) != NULL ||
3624	    vdev_walk_tree(tvd, online_vdev, NULL) != NULL ||
3625	    tvd->vdev_state != VDEV_STATE_HEALTHY) {
3626		if (ztest_opts.zo_verbose >= 5) {
3627			(void) printf("Could not expand LUN because "
3628			    "the vdev configuration changed.\n");
3629		}
3630		spa_config_exit(spa, SCL_STATE, spa);
3631		mutex_exit(&ztest_vdev_lock);
3632		mutex_exit(&ztest_checkpoint_lock);
3633		return;
3634	}
3635
3636	spa_config_exit(spa, SCL_STATE, spa);
3637
3638	/*
3639	 * Expanding the LUN will update the config asynchronously,
3640	 * thus we must wait for the async thread to complete any
3641	 * pending tasks before proceeding.
3642	 */
3643	for (;;) {
3644		boolean_t done;
3645		mutex_enter(&spa->spa_async_lock);
3646		done = (spa->spa_async_thread == NULL && !spa->spa_async_tasks);
3647		mutex_exit(&spa->spa_async_lock);
3648		if (done)
3649			break;
3650		txg_wait_synced(spa_get_dsl(spa), 0);
3651		(void) poll(NULL, 0, 100);
3652	}
3653
3654	spa_config_enter(spa, SCL_STATE, spa, RW_READER);
3655
3656	tvd = spa->spa_root_vdev->vdev_child[top];
3657	new_ms_count = tvd->vdev_ms_count;
3658	new_class_space = metaslab_class_get_space(mc);
3659
3660	if (tvd->vdev_mg != mg || mg->mg_class != mc) {
3661		if (ztest_opts.zo_verbose >= 5) {
3662			(void) printf("Could not verify LUN expansion due to "
3663			    "intervening vdev offline or remove.\n");
3664		}
3665		spa_config_exit(spa, SCL_STATE, spa);
3666		mutex_exit(&ztest_vdev_lock);
3667		mutex_exit(&ztest_checkpoint_lock);
3668		return;
3669	}
3670
3671	/*
3672	 * Make sure we were able to grow the vdev.
3673	 */
3674	if (new_ms_count <= old_ms_count) {
3675		fatal(0, "LUN expansion failed: ms_count %llu < %llu\n",
3676		    old_ms_count, new_ms_count);
3677	}
3678
3679	/*
3680	 * Make sure we were able to grow the pool.
3681	 */
3682	if (new_class_space <= old_class_space) {
3683		fatal(0, "LUN expansion failed: class_space %llu < %llu\n",
3684		    old_class_space, new_class_space);
3685	}
3686
3687	if (ztest_opts.zo_verbose >= 5) {
3688		char oldnumbuf[NN_NUMBUF_SZ], newnumbuf[NN_NUMBUF_SZ];
3689
3690		nicenum(old_class_space, oldnumbuf, sizeof (oldnumbuf));
3691		nicenum(new_class_space, newnumbuf, sizeof (newnumbuf));
3692		(void) printf("%s grew from %s to %s\n",
3693		    spa->spa_name, oldnumbuf, newnumbuf);
3694	}
3695
3696	spa_config_exit(spa, SCL_STATE, spa);
3697	mutex_exit(&ztest_vdev_lock);
3698	mutex_exit(&ztest_checkpoint_lock);
3699}
3700
3701/*
3702 * Verify that dmu_objset_{create,destroy,open,close} work as expected.
3703 */
3704/* ARGSUSED */
3705static void
3706ztest_objset_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
3707{
3708	/*
3709	 * Create the objects common to all ztest datasets.
3710	 */
3711	VERIFY(zap_create_claim(os, ZTEST_DIROBJ,
3712	    DMU_OT_ZAP_OTHER, DMU_OT_NONE, 0, tx) == 0);
3713}
3714
3715static int
3716ztest_dataset_create(char *dsname)
3717{
3718	uint64_t zilset = ztest_random(100);
3719	int err = dmu_objset_create(dsname, DMU_OST_OTHER, 0,
3720	    ztest_objset_create_cb, NULL);
3721
3722	if (err || zilset < 80)
3723		return (err);
3724
3725	if (ztest_opts.zo_verbose >= 6)
3726		(void) printf("Setting dataset %s to sync always\n", dsname);
3727	return (ztest_dsl_prop_set_uint64(dsname, ZFS_PROP_SYNC,
3728	    ZFS_SYNC_ALWAYS, B_FALSE));
3729}
3730
3731/* ARGSUSED */
3732static int
3733ztest_objset_destroy_cb(const char *name, void *arg)
3734{
3735	objset_t *os;
3736	dmu_object_info_t doi;
3737	int error;
3738
3739	/*
3740	 * Verify that the dataset contains a directory object.
3741	 */
3742	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_TRUE, FTAG, &os));
3743	error = dmu_object_info(os, ZTEST_DIROBJ, &doi);
3744	if (error != ENOENT) {
3745		/* We could have crashed in the middle of destroying it */
3746		ASSERT0(error);
3747		ASSERT3U(doi.doi_type, ==, DMU_OT_ZAP_OTHER);
3748		ASSERT3S(doi.doi_physical_blocks_512, >=, 0);
3749	}
3750	dmu_objset_disown(os, FTAG);
3751
3752	/*
3753	 * Destroy the dataset.
3754	 */
3755	if (strchr(name, '@') != NULL) {
3756		VERIFY0(dsl_destroy_snapshot(name, B_FALSE));
3757	} else {
3758		VERIFY0(dsl_destroy_head(name));
3759	}
3760	return (0);
3761}
3762
3763static boolean_t
3764ztest_snapshot_create(char *osname, uint64_t id)
3765{
3766	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3767	int error;
3768
3769	(void) snprintf(snapname, sizeof (snapname), "%llu", (u_longlong_t)id);
3770
3771	error = dmu_objset_snapshot_one(osname, snapname);
3772	if (error == ENOSPC) {
3773		ztest_record_enospc(FTAG);
3774		return (B_FALSE);
3775	}
3776	if (error != 0 && error != EEXIST) {
3777		fatal(0, "ztest_snapshot_create(%s@%s) = %d", osname,
3778		    snapname, error);
3779	}
3780	return (B_TRUE);
3781}
3782
3783static boolean_t
3784ztest_snapshot_destroy(char *osname, uint64_t id)
3785{
3786	char snapname[ZFS_MAX_DATASET_NAME_LEN];
3787	int error;
3788
3789	(void) snprintf(snapname, sizeof (snapname), "%s@%llu", osname,
3790	    (u_longlong_t)id);
3791
3792	error = dsl_destroy_snapshot(snapname, B_FALSE);
3793	if (error != 0 && error != ENOENT)
3794		fatal(0, "ztest_snapshot_destroy(%s) = %d", snapname, error);
3795	return (B_TRUE);
3796}
3797
3798/* ARGSUSED */
3799void
3800ztest_dmu_objset_create_destroy(ztest_ds_t *zd, uint64_t id)
3801{
3802	ztest_ds_t zdtmp;
3803	int iters;
3804	int error;
3805	objset_t *os, *os2;
3806	char name[ZFS_MAX_DATASET_NAME_LEN];
3807	zilog_t *zilog;
3808
3809	rw_enter(&ztest_name_lock, RW_READER);
3810
3811	(void) snprintf(name, sizeof (name), "%s/temp_%llu",
3812	    ztest_opts.zo_pool, (u_longlong_t)id);
3813
3814	/*
3815	 * If this dataset exists from a previous run, process its replay log
3816	 * half of the time.  If we don't replay it, then dmu_objset_destroy()
3817	 * (invoked from ztest_objset_destroy_cb()) should just throw it away.
3818	 */
3819	if (ztest_random(2) == 0 &&
3820	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os) == 0) {
3821		ztest_zd_init(&zdtmp, NULL, os);
3822		zil_replay(os, &zdtmp, ztest_replay_vector);
3823		ztest_zd_fini(&zdtmp);
3824		dmu_objset_disown(os, FTAG);
3825	}
3826
3827	/*
3828	 * There may be an old instance of the dataset we're about to
3829	 * create lying around from a previous run.  If so, destroy it
3830	 * and all of its snapshots.
3831	 */
3832	(void) dmu_objset_find(name, ztest_objset_destroy_cb, NULL,
3833	    DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
3834
3835	/*
3836	 * Verify that the destroyed dataset is no longer in the namespace.
3837	 */
3838	VERIFY3U(ENOENT, ==, dmu_objset_own(name, DMU_OST_OTHER, B_TRUE,
3839	    FTAG, &os));
3840
3841	/*
3842	 * Verify that we can create a new dataset.
3843	 */
3844	error = ztest_dataset_create(name);
3845	if (error) {
3846		if (error == ENOSPC) {
3847			ztest_record_enospc(FTAG);
3848			rw_exit(&ztest_name_lock);
3849			return;
3850		}
3851		fatal(0, "dmu_objset_create(%s) = %d", name, error);
3852	}
3853
3854	VERIFY0(dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os));
3855
3856	ztest_zd_init(&zdtmp, NULL, os);
3857
3858	/*
3859	 * Open the intent log for it.
3860	 */
3861	zilog = zil_open(os, ztest_get_data);
3862
3863	/*
3864	 * Put some objects in there, do a little I/O to them,
3865	 * and randomly take a couple of snapshots along the way.
3866	 */
3867	iters = ztest_random(5);
3868	for (int i = 0; i < iters; i++) {
3869		ztest_dmu_object_alloc_free(&zdtmp, id);
3870		if (ztest_random(iters) == 0)
3871			(void) ztest_snapshot_create(name, i);
3872	}
3873
3874	/*
3875	 * Verify that we cannot create an existing dataset.
3876	 */
3877	VERIFY3U(EEXIST, ==,
3878	    dmu_objset_create(name, DMU_OST_OTHER, 0, NULL, NULL));
3879
3880	/*
3881	 * Verify that we can hold an objset that is also owned.
3882	 */
3883	VERIFY3U(0, ==, dmu_objset_hold(name, FTAG, &os2));
3884	dmu_objset_rele(os2, FTAG);
3885
3886	/*
3887	 * Verify that we cannot own an objset that is already owned.
3888	 */
3889	VERIFY3U(EBUSY, ==,
3890	    dmu_objset_own(name, DMU_OST_OTHER, B_FALSE, FTAG, &os2));
3891
3892	zil_close(zilog);
3893	dmu_objset_disown(os, FTAG);
3894	ztest_zd_fini(&zdtmp);
3895
3896	rw_exit(&ztest_name_lock);
3897}
3898
3899/*
3900 * Verify that dmu_snapshot_{create,destroy,open,close} work as expected.
3901 */
3902void
3903ztest_dmu_snapshot_create_destroy(ztest_ds_t *zd, uint64_t id)
3904{
3905	rw_enter(&ztest_name_lock, RW_READER);
3906	(void) ztest_snapshot_destroy(zd->zd_name, id);
3907	(void) ztest_snapshot_create(zd->zd_name, id);
3908	rw_exit(&ztest_name_lock);
3909}
3910
3911/*
3912 * Cleanup non-standard snapshots and clones.
3913 */
3914void
3915ztest_dsl_dataset_cleanup(char *osname, uint64_t id)
3916{
3917	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3918	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3919	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3920	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3921	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3922	int error;
3923
3924	(void) snprintf(snap1name, sizeof (snap1name),
3925	    "%s@s1_%llu", osname, id);
3926	(void) snprintf(clone1name, sizeof (clone1name),
3927	    "%s/c1_%llu", osname, id);
3928	(void) snprintf(snap2name, sizeof (snap2name),
3929	    "%s@s2_%llu", clone1name, id);
3930	(void) snprintf(clone2name, sizeof (clone2name),
3931	    "%s/c2_%llu", osname, id);
3932	(void) snprintf(snap3name, sizeof (snap3name),
3933	    "%s@s3_%llu", clone1name, id);
3934
3935	error = dsl_destroy_head(clone2name);
3936	if (error && error != ENOENT)
3937		fatal(0, "dsl_destroy_head(%s) = %d", clone2name, error);
3938	error = dsl_destroy_snapshot(snap3name, B_FALSE);
3939	if (error && error != ENOENT)
3940		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap3name, error);
3941	error = dsl_destroy_snapshot(snap2name, B_FALSE);
3942	if (error && error != ENOENT)
3943		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap2name, error);
3944	error = dsl_destroy_head(clone1name);
3945	if (error && error != ENOENT)
3946		fatal(0, "dsl_destroy_head(%s) = %d", clone1name, error);
3947	error = dsl_destroy_snapshot(snap1name, B_FALSE);
3948	if (error && error != ENOENT)
3949		fatal(0, "dsl_destroy_snapshot(%s) = %d", snap1name, error);
3950}
3951
3952/*
3953 * Verify dsl_dataset_promote handles EBUSY
3954 */
3955void
3956ztest_dsl_dataset_promote_busy(ztest_ds_t *zd, uint64_t id)
3957{
3958	objset_t *os;
3959	char snap1name[ZFS_MAX_DATASET_NAME_LEN];
3960	char clone1name[ZFS_MAX_DATASET_NAME_LEN];
3961	char snap2name[ZFS_MAX_DATASET_NAME_LEN];
3962	char clone2name[ZFS_MAX_DATASET_NAME_LEN];
3963	char snap3name[ZFS_MAX_DATASET_NAME_LEN];
3964	char *osname = zd->zd_name;
3965	int error;
3966
3967	rw_enter(&ztest_name_lock, RW_READER);
3968
3969	ztest_dsl_dataset_cleanup(osname, id);
3970
3971	(void) snprintf(snap1name, sizeof (snap1name),
3972	    "%s@s1_%llu", osname, id);
3973	(void) snprintf(clone1name, sizeof (clone1name),
3974	    "%s/c1_%llu", osname, id);
3975	(void) snprintf(snap2name, sizeof (snap2name),
3976	    "%s@s2_%llu", clone1name, id);
3977	(void) snprintf(clone2name, sizeof (clone2name),
3978	    "%s/c2_%llu", osname, id);
3979	(void) snprintf(snap3name, sizeof (snap3name),
3980	    "%s@s3_%llu", clone1name, id);
3981
3982	error = dmu_objset_snapshot_one(osname, strchr(snap1name, '@') + 1);
3983	if (error && error != EEXIST) {
3984		if (error == ENOSPC) {
3985			ztest_record_enospc(FTAG);
3986			goto out;
3987		}
3988		fatal(0, "dmu_take_snapshot(%s) = %d", snap1name, error);
3989	}
3990
3991	error = dmu_objset_clone(clone1name, snap1name);
3992	if (error) {
3993		if (error == ENOSPC) {
3994			ztest_record_enospc(FTAG);
3995			goto out;
3996		}
3997		fatal(0, "dmu_objset_create(%s) = %d", clone1name, error);
3998	}
3999
4000	error = dmu_objset_snapshot_one(clone1name, strchr(snap2name, '@') + 1);
4001	if (error && error != EEXIST) {
4002		if (error == ENOSPC) {
4003			ztest_record_enospc(FTAG);
4004			goto out;
4005		}
4006		fatal(0, "dmu_open_snapshot(%s) = %d", snap2name, error);
4007	}
4008
4009	error = dmu_objset_snapshot_one(clone1name, strchr(snap3name, '@') + 1);
4010	if (error && error != EEXIST) {
4011		if (error == ENOSPC) {
4012			ztest_record_enospc(FTAG);
4013			goto out;
4014		}
4015		fatal(0, "dmu_open_snapshot(%s) = %d", snap3name, error);
4016	}
4017
4018	error = dmu_objset_clone(clone2name, snap3name);
4019	if (error) {
4020		if (error == ENOSPC) {
4021			ztest_record_enospc(FTAG);
4022			goto out;
4023		}
4024		fatal(0, "dmu_objset_create(%s) = %d", clone2name, error);
4025	}
4026
4027	error = dmu_objset_own(snap2name, DMU_OST_ANY, B_TRUE, FTAG, &os);
4028	if (error)
4029		fatal(0, "dmu_objset_own(%s) = %d", snap2name, error);
4030	error = dsl_dataset_promote(clone2name, NULL);
4031	if (error == ENOSPC) {
4032		dmu_objset_disown(os, FTAG);
4033		ztest_record_enospc(FTAG);
4034		goto out;
4035	}
4036	if (error != EBUSY)
4037		fatal(0, "dsl_dataset_promote(%s), %d, not EBUSY", clone2name,
4038		    error);
4039	dmu_objset_disown(os, FTAG);
4040
4041out:
4042	ztest_dsl_dataset_cleanup(osname, id);
4043
4044	rw_exit(&ztest_name_lock);
4045}
4046
4047/*
4048 * Verify that dmu_object_{alloc,free} work as expected.
4049 */
4050void
4051ztest_dmu_object_alloc_free(ztest_ds_t *zd, uint64_t id)
4052{
4053	ztest_od_t od[4];
4054	int batchsize = sizeof (od) / sizeof (od[0]);
4055
4056	for (int b = 0; b < batchsize; b++) {
4057		ztest_od_init(&od[b], id, FTAG, b, DMU_OT_UINT64_OTHER,
4058		    0, 0, 0);
4059	}
4060
4061	/*
4062	 * Destroy the previous batch of objects, create a new batch,
4063	 * and do some I/O on the new objects.
4064	 */
4065	if (ztest_object_init(zd, od, sizeof (od), B_TRUE) != 0)
4066		return;
4067
4068	while (ztest_random(4 * batchsize) != 0)
4069		ztest_io(zd, od[ztest_random(batchsize)].od_object,
4070		    ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4071}
4072
4073/*
4074 * Rewind the global allocator to verify object allocation backfilling.
4075 */
4076void
4077ztest_dmu_object_next_chunk(ztest_ds_t *zd, uint64_t id)
4078{
4079	objset_t *os = zd->zd_os;
4080	int dnodes_per_chunk = 1 << dmu_object_alloc_chunk_shift;
4081	uint64_t object;
4082
4083	/*
4084	 * Rewind the global allocator randomly back to a lower object number
4085	 * to force backfilling and reclamation of recently freed dnodes.
4086	 */
4087	mutex_enter(&os->os_obj_lock);
4088	object = ztest_random(os->os_obj_next_chunk);
4089	os->os_obj_next_chunk = P2ALIGN(object, dnodes_per_chunk);
4090	mutex_exit(&os->os_obj_lock);
4091}
4092
4093/*
4094 * Verify that dmu_{read,write} work as expected.
4095 */
4096void
4097ztest_dmu_read_write(ztest_ds_t *zd, uint64_t id)
4098{
4099	objset_t *os = zd->zd_os;
4100	ztest_od_t od[2];
4101	dmu_tx_t *tx;
4102	int i, freeit, error;
4103	uint64_t n, s, txg;
4104	bufwad_t *packbuf, *bigbuf, *pack, *bigH, *bigT;
4105	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4106	uint64_t chunksize = (1000 + ztest_random(1000)) * sizeof (uint64_t);
4107	uint64_t regions = 997;
4108	uint64_t stride = 123456789ULL;
4109	uint64_t width = 40;
4110	int free_percent = 5;
4111
4112	/*
4113	 * This test uses two objects, packobj and bigobj, that are always
4114	 * updated together (i.e. in the same tx) so that their contents are
4115	 * in sync and can be compared.  Their contents relate to each other
4116	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4117	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4118	 * for any index n, there are three bufwads that should be identical:
4119	 *
4120	 *	packobj, at offset n * sizeof (bufwad_t)
4121	 *	bigobj, at the head of the nth chunk
4122	 *	bigobj, at the tail of the nth chunk
4123	 *
4124	 * The chunk size is arbitrary. It doesn't have to be a power of two,
4125	 * and it doesn't have any relation to the object blocksize.
4126	 * The only requirement is that it can hold at least two bufwads.
4127	 *
4128	 * Normally, we write the bufwad to each of these locations.
4129	 * However, free_percent of the time we instead write zeroes to
4130	 * packobj and perform a dmu_free_range() on bigobj.  By comparing
4131	 * bigobj to packobj, we can verify that the DMU is correctly
4132	 * tracking which parts of an object are allocated and free,
4133	 * and that the contents of the allocated blocks are correct.
4134	 */
4135
4136	/*
4137	 * Read the directory info.  If it's the first time, set things up.
4138	 */
4139	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0,
4140	    chunksize);
4141	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4142	    chunksize);
4143
4144	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4145		return;
4146
4147	bigobj = od[0].od_object;
4148	packobj = od[1].od_object;
4149	chunksize = od[0].od_gen;
4150	ASSERT(chunksize == od[1].od_gen);
4151
4152	/*
4153	 * Prefetch a random chunk of the big object.
4154	 * Our aim here is to get some async reads in flight
4155	 * for blocks that we may free below; the DMU should
4156	 * handle this race correctly.
4157	 */
4158	n = ztest_random(regions) * stride + ztest_random(width);
4159	s = 1 + ztest_random(2 * width - 1);
4160	dmu_prefetch(os, bigobj, 0, n * chunksize, s * chunksize,
4161	    ZIO_PRIORITY_SYNC_READ);
4162
4163	/*
4164	 * Pick a random index and compute the offsets into packobj and bigobj.
4165	 */
4166	n = ztest_random(regions) * stride + ztest_random(width);
4167	s = 1 + ztest_random(width - 1);
4168
4169	packoff = n * sizeof (bufwad_t);
4170	packsize = s * sizeof (bufwad_t);
4171
4172	bigoff = n * chunksize;
4173	bigsize = s * chunksize;
4174
4175	packbuf = umem_alloc(packsize, UMEM_NOFAIL);
4176	bigbuf = umem_alloc(bigsize, UMEM_NOFAIL);
4177
4178	/*
4179	 * free_percent of the time, free a range of bigobj rather than
4180	 * overwriting it.
4181	 */
4182	freeit = (ztest_random(100) < free_percent);
4183
4184	/*
4185	 * Read the current contents of our objects.
4186	 */
4187	error = dmu_read(os, packobj, packoff, packsize, packbuf,
4188	    DMU_READ_PREFETCH);
4189	ASSERT0(error);
4190	error = dmu_read(os, bigobj, bigoff, bigsize, bigbuf,
4191	    DMU_READ_PREFETCH);
4192	ASSERT0(error);
4193
4194	/*
4195	 * Get a tx for the mods to both packobj and bigobj.
4196	 */
4197	tx = dmu_tx_create(os);
4198
4199	dmu_tx_hold_write(tx, packobj, packoff, packsize);
4200
4201	if (freeit)
4202		dmu_tx_hold_free(tx, bigobj, bigoff, bigsize);
4203	else
4204		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4205
4206	/* This accounts for setting the checksum/compression. */
4207	dmu_tx_hold_bonus(tx, bigobj);
4208
4209	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4210	if (txg == 0) {
4211		umem_free(packbuf, packsize);
4212		umem_free(bigbuf, bigsize);
4213		return;
4214	}
4215
4216	enum zio_checksum cksum;
4217	do {
4218		cksum = (enum zio_checksum)
4219		    ztest_random_dsl_prop(ZFS_PROP_CHECKSUM);
4220	} while (cksum >= ZIO_CHECKSUM_LEGACY_FUNCTIONS);
4221	dmu_object_set_checksum(os, bigobj, cksum, tx);
4222
4223	enum zio_compress comp;
4224	do {
4225		comp = (enum zio_compress)
4226		    ztest_random_dsl_prop(ZFS_PROP_COMPRESSION);
4227	} while (comp >= ZIO_COMPRESS_LEGACY_FUNCTIONS);
4228	dmu_object_set_compress(os, bigobj, comp, tx);
4229
4230	/*
4231	 * For each index from n to n + s, verify that the existing bufwad
4232	 * in packobj matches the bufwads at the head and tail of the
4233	 * corresponding chunk in bigobj.  Then update all three bufwads
4234	 * with the new values we want to write out.
4235	 */
4236	for (i = 0; i < s; i++) {
4237		/* LINTED */
4238		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4239		/* LINTED */
4240		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4241		/* LINTED */
4242		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4243
4244		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4245		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4246
4247		if (pack->bw_txg > txg)
4248			fatal(0, "future leak: got %llx, open txg is %llx",
4249			    pack->bw_txg, txg);
4250
4251		if (pack->bw_data != 0 && pack->bw_index != n + i)
4252			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4253			    pack->bw_index, n, i);
4254
4255		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4256			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4257
4258		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4259			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4260
4261		if (freeit) {
4262			bzero(pack, sizeof (bufwad_t));
4263		} else {
4264			pack->bw_index = n + i;
4265			pack->bw_txg = txg;
4266			pack->bw_data = 1 + ztest_random(-2ULL);
4267		}
4268		*bigH = *pack;
4269		*bigT = *pack;
4270	}
4271
4272	/*
4273	 * We've verified all the old bufwads, and made new ones.
4274	 * Now write them out.
4275	 */
4276	dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4277
4278	if (freeit) {
4279		if (ztest_opts.zo_verbose >= 7) {
4280			(void) printf("freeing offset %llx size %llx"
4281			    " txg %llx\n",
4282			    (u_longlong_t)bigoff,
4283			    (u_longlong_t)bigsize,
4284			    (u_longlong_t)txg);
4285		}
4286		VERIFY(0 == dmu_free_range(os, bigobj, bigoff, bigsize, tx));
4287	} else {
4288		if (ztest_opts.zo_verbose >= 7) {
4289			(void) printf("writing offset %llx size %llx"
4290			    " txg %llx\n",
4291			    (u_longlong_t)bigoff,
4292			    (u_longlong_t)bigsize,
4293			    (u_longlong_t)txg);
4294		}
4295		dmu_write(os, bigobj, bigoff, bigsize, bigbuf, tx);
4296	}
4297
4298	dmu_tx_commit(tx);
4299
4300	/*
4301	 * Sanity check the stuff we just wrote.
4302	 */
4303	{
4304		void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4305		void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4306
4307		VERIFY(0 == dmu_read(os, packobj, packoff,
4308		    packsize, packcheck, DMU_READ_PREFETCH));
4309		VERIFY(0 == dmu_read(os, bigobj, bigoff,
4310		    bigsize, bigcheck, DMU_READ_PREFETCH));
4311
4312		ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4313		ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4314
4315		umem_free(packcheck, packsize);
4316		umem_free(bigcheck, bigsize);
4317	}
4318
4319	umem_free(packbuf, packsize);
4320	umem_free(bigbuf, bigsize);
4321}
4322
4323void
4324compare_and_update_pbbufs(uint64_t s, bufwad_t *packbuf, bufwad_t *bigbuf,
4325    uint64_t bigsize, uint64_t n, uint64_t chunksize, uint64_t txg)
4326{
4327	uint64_t i;
4328	bufwad_t *pack;
4329	bufwad_t *bigH;
4330	bufwad_t *bigT;
4331
4332	/*
4333	 * For each index from n to n + s, verify that the existing bufwad
4334	 * in packobj matches the bufwads at the head and tail of the
4335	 * corresponding chunk in bigobj.  Then update all three bufwads
4336	 * with the new values we want to write out.
4337	 */
4338	for (i = 0; i < s; i++) {
4339		/* LINTED */
4340		pack = (bufwad_t *)((char *)packbuf + i * sizeof (bufwad_t));
4341		/* LINTED */
4342		bigH = (bufwad_t *)((char *)bigbuf + i * chunksize);
4343		/* LINTED */
4344		bigT = (bufwad_t *)((char *)bigH + chunksize) - 1;
4345
4346		ASSERT((uintptr_t)bigH - (uintptr_t)bigbuf < bigsize);
4347		ASSERT((uintptr_t)bigT - (uintptr_t)bigbuf < bigsize);
4348
4349		if (pack->bw_txg > txg)
4350			fatal(0, "future leak: got %llx, open txg is %llx",
4351			    pack->bw_txg, txg);
4352
4353		if (pack->bw_data != 0 && pack->bw_index != n + i)
4354			fatal(0, "wrong index: got %llx, wanted %llx+%llx",
4355			    pack->bw_index, n, i);
4356
4357		if (bcmp(pack, bigH, sizeof (bufwad_t)) != 0)
4358			fatal(0, "pack/bigH mismatch in %p/%p", pack, bigH);
4359
4360		if (bcmp(pack, bigT, sizeof (bufwad_t)) != 0)
4361			fatal(0, "pack/bigT mismatch in %p/%p", pack, bigT);
4362
4363		pack->bw_index = n + i;
4364		pack->bw_txg = txg;
4365		pack->bw_data = 1 + ztest_random(-2ULL);
4366
4367		*bigH = *pack;
4368		*bigT = *pack;
4369	}
4370}
4371
4372void
4373ztest_dmu_read_write_zcopy(ztest_ds_t *zd, uint64_t id)
4374{
4375	objset_t *os = zd->zd_os;
4376	ztest_od_t od[2];
4377	dmu_tx_t *tx;
4378	uint64_t i;
4379	int error;
4380	uint64_t n, s, txg;
4381	bufwad_t *packbuf, *bigbuf;
4382	uint64_t packobj, packoff, packsize, bigobj, bigoff, bigsize;
4383	uint64_t blocksize = ztest_random_blocksize();
4384	uint64_t chunksize = blocksize;
4385	uint64_t regions = 997;
4386	uint64_t stride = 123456789ULL;
4387	uint64_t width = 9;
4388	dmu_buf_t *bonus_db;
4389	arc_buf_t **bigbuf_arcbufs;
4390	dmu_object_info_t doi;
4391
4392	/*
4393	 * This test uses two objects, packobj and bigobj, that are always
4394	 * updated together (i.e. in the same tx) so that their contents are
4395	 * in sync and can be compared.  Their contents relate to each other
4396	 * in a simple way: packobj is a dense array of 'bufwad' structures,
4397	 * while bigobj is a sparse array of the same bufwads.  Specifically,
4398	 * for any index n, there are three bufwads that should be identical:
4399	 *
4400	 *	packobj, at offset n * sizeof (bufwad_t)
4401	 *	bigobj, at the head of the nth chunk
4402	 *	bigobj, at the tail of the nth chunk
4403	 *
4404	 * The chunk size is set equal to bigobj block size so that
4405	 * dmu_assign_arcbuf() can be tested for object updates.
4406	 */
4407
4408	/*
4409	 * Read the directory info.  If it's the first time, set things up.
4410	 */
4411	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
4412	    0, 0);
4413	ztest_od_init(&od[1], id, FTAG, 1, DMU_OT_UINT64_OTHER, 0, 0,
4414	    chunksize);
4415
4416	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4417		return;
4418
4419	bigobj = od[0].od_object;
4420	packobj = od[1].od_object;
4421	blocksize = od[0].od_blocksize;
4422	chunksize = blocksize;
4423	ASSERT(chunksize == od[1].od_gen);
4424
4425	VERIFY(dmu_object_info(os, bigobj, &doi) == 0);
4426	VERIFY(ISP2(doi.doi_data_block_size));
4427	VERIFY(chunksize == doi.doi_data_block_size);
4428	VERIFY(chunksize >= 2 * sizeof (bufwad_t));
4429
4430	/*
4431	 * Pick a random index and compute the offsets into packobj and bigobj.
4432	 */
4433	n = ztest_random(regions) * stride + ztest_random(width);
4434	s = 1 + ztest_random(width - 1);
4435
4436	packoff = n * sizeof (bufwad_t);
4437	packsize = s * sizeof (bufwad_t);
4438
4439	bigoff = n * chunksize;
4440	bigsize = s * chunksize;
4441
4442	packbuf = umem_zalloc(packsize, UMEM_NOFAIL);
4443	bigbuf = umem_zalloc(bigsize, UMEM_NOFAIL);
4444
4445	VERIFY3U(0, ==, dmu_bonus_hold(os, bigobj, FTAG, &bonus_db));
4446
4447	bigbuf_arcbufs = umem_zalloc(2 * s * sizeof (arc_buf_t *), UMEM_NOFAIL);
4448
4449	/*
4450	 * Iteration 0 test zcopy for DB_UNCACHED dbufs.
4451	 * Iteration 1 test zcopy to already referenced dbufs.
4452	 * Iteration 2 test zcopy to dirty dbuf in the same txg.
4453	 * Iteration 3 test zcopy to dbuf dirty in previous txg.
4454	 * Iteration 4 test zcopy when dbuf is no longer dirty.
4455	 * Iteration 5 test zcopy when it can't be done.
4456	 * Iteration 6 one more zcopy write.
4457	 */
4458	for (i = 0; i < 7; i++) {
4459		uint64_t j;
4460		uint64_t off;
4461
4462		/*
4463		 * In iteration 5 (i == 5) use arcbufs
4464		 * that don't match bigobj blksz to test
4465		 * dmu_assign_arcbuf() when it can't directly
4466		 * assign an arcbuf to a dbuf.
4467		 */
4468		for (j = 0; j < s; j++) {
4469			if (i != 5) {
4470				bigbuf_arcbufs[j] =
4471				    dmu_request_arcbuf(bonus_db, chunksize);
4472			} else {
4473				bigbuf_arcbufs[2 * j] =
4474				    dmu_request_arcbuf(bonus_db, chunksize / 2);
4475				bigbuf_arcbufs[2 * j + 1] =
4476				    dmu_request_arcbuf(bonus_db, chunksize / 2);
4477			}
4478		}
4479
4480		/*
4481		 * Get a tx for the mods to both packobj and bigobj.
4482		 */
4483		tx = dmu_tx_create(os);
4484
4485		dmu_tx_hold_write(tx, packobj, packoff, packsize);
4486		dmu_tx_hold_write(tx, bigobj, bigoff, bigsize);
4487
4488		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4489		if (txg == 0) {
4490			umem_free(packbuf, packsize);
4491			umem_free(bigbuf, bigsize);
4492			for (j = 0; j < s; j++) {
4493				if (i != 5) {
4494					dmu_return_arcbuf(bigbuf_arcbufs[j]);
4495				} else {
4496					dmu_return_arcbuf(
4497					    bigbuf_arcbufs[2 * j]);
4498					dmu_return_arcbuf(
4499					    bigbuf_arcbufs[2 * j + 1]);
4500				}
4501			}
4502			umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4503			dmu_buf_rele(bonus_db, FTAG);
4504			return;
4505		}
4506
4507		/*
4508		 * 50% of the time don't read objects in the 1st iteration to
4509		 * test dmu_assign_arcbuf() for the case when there're no
4510		 * existing dbufs for the specified offsets.
4511		 */
4512		if (i != 0 || ztest_random(2) != 0) {
4513			error = dmu_read(os, packobj, packoff,
4514			    packsize, packbuf, DMU_READ_PREFETCH);
4515			ASSERT0(error);
4516			error = dmu_read(os, bigobj, bigoff, bigsize,
4517			    bigbuf, DMU_READ_PREFETCH);
4518			ASSERT0(error);
4519		}
4520		compare_and_update_pbbufs(s, packbuf, bigbuf, bigsize,
4521		    n, chunksize, txg);
4522
4523		/*
4524		 * We've verified all the old bufwads, and made new ones.
4525		 * Now write them out.
4526		 */
4527		dmu_write(os, packobj, packoff, packsize, packbuf, tx);
4528		if (ztest_opts.zo_verbose >= 7) {
4529			(void) printf("writing offset %llx size %llx"
4530			    " txg %llx\n",
4531			    (u_longlong_t)bigoff,
4532			    (u_longlong_t)bigsize,
4533			    (u_longlong_t)txg);
4534		}
4535		for (off = bigoff, j = 0; j < s; j++, off += chunksize) {
4536			dmu_buf_t *dbt;
4537			if (i != 5) {
4538				bcopy((caddr_t)bigbuf + (off - bigoff),
4539				    bigbuf_arcbufs[j]->b_data, chunksize);
4540			} else {
4541				bcopy((caddr_t)bigbuf + (off - bigoff),
4542				    bigbuf_arcbufs[2 * j]->b_data,
4543				    chunksize / 2);
4544				bcopy((caddr_t)bigbuf + (off - bigoff) +
4545				    chunksize / 2,
4546				    bigbuf_arcbufs[2 * j + 1]->b_data,
4547				    chunksize / 2);
4548			}
4549
4550			if (i == 1) {
4551				VERIFY(dmu_buf_hold(os, bigobj, off,
4552				    FTAG, &dbt, DMU_READ_NO_PREFETCH) == 0);
4553			}
4554			if (i != 5) {
4555				dmu_assign_arcbuf(bonus_db, off,
4556				    bigbuf_arcbufs[j], tx);
4557			} else {
4558				dmu_assign_arcbuf(bonus_db, off,
4559				    bigbuf_arcbufs[2 * j], tx);
4560				dmu_assign_arcbuf(bonus_db,
4561				    off + chunksize / 2,
4562				    bigbuf_arcbufs[2 * j + 1], tx);
4563			}
4564			if (i == 1) {
4565				dmu_buf_rele(dbt, FTAG);
4566			}
4567		}
4568		dmu_tx_commit(tx);
4569
4570		/*
4571		 * Sanity check the stuff we just wrote.
4572		 */
4573		{
4574			void *packcheck = umem_alloc(packsize, UMEM_NOFAIL);
4575			void *bigcheck = umem_alloc(bigsize, UMEM_NOFAIL);
4576
4577			VERIFY(0 == dmu_read(os, packobj, packoff,
4578			    packsize, packcheck, DMU_READ_PREFETCH));
4579			VERIFY(0 == dmu_read(os, bigobj, bigoff,
4580			    bigsize, bigcheck, DMU_READ_PREFETCH));
4581
4582			ASSERT(bcmp(packbuf, packcheck, packsize) == 0);
4583			ASSERT(bcmp(bigbuf, bigcheck, bigsize) == 0);
4584
4585			umem_free(packcheck, packsize);
4586			umem_free(bigcheck, bigsize);
4587		}
4588		if (i == 2) {
4589			txg_wait_open(dmu_objset_pool(os), 0);
4590		} else if (i == 3) {
4591			txg_wait_synced(dmu_objset_pool(os), 0);
4592		}
4593	}
4594
4595	dmu_buf_rele(bonus_db, FTAG);
4596	umem_free(packbuf, packsize);
4597	umem_free(bigbuf, bigsize);
4598	umem_free(bigbuf_arcbufs, 2 * s * sizeof (arc_buf_t *));
4599}
4600
4601/* ARGSUSED */
4602void
4603ztest_dmu_write_parallel(ztest_ds_t *zd, uint64_t id)
4604{
4605	ztest_od_t od[1];
4606	uint64_t offset = (1ULL << (ztest_random(20) + 43)) +
4607	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4608
4609	/*
4610	 * Have multiple threads write to large offsets in an object
4611	 * to verify that parallel writes to an object -- even to the
4612	 * same blocks within the object -- doesn't cause any trouble.
4613	 */
4614	ztest_od_init(&od[0], ID_PARALLEL, FTAG, 0, DMU_OT_UINT64_OTHER,
4615	    0, 0, 0);
4616
4617	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4618		return;
4619
4620	while (ztest_random(10) != 0)
4621		ztest_io(zd, od[0].od_object, offset);
4622}
4623
4624void
4625ztest_dmu_prealloc(ztest_ds_t *zd, uint64_t id)
4626{
4627	ztest_od_t od[1];
4628	uint64_t offset = (1ULL << (ztest_random(4) + SPA_MAXBLOCKSHIFT)) +
4629	    (ztest_random(ZTEST_RANGE_LOCKS) << SPA_MAXBLOCKSHIFT);
4630	uint64_t count = ztest_random(20) + 1;
4631	uint64_t blocksize = ztest_random_blocksize();
4632	void *data;
4633
4634	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
4635	    0, 0);
4636
4637	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4638		return;
4639
4640	if (ztest_truncate(zd, od[0].od_object, offset, count * blocksize) != 0)
4641		return;
4642
4643	ztest_prealloc(zd, od[0].od_object, offset, count * blocksize);
4644
4645	data = umem_zalloc(blocksize, UMEM_NOFAIL);
4646
4647	while (ztest_random(count) != 0) {
4648		uint64_t randoff = offset + (ztest_random(count) * blocksize);
4649		if (ztest_write(zd, od[0].od_object, randoff, blocksize,
4650		    data) != 0)
4651			break;
4652		while (ztest_random(4) != 0)
4653			ztest_io(zd, od[0].od_object, randoff);
4654	}
4655
4656	umem_free(data, blocksize);
4657}
4658
4659/*
4660 * Verify that zap_{create,destroy,add,remove,update} work as expected.
4661 */
4662#define	ZTEST_ZAP_MIN_INTS	1
4663#define	ZTEST_ZAP_MAX_INTS	4
4664#define	ZTEST_ZAP_MAX_PROPS	1000
4665
4666void
4667ztest_zap(ztest_ds_t *zd, uint64_t id)
4668{
4669	objset_t *os = zd->zd_os;
4670	ztest_od_t od[1];
4671	uint64_t object;
4672	uint64_t txg, last_txg;
4673	uint64_t value[ZTEST_ZAP_MAX_INTS];
4674	uint64_t zl_ints, zl_intsize, prop;
4675	int i, ints;
4676	dmu_tx_t *tx;
4677	char propname[100], txgname[100];
4678	int error;
4679	char *hc[2] = { "s.acl.h", ".s.open.h.hyLZlg" };
4680
4681	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4682
4683	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4684		return;
4685
4686	object = od[0].od_object;
4687
4688	/*
4689	 * Generate a known hash collision, and verify that
4690	 * we can lookup and remove both entries.
4691	 */
4692	tx = dmu_tx_create(os);
4693	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4694	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4695	if (txg == 0)
4696		return;
4697	for (i = 0; i < 2; i++) {
4698		value[i] = i;
4699		VERIFY3U(0, ==, zap_add(os, object, hc[i], sizeof (uint64_t),
4700		    1, &value[i], tx));
4701	}
4702	for (i = 0; i < 2; i++) {
4703		VERIFY3U(EEXIST, ==, zap_add(os, object, hc[i],
4704		    sizeof (uint64_t), 1, &value[i], tx));
4705		VERIFY3U(0, ==,
4706		    zap_length(os, object, hc[i], &zl_intsize, &zl_ints));
4707		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4708		ASSERT3U(zl_ints, ==, 1);
4709	}
4710	for (i = 0; i < 2; i++) {
4711		VERIFY3U(0, ==, zap_remove(os, object, hc[i], tx));
4712	}
4713	dmu_tx_commit(tx);
4714
4715	/*
4716	 * Generate a buch of random entries.
4717	 */
4718	ints = MAX(ZTEST_ZAP_MIN_INTS, object % ZTEST_ZAP_MAX_INTS);
4719
4720	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4721	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4722	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4723	bzero(value, sizeof (value));
4724	last_txg = 0;
4725
4726	/*
4727	 * If these zap entries already exist, validate their contents.
4728	 */
4729	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4730	if (error == 0) {
4731		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4732		ASSERT3U(zl_ints, ==, 1);
4733
4734		VERIFY(zap_lookup(os, object, txgname, zl_intsize,
4735		    zl_ints, &last_txg) == 0);
4736
4737		VERIFY(zap_length(os, object, propname, &zl_intsize,
4738		    &zl_ints) == 0);
4739
4740		ASSERT3U(zl_intsize, ==, sizeof (uint64_t));
4741		ASSERT3U(zl_ints, ==, ints);
4742
4743		VERIFY(zap_lookup(os, object, propname, zl_intsize,
4744		    zl_ints, value) == 0);
4745
4746		for (i = 0; i < ints; i++) {
4747			ASSERT3U(value[i], ==, last_txg + object + i);
4748		}
4749	} else {
4750		ASSERT3U(error, ==, ENOENT);
4751	}
4752
4753	/*
4754	 * Atomically update two entries in our zap object.
4755	 * The first is named txg_%llu, and contains the txg
4756	 * in which the property was last updated.  The second
4757	 * is named prop_%llu, and the nth element of its value
4758	 * should be txg + object + n.
4759	 */
4760	tx = dmu_tx_create(os);
4761	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4762	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4763	if (txg == 0)
4764		return;
4765
4766	if (last_txg > txg)
4767		fatal(0, "zap future leak: old %llu new %llu", last_txg, txg);
4768
4769	for (i = 0; i < ints; i++)
4770		value[i] = txg + object + i;
4771
4772	VERIFY3U(0, ==, zap_update(os, object, txgname, sizeof (uint64_t),
4773	    1, &txg, tx));
4774	VERIFY3U(0, ==, zap_update(os, object, propname, sizeof (uint64_t),
4775	    ints, value, tx));
4776
4777	dmu_tx_commit(tx);
4778
4779	/*
4780	 * Remove a random pair of entries.
4781	 */
4782	prop = ztest_random(ZTEST_ZAP_MAX_PROPS);
4783	(void) sprintf(propname, "prop_%llu", (u_longlong_t)prop);
4784	(void) sprintf(txgname, "txg_%llu", (u_longlong_t)prop);
4785
4786	error = zap_length(os, object, txgname, &zl_intsize, &zl_ints);
4787
4788	if (error == ENOENT)
4789		return;
4790
4791	ASSERT0(error);
4792
4793	tx = dmu_tx_create(os);
4794	dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4795	txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4796	if (txg == 0)
4797		return;
4798	VERIFY3U(0, ==, zap_remove(os, object, txgname, tx));
4799	VERIFY3U(0, ==, zap_remove(os, object, propname, tx));
4800	dmu_tx_commit(tx);
4801}
4802
4803/*
4804 * Testcase to test the upgrading of a microzap to fatzap.
4805 */
4806void
4807ztest_fzap(ztest_ds_t *zd, uint64_t id)
4808{
4809	objset_t *os = zd->zd_os;
4810	ztest_od_t od[1];
4811	uint64_t object, txg;
4812
4813	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_ZAP_OTHER, 0, 0, 0);
4814
4815	if (ztest_object_init(zd, od, sizeof (od), !ztest_random(2)) != 0)
4816		return;
4817
4818	object = od[0].od_object;
4819
4820	/*
4821	 * Add entries to this ZAP and make sure it spills over
4822	 * and gets upgraded to a fatzap. Also, since we are adding
4823	 * 2050 entries we should see ptrtbl growth and leaf-block split.
4824	 */
4825	for (int i = 0; i < 2050; i++) {
4826		char name[ZFS_MAX_DATASET_NAME_LEN];
4827		uint64_t value = i;
4828		dmu_tx_t *tx;
4829		int error;
4830
4831		(void) snprintf(name, sizeof (name), "fzap-%llu-%llu",
4832		    id, value);
4833
4834		tx = dmu_tx_create(os);
4835		dmu_tx_hold_zap(tx, object, B_TRUE, name);
4836		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4837		if (txg == 0)
4838			return;
4839		error = zap_add(os, object, name, sizeof (uint64_t), 1,
4840		    &value, tx);
4841		ASSERT(error == 0 || error == EEXIST);
4842		dmu_tx_commit(tx);
4843	}
4844}
4845
4846/* ARGSUSED */
4847void
4848ztest_zap_parallel(ztest_ds_t *zd, uint64_t id)
4849{
4850	objset_t *os = zd->zd_os;
4851	ztest_od_t od[1];
4852	uint64_t txg, object, count, wsize, wc, zl_wsize, zl_wc;
4853	dmu_tx_t *tx;
4854	int i, namelen, error;
4855	int micro = ztest_random(2);
4856	char name[20], string_value[20];
4857	void *data;
4858
4859	ztest_od_init(&od[0], ID_PARALLEL, FTAG, micro, DMU_OT_ZAP_OTHER,
4860	    0, 0, 0);
4861
4862	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
4863		return;
4864
4865	object = od[0].od_object;
4866
4867	/*
4868	 * Generate a random name of the form 'xxx.....' where each
4869	 * x is a random printable character and the dots are dots.
4870	 * There are 94 such characters, and the name length goes from
4871	 * 6 to 20, so there are 94^3 * 15 = 12,458,760 possible names.
4872	 */
4873	namelen = ztest_random(sizeof (name) - 5) + 5 + 1;
4874
4875	for (i = 0; i < 3; i++)
4876		name[i] = '!' + ztest_random('~' - '!' + 1);
4877	for (; i < namelen - 1; i++)
4878		name[i] = '.';
4879	name[i] = '\0';
4880
4881	if ((namelen & 1) || micro) {
4882		wsize = sizeof (txg);
4883		wc = 1;
4884		data = &txg;
4885	} else {
4886		wsize = 1;
4887		wc = namelen;
4888		data = string_value;
4889	}
4890
4891	count = -1ULL;
4892	VERIFY0(zap_count(os, object, &count));
4893	ASSERT(count != -1ULL);
4894
4895	/*
4896	 * Select an operation: length, lookup, add, update, remove.
4897	 */
4898	i = ztest_random(5);
4899
4900	if (i >= 2) {
4901		tx = dmu_tx_create(os);
4902		dmu_tx_hold_zap(tx, object, B_TRUE, NULL);
4903		txg = ztest_tx_assign(tx, TXG_MIGHTWAIT, FTAG);
4904		if (txg == 0)
4905			return;
4906		bcopy(name, string_value, namelen);
4907	} else {
4908		tx = NULL;
4909		txg = 0;
4910		bzero(string_value, namelen);
4911	}
4912
4913	switch (i) {
4914
4915	case 0:
4916		error = zap_length(os, object, name, &zl_wsize, &zl_wc);
4917		if (error == 0) {
4918			ASSERT3U(wsize, ==, zl_wsize);
4919			ASSERT3U(wc, ==, zl_wc);
4920		} else {
4921			ASSERT3U(error, ==, ENOENT);
4922		}
4923		break;
4924
4925	case 1:
4926		error = zap_lookup(os, object, name, wsize, wc, data);
4927		if (error == 0) {
4928			if (data == string_value &&
4929			    bcmp(name, data, namelen) != 0)
4930				fatal(0, "name '%s' != val '%s' len %d",
4931				    name, data, namelen);
4932		} else {
4933			ASSERT3U(error, ==, ENOENT);
4934		}
4935		break;
4936
4937	case 2:
4938		error = zap_add(os, object, name, wsize, wc, data, tx);
4939		ASSERT(error == 0 || error == EEXIST);
4940		break;
4941
4942	case 3:
4943		VERIFY(zap_update(os, object, name, wsize, wc, data, tx) == 0);
4944		break;
4945
4946	case 4:
4947		error = zap_remove(os, object, name, tx);
4948		ASSERT(error == 0 || error == ENOENT);
4949		break;
4950	}
4951
4952	if (tx != NULL)
4953		dmu_tx_commit(tx);
4954}
4955
4956/*
4957 * Commit callback data.
4958 */
4959typedef struct ztest_cb_data {
4960	list_node_t		zcd_node;
4961	uint64_t		zcd_txg;
4962	int			zcd_expected_err;
4963	boolean_t		zcd_added;
4964	boolean_t		zcd_called;
4965	spa_t			*zcd_spa;
4966} ztest_cb_data_t;
4967
4968/* This is the actual commit callback function */
4969static void
4970ztest_commit_callback(void *arg, int error)
4971{
4972	ztest_cb_data_t *data = arg;
4973	uint64_t synced_txg;
4974
4975	VERIFY(data != NULL);
4976	VERIFY3S(data->zcd_expected_err, ==, error);
4977	VERIFY(!data->zcd_called);
4978
4979	synced_txg = spa_last_synced_txg(data->zcd_spa);
4980	if (data->zcd_txg > synced_txg)
4981		fatal(0, "commit callback of txg %" PRIu64 " called prematurely"
4982		    ", last synced txg = %" PRIu64 "\n", data->zcd_txg,
4983		    synced_txg);
4984
4985	data->zcd_called = B_TRUE;
4986
4987	if (error == ECANCELED) {
4988		ASSERT0(data->zcd_txg);
4989		ASSERT(!data->zcd_added);
4990
4991		/*
4992		 * The private callback data should be destroyed here, but
4993		 * since we are going to check the zcd_called field after
4994		 * dmu_tx_abort(), we will destroy it there.
4995		 */
4996		return;
4997	}
4998
4999	/* Was this callback added to the global callback list? */
5000	if (!data->zcd_added)
5001		goto out;
5002
5003	ASSERT3U(data->zcd_txg, !=, 0);
5004
5005	/* Remove our callback from the list */
5006	mutex_enter(&zcl.zcl_callbacks_lock);
5007	list_remove(&zcl.zcl_callbacks, data);
5008	mutex_exit(&zcl.zcl_callbacks_lock);
5009
5010out:
5011	umem_free(data, sizeof (ztest_cb_data_t));
5012}
5013
5014/* Allocate and initialize callback data structure */
5015static ztest_cb_data_t *
5016ztest_create_cb_data(objset_t *os, uint64_t txg)
5017{
5018	ztest_cb_data_t *cb_data;
5019
5020	cb_data = umem_zalloc(sizeof (ztest_cb_data_t), UMEM_NOFAIL);
5021
5022	cb_data->zcd_txg = txg;
5023	cb_data->zcd_spa = dmu_objset_spa(os);
5024
5025	return (cb_data);
5026}
5027
5028/*
5029 * If a number of txgs equal to this threshold have been created after a commit
5030 * callback has been registered but not called, then we assume there is an
5031 * implementation bug.
5032 */
5033#define	ZTEST_COMMIT_CALLBACK_THRESH	(TXG_CONCURRENT_STATES + 2)
5034
5035/*
5036 * Commit callback test.
5037 */
5038void
5039ztest_dmu_commit_callbacks(ztest_ds_t *zd, uint64_t id)
5040{
5041	objset_t *os = zd->zd_os;
5042	ztest_od_t od[1];
5043	dmu_tx_t *tx;
5044	ztest_cb_data_t *cb_data[3], *tmp_cb;
5045	uint64_t old_txg, txg;
5046	int i, error;
5047
5048	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, 0, 0, 0);
5049
5050	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5051		return;
5052
5053	tx = dmu_tx_create(os);
5054
5055	cb_data[0] = ztest_create_cb_data(os, 0);
5056	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[0]);
5057
5058	dmu_tx_hold_write(tx, od[0].od_object, 0, sizeof (uint64_t));
5059
5060	/* Every once in a while, abort the transaction on purpose */
5061	if (ztest_random(100) == 0)
5062		error = -1;
5063
5064	if (!error)
5065		error = dmu_tx_assign(tx, TXG_NOWAIT);
5066
5067	txg = error ? 0 : dmu_tx_get_txg(tx);
5068
5069	cb_data[0]->zcd_txg = txg;
5070	cb_data[1] = ztest_create_cb_data(os, txg);
5071	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[1]);
5072
5073	if (error) {
5074		/*
5075		 * It's not a strict requirement to call the registered
5076		 * callbacks from inside dmu_tx_abort(), but that's what
5077		 * it's supposed to happen in the current implementation
5078		 * so we will check for that.
5079		 */
5080		for (i = 0; i < 2; i++) {
5081			cb_data[i]->zcd_expected_err = ECANCELED;
5082			VERIFY(!cb_data[i]->zcd_called);
5083		}
5084
5085		dmu_tx_abort(tx);
5086
5087		for (i = 0; i < 2; i++) {
5088			VERIFY(cb_data[i]->zcd_called);
5089			umem_free(cb_data[i], sizeof (ztest_cb_data_t));
5090		}
5091
5092		return;
5093	}
5094
5095	cb_data[2] = ztest_create_cb_data(os, txg);
5096	dmu_tx_callback_register(tx, ztest_commit_callback, cb_data[2]);
5097
5098	/*
5099	 * Read existing data to make sure there isn't a future leak.
5100	 */
5101	VERIFY(0 == dmu_read(os, od[0].od_object, 0, sizeof (uint64_t),
5102	    &old_txg, DMU_READ_PREFETCH));
5103
5104	if (old_txg > txg)
5105		fatal(0, "future leak: got %" PRIu64 ", open txg is %" PRIu64,
5106		    old_txg, txg);
5107
5108	dmu_write(os, od[0].od_object, 0, sizeof (uint64_t), &txg, tx);
5109
5110	mutex_enter(&zcl.zcl_callbacks_lock);
5111
5112	/*
5113	 * Since commit callbacks don't have any ordering requirement and since
5114	 * it is theoretically possible for a commit callback to be called
5115	 * after an arbitrary amount of time has elapsed since its txg has been
5116	 * synced, it is difficult to reliably determine whether a commit
5117	 * callback hasn't been called due to high load or due to a flawed
5118	 * implementation.
5119	 *
5120	 * In practice, we will assume that if after a certain number of txgs a
5121	 * commit callback hasn't been called, then most likely there's an
5122	 * implementation bug..
5123	 */
5124	tmp_cb = list_head(&zcl.zcl_callbacks);
5125	if (tmp_cb != NULL &&
5126	    (txg - ZTEST_COMMIT_CALLBACK_THRESH) > tmp_cb->zcd_txg) {
5127		fatal(0, "Commit callback threshold exceeded, oldest txg: %"
5128		    PRIu64 ", open txg: %" PRIu64 "\n", tmp_cb->zcd_txg, txg);
5129	}
5130
5131	/*
5132	 * Let's find the place to insert our callbacks.
5133	 *
5134	 * Even though the list is ordered by txg, it is possible for the
5135	 * insertion point to not be the end because our txg may already be
5136	 * quiescing at this point and other callbacks in the open txg
5137	 * (from other objsets) may have sneaked in.
5138	 */
5139	tmp_cb = list_tail(&zcl.zcl_callbacks);
5140	while (tmp_cb != NULL && tmp_cb->zcd_txg > txg)
5141		tmp_cb = list_prev(&zcl.zcl_callbacks, tmp_cb);
5142
5143	/* Add the 3 callbacks to the list */
5144	for (i = 0; i < 3; i++) {
5145		if (tmp_cb == NULL)
5146			list_insert_head(&zcl.zcl_callbacks, cb_data[i]);
5147		else
5148			list_insert_after(&zcl.zcl_callbacks, tmp_cb,
5149			    cb_data[i]);
5150
5151		cb_data[i]->zcd_added = B_TRUE;
5152		VERIFY(!cb_data[i]->zcd_called);
5153
5154		tmp_cb = cb_data[i];
5155	}
5156
5157	mutex_exit(&zcl.zcl_callbacks_lock);
5158
5159	dmu_tx_commit(tx);
5160}
5161
5162/*
5163 * Visit each object in the dataset. Verify that its properties
5164 * are consistent what was stored in the block tag when it was created,
5165 * and that its unused bonus buffer space has not been overwritten.
5166 */
5167void
5168ztest_verify_dnode_bt(ztest_ds_t *zd, uint64_t id)
5169{
5170	objset_t *os = zd->zd_os;
5171	uint64_t obj;
5172	int err = 0;
5173
5174	for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
5175		ztest_block_tag_t *bt = NULL;
5176		dmu_object_info_t doi;
5177		dmu_buf_t *db;
5178
5179		if (dmu_bonus_hold(os, obj, FTAG, &db) != 0)
5180			continue;
5181
5182		dmu_object_info_from_db(db, &doi);
5183		if (doi.doi_bonus_size >= sizeof (*bt))
5184			bt = ztest_bt_bonus(db);
5185
5186		if (bt && bt->bt_magic == BT_MAGIC) {
5187			ztest_bt_verify(bt, os, obj, doi.doi_dnodesize,
5188			    bt->bt_offset, bt->bt_gen, bt->bt_txg,
5189			    bt->bt_crtxg);
5190			ztest_verify_unused_bonus(db, bt, obj, os, bt->bt_gen);
5191		}
5192
5193		dmu_buf_rele(db, FTAG);
5194	}
5195}
5196
5197/* ARGSUSED */
5198void
5199ztest_dsl_prop_get_set(ztest_ds_t *zd, uint64_t id)
5200{
5201	zfs_prop_t proplist[] = {
5202		ZFS_PROP_CHECKSUM,
5203		ZFS_PROP_COMPRESSION,
5204		ZFS_PROP_COPIES,
5205		ZFS_PROP_DEDUP
5206	};
5207
5208	rw_enter(&ztest_name_lock, RW_READER);
5209
5210	for (int p = 0; p < sizeof (proplist) / sizeof (proplist[0]); p++)
5211		(void) ztest_dsl_prop_set_uint64(zd->zd_name, proplist[p],
5212		    ztest_random_dsl_prop(proplist[p]), (int)ztest_random(2));
5213
5214	rw_exit(&ztest_name_lock);
5215}
5216
5217/* ARGSUSED */
5218void
5219ztest_remap_blocks(ztest_ds_t *zd, uint64_t id)
5220{
5221	rw_enter(&ztest_name_lock, RW_READER);
5222
5223	int error = dmu_objset_remap_indirects(zd->zd_name);
5224	if (error == ENOSPC)
5225		error = 0;
5226	ASSERT0(error);
5227
5228	rw_exit(&ztest_name_lock);
5229}
5230
5231/* ARGSUSED */
5232void
5233ztest_spa_prop_get_set(ztest_ds_t *zd, uint64_t id)
5234{
5235	nvlist_t *props = NULL;
5236
5237	rw_enter(&ztest_name_lock, RW_READER);
5238
5239	(void) ztest_spa_prop_set_uint64(ZPOOL_PROP_DEDUPDITTO,
5240	    ZIO_DEDUPDITTO_MIN + ztest_random(ZIO_DEDUPDITTO_MIN));
5241
5242	VERIFY0(spa_prop_get(ztest_spa, &props));
5243
5244	if (ztest_opts.zo_verbose >= 6)
5245		dump_nvlist(props, 4);
5246
5247	nvlist_free(props);
5248
5249	rw_exit(&ztest_name_lock);
5250}
5251
5252static int
5253user_release_one(const char *snapname, const char *holdname)
5254{
5255	nvlist_t *snaps, *holds;
5256	int error;
5257
5258	snaps = fnvlist_alloc();
5259	holds = fnvlist_alloc();
5260	fnvlist_add_boolean(holds, holdname);
5261	fnvlist_add_nvlist(snaps, snapname, holds);
5262	fnvlist_free(holds);
5263	error = dsl_dataset_user_release(snaps, NULL);
5264	fnvlist_free(snaps);
5265	return (error);
5266}
5267
5268/*
5269 * Test snapshot hold/release and deferred destroy.
5270 */
5271void
5272ztest_dmu_snapshot_hold(ztest_ds_t *zd, uint64_t id)
5273{
5274	int error;
5275	objset_t *os = zd->zd_os;
5276	objset_t *origin;
5277	char snapname[100];
5278	char fullname[100];
5279	char clonename[100];
5280	char tag[100];
5281	char osname[ZFS_MAX_DATASET_NAME_LEN];
5282	nvlist_t *holds;
5283
5284	rw_enter(&ztest_name_lock, RW_READER);
5285
5286	dmu_objset_name(os, osname);
5287
5288	(void) snprintf(snapname, sizeof (snapname), "sh1_%llu", id);
5289	(void) snprintf(fullname, sizeof (fullname), "%s@%s", osname, snapname);
5290	(void) snprintf(clonename, sizeof (clonename),
5291	    "%s/ch1_%llu", osname, id);
5292	(void) snprintf(tag, sizeof (tag), "tag_%llu", id);
5293
5294	/*
5295	 * Clean up from any previous run.
5296	 */
5297	error = dsl_destroy_head(clonename);
5298	if (error != ENOENT)
5299		ASSERT0(error);
5300	error = user_release_one(fullname, tag);
5301	if (error != ESRCH && error != ENOENT)
5302		ASSERT0(error);
5303	error = dsl_destroy_snapshot(fullname, B_FALSE);
5304	if (error != ENOENT)
5305		ASSERT0(error);
5306
5307	/*
5308	 * Create snapshot, clone it, mark snap for deferred destroy,
5309	 * destroy clone, verify snap was also destroyed.
5310	 */
5311	error = dmu_objset_snapshot_one(osname, snapname);
5312	if (error) {
5313		if (error == ENOSPC) {
5314			ztest_record_enospc("dmu_objset_snapshot");
5315			goto out;
5316		}
5317		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5318	}
5319
5320	error = dmu_objset_clone(clonename, fullname);
5321	if (error) {
5322		if (error == ENOSPC) {
5323			ztest_record_enospc("dmu_objset_clone");
5324			goto out;
5325		}
5326		fatal(0, "dmu_objset_clone(%s) = %d", clonename, error);
5327	}
5328
5329	error = dsl_destroy_snapshot(fullname, B_TRUE);
5330	if (error) {
5331		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5332		    fullname, error);
5333	}
5334
5335	error = dsl_destroy_head(clonename);
5336	if (error)
5337		fatal(0, "dsl_destroy_head(%s) = %d", clonename, error);
5338
5339	error = dmu_objset_hold(fullname, FTAG, &origin);
5340	if (error != ENOENT)
5341		fatal(0, "dmu_objset_hold(%s) = %d", fullname, error);
5342
5343	/*
5344	 * Create snapshot, add temporary hold, verify that we can't
5345	 * destroy a held snapshot, mark for deferred destroy,
5346	 * release hold, verify snapshot was destroyed.
5347	 */
5348	error = dmu_objset_snapshot_one(osname, snapname);
5349	if (error) {
5350		if (error == ENOSPC) {
5351			ztest_record_enospc("dmu_objset_snapshot");
5352			goto out;
5353		}
5354		fatal(0, "dmu_objset_snapshot(%s) = %d", fullname, error);
5355	}
5356
5357	holds = fnvlist_alloc();
5358	fnvlist_add_string(holds, fullname, tag);
5359	error = dsl_dataset_user_hold(holds, 0, NULL);
5360	fnvlist_free(holds);
5361
5362	if (error == ENOSPC) {
5363		ztest_record_enospc("dsl_dataset_user_hold");
5364		goto out;
5365	} else if (error) {
5366		fatal(0, "dsl_dataset_user_hold(%s, %s) = %u",
5367		    fullname, tag, error);
5368	}
5369
5370	error = dsl_destroy_snapshot(fullname, B_FALSE);
5371	if (error != EBUSY) {
5372		fatal(0, "dsl_destroy_snapshot(%s, B_FALSE) = %d",
5373		    fullname, error);
5374	}
5375
5376	error = dsl_destroy_snapshot(fullname, B_TRUE);
5377	if (error) {
5378		fatal(0, "dsl_destroy_snapshot(%s, B_TRUE) = %d",
5379		    fullname, error);
5380	}
5381
5382	error = user_release_one(fullname, tag);
5383	if (error)
5384		fatal(0, "user_release_one(%s, %s) = %d", fullname, tag, error);
5385
5386	VERIFY3U(dmu_objset_hold(fullname, FTAG, &origin), ==, ENOENT);
5387
5388out:
5389	rw_exit(&ztest_name_lock);
5390}
5391
5392/*
5393 * Inject random faults into the on-disk data.
5394 */
5395/* ARGSUSED */
5396void
5397ztest_fault_inject(ztest_ds_t *zd, uint64_t id)
5398{
5399	ztest_shared_t *zs = ztest_shared;
5400	spa_t *spa = ztest_spa;
5401	int fd;
5402	uint64_t offset;
5403	uint64_t leaves;
5404	uint64_t bad = 0x1990c0ffeedecadeULL;
5405	uint64_t top, leaf;
5406	char path0[MAXPATHLEN];
5407	char pathrand[MAXPATHLEN];
5408	size_t fsize;
5409	int bshift = SPA_MAXBLOCKSHIFT + 2;
5410	int iters = 1000;
5411	int maxfaults;
5412	int mirror_save;
5413	vdev_t *vd0 = NULL;
5414	uint64_t guid0 = 0;
5415	boolean_t islog = B_FALSE;
5416
5417	mutex_enter(&ztest_vdev_lock);
5418
5419	/*
5420	 * Device removal is in progress, fault injection must be disabled
5421	 * until it completes and the pool is scrubbed.  The fault injection
5422	 * strategy for damaging blocks does not take in to account evacuated
5423	 * blocks which may have already been damaged.
5424	 */
5425	if (ztest_device_removal_active) {
5426		mutex_exit(&ztest_vdev_lock);
5427		return;
5428	}
5429
5430	maxfaults = MAXFAULTS();
5431	leaves = MAX(zs->zs_mirrors, 1) * ztest_opts.zo_raidz;
5432	mirror_save = zs->zs_mirrors;
5433	mutex_exit(&ztest_vdev_lock);
5434
5435	ASSERT(leaves >= 1);
5436
5437	/*
5438	 * Grab the name lock as reader. There are some operations
5439	 * which don't like to have their vdevs changed while
5440	 * they are in progress (i.e. spa_change_guid). Those
5441	 * operations will have grabbed the name lock as writer.
5442	 */
5443	rw_enter(&ztest_name_lock, RW_READER);
5444
5445	/*
5446	 * We need SCL_STATE here because we're going to look at vd0->vdev_tsd.
5447	 */
5448	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5449
5450	if (ztest_random(2) == 0) {
5451		/*
5452		 * Inject errors on a normal data device or slog device.
5453		 */
5454		top = ztest_random_vdev_top(spa, B_TRUE);
5455		leaf = ztest_random(leaves) + zs->zs_splits;
5456
5457		/*
5458		 * Generate paths to the first leaf in this top-level vdev,
5459		 * and to the random leaf we selected.  We'll induce transient
5460		 * write failures and random online/offline activity on leaf 0,
5461		 * and we'll write random garbage to the randomly chosen leaf.
5462		 */
5463		(void) snprintf(path0, sizeof (path0), ztest_dev_template,
5464		    ztest_opts.zo_dir, ztest_opts.zo_pool,
5465		    top * leaves + zs->zs_splits);
5466		(void) snprintf(pathrand, sizeof (pathrand), ztest_dev_template,
5467		    ztest_opts.zo_dir, ztest_opts.zo_pool,
5468		    top * leaves + leaf);
5469
5470		vd0 = vdev_lookup_by_path(spa->spa_root_vdev, path0);
5471		if (vd0 != NULL && vd0->vdev_top->vdev_islog)
5472			islog = B_TRUE;
5473
5474		/*
5475		 * If the top-level vdev needs to be resilvered
5476		 * then we only allow faults on the device that is
5477		 * resilvering.
5478		 */
5479		if (vd0 != NULL && maxfaults != 1 &&
5480		    (!vdev_resilver_needed(vd0->vdev_top, NULL, NULL) ||
5481		    vd0->vdev_resilver_txg != 0)) {
5482			/*
5483			 * Make vd0 explicitly claim to be unreadable,
5484			 * or unwriteable, or reach behind its back
5485			 * and close the underlying fd.  We can do this if
5486			 * maxfaults == 0 because we'll fail and reexecute,
5487			 * and we can do it if maxfaults >= 2 because we'll
5488			 * have enough redundancy.  If maxfaults == 1, the
5489			 * combination of this with injection of random data
5490			 * corruption below exceeds the pool's fault tolerance.
5491			 */
5492			vdev_file_t *vf = vd0->vdev_tsd;
5493
5494			zfs_dbgmsg("injecting fault to vdev %llu; maxfaults=%d",
5495			    (long long)vd0->vdev_id, (int)maxfaults);
5496
5497			if (vf != NULL && ztest_random(3) == 0) {
5498				(void) close(vf->vf_vnode->v_fd);
5499				vf->vf_vnode->v_fd = -1;
5500			} else if (ztest_random(2) == 0) {
5501				vd0->vdev_cant_read = B_TRUE;
5502			} else {
5503				vd0->vdev_cant_write = B_TRUE;
5504			}
5505			guid0 = vd0->vdev_guid;
5506		}
5507	} else {
5508		/*
5509		 * Inject errors on an l2cache device.
5510		 */
5511		spa_aux_vdev_t *sav = &spa->spa_l2cache;
5512
5513		if (sav->sav_count == 0) {
5514			spa_config_exit(spa, SCL_STATE, FTAG);
5515			rw_exit(&ztest_name_lock);
5516			return;
5517		}
5518		vd0 = sav->sav_vdevs[ztest_random(sav->sav_count)];
5519		guid0 = vd0->vdev_guid;
5520		(void) strcpy(path0, vd0->vdev_path);
5521		(void) strcpy(pathrand, vd0->vdev_path);
5522
5523		leaf = 0;
5524		leaves = 1;
5525		maxfaults = INT_MAX;	/* no limit on cache devices */
5526	}
5527
5528	spa_config_exit(spa, SCL_STATE, FTAG);
5529	rw_exit(&ztest_name_lock);
5530
5531	/*
5532	 * If we can tolerate two or more faults, or we're dealing
5533	 * with a slog, randomly online/offline vd0.
5534	 */
5535	if ((maxfaults >= 2 || islog) && guid0 != 0) {
5536		if (ztest_random(10) < 6) {
5537			int flags = (ztest_random(2) == 0 ?
5538			    ZFS_OFFLINE_TEMPORARY : 0);
5539
5540			/*
5541			 * We have to grab the zs_name_lock as writer to
5542			 * prevent a race between offlining a slog and
5543			 * destroying a dataset. Offlining the slog will
5544			 * grab a reference on the dataset which may cause
5545			 * dmu_objset_destroy() to fail with EBUSY thus
5546			 * leaving the dataset in an inconsistent state.
5547			 */
5548			if (islog)
5549				rw_enter(&ztest_name_lock, RW_WRITER);
5550
5551			VERIFY(vdev_offline(spa, guid0, flags) != EBUSY);
5552
5553			if (islog)
5554				rw_exit(&ztest_name_lock);
5555		} else {
5556			/*
5557			 * Ideally we would like to be able to randomly
5558			 * call vdev_[on|off]line without holding locks
5559			 * to force unpredictable failures but the side
5560			 * effects of vdev_[on|off]line prevent us from
5561			 * doing so. We grab the ztest_vdev_lock here to
5562			 * prevent a race between injection testing and
5563			 * aux_vdev removal.
5564			 */
5565			mutex_enter(&ztest_vdev_lock);
5566			(void) vdev_online(spa, guid0, 0, NULL);
5567			mutex_exit(&ztest_vdev_lock);
5568		}
5569	}
5570
5571	if (maxfaults == 0)
5572		return;
5573
5574	/*
5575	 * We have at least single-fault tolerance, so inject data corruption.
5576	 */
5577	fd = open(pathrand, O_RDWR);
5578
5579	if (fd == -1) /* we hit a gap in the device namespace */
5580		return;
5581
5582	fsize = lseek(fd, 0, SEEK_END);
5583
5584	while (--iters != 0) {
5585		/*
5586		 * The offset must be chosen carefully to ensure that
5587		 * we do not inject a given logical block with errors
5588		 * on two different leaf devices, because ZFS can not
5589		 * tolerate that (if maxfaults==1).
5590		 *
5591		 * We divide each leaf into chunks of size
5592		 * (# leaves * SPA_MAXBLOCKSIZE * 4).  Within each chunk
5593		 * there is a series of ranges to which we can inject errors.
5594		 * Each range can accept errors on only a single leaf vdev.
5595		 * The error injection ranges are separated by ranges
5596		 * which we will not inject errors on any device (DMZs).
5597		 * Each DMZ must be large enough such that a single block
5598		 * can not straddle it, so that a single block can not be
5599		 * a target in two different injection ranges (on different
5600		 * leaf vdevs).
5601		 *
5602		 * For example, with 3 leaves, each chunk looks like:
5603		 *    0 to  32M: injection range for leaf 0
5604		 *  32M to  64M: DMZ - no injection allowed
5605		 *  64M to  96M: injection range for leaf 1
5606		 *  96M to 128M: DMZ - no injection allowed
5607		 * 128M to 160M: injection range for leaf 2
5608		 * 160M to 192M: DMZ - no injection allowed
5609		 */
5610		offset = ztest_random(fsize / (leaves << bshift)) *
5611		    (leaves << bshift) + (leaf << bshift) +
5612		    (ztest_random(1ULL << (bshift - 1)) & -8ULL);
5613
5614		/*
5615		 * Only allow damage to the labels at one end of the vdev.
5616		 *
5617		 * If all labels are damaged, the device will be totally
5618		 * inaccessible, which will result in loss of data,
5619		 * because we also damage (parts of) the other side of
5620		 * the mirror/raidz.
5621		 *
5622		 * Additionally, we will always have both an even and an
5623		 * odd label, so that we can handle crashes in the
5624		 * middle of vdev_config_sync().
5625		 */
5626		if ((leaf & 1) == 0 && offset < VDEV_LABEL_START_SIZE)
5627			continue;
5628
5629		/*
5630		 * The two end labels are stored at the "end" of the disk, but
5631		 * the end of the disk (vdev_psize) is aligned to
5632		 * sizeof (vdev_label_t).
5633		 */
5634		uint64_t psize = P2ALIGN(fsize, sizeof (vdev_label_t));
5635		if ((leaf & 1) == 1 &&
5636		    offset + sizeof (bad) > psize - VDEV_LABEL_END_SIZE)
5637			continue;
5638
5639		mutex_enter(&ztest_vdev_lock);
5640		if (mirror_save != zs->zs_mirrors) {
5641			mutex_exit(&ztest_vdev_lock);
5642			(void) close(fd);
5643			return;
5644		}
5645
5646		if (pwrite(fd, &bad, sizeof (bad), offset) != sizeof (bad))
5647			fatal(1, "can't inject bad word at 0x%llx in %s",
5648			    offset, pathrand);
5649
5650		mutex_exit(&ztest_vdev_lock);
5651
5652		if (ztest_opts.zo_verbose >= 7)
5653			(void) printf("injected bad word into %s,"
5654			    " offset 0x%llx\n", pathrand, (u_longlong_t)offset);
5655	}
5656
5657	(void) close(fd);
5658}
5659
5660/*
5661 * Verify that DDT repair works as expected.
5662 */
5663void
5664ztest_ddt_repair(ztest_ds_t *zd, uint64_t id)
5665{
5666	ztest_shared_t *zs = ztest_shared;
5667	spa_t *spa = ztest_spa;
5668	objset_t *os = zd->zd_os;
5669	ztest_od_t od[1];
5670	uint64_t object, blocksize, txg, pattern, psize;
5671	enum zio_checksum checksum = spa_dedup_checksum(spa);
5672	dmu_buf_t *db;
5673	dmu_tx_t *tx;
5674	abd_t *abd;
5675	blkptr_t blk;
5676	int copies = 2 * ZIO_DEDUPDITTO_MIN;
5677
5678	blocksize = ztest_random_blocksize();
5679	blocksize = MIN(blocksize, 2048);	/* because we write so many */
5680
5681	ztest_od_init(&od[0], id, FTAG, 0, DMU_OT_UINT64_OTHER, blocksize,
5682	    0, 0);
5683
5684	if (ztest_object_init(zd, od, sizeof (od), B_FALSE) != 0)
5685		return;
5686
5687	/*
5688	 * Take the name lock as writer to prevent anyone else from changing
5689	 * the pool and dataset properies we need to maintain during this test.
5690	 */
5691	rw_enter(&ztest_name_lock, RW_WRITER);
5692
5693	if (ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_DEDUP, checksum,
5694	    B_FALSE) != 0 ||
5695	    ztest_dsl_prop_set_uint64(zd->zd_name, ZFS_PROP_COPIES, 1,
5696	    B_FALSE) != 0) {
5697		rw_exit(&ztest_name_lock);
5698		return;
5699	}
5700
5701	dmu_objset_stats_t dds;
5702	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
5703	dmu_objset_fast_stat(os, &dds);
5704	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
5705
5706	object = od[0].od_object;
5707	blocksize = od[0].od_blocksize;
5708	pattern = zs->zs_guid ^ dds.dds_guid;
5709
5710	ASSERT(object != 0);
5711
5712	tx = dmu_tx_create(os);
5713	dmu_tx_hold_write(tx, object, 0, copies * blocksize);
5714	txg = ztest_tx_assign(tx, TXG_WAIT, FTAG);
5715	if (txg == 0) {
5716		rw_exit(&ztest_name_lock);
5717		return;
5718	}
5719
5720	/*
5721	 * Write all the copies of our block.
5722	 */
5723	for (int i = 0; i < copies; i++) {
5724		uint64_t offset = i * blocksize;
5725		int error = dmu_buf_hold(os, object, offset, FTAG, &db,
5726		    DMU_READ_NO_PREFETCH);
5727		if (error != 0) {
5728			fatal(B_FALSE, "dmu_buf_hold(%p, %llu, %llu) = %u",
5729			    os, (long long)object, (long long) offset, error);
5730		}
5731		ASSERT(db->db_offset == offset);
5732		ASSERT(db->db_size == blocksize);
5733		ASSERT(ztest_pattern_match(db->db_data, db->db_size, pattern) ||
5734		    ztest_pattern_match(db->db_data, db->db_size, 0ULL));
5735		dmu_buf_will_fill(db, tx);
5736		ztest_pattern_set(db->db_data, db->db_size, pattern);
5737		dmu_buf_rele(db, FTAG);
5738	}
5739
5740	dmu_tx_commit(tx);
5741	txg_wait_synced(spa_get_dsl(spa), txg);
5742
5743	/*
5744	 * Find out what block we got.
5745	 */
5746	VERIFY0(dmu_buf_hold(os, object, 0, FTAG, &db,
5747	    DMU_READ_NO_PREFETCH));
5748	blk = *((dmu_buf_impl_t *)db)->db_blkptr;
5749	dmu_buf_rele(db, FTAG);
5750
5751	/*
5752	 * Damage the block.  Dedup-ditto will save us when we read it later.
5753	 */
5754	psize = BP_GET_PSIZE(&blk);
5755	abd = abd_alloc_linear(psize, B_TRUE);
5756	ztest_pattern_set(abd_to_buf(abd), psize, ~pattern);
5757
5758	(void) zio_wait(zio_rewrite(NULL, spa, 0, &blk,
5759	    abd, psize, NULL, NULL, ZIO_PRIORITY_SYNC_WRITE,
5760	    ZIO_FLAG_CANFAIL | ZIO_FLAG_INDUCE_DAMAGE, NULL));
5761
5762	abd_free(abd);
5763
5764	rw_exit(&ztest_name_lock);
5765}
5766
5767/*
5768 * Scrub the pool.
5769 */
5770/* ARGSUSED */
5771void
5772ztest_scrub(ztest_ds_t *zd, uint64_t id)
5773{
5774	spa_t *spa = ztest_spa;
5775
5776	/*
5777	 * Scrub in progress by device removal.
5778	 */
5779	if (ztest_device_removal_active)
5780		return;
5781
5782	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5783	(void) poll(NULL, 0, 100); /* wait a moment, then force a restart */
5784	(void) spa_scan(spa, POOL_SCAN_SCRUB);
5785}
5786
5787/*
5788 * Change the guid for the pool.
5789 */
5790/* ARGSUSED */
5791void
5792ztest_reguid(ztest_ds_t *zd, uint64_t id)
5793{
5794	spa_t *spa = ztest_spa;
5795	uint64_t orig, load;
5796	int error;
5797
5798	if (ztest_opts.zo_mmp_test)
5799		return;
5800
5801	orig = spa_guid(spa);
5802	load = spa_load_guid(spa);
5803
5804	rw_enter(&ztest_name_lock, RW_WRITER);
5805	error = spa_change_guid(spa);
5806	rw_exit(&ztest_name_lock);
5807
5808	if (error != 0)
5809		return;
5810
5811	if (ztest_opts.zo_verbose >= 4) {
5812		(void) printf("Changed guid old %llu -> %llu\n",
5813		    (u_longlong_t)orig, (u_longlong_t)spa_guid(spa));
5814	}
5815
5816	VERIFY3U(orig, !=, spa_guid(spa));
5817	VERIFY3U(load, ==, spa_load_guid(spa));
5818}
5819
5820static vdev_t *
5821ztest_random_concrete_vdev_leaf(vdev_t *vd)
5822{
5823	if (vd == NULL)
5824		return (NULL);
5825
5826	if (vd->vdev_children == 0)
5827		return (vd);
5828
5829	vdev_t *eligible[vd->vdev_children];
5830	int eligible_idx = 0, i;
5831	for (i = 0; i < vd->vdev_children; i++) {
5832		vdev_t *cvd = vd->vdev_child[i];
5833		if (cvd->vdev_top->vdev_removing)
5834			continue;
5835		if (cvd->vdev_children > 0 ||
5836		    (vdev_is_concrete(cvd) && !cvd->vdev_detached)) {
5837			eligible[eligible_idx++] = cvd;
5838		}
5839	}
5840	VERIFY(eligible_idx > 0);
5841
5842	uint64_t child_no = ztest_random(eligible_idx);
5843	return (ztest_random_concrete_vdev_leaf(eligible[child_no]));
5844}
5845
5846/* ARGSUSED */
5847void
5848ztest_initialize(ztest_ds_t *zd, uint64_t id)
5849{
5850	spa_t *spa = ztest_spa;
5851	int error = 0;
5852
5853	mutex_enter(&ztest_vdev_lock);
5854
5855	spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
5856
5857	/* Random leaf vdev */
5858	vdev_t *rand_vd = ztest_random_concrete_vdev_leaf(spa->spa_root_vdev);
5859	if (rand_vd == NULL) {
5860		spa_config_exit(spa, SCL_VDEV, FTAG);
5861		mutex_exit(&ztest_vdev_lock);
5862		return;
5863	}
5864
5865	/*
5866	 * The random vdev we've selected may change as soon as we
5867	 * drop the spa_config_lock. We create local copies of things
5868	 * we're interested in.
5869	 */
5870	uint64_t guid = rand_vd->vdev_guid;
5871	char *path = strdup(rand_vd->vdev_path);
5872	boolean_t active = rand_vd->vdev_initialize_thread != NULL;
5873
5874	zfs_dbgmsg("vd %p, guid %llu", rand_vd, guid);
5875	spa_config_exit(spa, SCL_VDEV, FTAG);
5876
5877	uint64_t cmd = ztest_random(POOL_INITIALIZE_FUNCS);
5878	error = spa_vdev_initialize(spa, guid, cmd);
5879	switch (cmd) {
5880	case POOL_INITIALIZE_CANCEL:
5881		if (ztest_opts.zo_verbose >= 4) {
5882			(void) printf("Cancel initialize %s", path);
5883			if (!active)
5884				(void) printf(" failed (no initialize active)");
5885			(void) printf("\n");
5886		}
5887		break;
5888	case POOL_INITIALIZE_DO:
5889		if (ztest_opts.zo_verbose >= 4) {
5890			(void) printf("Start initialize %s", path);
5891			if (active && error == 0)
5892				(void) printf(" failed (already active)");
5893			else if (error != 0)
5894				(void) printf(" failed (error %d)", error);
5895			(void) printf("\n");
5896		}
5897		break;
5898	case POOL_INITIALIZE_SUSPEND:
5899		if (ztest_opts.zo_verbose >= 4) {
5900			(void) printf("Suspend initialize %s", path);
5901			if (!active)
5902				(void) printf(" failed (no initialize active)");
5903			(void) printf("\n");
5904		}
5905		break;
5906	}
5907	free(path);
5908	mutex_exit(&ztest_vdev_lock);
5909}
5910
5911/*
5912 * Verify pool integrity by running zdb.
5913 */
5914static void
5915ztest_run_zdb(char *pool)
5916{
5917	int status;
5918	char zdb[MAXPATHLEN + MAXNAMELEN + 20];
5919	char zbuf[1024];
5920	char *bin;
5921	char *ztest;
5922	char *isa;
5923	int isalen;
5924	FILE *fp;
5925
5926	strlcpy(zdb, "/usr/bin/ztest", sizeof(zdb));
5927
5928	/* zdb lives in /usr/sbin, while ztest lives in /usr/bin */
5929	bin = strstr(zdb, "/usr/bin/");
5930	ztest = strstr(bin, "/ztest");
5931	isa = bin + 8;
5932	isalen = ztest - isa;
5933	isa = strdup(isa);
5934	/* LINTED */
5935	(void) sprintf(bin,
5936	    "/usr/sbin%.*s/zdb -bcc%s%s -G -d -U %s "
5937	    "-o zfs_reconstruct_indirect_combinations_max=65536 %s",
5938	    isalen,
5939	    isa,
5940	    ztest_opts.zo_verbose >= 3 ? "s" : "",
5941	    ztest_opts.zo_verbose >= 4 ? "v" : "",
5942	    spa_config_path,
5943	    pool);
5944	free(isa);
5945
5946	if (ztest_opts.zo_verbose >= 5)
5947		(void) printf("Executing %s\n", strstr(zdb, "zdb "));
5948
5949	fp = popen(zdb, "r");
5950	assert(fp != NULL);
5951
5952	while (fgets(zbuf, sizeof (zbuf), fp) != NULL)
5953		if (ztest_opts.zo_verbose >= 3)
5954			(void) printf("%s", zbuf);
5955
5956	status = pclose(fp);
5957
5958	if (status == 0)
5959		return;
5960
5961	ztest_dump_core = 0;
5962	if (WIFEXITED(status))
5963		fatal(0, "'%s' exit code %d", zdb, WEXITSTATUS(status));
5964	else
5965		fatal(0, "'%s' died with signal %d", zdb, WTERMSIG(status));
5966}
5967
5968static void
5969ztest_walk_pool_directory(char *header)
5970{
5971	spa_t *spa = NULL;
5972
5973	if (ztest_opts.zo_verbose >= 6)
5974		(void) printf("%s\n", header);
5975
5976	mutex_enter(&spa_namespace_lock);
5977	while ((spa = spa_next(spa)) != NULL)
5978		if (ztest_opts.zo_verbose >= 6)
5979			(void) printf("\t%s\n", spa_name(spa));
5980	mutex_exit(&spa_namespace_lock);
5981}
5982
5983static void
5984ztest_spa_import_export(char *oldname, char *newname)
5985{
5986	nvlist_t *config, *newconfig;
5987	uint64_t pool_guid;
5988	spa_t *spa;
5989	int error;
5990
5991	if (ztest_opts.zo_verbose >= 4) {
5992		(void) printf("import/export: old = %s, new = %s\n",
5993		    oldname, newname);
5994	}
5995
5996	/*
5997	 * Clean up from previous runs.
5998	 */
5999	(void) spa_destroy(newname);
6000
6001	/*
6002	 * Get the pool's configuration and guid.
6003	 */
6004	VERIFY3U(0, ==, spa_open(oldname, &spa, FTAG));
6005
6006	/*
6007	 * Kick off a scrub to tickle scrub/export races.
6008	 */
6009	if (ztest_random(2) == 0)
6010		(void) spa_scan(spa, POOL_SCAN_SCRUB);
6011
6012	pool_guid = spa_guid(spa);
6013