spa.c revision e9103aaee0c546d4644791198c54abb03c89969e
1fa9e406ahrens/*
2fa9e406ahrens * CDDL HEADER START
3fa9e406ahrens *
4fa9e406ahrens * The contents of this file are subject to the terms of the
5ea8dc4beschrock * Common Development and Distribution License (the "License").
6ea8dc4beschrock * You may not use this file except in compliance with the License.
7fa9e406ahrens *
8fa9e406ahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9fa9e406ahrens * or http://www.opensolaris.org/os/licensing.
10fa9e406ahrens * See the License for the specific language governing permissions
11fa9e406ahrens * and limitations under the License.
12fa9e406ahrens *
13fa9e406ahrens * When distributing Covered Code, include this CDDL HEADER in each
14fa9e406ahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15fa9e406ahrens * If applicable, add the following below this CDDL HEADER, with the
16fa9e406ahrens * fields enclosed by brackets "[]" replaced with your own identifying
17fa9e406ahrens * information: Portions Copyright [yyyy] [name of copyright owner]
18fa9e406ahrens *
19fa9e406ahrens * CDDL HEADER END
20fa9e406ahrens */
2199653d4eschrock
22fa9e406ahrens/*
2398d1cbfGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
245aeb947Garrett D'Amore * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25e9103aaGarrett D'Amore * Copyright (c) 2011 by Delphix. All rights reserved.
265aeb947Garrett D'Amore */
27fa9e406ahrens
28fa9e406ahrens/*
29fa9e406ahrens * This file contains all the routines used when modifying on-disk SPA state.
30fa9e406ahrens * This includes opening, importing, destroying, exporting a pool, and syncing a
31fa9e406ahrens * pool.
32fa9e406ahrens */
33fa9e406ahrens
34fa9e406ahrens#include <sys/zfs_context.h>
35ea8dc4beschrock#include <sys/fm/fs/zfs.h>
36fa9e406ahrens#include <sys/spa_impl.h>
37fa9e406ahrens#include <sys/zio.h>
38fa9e406ahrens#include <sys/zio_checksum.h>
39fa9e406ahrens#include <sys/dmu.h>
40fa9e406ahrens#include <sys/dmu_tx.h>
41fa9e406ahrens#include <sys/zap.h>
42fa9e406ahrens#include <sys/zil.h>
43b24ab67Jeff Bonwick#include <sys/ddt.h>
44fa9e406ahrens#include <sys/vdev_impl.h>
45fa9e406ahrens#include <sys/metaslab.h>
4688ecc94George Wilson#include <sys/metaslab_impl.h>
47fa9e406ahrens#include <sys/uberblock_impl.h>
48fa9e406ahrens#include <sys/txg.h>
49fa9e406ahrens#include <sys/avl.h>
50fa9e406ahrens#include <sys/dmu_traverse.h>
51b1b8ab3lling#include <sys/dmu_objset.h>
52fa9e406ahrens#include <sys/unique.h>
53fa9e406ahrens#include <sys/dsl_pool.h>
54b1b8ab3lling#include <sys/dsl_dataset.h>
55fa9e406ahrens#include <sys/dsl_dir.h>
56fa9e406ahrens#include <sys/dsl_prop.h>
57b1b8ab3lling#include <sys/dsl_synctask.h>
58fa9e406ahrens#include <sys/fs/zfs.h>
59fa94a07brendan#include <sys/arc.h>
60fa9e406ahrens#include <sys/callb.h>
619517395ek#include <sys/systeminfo.h>
62e7cbe64gw#include <sys/spa_boot.h>
63573ca77George Wilson#include <sys/zfs_ioctl.h>
643f9d6adLin Ling#include <sys/dsl_scan.h>
65fa9e406ahrens
665679c89jv#ifdef	_KERNEL
67dedec47Jack Meng#include <sys/bootprops.h>
6835a5a35Jonathan Adams#include <sys/callb.h>
6935a5a35Jonathan Adams#include <sys/cpupart.h>
7035a5a35Jonathan Adams#include <sys/pool.h>
7135a5a35Jonathan Adams#include <sys/sysdc.h>
7235a5a35Jonathan Adams#include <sys/zone.h>
735679c89jv#endif	/* _KERNEL */
745679c89jv
75990b485lling#include "zfs_prop.h"
76b7b9745perrin#include "zfs_comutil.h"
77990b485lling
7835a5a35Jonathan Adamstypedef enum zti_modes {
792e0c549Jonathan Adams	zti_mode_fixed,			/* value is # of threads (min 1) */
802e0c549Jonathan Adams	zti_mode_online_percent,	/* value is % of online CPUs */
8135a5a35Jonathan Adams	zti_mode_batch,			/* cpu-intensive; value is ignored */
8280eb36fGeorge Wilson	zti_mode_null,			/* don't create a taskq */
832e0c549Jonathan Adams	zti_nmodes
8435a5a35Jonathan Adams} zti_modes_t;
85416e0cdek
8680eb36fGeorge Wilson#define	ZTI_FIX(n)	{ zti_mode_fixed, (n) }
8780eb36fGeorge Wilson#define	ZTI_PCT(n)	{ zti_mode_online_percent, (n) }
8835a5a35Jonathan Adams#define	ZTI_BATCH	{ zti_mode_batch, 0 }
8980eb36fGeorge Wilson#define	ZTI_NULL	{ zti_mode_null, 0 }
902e0c549Jonathan Adams
9180eb36fGeorge Wilson#define	ZTI_ONE		ZTI_FIX(1)
922e0c549Jonathan Adams
932e0c549Jonathan Adamstypedef struct zio_taskq_info {
9480eb36fGeorge Wilson	enum zti_modes zti_mode;
9580eb36fGeorge Wilson	uint_t zti_value;
962e0c549Jonathan Adams} zio_taskq_info_t;
972e0c549Jonathan Adams
982e0c549Jonathan Adamsstatic const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
9935a5a35Jonathan Adams	"issue", "issue_high", "intr", "intr_high"
1002e0c549Jonathan Adams};
1012e0c549Jonathan Adams
10280eb36fGeorge Wilson/*
10380eb36fGeorge Wilson * Define the taskq threads for the following I/O types:
10480eb36fGeorge Wilson * 	NULL, READ, WRITE, FREE, CLAIM, and IOCTL
10580eb36fGeorge Wilson */
10680eb36fGeorge Wilsonconst zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
10780eb36fGeorge Wilson	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
10880eb36fGeorge Wilson	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
10935a5a35Jonathan Adams	{ ZTI_FIX(8),	ZTI_NULL,	ZTI_BATCH,	ZTI_NULL },
11035a5a35Jonathan Adams	{ ZTI_BATCH,	ZTI_FIX(5),	ZTI_FIX(8),	ZTI_FIX(5) },
111c7cd242George Wilson	{ ZTI_FIX(100),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
11280eb36fGeorge Wilson	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
11380eb36fGeorge Wilson	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL },
1142e0c549Jonathan Adams};
1152e0c549Jonathan Adams
1163f9d6adLin Lingstatic dsl_syncfunc_t spa_sync_props;
11789a89ebllingstatic boolean_t spa_has_active_shared_spare(spa_t *spa);
1181195e68Mark J Musantestatic int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
1191195e68Mark J Musante    spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1201195e68Mark J Musante    char **ereport);
121cb04b87Mark J Musantestatic void spa_vdev_resilver_done(spa_t *spa);
122990b485lling
12335a5a35Jonathan Adamsuint_t		zio_taskq_batch_pct = 100;	/* 1 thread per cpu in pset */
12435a5a35Jonathan Adamsid_t		zio_taskq_psrset_bind = PS_NONE;
12535a5a35Jonathan Adamsboolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
12635a5a35Jonathan Adamsuint_t		zio_taskq_basedc = 80;		/* base duty cycle */
12735a5a35Jonathan Adams
12835a5a35Jonathan Adamsboolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
12935a5a35Jonathan Adams
13035a5a35Jonathan Adams/*
13135a5a35Jonathan Adams * This (illegal) pool name is used when temporarily importing a spa_t in order
13235a5a35Jonathan Adams * to get the vdev stats associated with the imported devices.
13335a5a35Jonathan Adams */
13435a5a35Jonathan Adams#define	TRYIMPORT_NAME	"$import"
13535a5a35Jonathan Adams
136990b485lling/*
137990b485lling * ==========================================================================
138990b485lling * SPA properties routines
139990b485lling * ==========================================================================
140990b485lling */
141990b485lling
142990b485lling/*
143990b485lling * Add a (source=src, propname=propval) list to an nvlist.
144990b485lling */
1459d82f4fllingstatic void
146990b485llingspa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
147990b485lling    uint64_t intval, zprop_source_t src)
148990b485lling{
149990b485lling	const char *propname = zpool_prop_to_name(prop);
150990b485lling	nvlist_t *propval;
151990b485lling
1529d82f4flling	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1539d82f4flling	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
154990b485lling
1559d82f4flling	if (strval != NULL)
1569d82f4flling		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
1579d82f4flling	else
1589d82f4flling		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
159990b485lling
1609d82f4flling	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
161990b485lling	nvlist_free(propval);
162990b485lling}
163990b485lling
164990b485lling/*
165990b485lling * Get property values from the spa configuration.
166990b485lling */
1679d82f4fllingstatic void
168990b485llingspa_prop_get_config(spa_t *spa, nvlist_t **nvp)
169990b485lling{
170379c004Eric Schrock	uint64_t size;
171485bbbfGeorge Wilson	uint64_t alloc;
172990b485lling	uint64_t cap, version;
173990b485lling	zprop_source_t src = ZPROP_SRC_NONE;
174c5904d1eschrock	spa_config_dirent_t *dp;
175990b485lling
176e14bb32Jeff Bonwick	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
177e14bb32Jeff Bonwick
178379c004Eric Schrock	if (spa->spa_root_vdev != NULL) {
179485bbbfGeorge Wilson		alloc = metaslab_class_get_alloc(spa_normal_class(spa));
180b24ab67Jeff Bonwick		size = metaslab_class_get_space(spa_normal_class(spa));
181379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
182379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
183485bbbfGeorge Wilson		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
184485bbbfGeorge Wilson		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
185485bbbfGeorge Wilson		    size - alloc, src);
186f9af39bGeorge Wilson		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
187f9af39bGeorge Wilson		    (spa_mode(spa) == FREAD), src);
188379c004Eric Schrock
189485bbbfGeorge Wilson		cap = (size == 0) ? 0 : (alloc * 100 / size);
190379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
191379c004Eric Schrock
192b24ab67Jeff Bonwick		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
193b24ab67Jeff Bonwick		    ddt_get_pool_dedup_ratio(spa), src);
194b24ab67Jeff Bonwick
195379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
196379c004Eric Schrock		    spa->spa_root_vdev->vdev_state, src);
197379c004Eric Schrock
198379c004Eric Schrock		version = spa_version(spa);
199379c004Eric Schrock		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
200379c004Eric Schrock			src = ZPROP_SRC_DEFAULT;
201379c004Eric Schrock		else
202379c004Eric Schrock			src = ZPROP_SRC_LOCAL;
203379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
204379c004Eric Schrock	}
205990b485lling
2069d82f4flling	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
207990b485lling
2089d82f4flling	if (spa->spa_root != NULL)
2099d82f4flling		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
2109d82f4flling		    0, ZPROP_SRC_LOCAL);
211990b485lling
212c5904d1eschrock	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
213c5904d1eschrock		if (dp->scd_path == NULL) {
2149d82f4flling			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
215c5904d1eschrock			    "none", 0, ZPROP_SRC_LOCAL);
216c5904d1eschrock		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
2179d82f4flling			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
218c5904d1eschrock			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
2192f8aaabeschrock		}
2202f8aaabeschrock	}
221990b485lling}
222990b485lling
223990b485lling/*
224990b485lling * Get zpool property values.
225990b485lling */
226990b485llingint
227990b485llingspa_prop_get(spa_t *spa, nvlist_t **nvp)
228990b485lling{
229b24ab67Jeff Bonwick	objset_t *mos = spa->spa_meta_objset;
230990b485lling	zap_cursor_t zc;
231990b485lling	zap_attribute_t za;
232990b485lling	int err;
233990b485lling
2349d82f4flling	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
235990b485lling
236e14bb32Jeff Bonwick	mutex_enter(&spa->spa_props_lock);
237e14bb32Jeff Bonwick
238990b485lling	/*
239990b485lling	 * Get properties from the spa config.
240990b485lling	 */
2419d82f4flling	spa_prop_get_config(spa, nvp);
242990b485lling
243990b485lling	/* If no pool property object, no more prop to get. */
244afee20eGeorge Wilson	if (mos == NULL || spa->spa_pool_props_object == 0) {
245990b485lling		mutex_exit(&spa->spa_props_lock);
246990b485lling		return (0);
247990b485lling	}
248990b485lling
249990b485lling	/*
250990b485lling	 * Get properties from the MOS pool property object.
251990b485lling	 */
252990b485lling	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
253990b485lling	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
254990b485lling	    zap_cursor_advance(&zc)) {
255990b485lling		uint64_t intval = 0;
256990b485lling		char *strval = NULL;
257990b485lling		zprop_source_t src = ZPROP_SRC_DEFAULT;
258990b485lling		zpool_prop_t prop;
259990b485lling
260990b485lling		if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
261990b485lling			continue;
262990b485lling
263990b485lling		switch (za.za_integer_length) {
264990b485lling		case 8:
265990b485lling			/* integer property */
266990b485lling			if (za.za_first_integer !=
267990b485lling			    zpool_prop_default_numeric(prop))
268990b485lling				src = ZPROP_SRC_LOCAL;
269990b485lling
270990b485lling			if (prop == ZPOOL_PROP_BOOTFS) {
271990b485lling				dsl_pool_t *dp;
272990b485lling				dsl_dataset_t *ds = NULL;
273990b485lling
274990b485lling				dp = spa_get_dsl(spa);
275990b485lling				rw_enter(&dp->dp_config_rwlock, RW_READER);
276745cd3cmaybee				if (err = dsl_dataset_hold_obj(dp,
277745cd3cmaybee				    za.za_first_integer, FTAG, &ds)) {
278990b485lling					rw_exit(&dp->dp_config_rwlock);
279990b485lling					break;
280990b485lling				}
281990b485lling
282990b485lling				strval = kmem_alloc(
283990b485lling				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
284990b485lling				    KM_SLEEP);
285990b485lling				dsl_dataset_name(ds, strval);
286745cd3cmaybee				dsl_dataset_rele(ds, FTAG);
287990b485lling				rw_exit(&dp->dp_config_rwlock);
288990b485lling			} else {
289990b485lling				strval = NULL;
290990b485lling				intval = za.za_first_integer;
291990b485lling			}
292990b485lling
2939d82f4flling			spa_prop_add_list(*nvp, prop, strval, intval, src);
294990b485lling
295990b485lling			if (strval != NULL)
296990b485lling				kmem_free(strval,
297990b485lling				    MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
298990b485lling
299990b485lling			break;
300990b485lling
301990b485lling		case 1:
302990b485lling			/* string property */
303990b485lling			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
304990b485lling			err = zap_lookup(mos, spa->spa_pool_props_object,
305990b485lling			    za.za_name, 1, za.za_num_integers, strval);
306990b485lling			if (err) {
307990b485lling				kmem_free(strval, za.za_num_integers);
308990b485lling				break;
309990b485lling			}
3109d82f4flling			spa_prop_add_list(*nvp, prop, strval, 0, src);
311990b485lling			kmem_free(strval, za.za_num_integers);
312990b485lling			break;
313990b485lling
314990b485lling		default:
315990b485lling			break;
316990b485lling		}
317990b485lling	}
318990b485lling	zap_cursor_fini(&zc);
319990b485lling	mutex_exit(&spa->spa_props_lock);
320990b485llingout:
321990b485lling	if (err && err != ENOENT) {
322990b485lling		nvlist_free(*nvp);
3239d82f4flling		*nvp = NULL;
324990b485lling		return (err);
325990b485lling	}
326990b485lling
327990b485lling	return (0);
328990b485lling}
329990b485lling
330990b485lling/*
331990b485lling * Validate the given pool properties nvlist and modify the list
332990b485lling * for the property values to be set.
333990b485lling */
334990b485llingstatic int
335990b485llingspa_prop_validate(spa_t *spa, nvlist_t *props)
336990b485lling{
337990b485lling	nvpair_t *elem;
338990b485lling	int error = 0, reset_bootfs = 0;
339990b485lling	uint64_t objnum;
340990b485lling
341990b485lling	elem = NULL;
342990b485lling	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
343990b485lling		zpool_prop_t prop;
344990b485lling		char *propname, *strval;
345990b485lling		uint64_t intval;
346990b485lling		objset_t *os;
3472f8aaabeschrock		char *slash;
348990b485lling
349990b485lling		propname = nvpair_name(elem);
350990b485lling
351990b485lling		if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL)
352990b485lling			return (EINVAL);
353990b485lling
354990b485lling		switch (prop) {
355990b485lling		case ZPOOL_PROP_VERSION:
356990b485lling			error = nvpair_value_uint64(elem, &intval);
357990b485lling			if (!error &&
358990b485lling			    (intval < spa_version(spa) || intval > SPA_VERSION))
359990b485lling				error = EINVAL;
360990b485lling			break;
361990b485lling
362990b485lling		case ZPOOL_PROP_DELEGATION:
363990b485lling		case ZPOOL_PROP_AUTOREPLACE:
364d5b5bb2Rich Morris		case ZPOOL_PROP_LISTSNAPS:
365573ca77George Wilson		case ZPOOL_PROP_AUTOEXPAND:
366990b485lling			error = nvpair_value_uint64(elem, &intval);
367990b485lling			if (!error && intval > 1)
368990b485lling				error = EINVAL;
369990b485lling			break;
370990b485lling
371990b485lling		case ZPOOL_PROP_BOOTFS:
37225f89eeJeff Bonwick			/*
37325f89eeJeff Bonwick			 * If the pool version is less than SPA_VERSION_BOOTFS,
37425f89eeJeff Bonwick			 * or the pool is still being created (version == 0),
37525f89eeJeff Bonwick			 * the bootfs property cannot be set.
37625f89eeJeff Bonwick			 */
377990b485lling			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
378990b485lling				error = ENOTSUP;
379990b485lling				break;
380990b485lling			}
381990b485lling
382990b485lling			/*
38315e6edfgw			 * Make sure the vdev config is bootable
384990b485lling			 */
38515e6edfgw			if (!vdev_is_bootable(spa->spa_root_vdev)) {
386990b485lling				error = ENOTSUP;
387990b485lling				break;
388990b485lling			}
389990b485lling
390990b485lling			reset_bootfs = 1;
391990b485lling
392990b485lling			error = nvpair_value_string(elem, &strval);
393990b485lling
394990b485lling			if (!error) {
39515e6edfgw				uint64_t compress;
39615e6edfgw
397990b485lling				if (strval == NULL || strval[0] == '\0') {
398990b485lling					objnum = zpool_prop_default_numeric(
399990b485lling					    ZPOOL_PROP_BOOTFS);
400990b485lling					break;
401990b485lling				}
402990b485lling
403503ad85Matthew Ahrens				if (error = dmu_objset_hold(strval, FTAG, &os))
404990b485lling					break;
40515e6edfgw
406503ad85Matthew Ahrens				/* Must be ZPL and not gzip compressed. */
407503ad85Matthew Ahrens
408503ad85Matthew Ahrens				if (dmu_objset_type(os) != DMU_OST_ZFS) {
409503ad85Matthew Ahrens					error = ENOTSUP;
410503ad85Matthew Ahrens				} else if ((error = dsl_prop_get_integer(strval,
41115e6edfgw				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
41215e6edfgw				    &compress, NULL)) == 0 &&
41315e6edfgw				    !BOOTFS_COMPRESS_VALID(compress)) {
41415e6edfgw					error = ENOTSUP;
41515e6edfgw				} else {
41615e6edfgw					objnum = dmu_objset_id(os);
41715e6edfgw				}
418503ad85Matthew Ahrens				dmu_objset_rele(os, FTAG);
419990b485lling			}
420990b485lling			break;
421e14bb32Jeff Bonwick
4220a4e951gw		case ZPOOL_PROP_FAILUREMODE:
4230a4e951gw			error = nvpair_value_uint64(elem, &intval);
4240a4e951gw			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
4250a4e951gw			    intval > ZIO_FAILURE_MODE_PANIC))
4260a4e951gw				error = EINVAL;
4270a4e951gw
4280a4e951gw			/*
4290a4e951gw			 * This is a special case which only occurs when
4300a4e951gw			 * the pool has completely failed. This allows
4310a4e951gw			 * the user to change the in-core failmode property
4320a4e951gw			 * without syncing it out to disk (I/Os might
4330a4e951gw			 * currently be blocked). We do this by returning
4340a4e951gw			 * EIO to the caller (spa_prop_set) to trick it
4350a4e951gw			 * into thinking we encountered a property validation
4360a4e951gw			 * error.
4370a4e951gw			 */
438e14bb32Jeff Bonwick			if (!error && spa_suspended(spa)) {
4390a4e951gw				spa->spa_failmode = intval;
4400a4e951gw				error = EIO;
4410a4e951gw			}
4420a4e951gw			break;
4432f8aaabeschrock
4442f8aaabeschrock		case ZPOOL_PROP_CACHEFILE:
4452f8aaabeschrock			if ((error = nvpair_value_string(elem, &strval)) != 0)
4462f8aaabeschrock				break;
4472f8aaabeschrock
4482f8aaabeschrock			if (strval[0] == '\0')
4492f8aaabeschrock				break;
4502f8aaabeschrock
4512f8aaabeschrock			if (strcmp(strval, "none") == 0)
4522f8aaabeschrock				break;
4532f8aaabeschrock
4542f8aaabeschrock			if (strval[0] != '/') {
4552f8aaabeschrock				error = EINVAL;
4562f8aaabeschrock				break;
4572f8aaabeschrock			}
4582f8aaabeschrock
4592f8aaabeschrock			slash = strrchr(strval, '/');
4602f8aaabeschrock			ASSERT(slash != NULL);
4612f8aaabeschrock
4622f8aaabeschrock			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
4632f8aaabeschrock			    strcmp(slash, "/..") == 0)
4642f8aaabeschrock				error = EINVAL;
4652f8aaabeschrock			break;
466b24ab67Jeff Bonwick
467b24ab67Jeff Bonwick		case ZPOOL_PROP_DEDUPDITTO:
468b24ab67Jeff Bonwick			if (spa_version(spa) < SPA_VERSION_DEDUP)
469b24ab67Jeff Bonwick				error = ENOTSUP;
470b24ab67Jeff Bonwick			else
471b24ab67Jeff Bonwick				error = nvpair_value_uint64(elem, &intval);
472b24ab67Jeff Bonwick			if (error == 0 &&
473b24ab67Jeff Bonwick			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
474b24ab67Jeff Bonwick				error = EINVAL;
475b24ab67Jeff Bonwick			break;
476990b485lling		}
477990b485lling
478990b485lling		if (error)
479990b485lling			break;
480990b485lling	}
481990b485lling
482990b485lling	if (!error && reset_bootfs) {
483990b485lling		error = nvlist_remove(props,
484990b485lling		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
485990b485lling
486990b485lling		if (!error) {
487990b485lling			error = nvlist_add_uint64(props,
488990b485lling			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
489990b485lling		}
490990b485lling	}
491990b485lling
492990b485lling	return (error);
493990b485lling}
494990b485lling
495379c004Eric Schrockvoid
496379c004Eric Schrockspa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
497379c004Eric Schrock{
498379c004Eric Schrock	char *cachefile;
499379c004Eric Schrock	spa_config_dirent_t *dp;
500379c004Eric Schrock
501379c004Eric Schrock	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
502379c004Eric Schrock	    &cachefile) != 0)
503379c004Eric Schrock		return;
504379c004Eric Schrock
505379c004Eric Schrock	dp = kmem_alloc(sizeof (spa_config_dirent_t),
506379c004Eric Schrock	    KM_SLEEP);
507379c004Eric Schrock
508379c004Eric Schrock	if (cachefile[0] == '\0')
509379c004Eric Schrock		dp->scd_path = spa_strdup(spa_config_path);
510379c004Eric Schrock	else if (strcmp(cachefile, "none") == 0)
511379c004Eric Schrock		dp->scd_path = NULL;
512379c004Eric Schrock	else
513379c004Eric Schrock		dp->scd_path = spa_strdup(cachefile);
514379c004Eric Schrock
515379c004Eric Schrock	list_insert_head(&spa->spa_config_list, dp);
516379c004Eric Schrock	if (need_sync)
517379c004Eric Schrock		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
518379c004Eric Schrock}
519379c004Eric Schrock
520990b485llingint
521990b485llingspa_prop_set(spa_t *spa, nvlist_t *nvp)
522990b485lling{
523990b485lling	int error;
524379c004Eric Schrock	nvpair_t *elem;
525379c004Eric Schrock	boolean_t need_sync = B_FALSE;
526379c004Eric Schrock	zpool_prop_t prop;
527990b485lling
528990b485lling	if ((error = spa_prop_validate(spa, nvp)) != 0)
529990b485lling		return (error);
530990b485lling
531379c004Eric Schrock	elem = NULL;
532379c004Eric Schrock	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
533379c004Eric Schrock		if ((prop = zpool_name_to_prop(
534379c004Eric Schrock		    nvpair_name(elem))) == ZPROP_INVAL)
535379c004Eric Schrock			return (EINVAL);
536379c004Eric Schrock
537f9af39bGeorge Wilson		if (prop == ZPOOL_PROP_CACHEFILE ||
538f9af39bGeorge Wilson		    prop == ZPOOL_PROP_ALTROOT ||
539f9af39bGeorge Wilson		    prop == ZPOOL_PROP_READONLY)
540379c004Eric Schrock			continue;
541379c004Eric Schrock
542379c004Eric Schrock		need_sync = B_TRUE;
543379c004Eric Schrock		break;
544379c004Eric Schrock	}
545379c004Eric Schrock
546379c004Eric Schrock	if (need_sync)
547379c004Eric Schrock		return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
548379c004Eric Schrock		    spa, nvp, 3));
549379c004Eric Schrock	else
550379c004Eric Schrock		return (0);
551990b485lling}
552990b485lling
553990b485lling/*
554990b485lling * If the bootfs property value is dsobj, clear it.
555990b485lling */
556990b485llingvoid
557990b485llingspa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
558990b485lling{
559990b485lling	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
560990b485lling		VERIFY(zap_remove(spa->spa_meta_objset,
561990b485lling		    spa->spa_pool_props_object,
562990b485lling		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
563990b485lling		spa->spa_bootfs = 0;
564990b485lling	}
565990b485lling}
566990b485lling
567fa9e406ahrens/*
568e9103aaGarrett D'Amore * Change the GUID for the pool.  This is done so that we can later
569e9103aaGarrett D'Amore * re-import a pool built from a clone of our own vdevs.  We will modify
570e9103aaGarrett D'Amore * the root vdev's guid, our own pool guid, and then mark all of our
571e9103aaGarrett D'Amore * vdevs dirty.  Note that we must make sure that all our vdevs are
572e9103aaGarrett D'Amore * online when we do this, or else any vdevs that weren't present
573e9103aaGarrett D'Amore * would be orphaned from our pool.  We are also going to issue a
574e9103aaGarrett D'Amore * sysevent to update any watchers.
575e9103aaGarrett D'Amore */
576e9103aaGarrett D'Amoreint
577e9103aaGarrett D'Amorespa_change_guid(spa_t *spa)
578e9103aaGarrett D'Amore{
579e9103aaGarrett D'Amore	uint64_t	oldguid, newguid;
580e9103aaGarrett D'Amore	uint64_t	txg;
581e9103aaGarrett D'Amore
582e9103aaGarrett D'Amore	if (!(spa_mode_global & FWRITE))
583e9103aaGarrett D'Amore		return (EROFS);
584e9103aaGarrett D'Amore
585e9103aaGarrett D'Amore	txg = spa_vdev_enter(spa);
586e9103aaGarrett D'Amore
587e9103aaGarrett D'Amore	if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
588e9103aaGarrett D'Amore		return (spa_vdev_exit(spa, NULL, txg, ENXIO));
589e9103aaGarrett D'Amore
590e9103aaGarrett D'Amore	oldguid = spa_guid(spa);
591e9103aaGarrett D'Amore	newguid = spa_generate_guid(NULL);
592e9103aaGarrett D'Amore	ASSERT3U(oldguid, !=, newguid);
593e9103aaGarrett D'Amore
594e9103aaGarrett D'Amore	spa->spa_root_vdev->vdev_guid = newguid;
595e9103aaGarrett D'Amore	spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
596e9103aaGarrett D'Amore
597e9103aaGarrett D'Amore	vdev_config_dirty(spa->spa_root_vdev);
598e9103aaGarrett D'Amore
599e9103aaGarrett D'Amore	spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
600e9103aaGarrett D'Amore
601e9103aaGarrett D'Amore	return (spa_vdev_exit(spa, NULL, txg, 0));
602e9103aaGarrett D'Amore}
603e9103aaGarrett D'Amore
604e9103aaGarrett D'Amore/*
605fa9e406ahrens * ==========================================================================
606fa9e406ahrens * SPA state manipulation (open/create/destroy/import/export)
607fa9e406ahrens * ==========================================================================
608fa9e406ahrens */
609fa9e406ahrens
610ea8dc4beschrockstatic int
611ea8dc4beschrockspa_error_entry_compare(const void *a, const void *b)
612ea8dc4beschrock{
613ea8dc4beschrock	spa_error_entry_t *sa = (spa_error_entry_t *)a;
614ea8dc4beschrock	spa_error_entry_t *sb = (spa_error_entry_t *)b;
615ea8dc4beschrock	int ret;
616ea8dc4beschrock
617ea8dc4beschrock	ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
618ea8dc4beschrock	    sizeof (zbookmark_t));
619ea8dc4beschrock
620ea8dc4beschrock	if (ret < 0)
621ea8dc4beschrock		return (-1);
622ea8dc4beschrock	else if (ret > 0)
623ea8dc4beschrock		return (1);
624ea8dc4beschrock	else
625ea8dc4beschrock		return (0);
626ea8dc4beschrock}
627ea8dc4beschrock
628ea8dc4beschrock/*
629ea8dc4beschrock * Utility function which retrieves copies of the current logs and
630ea8dc4beschrock * re-initializes them in the process.
631ea8dc4beschrock */
632ea8dc4beschrockvoid
633ea8dc4beschrockspa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
634ea8dc4beschrock{
635ea8dc4beschrock	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
636ea8dc4beschrock
637ea8dc4beschrock	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
638ea8dc4beschrock	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
639ea8dc4beschrock
640ea8dc4beschrock	avl_create(&spa->spa_errlist_scrub,
641ea8dc4beschrock	    spa_error_entry_compare, sizeof (spa_error_entry_t),
642ea8dc4beschrock	    offsetof(spa_error_entry_t, se_avl));
643ea8dc4beschrock	avl_create(&spa->spa_errlist_last,
644ea8dc4beschrock	    spa_error_entry_compare, sizeof (spa_error_entry_t),
645ea8dc4beschrock	    offsetof(spa_error_entry_t, se_avl));
646ea8dc4beschrock}
647ea8dc4beschrock
64835a5a35Jonathan Adamsstatic taskq_t *
64935a5a35Jonathan Adamsspa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
65035a5a35Jonathan Adams    uint_t value)
651fa9e406ahrens{
6525aeb947Garrett D'Amore	uint_t flags = 0;
65335a5a35Jonathan Adams	boolean_t batch = B_FALSE;
654fa9e406ahrens
65535a5a35Jonathan Adams	switch (mode) {
65635a5a35Jonathan Adams	case zti_mode_null:
65735a5a35Jonathan Adams		return (NULL);		/* no taskq needed */
658fa9e406ahrens
65935a5a35Jonathan Adams	case zti_mode_fixed:
66035a5a35Jonathan Adams		ASSERT3U(value, >=, 1);
66135a5a35Jonathan Adams		value = MAX(value, 1);
66235a5a35Jonathan Adams		break;
663fa9e406ahrens
66435a5a35Jonathan Adams	case zti_mode_batch:
66535a5a35Jonathan Adams		batch = B_TRUE;
66635a5a35Jonathan Adams		flags |= TASKQ_THREADS_CPU_PCT;
66735a5a35Jonathan Adams		value = zio_taskq_batch_pct;
66835a5a35Jonathan Adams		break;
66935a5a35Jonathan Adams
67035a5a35Jonathan Adams	case zti_mode_online_percent:
67135a5a35Jonathan Adams		flags |= TASKQ_THREADS_CPU_PCT;
67235a5a35Jonathan Adams		break;
67335a5a35Jonathan Adams
67435a5a35Jonathan Adams	default:
67535a5a35Jonathan Adams		panic("unrecognized mode for %s taskq (%u:%u) in "
67635a5a35Jonathan Adams		    "spa_activate()",
67735a5a35Jonathan Adams		    name, mode, value);
67835a5a35Jonathan Adams		break;
67935a5a35Jonathan Adams	}
68035a5a35Jonathan Adams
68135a5a35Jonathan Adams	if (zio_taskq_sysdc && spa->spa_proc != &p0) {
68235a5a35Jonathan Adams		if (batch)
68335a5a35Jonathan Adams			flags |= TASKQ_DC_BATCH;
68435a5a35Jonathan Adams
68535a5a35Jonathan Adams		return (taskq_create_sysdc(name, value, 50, INT_MAX,
68635a5a35Jonathan Adams		    spa->spa_proc, zio_taskq_basedc, flags));
68735a5a35Jonathan Adams	}
68835a5a35Jonathan Adams	return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
68935a5a35Jonathan Adams	    spa->spa_proc, flags));
69035a5a35Jonathan Adams}
69135a5a35Jonathan Adams
69235a5a35Jonathan Adamsstatic void
69335a5a35Jonathan Adamsspa_create_zio_taskqs(spa_t *spa)
694