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 */
2398d1cbfGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
240ce4bbcMatthew Ahrens * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
251437283Hans Rosenfeld * Copyright (c) 2015, Nexenta Systems, Inc.  All rights reserved.
26bc9014eJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
2745818eeMatthew Ahrens * Copyright 2013 Saso Kiselkov. All rights reserved.
28c3d26abMatthew Ahrens * Copyright (c) 2014 Integros [integros.com]
29c8811bdToomas Soome * Copyright 2016 Toomas Soome <tsoome@me.com>
300c06d38jwpoduska * Copyright (c) 2017, 2019, Datto Inc. All rights reserved.
31e830fb1Kody A Kantor * Copyright 2019 Joyent, Inc.
32663207aDon Brady * Copyright (c) 2017, Intel Corporation.
330fb055eAndy Fiddaman * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
3430c304dJoshua M. Clulow * Copyright 2020 Joshua M. Clulow <josh@sysmgr.org>
355aeb947Garrett D'Amore */
383e30c24Will Andrews * SPA: Storage Pool Allocator
393e30c24Will Andrews *
40fa9e406ahrens * This file contains all the routines used when modifying on-disk SPA state.
41fa9e406ahrens * This includes opening, importing, destroying, exporting a pool, and syncing a
42fa9e406ahrens * pool.
43fa9e406ahrens */
45fa9e406ahrens#include <sys/zfs_context.h>
46ea8dc4beschrock#include <sys/fm/fs/zfs.h>
47fa9e406ahrens#include <sys/spa_impl.h>
48fa9e406ahrens#include <sys/zio.h>
49fa9e406ahrens#include <sys/zio_checksum.h>
50fa9e406ahrens#include <sys/dmu.h>
51fa9e406ahrens#include <sys/dmu_tx.h>
52fa9e406ahrens#include <sys/zap.h>
53fa9e406ahrens#include <sys/zil.h>
54b24ab67Jeff Bonwick#include <sys/ddt.h>
55fa9e406ahrens#include <sys/vdev_impl.h>
565cabbc6Prashanth Sreenivasa#include <sys/vdev_removal.h>
575cabbc6Prashanth Sreenivasa#include <sys/vdev_indirect_mapping.h>
585cabbc6Prashanth Sreenivasa#include <sys/vdev_indirect_births.h>
59094e47eGeorge Wilson#include <sys/vdev_initialize.h>
60084fd14Brian Behlendorf#include <sys/vdev_trim.h>
61fa9e406ahrens#include <sys/metaslab.h>
6288ecc94George Wilson#include <sys/metaslab_impl.h>
63e0f1c0aOlaf Faaland#include <sys/mmp.h>
64fa9e406ahrens#include <sys/uberblock_impl.h>
65fa9e406ahrens#include <sys/txg.h>
66fa9e406ahrens#include <sys/avl.h>
675cabbc6Prashanth Sreenivasa#include <sys/bpobj.h>
68fa9e406ahrens#include <sys/dmu_traverse.h>
69b1b8ab3lling#include <sys/dmu_objset.h>
70fa9e406ahrens#include <sys/unique.h>
71fa9e406ahrens#include <sys/dsl_pool.h>
72b1b8ab3lling#include <sys/dsl_dataset.h>
73fa9e406ahrens#include <sys/dsl_dir.h>
74fa9e406ahrens#include <sys/dsl_prop.h>
75b1b8ab3lling#include <sys/dsl_synctask.h>
76fa9e406ahrens#include <sys/fs/zfs.h>
77fa94a07brendan#include <sys/arc.h>
78fa9e406ahrens#include <sys/callb.h>
799517395ek#include <sys/systeminfo.h>
80e7cbe64gw#include <sys/spa_boot.h>
81573ca77George Wilson#include <sys/zfs_ioctl.h>
823f9d6adLin Ling#include <sys/dsl_scan.h>
83ad135b5Christopher Siden#include <sys/zfeature.h>
843b2aab1Matthew Ahrens#include <sys/dsl_destroy.h>
85770499eDan Kimmel#include <sys/abd.h>
875679c89jv#ifdef	_KERNEL
88dedec47Jack Meng#include <sys/bootprops.h>
8935a5a35Jonathan Adams#include <sys/callb.h>
9035a5a35Jonathan Adams#include <sys/cpupart.h>
9135a5a35Jonathan Adams#include <sys/pool.h>
9235a5a35Jonathan Adams#include <sys/sysdc.h>
9335a5a35Jonathan Adams#include <sys/zone.h>
945679c89jv#endif	/* _KERNEL */
96990b485lling#include "zfs_prop.h"
97b7b9745perrin#include "zfs_comutil.h"
993cb69f7Will Andrews/*
1003cb69f7Will Andrews * The interval, in seconds, at which failed configuration cache file writes
1013cb69f7Will Andrews * should be retried.
1023cb69f7Will Andrews */
1035cabbc6Prashanth Sreenivasaint zfs_ccw_retry_interval = 300;
1043cb69f7Will Andrews
10535a5a35Jonathan Adamstypedef enum zti_modes {
106ec94d32Adam Leventhal	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
107ec94d32Adam Leventhal	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
108ec94d32Adam Leventhal	ZTI_MODE_NULL,			/* don't create a taskq */
109ec94d32Adam Leventhal	ZTI_NMODES
11035a5a35Jonathan Adams} zti_modes_t;
112ec94d32Adam Leventhal#define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
113ec94d32Adam Leventhal#define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
114ec94d32Adam Leventhal#define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
1152e0c549Jonathan Adams
116ec94d32Adam Leventhal#define	ZTI_N(n)	ZTI_P(n, 1)
117ec94d32Adam Leventhal#define	ZTI_ONE		ZTI_N(1)
1182e0c549Jonathan Adams
1192e0c549Jonathan Adamstypedef struct zio_taskq_info {
120ec94d32Adam Leventhal	zti_modes_t zti_mode;
12180eb36fGeorge Wilson	uint_t zti_value;
122ec94d32Adam Leventhal	uint_t zti_count;
1232e0c549Jonathan Adams} zio_taskq_info_t;
1242e0c549Jonathan Adams
1252e0c549Jonathan Adamsstatic const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
12635a5a35Jonathan Adams	"issue", "issue_high", "intr", "intr_high"
1272e0c549Jonathan Adams};
1282e0c549Jonathan Adams
12980eb36fGeorge Wilson/*
130ec94d32Adam Leventhal * This table defines the taskq settings for each ZFS I/O type. When
131ec94d32Adam Leventhal * initializing a pool, we use this table to create an appropriately sized
132ec94d32Adam Leventhal * taskq. Some operations are low volume and therefore have a small, static
133ec94d32Adam Leventhal * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
134ec94d32Adam Leventhal * macros. Other operations process a large amount of data; the ZTI_BATCH
135ec94d32Adam Leventhal * macro causes us to create a taskq oriented for throughput. Some operations
136084fd14Brian Behlendorf * are so high frequency and short-lived that the taskq itself can become a
137ec94d32Adam Leventhal * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
138ec94d32Adam Leventhal * additional degree of parallelism specified by the number of threads per-
139ec94d32Adam Leventhal * taskq and the number of taskqs; when dispatching an event in this case, the
140ec94d32Adam Leventhal * particular taskq is chosen at random.
141ec94d32Adam Leventhal *
142ec94d32Adam Leventhal * The different taskq priorities are to handle the different contexts (issue
143ec94d32Adam Leventhal * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
144ec94d32Adam Leventhal * need to be handled with minimum delay.
14580eb36fGeorge Wilson */
14680eb36fGeorge Wilsonconst zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
14780eb36fGeorge Wilson	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
148ec94d32Adam Leventhal	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
1491b497abAdam H. Leventhal	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
150ec94d32Adam Leventhal	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
151ec94d32Adam Leventhal	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
152ec94d32Adam Leventhal	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
153ec94d32Adam Leventhal	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
154084fd14Brian Behlendorf	{ ZTI_N(4),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* TRIM */
1552e0c549Jonathan Adams};
1562e0c549Jonathan Adams
1573b2aab1Matthew Ahrensstatic void spa_sync_version(void *arg, dmu_tx_t *tx);
1583b2aab1Matthew Ahrensstatic void spa_sync_props(void *arg, dmu_tx_t *tx);
15989a89ebllingstatic boolean_t spa_has_active_shared_spare(spa_t *spa);
1608671400Serapheim Dimitropoulosstatic int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
161cb04b87Mark J Musantestatic void spa_vdev_resilver_done(spa_t *spa);
16369962b5Matthew Ahrensuint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
16435a5a35Jonathan Adamsid_t		zio_taskq_psrset_bind = PS_NONE;
16535a5a35Jonathan Adamsboolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
16635a5a35Jonathan Adamsuint_t		zio_taskq_basedc = 80;		/* base duty cycle */
16735a5a35Jonathan Adams
16835a5a35Jonathan Adamsboolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
16901f55e4George Wilsonextern int	zfs_sync_pass_deferred_free;
17035a5a35Jonathan Adams
17135a5a35Jonathan Adams/*
172e144c4ePavel Zakharov * Report any spa_load_verify errors found, but do not fail spa_load.
173e144c4ePavel Zakharov * This is used by zdb to analyze non-idle pools.
174e144c4ePavel Zakharov */
175e144c4ePavel Zakharovboolean_t	spa_load_verify_dryrun = B_FALSE;
176e144c4ePavel Zakharov
177e144c4ePavel Zakharov/*
17835a5a35Jonathan Adams * This (illegal) pool name is used when temporarily importing a spa_t in order
17935a5a35Jonathan Adams * to get the vdev stats associated with the imported devices.
18035a5a35Jonathan Adams */
18135a5a35Jonathan Adams#define	TRYIMPORT_NAME	"$import"
18235a5a35Jonathan Adams
1846f79381Pavel Zakharov * For debugging purposes: print out vdev tree during pool import.
1856f79381Pavel Zakharov */
1866f79381Pavel Zakharovboolean_t	spa_load_print_vdev_tree = B_FALSE;
1876f79381Pavel Zakharov
1886f79381Pavel Zakharov/*
1896f79381Pavel Zakharov * A non-zero value for zfs_max_missing_tvds means that we allow importing
1906f79381Pavel Zakharov * pools with missing top-level vdevs. This is strictly intended for advanced
1916f79381Pavel Zakharov * pool recovery cases since missing data is almost inevitable. Pools with
1926f79381Pavel Zakharov * missing devices can only be imported read-only for safety reasons, and their
1936f79381Pavel Zakharov * fail-mode will be automatically set to "continue".
1946f79381Pavel Zakharov *
1956f79381Pavel Zakharov * With 1 missing vdev we should be able to import the pool and mount all
1966f79381Pavel Zakharov * datasets. User data that was not modified after the missing device has been
1976f79381Pavel Zakharov * added should be recoverable. This means that snapshots created prior to the
1986f79381Pavel Zakharov * addition of that device should be completely intact.
1996f79381Pavel Zakharov *
2006f79381Pavel Zakharov * With 2 missing vdevs, some datasets may fail to mount since there are
2016f79381Pavel Zakharov * dataset statistics that are stored as regular metadata. Some data might be
2026f79381Pavel Zakharov * recoverable if those vdevs were added recently.
2036f79381Pavel Zakharov *
2046f79381Pavel Zakharov * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
2056f79381Pavel Zakharov * may be missing entirely. Chances of data recovery are very low. Note that
2066f79381Pavel Zakharov * there are also risks of performing an inadvertent rewind as we might be
2076f79381Pavel Zakharov * missing all the vdevs with the latest uberblocks.
2086f79381Pavel Zakharov */
2096f79381Pavel Zakharovuint64_t	zfs_max_missing_tvds = 0;
2106f79381Pavel Zakharov
2116f79381Pavel Zakharov/*
2126f79381Pavel Zakharov * The parameters below are similar to zfs_max_missing_tvds but are only
2136f79381Pavel Zakharov * intended for a preliminary open of the pool with an untrusted config which
2146f79381Pavel Zakharov * might be incomplete or out-dated.
2156f79381Pavel Zakharov *
2166f79381Pavel Zakharov * We are more tolerant for pools opened from a cachefile since we could have
2176f79381Pavel Zakharov * an out-dated cachefile where a device removal was not registered.
2186f79381Pavel Zakharov * We could have set the limit arbitrarily high but in the case where devices
2196f79381Pavel Zakharov * are really missing we would want to return the proper error codes; we chose
2206f79381Pavel Zakharov * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
2216f79381Pavel Zakharov * and we get a chance to retrieve the trusted config.
2226f79381Pavel Zakharov */
2236f79381Pavel Zakharovuint64_t	zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
2248671400Serapheim Dimitropoulos
2256f79381Pavel Zakharov/*
2266f79381Pavel Zakharov * In the case where config was assembled by scanning device paths (/dev/dsks
2276f79381Pavel Zakharov * by default) we are less tolerant since all the existing devices should have
2286f79381Pavel Zakharov * been detected and we want spa_load to return the right error codes.
2296f79381Pavel Zakharov */
2306f79381Pavel Zakharovuint64_t	zfs_max_missing_tvds_scan = 0;
2316f79381Pavel Zakharov
2326f79381Pavel Zakharov/*
233e830fb1Kody A Kantor * Interval in seconds at which to poll spare vdevs for health.
234e830fb1Kody A Kantor * Setting this to zero disables spare polling.
235e830fb1Kody A Kantor * Set to three hours by default.
236e830fb1Kody A Kantor */
237e830fb1Kody A Kantoruint_t		spa_spare_poll_interval_seconds = 60 * 60 * 3;
238e830fb1Kody A Kantor
239e830fb1Kody A Kantor/*
2408671400Serapheim Dimitropoulos * Debugging aid that pauses spa_sync() towards the end.
2418671400Serapheim Dimitropoulos */
2428671400Serapheim Dimitropoulosboolean_t	zfs_pause_spa_sync = B_FALSE;
2438671400Serapheim Dimitropoulos
2448671400Serapheim Dimitropoulos/*
245990b485lling * ==========================================================================
246990b485lling * SPA properties routines
247990b485lling * ==========================================================================
248990b485lling */
251990b485lling * Add a (source=src, propname=propval) list to an nvlist.
252990b485lling */
2539d82f4fllingstatic void
254990b485llingspa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
255990b485lling    uint64_t intval, zprop_source_t src)
257990b485lling	const char *propname = zpool_prop_to_name(prop);
258990b485lling	nvlist_t *propval;
2609d82f4flling	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2619d82f4flling	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
2639d82f4flling	if (strval != NULL)
2649d82f4flling		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
2659d82f4flling	else
2669d82f4flling		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
2689d82f4flling	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
269990b485lling	nvlist_free(propval);
273990b485lling * Get property values from the spa configuration.
274990b485lling */
2759d82f4fllingstatic void
276990b485llingspa_prop_get_config(spa_t *spa, nvlist_t **nvp)
2784263d13George Wilson	vdev_t *rvd = spa->spa_root_vdev;
279ad135b5Christopher Siden	dsl_pool_t *pool = spa->spa_dsl_pool;
2802e4c998George Wilson	uint64_t size, alloc, cap, version;
281990b485lling	zprop_source_t src = ZPROP_SRC_NONE;
282c5904d1eschrock	spa_config_dirent_t *dp;
2832e4c998George Wilson	metaslab_class_t *mc = spa_normal_class(spa);
285e14bb32Jeff Bonwick	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
286e14bb32Jeff Bonwick
2874263d13George Wilson	if (rvd != NULL) {
288663207aDon Brady		alloc = metaslab_class_get_alloc(mc);
289663207aDon Brady		alloc += metaslab_class_get_alloc(spa_special_class(spa));
290663207aDon Brady		alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
291663207aDon Brady
292663207aDon Brady		size = metaslab_class_get_space(mc);
293663207aDon Brady		size += metaslab_class_get_space(spa_special_class(spa));
294663207aDon Brady		size += metaslab_class_get_space(spa_dedup_class(spa));
295663207aDon Brady
296379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
297379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
298485bbbfGeorge Wilson		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
299485bbbfGeorge Wilson		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
300485bbbfGeorge Wilson		    size - alloc, src);
3018671400Serapheim Dimitropoulos		spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
3028671400Serapheim Dimitropoulos		    spa->spa_checkpoint_info.sci_dspace, src);
3034263d13George Wilson
3042e4c998George Wilson		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
3052e4c998George Wilson		    metaslab_class_fragmentation(mc), src);
3062e4c998George Wilson		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
3072e4c998George Wilson		    metaslab_class_expandable_space(mc), src);
308f9af39bGeorge Wilson		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
309f9af39bGeorge Wilson		    (spa_mode(spa) == FREAD), src);
310379c004Eric Schrock
311485bbbfGeorge Wilson		cap = (size == 0) ? 0 : (alloc * 100 / size);
312379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
313379c004Eric Schrock
314b24ab67Jeff Bonwick		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
315b24ab67Jeff Bonwick		    ddt_get_pool_dedup_ratio(spa), src);
316b24ab67Jeff Bonwick
317379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
3184263d13George Wilson		    rvd->vdev_state, src);
319379c004Eric Schrock
320379c004Eric Schrock		version = spa_version(spa);
321379c004Eric Schrock		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
322379c004Eric Schrock			src = ZPROP_SRC_DEFAULT;
323379c004Eric Schrock		else
324379c004Eric Schrock			src = ZPROP_SRC_LOCAL;
325379c004Eric Schrock		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
326379c004Eric Schrock	}
328ad135b5Christopher Siden	if (pool != NULL) {
329ad135b5Christopher Siden		/*
330ad135b5Christopher Siden		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
331ad135b5Christopher Siden		 * when opening pools before this version freedir will be NULL.
332ad135b5Christopher Siden		 */
3337fd05acMatthew Ahrens		if (pool->dp_free_dir != NULL) {
334ad135b5Christopher Siden			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
335c137962Justin T. Gibbs			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
336c137962Justin T. Gibbs			    src);
337ad135b5Christopher Siden		} else {
338ad135b5Christopher Siden			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
339ad135b5Christopher Siden			    NULL, 0, src);
340ad135b5Christopher Siden		}
3417fd05acMatthew Ahrens
3427fd05acMatthew Ahrens		if (pool->dp_leak_dir != NULL) {
3437fd05acMatthew Ahrens			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
344c137962Justin T. Gibbs			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
345c137962Justin T. Gibbs			    src);
3467fd05acMatthew Ahrens		} else {
3477fd05acMatthew Ahrens			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
3487fd05acMatthew Ahrens			    NULL, 0, src);
3497fd05acMatthew Ahrens		}
350ad135b5Christopher Siden	}
351ad135b5Christopher Siden
3529d82f4flling	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
3548704186Dan McDonald	if (spa->spa_comment != NULL) {
3558704186Dan McDonald		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
3568704186Dan McDonald		    0, ZPROP_SRC_LOCAL);
3578704186Dan McDonald	}
3588704186Dan McDonald
3599d82f4flling	if (spa->spa_root != NULL)
3609d82f4flling		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
3619d82f4flling		    0, ZPROP_SRC_LOCAL);
363b515258Matthew Ahrens	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
364b515258Matthew Ahrens		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
365b515258Matthew Ahrens		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
366b515258Matthew Ahrens	} else {
367b515258Matthew Ahrens		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
368b515258Matthew Ahrens		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
369b515258Matthew Ahrens	}
370b515258Matthew Ahrens
37154811daToomas Soome	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
37254811daToomas Soome		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
37354811daToomas Soome		    DNODE_MAX_SIZE, ZPROP_SRC_NONE);
37454811daToomas Soome	} else {
37554811daToomas Soome		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
37654811daToomas Soome		    DNODE_MIN_SIZE, ZPROP_SRC_NONE);
37754811daToomas Soome	}
37854811daToomas Soome
379c5904d1eschrock	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
380c5904d1eschrock		if (dp->scd_path == NULL) {
3819d82f4flling			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
382c5904d1eschrock			    "none", 0, ZPROP_SRC_LOCAL);
383c5904d1eschrock		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
3849d82f4flling			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
385c5904d1eschrock			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
3862f8aaabeschrock		}
3872f8aaabeschrock	}
391990b485lling * Get zpool property values.
392990b485lling */
394990b485llingspa_prop_get(spa_t *spa, nvlist_t **nvp)
396b24ab67Jeff Bonwick	objset_t *mos = spa->spa_meta_objset;
397990b485lling	zap_cursor_t zc;
398990b485lling	zap_attribute_t za;
399990b485lling	int err;
4019d82f4flling	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
403e14bb32Jeff Bonwick	mutex_enter(&spa->spa_props_lock);
404e14bb32Jeff Bonwick
405990b485lling	/*
406990b485lling	 * Get properties from the spa config.
407990b485lling	 */
4089d82f4flling	spa_prop_get_config(spa, nvp);
410990b485lling	/* If no pool property object, no more prop to get. */
411afee20eGeorge Wilson	if (mos == NULL || spa->spa_pool_props_object == 0) {
412990b485lling		mutex_exit(&spa->spa_props_lock);
413990b485lling		return (0);
414990b485lling	}
416990b485lling	/*
417990b485lling	 * Get properties from the MOS pool property object.
418990b485lling	 */
419990b485lling	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
420990b485lling	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
421990b485lling	    zap_cursor_advance(&zc)) {
422990b485lling		uint64_t intval = 0;
423990b485lling		char *strval = NULL;
424990b485lling		zprop_source_t src = ZPROP_SRC_DEFAULT;
425990b485lling		zpool_prop_t prop;
4274ae5f5fAlan Somers		if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
428990b485lling			continue;
430990b485lling		switch (za.za_integer_length) {
431990b485lling		case 8:
432990b485lling			/* integer property */
433990b485lling			if (za.za_first_integer !=
434990b485lling			    zpool_prop_default_numeric(prop))
435990b485lling				src = ZPROP_SRC_LOCAL;
437990b485lling			if (prop == ZPOOL_PROP_BOOTFS) {
438990b485lling				dsl_pool_t *dp;
439990b485lling				dsl_dataset_t *ds = NULL;
441990b485lling				dp = spa_get_dsl(spa);
4423b2aab1Matthew Ahrens				dsl_pool_config_enter(dp, FTAG);
443094e47eGeorge Wilson				err = dsl_dataset_hold_obj(dp,
444094e47eGeorge Wilson				    za.za_first_integer, FTAG, &ds);
445094e47eGeorge Wilson				if (err != 0) {
4463b2aab1Matthew Ahrens					dsl_pool_config_exit(dp, FTAG);
447990b485lling					break;
448990b485lling				}
4509adfa60Matthew Ahrens				strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
451990b485lling				    KM_SLEEP);
452990b485lling				dsl_dataset_name(ds, strval);
453745cd3cmaybee				dsl_dataset_rele(ds, FTAG);
4543b2aab1Matthew Ahrens				dsl_pool_config_exit(dp, FTAG);
455990b485lling			} else {
456990b485lling				strval = NULL;
457990b485lling				intval = za.za_first_integer;
458990b485lling			}
4609d82f4flling			spa_prop_add_list(*nvp, prop, strval, intval, src);
462990b485lling			if (strval != NULL)
4639adfa60Matthew Ahrens				kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
465990b485lling			break;
467990b485lling		case 1:
468990b485lling			/* string property */
469990b485lling			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
470990b485lling			err = zap_lookup(mos, spa->spa_pool_props_object,
471990b485lling			    za.za_name, 1, za.za_num_integers, strval);
472990b485lling			if (err) {
473990b485lling				kmem_free(strval, za.za_num_integers);
474990b485lling				break;
475990b485lling			}
4769d82f4flling			spa_prop_add_list(*nvp, prop, strval, 0, src);
477990b485lling			kmem_free(strval, za.za_num_integers);
478990b485lling			break;
480990b485lling		default:
481990b485lling			break;
482990b485lling		}
483990b485lling	}
484990b485lling	zap_cursor_fini(&zc);
485990b485lling	mutex_exit(&spa->spa_props_lock);
487990b485lling	if (err && err != ENOENT) {
488990b485lling		nvlist_free(*nvp);
4899d82f4flling		*nvp = NULL;
490990b485lling		return (err);
491990b485lling	}
493990b485lling	return (0);
497990b485lling * Validate the given pool properties nvlist and modify the list
498990b485lling * for the property values to be set.
499990b485lling */
500990b485llingstatic int
501990b485llingspa_prop_validate(spa_t *spa, nvlist_t *props)
503990b485lling	nvpair_t *elem;
504990b485lling	int error = 0, reset_bootfs = 0;
505d5285caGeorge Wilson	uint64_t objnum = 0;
506ad135b5Christopher Siden	boolean_t has_feature = B_FALSE;
508990b485lling	elem = NULL;
509990b485lling	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
510990b485lling		uint64_t intval;
511ad135b5Christopher Siden		char *strval, *slash, *check, *fname;
512ad135b5Christopher Siden		const char *propname = nvpair_name(elem);
513ad135b5Christopher Siden		zpool_prop_t prop = zpool_name_to_prop(propname);
514ad135b5Christopher Siden
515ad135b5Christopher Siden		switch (prop) {
5164ae5f5fAlan Somers		case ZPOOL_PROP_INVAL:
517ad135b5Christopher Siden			if (!zpool_prop_feature(propname)) {
518be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
519ad135b5Christopher Siden				break;
520ad135b5Christopher Siden			}
522ad135b5Christopher Siden			/*
523ad135b5Christopher Siden			 * Sanitize the input.
524ad135b5Christopher Siden			 */
525ad135b5Christopher Siden			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
526be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
527ad135b5Christopher Siden				break;
528ad135b5Christopher Siden			}
530ad135b5Christopher Siden			if (nvpair_value_uint64(elem, &intval) != 0) {
531be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
532ad135b5Christopher Siden				break;
533ad135b5Christopher Siden			}
534ad135b5Christopher Siden
535ad135b5Christopher Siden			if (intval != 0) {
536be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
537ad135b5Christopher Siden				break;
538ad135b5Christopher Siden			}
539ad135b5Christopher Siden
540ad135b5Christopher Siden			fname = strchr(propname, '@') + 1;
541ad135b5Christopher Siden			if (zfeature_lookup_name(fname, NULL) != 0) {
542be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
543ad135b5Christopher Siden				break;
544ad135b5Christopher Siden			}
545ad135b5Christopher Siden
546ad135b5Christopher Siden			has_feature = B_TRUE;
547ad135b5Christopher Siden			break;
549990b485lling		case ZPOOL_PROP_VERSION:
550990b485lling			error = nvpair_value_uint64(elem, &intval);
551990b485lling			if (!error &&
552ad135b5Christopher Siden			    (intval < spa_version(spa) ||
553ad135b5Christopher Siden			    intval > SPA_VERSION_BEFORE_FEATURES ||
554ad135b5Christopher Siden			    has_feature))
555be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
556990b485lling			break;
558990b485lling		case ZPOOL_PROP_DELEGATION:
559990b485lling		case ZPOOL_PROP_AUTOREPLACE:
560d5b5bb2Rich Morris		case ZPOOL_PROP_LISTSNAPS:
561573ca77George Wilson		case ZPOOL_PROP_AUTOEXPAND:
562084fd14Brian Behlendorf		case ZPOOL_PROP_AUTOTRIM:
563990b485lling			error = nvpair_value_uint64(elem, &intval);
564990b485lling			if (!error && intval > 1)
565be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
566990b485lling			break;
568e0f1c0aOlaf Faaland		case ZPOOL_PROP_MULTIHOST:
569e0f1c0aOlaf Faaland			error = nvpair_value_uint64(elem, &intval);
570e0f1c0aOlaf Faaland			if (!error && intval > 1)
571e0f1c0aOlaf Faaland				error = SET_ERROR(EINVAL);
572e0f1c0aOlaf Faaland
573e0f1c0aOlaf Faaland			if (!error && !spa_get_hostid())
574e0f1c0aOlaf Faaland				error = SET_ERROR(ENOTSUP);
575e0f1c0aOlaf Faaland
576e0f1c0aOlaf Faaland			break;
577e0f1c0aOlaf Faaland
578990b485lling		case ZPOOL_PROP_BOOTFS:
57925f89eeJeff Bonwick			/*
58025f89eeJeff Bonwick			 * If the pool version is less than SPA_VERSION_BOOTFS,
58125f89eeJeff Bonwick			 * or the pool is still being created (version == 0),
58225f89eeJeff Bonwick			 * the bootfs property cannot be set.
58325f89eeJeff Bonwick			 */
584990b485lling			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
585be6fd75Matthew Ahrens				error = SET_ERROR(ENOTSUP);
586990b485lling				break;
587990b485lling			}
589990b485lling			/*
59015e6edfgw			 * Make sure the vdev config is bootable
591990b485lling			 */
59215e6edfgw			if (!vdev_is_bootable(spa->spa_root_vdev)) {
593be6fd75Matthew Ahrens				error = SET_ERROR(ENOTSUP);
594990b485lling				break;
595990b485lling			}
597990b485lling			reset_bootfs = 1;
599990b485lling			error = nvpair_value_string(elem, &strval);
601990b485lling			if (!error) {
602ad135b5Christopher Siden				objset_t *os;
603b515258Matthew Ahrens				uint64_t propval;
605990b485lling				if (strval == NULL || strval[0] == '\0') {
606990b485lling					objnum = zpool_prop_default_numeric(
607990b485lling					    ZPOOL_PROP_BOOTFS);
608990b485lling					break;
609990b485lling				}
611094e47eGeorge Wilson				error = dmu_objset_hold(strval, FTAG, &os);
612094e47eGeorge Wilson				if (error != 0)
613990b485lling					break;
615b515258Matthew Ahrens				/*
616b515258Matthew Ahrens				 * Must be ZPL, and its property settings
61754811daToomas Soome				 * must be supported.
618b515258Matthew Ahrens				 */
619503ad85Matthew Ahrens
620503ad85Matthew Ahrens				if (dmu_objset_type(os) != DMU_OST_ZFS) {
621be6fd75Matthew Ahrens					error = SET_ERROR(ENOTSUP);
6223b2aab1Matthew Ahrens				} else if ((error =
6233b2aab1Matthew Ahrens				    dsl_prop_get_int_ds(dmu_objset_ds(os),
62415e6edfgw				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
625b515258Matthew Ahrens				    &propval)) == 0 &&
626b515258Matthew Ahrens				    !BOOTFS_COMPRESS_VALID(propval)) {
627b515258Matthew Ahrens					error = SET_ERROR(ENOTSUP);
62815e6edfgw				} else {
62915e6edfgw					objnum = dmu_objset_id(os);
63015e6edfgw				}
631503ad85Matthew Ahrens				dmu_objset_rele(os, FTAG);
632990b485lling			}
633990b485lling			break;
634e14bb32Jeff Bonwick
6350a4e951gw		case ZPOOL_PROP_FAILUREMODE:
6360a4e951gw			error = nvpair_value_uint64(elem, &intval);
6370a4e951gw			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
6380a4e951gw			    intval > ZIO_FAILURE_MODE_PANIC))
639be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
6410a4e951gw			/*
6420a4e951gw			 * This is a special case which only occurs when
6430a4e951gw			 * the pool has completely failed. This allows
6440a4e951gw			 * the user to change the in-core failmode property
6450a4e951gw			 * without syncing it out to disk (I/Os might
6460a4e951gw			 * currently be blocked). We do this by returning
6470a4e951gw			 * EIO to the caller (spa_prop_set) to trick it
6480a4e951gw			 * into thinking we encountered a property validation
6490a4e951gw			 * error.
6500a4e951gw			 */
651e14bb32Jeff Bonwick			if (!error && spa_suspended(spa)) {
6520a4e951gw				spa->spa_failmode = intval;
653be6fd75Matthew Ahrens				error = SET_ERROR(EIO);
6540a4e951gw			}
6550a4e951gw			break;
6572f8aaabeschrock		case ZPOOL_PROP_CACHEFILE:
6582f8aaabeschrock			if ((error = nvpair_value_string(elem, &strval)) != 0)
6592f8aaabeschrock				break;
6612f8aaabeschrock			if (strval[0] == '\0')
6622f8aaabeschrock				break;
6642f8aaabeschrock			if (strcmp(strval, "none") == 0)
6652f8aaabeschrock				break;
6672f8aaabeschrock			if (strval[0] != '/') {
668be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
6692f8aaabeschrock				break;
6702f8aaabeschrock			}
6722f8aaabeschrock			slash = strrchr(strval, '/');
6732f8aaabeschrock			ASSERT(slash != NULL);
6752f8aaabeschrock			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
6762f8aaabeschrock			    strcmp(slash, "/..") == 0)
677be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
6782f8aaabeschrock			break;
679b24ab67Jeff Bonwick
6808704186Dan McDonald		case ZPOOL_PROP_COMMENT:
6818704186Dan McDonald			if ((error = nvpair_value_string(elem, &strval)) != 0)
6828704186Dan McDonald				break;
6838704186Dan McDonald			for (check = strval; *check != '\0'; check++) {
6848704186Dan McDonald				/*
6858704186Dan McDonald				 * The kernel doesn't have an easy isprint()
6868704186Dan McDonald				 * check.  For this kernel check, we merely
6878704186Dan McDonald				 * check ASCII apart from DEL.  Fix this if
6888704186Dan McDonald				 * there is an easy-to-use kernel isprint().
6898704186Dan McDonald				 */
6908704186Dan McDonald				if (*check >= 0x7f) {
691be6fd75Matthew Ahrens					error = SET_ERROR(EINVAL);
6928704186Dan McDonald					break;
6938704186Dan McDonald				}
6948704186Dan McDonald			}
6958704186Dan McDonald			if (strlen(strval) > ZPROP_MAX_COMMENT)
6968704186Dan McDonald				error = E2BIG;
6978704186Dan McDonald			break;
6988704186Dan McDonald
699b24ab67Jeff Bonwick		case ZPOOL_PROP_DEDUPDITTO:
700b24ab67Jeff Bonwick			if (spa_version(spa) < SPA_VERSION_DEDUP)
701be6fd75Matthew Ahrens				error = SET_ERROR(ENOTSUP);
702b24ab67Jeff Bonwick			else
703b24ab67Jeff Bonwick				error = nvpair_value_uint64(elem, &intval);
704b24ab67Jeff Bonwick			if (error == 0 &&
705b24ab67Jeff Bonwick			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
706be6fd75Matthew Ahrens				error = SET_ERROR(EINVAL);
707b24ab67Jeff Bonwick			break;
708990b485lling		}
710990b485lling		if (error)
711990b485lling			break;
712990b485lling	}
714990b485lling	if (!error && reset_bootfs) {
715990b485lling		error = nvlist_remove(props,
716990b485lling		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
718990b485lling		if (!error) {
719990b485lling			error = nvlist_add_uint64(props,
720990b485lling			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
721990b485lling		}
722990b485lling	}
724990b485lling	return (error);
727379c004Eric Schrockvoid
728379c004Eric Schrockspa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
729379c004Eric Schrock{
730379c004Eric Schrock	char *cachefile;
731379c004Eric Schrock	spa_config_dirent_t *dp;
732379c004Eric Schrock
733379c004Eric Schrock	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
734379c004Eric Schrock	    &cachefile) != 0)
735379c004Eric Schrock		return;
736379c004Eric Schrock
737379c004Eric Schrock	dp = kmem_alloc(sizeof (spa_config_dirent_t),
738379c004Eric Schrock	    KM_SLEEP);
739379c004Eric Schrock
740379c004Eric Schrock	if (cachefile[0] == '\0')
741379c004Eric Schrock		dp->scd_path = spa_strdup(spa_config_path);
742379c004Eric Schrock	else if (strcmp(cachefile, "none") == 0)
743379c004Eric Schrock		dp->scd_path = NULL;
744379c004Eric Schrock	else
745379c004Eric Schrock		dp->scd_path = spa_strdup(cachefile);
746379c004Eric Schrock
747379c004Eric Schrock	list_insert_head(&spa->spa_config_list, dp);
748379c004Eric Schrock	if (need_sync)
749379c004Eric Schrock		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
750379c004Eric Schrock}
751379c004Eric Schrock
753990b485llingspa_prop_set(spa_t *spa, nvlist_t *nvp)
755990b485lling	int error;
756ad135b5Christopher Siden	nvpair_t *elem = NULL;
757379c004Eric Schrock	boolean_t need_sync = B_FALSE;
759990b485lling	if ((error = spa_prop_validate(spa, nvp)) != 0)
760990b485lling		return (error);
762379c004Eric Schrock	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
763ad135b5Christopher Siden		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
764379c004Eric Schrock
765f9af39bGeorge Wilson		if (prop == ZPOOL_PROP_CACHEFILE ||
766f9af39bGeorge Wilson		    prop == ZPOOL_PROP_ALTROOT ||
767f9af39bGeorge Wilson		    prop == ZPOOL_PROP_READONLY)
768379c004Eric Schrock			continue;
769379c004Eric Schrock
7704ae5f5fAlan Somers		if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
771ad135b5Christopher Siden			uint64_t ver;
772ad135b5Christopher Siden
773ad135b5Christopher Siden			if (prop == ZPOOL_PROP_VERSION) {
774ad135b5Christopher Siden				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
775ad135b5Christopher Siden			} else {