xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa.c (revision bb78539e)
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 /*
23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2011, 2019 by Delphix. All rights reserved.
25  * Copyright (c) 2015, Nexenta Systems, Inc.  All rights reserved.
26  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27  * Copyright 2013 Saso Kiselkov. All rights reserved.
28  * Copyright (c) 2014 Integros [integros.com]
29  * Copyright 2016 Toomas Soome <tsoome@me.com>
30  * Copyright (c) 2017, 2019, Datto Inc. All rights reserved.
31  * Copyright 2019 Joyent, Inc.
32  * Copyright (c) 2017, Intel Corporation.
33  * Copyright 2020 Joshua M. Clulow <josh@sysmgr.org>
34  * Copyright 2021 OmniOS Community Edition (OmniOSce) Association.
35  * Copyright 2022 Oxide Computer Company
36  * Copyright 2023 MNX Cloud, Inc.
37  */
38 
39 /*
40  * SPA: Storage Pool Allocator
41  *
42  * This file contains all the routines used when modifying on-disk SPA state.
43  * This includes opening, importing, destroying, exporting a pool, and syncing a
44  * pool.
45  */
46 
47 #include <sys/zfs_context.h>
48 #include <sys/fm/fs/zfs.h>
49 #include <sys/spa_impl.h>
50 #include <sys/zio.h>
51 #include <sys/zio_checksum.h>
52 #include <sys/dmu.h>
53 #include <sys/dmu_tx.h>
54 #include <sys/zap.h>
55 #include <sys/zil.h>
56 #include <sys/ddt.h>
57 #include <sys/vdev_impl.h>
58 #include <sys/vdev_removal.h>
59 #include <sys/vdev_indirect_mapping.h>
60 #include <sys/vdev_indirect_births.h>
61 #include <sys/vdev_initialize.h>
62 #include <sys/vdev_trim.h>
63 #include <sys/metaslab.h>
64 #include <sys/metaslab_impl.h>
65 #include <sys/mmp.h>
66 #include <sys/uberblock_impl.h>
67 #include <sys/txg.h>
68 #include <sys/avl.h>
69 #include <sys/bpobj.h>
70 #include <sys/dmu_traverse.h>
71 #include <sys/dmu_objset.h>
72 #include <sys/unique.h>
73 #include <sys/dsl_pool.h>
74 #include <sys/dsl_dataset.h>
75 #include <sys/dsl_dir.h>
76 #include <sys/dsl_prop.h>
77 #include <sys/dsl_synctask.h>
78 #include <sys/fs/zfs.h>
79 #include <sys/arc.h>
80 #include <sys/callb.h>
81 #include <sys/systeminfo.h>
82 #include <sys/spa_boot.h>
83 #include <sys/zfs_ioctl.h>
84 #include <sys/dsl_scan.h>
85 #include <sys/zfeature.h>
86 #include <sys/dsl_destroy.h>
87 #include <sys/abd.h>
88 
89 #ifdef	_KERNEL
90 #include <sys/bootprops.h>
91 #include <sys/callb.h>
92 #include <sys/cpupart.h>
93 #include <sys/pool.h>
94 #include <sys/sysdc.h>
95 #include <sys/zone.h>
96 #endif	/* _KERNEL */
97 
98 #include "zfs_prop.h"
99 #include "zfs_comutil.h"
100 
101 /*
102  * The interval, in seconds, at which failed configuration cache file writes
103  * should be retried.
104  */
105 int zfs_ccw_retry_interval = 300;
106 
107 typedef enum zti_modes {
108 	ZTI_MODE_FIXED,			/* value is # of threads (min 1) */
109 	ZTI_MODE_BATCH,			/* cpu-intensive; value is ignored */
110 	ZTI_MODE_NULL,			/* don't create a taskq */
111 	ZTI_NMODES
112 } zti_modes_t;
113 
114 #define	ZTI_P(n, q)	{ ZTI_MODE_FIXED, (n), (q) }
115 #define	ZTI_BATCH	{ ZTI_MODE_BATCH, 0, 1 }
116 #define	ZTI_NULL	{ ZTI_MODE_NULL, 0, 0 }
117 
118 #define	ZTI_N(n)	ZTI_P(n, 1)
119 #define	ZTI_ONE		ZTI_N(1)
120 
121 typedef struct zio_taskq_info {
122 	zti_modes_t zti_mode;
123 	uint_t zti_value;
124 	uint_t zti_count;
125 } zio_taskq_info_t;
126 
127 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
128 	"issue", "issue_high", "intr", "intr_high"
129 };
130 
131 /*
132  * This table defines the taskq settings for each ZFS I/O type. When
133  * initializing a pool, we use this table to create an appropriately sized
134  * taskq. Some operations are low volume and therefore have a small, static
135  * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
136  * macros. Other operations process a large amount of data; the ZTI_BATCH
137  * macro causes us to create a taskq oriented for throughput. Some operations
138  * are so high frequency and short-lived that the taskq itself can become a
139  * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
140  * additional degree of parallelism specified by the number of threads per-
141  * taskq and the number of taskqs; when dispatching an event in this case, the
142  * particular taskq is chosen at random.
143  *
144  * The different taskq priorities are to handle the different contexts (issue
145  * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
146  * need to be handled with minimum delay.
147  */
148 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
149 	/* ISSUE	ISSUE_HIGH	INTR		INTR_HIGH */
150 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* NULL */
151 	{ ZTI_N(8),	ZTI_NULL,	ZTI_P(12, 8),	ZTI_NULL }, /* READ */
152 	{ ZTI_BATCH,	ZTI_N(5),	ZTI_N(8),	ZTI_N(5) }, /* WRITE */
153 	{ ZTI_P(12, 8),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* FREE */
154 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* CLAIM */
155 	{ ZTI_ONE,	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* IOCTL */
156 	{ ZTI_N(4),	ZTI_NULL,	ZTI_ONE,	ZTI_NULL }, /* TRIM */
157 };
158 
159 static void spa_sync_version(void *arg, dmu_tx_t *tx);
160 static void spa_sync_props(void *arg, dmu_tx_t *tx);
161 static boolean_t spa_has_active_shared_spare(spa_t *spa);
162 static int spa_load_impl(spa_t *spa, spa_import_type_t type, char **ereport);
163 static void spa_vdev_resilver_done(spa_t *spa);
164 
165 uint_t		zio_taskq_batch_pct = 75;	/* 1 thread per cpu in pset */
166 id_t		zio_taskq_psrset_bind = PS_NONE;
167 boolean_t	zio_taskq_sysdc = B_TRUE;	/* use SDC scheduling class */
168 uint_t		zio_taskq_basedc = 80;		/* base duty cycle */
169 
170 boolean_t	spa_create_process = B_TRUE;	/* no process ==> no sysdc */
171 extern int	zfs_sync_pass_deferred_free;
172 
173 /*
174  * Report any spa_load_verify errors found, but do not fail spa_load.
175  * This is used by zdb to analyze non-idle pools.
176  */
177 boolean_t	spa_load_verify_dryrun = B_FALSE;
178 
179 /*
180  * This (illegal) pool name is used when temporarily importing a spa_t in order
181  * to get the vdev stats associated with the imported devices.
182  */
183 #define	TRYIMPORT_NAME	"$import"
184 
185 /*
186  * For debugging purposes: print out vdev tree during pool import.
187  */
188 boolean_t	spa_load_print_vdev_tree = B_FALSE;
189 
190 /*
191  * A non-zero value for zfs_max_missing_tvds means that we allow importing
192  * pools with missing top-level vdevs. This is strictly intended for advanced
193  * pool recovery cases since missing data is almost inevitable. Pools with
194  * missing devices can only be imported read-only for safety reasons, and their
195  * fail-mode will be automatically set to "continue".
196  *
197  * With 1 missing vdev we should be able to import the pool and mount all
198  * datasets. User data that was not modified after the missing device has been
199  * added should be recoverable. This means that snapshots created prior to the
200  * addition of that device should be completely intact.
201  *
202  * With 2 missing vdevs, some datasets may fail to mount since there are
203  * dataset statistics that are stored as regular metadata. Some data might be
204  * recoverable if those vdevs were added recently.
205  *
206  * With 3 or more missing vdevs, the pool is severely damaged and MOS entries
207  * may be missing entirely. Chances of data recovery are very low. Note that
208  * there are also risks of performing an inadvertent rewind as we might be
209  * missing all the vdevs with the latest uberblocks.
210  */
211 uint64_t	zfs_max_missing_tvds = 0;
212 
213 /*
214  * The parameters below are similar to zfs_max_missing_tvds but are only
215  * intended for a preliminary open of the pool with an untrusted config which
216  * might be incomplete or out-dated.
217  *
218  * We are more tolerant for pools opened from a cachefile since we could have
219  * an out-dated cachefile where a device removal was not registered.
220  * We could have set the limit arbitrarily high but in the case where devices
221  * are really missing we would want to return the proper error codes; we chose
222  * SPA_DVAS_PER_BP - 1 so that some copies of the MOS would still be available
223  * and we get a chance to retrieve the trusted config.
224  */
225 uint64_t	zfs_max_missing_tvds_cachefile = SPA_DVAS_PER_BP - 1;
226 
227 /*
228  * In the case where config was assembled by scanning device paths (/dev/dsks
229  * by default) we are less tolerant since all the existing devices should have
230  * been detected and we want spa_load to return the right error codes.
231  */
232 uint64_t	zfs_max_missing_tvds_scan = 0;
233 
234 /*
235  * Interval in seconds at which to poll spare vdevs for health.
236  * Setting this to zero disables spare polling.
237  * Set to three hours by default.
238  */
239 uint_t		spa_spare_poll_interval_seconds = 60 * 60 * 3;
240 
241 /*
242  * Debugging aid that pauses spa_sync() towards the end.
243  */
244 boolean_t	zfs_pause_spa_sync = B_FALSE;
245 
246 /*
247  * ==========================================================================
248  * SPA properties routines
249  * ==========================================================================
250  */
251 
252 /*
253  * Add a (source=src, propname=propval) list to an nvlist.
254  */
255 static void
spa_prop_add_list(nvlist_t * nvl,zpool_prop_t prop,char * strval,uint64_t intval,zprop_source_t src)256 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
257     uint64_t intval, zprop_source_t src)
258 {
259 	const char *propname = zpool_prop_to_name(prop);
260 	nvlist_t *propval;
261 
262 	VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
263 	VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
264 
265 	if (strval != NULL)
266 		VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
267 	else
268 		VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
269 
270 	VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
271 	nvlist_free(propval);
272 }
273 
274 /*
275  * Get property values from the spa configuration.
276  */
277 static void
spa_prop_get_config(spa_t * spa,nvlist_t ** nvp)278 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
279 {
280 	vdev_t *rvd = spa->spa_root_vdev;
281 	dsl_pool_t *pool = spa->spa_dsl_pool;
282 	uint64_t size, alloc, cap, version;
283 	zprop_source_t src = ZPROP_SRC_NONE;
284 	spa_config_dirent_t *dp;
285 	metaslab_class_t *mc = spa_normal_class(spa);
286 
287 	ASSERT(MUTEX_HELD(&spa->spa_props_lock));
288 
289 	if (rvd != NULL) {
290 		alloc = metaslab_class_get_alloc(mc);
291 		alloc += metaslab_class_get_alloc(spa_special_class(spa));
292 		alloc += metaslab_class_get_alloc(spa_dedup_class(spa));
293 
294 		size = metaslab_class_get_space(mc);
295 		size += metaslab_class_get_space(spa_special_class(spa));
296 		size += metaslab_class_get_space(spa_dedup_class(spa));
297 
298 		spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
299 		spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
300 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
301 		spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
302 		    size - alloc, src);
303 		spa_prop_add_list(*nvp, ZPOOL_PROP_CHECKPOINT, NULL,
304 		    spa->spa_checkpoint_info.sci_dspace, src);
305 
306 		spa_prop_add_list(*nvp, ZPOOL_PROP_FRAGMENTATION, NULL,
307 		    metaslab_class_fragmentation(mc), src);
308 		spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL,
309 		    metaslab_class_expandable_space(mc), src);
310 		spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
311 		    (spa_mode(spa) == FREAD), src);
312 
313 		cap = (size == 0) ? 0 : (alloc * 100 / size);
314 		spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
315 
316 		spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
317 		    ddt_get_pool_dedup_ratio(spa), src);
318 
319 		spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
320 		    rvd->vdev_state, src);
321 
322 		version = spa_version(spa);
323 		if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
324 			src = ZPROP_SRC_DEFAULT;
325 		else
326 			src = ZPROP_SRC_LOCAL;
327 		spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
328 	}
329 
330 	if (pool != NULL) {
331 		/*
332 		 * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
333 		 * when opening pools before this version freedir will be NULL.
334 		 */
335 		if (pool->dp_free_dir != NULL) {
336 			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
337 			    dsl_dir_phys(pool->dp_free_dir)->dd_used_bytes,
338 			    src);
339 		} else {
340 			spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
341 			    NULL, 0, src);
342 		}
343 
344 		if (pool->dp_leak_dir != NULL) {
345 			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED, NULL,
346 			    dsl_dir_phys(pool->dp_leak_dir)->dd_used_bytes,
347 			    src);
348 		} else {
349 			spa_prop_add_list(*nvp, ZPOOL_PROP_LEAKED,
350 			    NULL, 0, src);
351 		}
352 	}
353 
354 	spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
355 
356 	if (spa->spa_comment != NULL) {
357 		spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
358 		    0, ZPROP_SRC_LOCAL);
359 	}
360 
361 	if (spa->spa_root != NULL)
362 		spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
363 		    0, ZPROP_SRC_LOCAL);
364 
365 	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
366 		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
367 		    MIN(zfs_max_recordsize, SPA_MAXBLOCKSIZE), ZPROP_SRC_NONE);
368 	} else {
369 		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXBLOCKSIZE, NULL,
370 		    SPA_OLD_MAXBLOCKSIZE, ZPROP_SRC_NONE);
371 	}
372 
373 	if (spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_DNODE)) {
374 		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
375 		    DNODE_MAX_SIZE, ZPROP_SRC_NONE);
376 	} else {
377 		spa_prop_add_list(*nvp, ZPOOL_PROP_MAXDNODESIZE, NULL,
378 		    DNODE_MIN_SIZE, ZPROP_SRC_NONE);
379 	}
380 
381 	if ((dp = list_head(&spa->spa_config_list)) != NULL) {
382 		if (dp->scd_path == NULL) {
383 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
384 			    "none", 0, ZPROP_SRC_LOCAL);
385 		} else if (strcmp(dp->scd_path, spa_config_path) != 0) {
386 			spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
387 			    dp->scd_path, 0, ZPROP_SRC_LOCAL);
388 		}
389 	}
390 }
391 
392 /*
393  * Get zpool property values.
394  */
395 int
spa_prop_get(spa_t * spa,nvlist_t ** nvp)396 spa_prop_get(spa_t *spa, nvlist_t **nvp)
397 {
398 	objset_t *mos = spa->spa_meta_objset;
399 	zap_cursor_t zc;
400 	zap_attribute_t za;
401 	int err;
402 
403 	VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
404 
405 	mutex_enter(&spa->spa_props_lock);
406 
407 	/*
408 	 * Get properties from the spa config.
409 	 */
410 	spa_prop_get_config(spa, nvp);
411 
412 	/* If no pool property object, no more prop to get. */
413 	if (mos == NULL || spa->spa_pool_props_object == 0) {
414 		mutex_exit(&spa->spa_props_lock);
415 		return (0);
416 	}
417 
418 	/*
419 	 * Get properties from the MOS pool property object.
420 	 */
421 	for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
422 	    (err = zap_cursor_retrieve(&zc, &za)) == 0;
423 	    zap_cursor_advance(&zc)) {
424 		uint64_t intval = 0;
425 		char *strval = NULL;
426 		zprop_source_t src = ZPROP_SRC_DEFAULT;
427 		zpool_prop_t prop;
428 
429 		if ((prop = zpool_name_to_prop(za.za_name)) == ZPOOL_PROP_INVAL)
430 			continue;
431 
432 		switch (za.za_integer_length) {
433 		case 8:
434 			/* integer property */
435 			if (za.za_first_integer !=
436 			    zpool_prop_default_numeric(prop))
437 				src = ZPROP_SRC_LOCAL;
438 
439 			if (prop == ZPOOL_PROP_BOOTFS) {
440 				dsl_pool_t *dp;
441 				dsl_dataset_t *ds = NULL;
442 
443 				dp = spa_get_dsl(spa);
444 				dsl_pool_config_enter(dp, FTAG);
445 				err = dsl_dataset_hold_obj(dp,
446 				    za.za_first_integer, FTAG, &ds);
447 				if (err != 0) {
448 					dsl_pool_config_exit(dp, FTAG);
449 					break;
450 				}
451 
452 				strval = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN,
453 				    KM_SLEEP);
454 				dsl_dataset_name(ds, strval);
455 				dsl_dataset_rele(ds, FTAG);
456 				dsl_pool_config_exit(dp, FTAG);
457 			} else {
458 				strval = NULL;
459 				intval = za.za_first_integer;
460 			}
461 
462 			spa_prop_add_list(*nvp, prop, strval, intval, src);
463 
464 			if (strval != NULL)
465 				kmem_free(strval, ZFS_MAX_DATASET_NAME_LEN);
466 
467 			break;
468 
469 		case 1:
470 			/* string property */
471 			strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
472 			err = zap_lookup(mos, spa->spa_pool_props_object,
473 			    za.za_name, 1, za.za_num_integers, strval);
474 			if (err) {
475 				kmem_free(strval, za.za_num_integers);
476 				break;
477 			}
478 			spa_prop_add_list(*nvp, prop, strval, 0, src);
479 			kmem_free(strval, za.za_num_integers);
480 			break;
481 
482 		default:
483 			break;
484 		}
485 	}
486 	zap_cursor_fini(&zc);
487 	mutex_exit(&spa->spa_props_lock);
488 out:
489 	if (err && err != ENOENT) {
490 		nvlist_free(*nvp);
491 		*nvp = NULL;
492 		return (err);
493 	}
494 
495 	return (0);
496 }
497 
498 /*
499  * Validate the given pool properties nvlist and modify the list
500  * for the property values to be set.
501  */
502 static int
spa_prop_validate(spa_t * spa,nvlist_t * props)503 spa_prop_validate(spa_t *spa, nvlist_t *props)
504 {
505 	nvpair_t *elem;
506 	int error = 0, reset_bootfs = 0;
507 	uint64_t objnum = 0;
508 	boolean_t has_feature = B_FALSE;
509 
510 	elem = NULL;
511 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
512 		uint64_t intval;
513 		char *strval, *slash, *check, *fname;
514 		const char *propname = nvpair_name(elem);
515 		zpool_prop_t prop = zpool_name_to_prop(propname);
516 
517 		switch (prop) {
518 		case ZPOOL_PROP_INVAL:
519 			if (!zpool_prop_feature(propname)) {
520 				error = SET_ERROR(EINVAL);
521 				break;
522 			}
523 
524 			/*
525 			 * Sanitize the input.
526 			 */
527 			if (nvpair_type(elem) != DATA_TYPE_UINT64) {
528 				error = SET_ERROR(EINVAL);
529 				break;
530 			}
531 
532 			if (nvpair_value_uint64(elem, &intval) != 0) {
533 				error = SET_ERROR(EINVAL);
534 				break;
535 			}
536 
537 			if (intval != 0) {
538 				error = SET_ERROR(EINVAL);
539 				break;
540 			}
541 
542 			fname = strchr(propname, '@') + 1;
543 			if (zfeature_lookup_name(fname, NULL) != 0) {
544 				error = SET_ERROR(EINVAL);
545 				break;
546 			}
547 
548 			has_feature = B_TRUE;
549 			break;
550 
551 		case ZPOOL_PROP_VERSION:
552 			error = nvpair_value_uint64(elem, &intval);
553 			if (!error &&
554 			    (intval < spa_version(spa) ||
555 			    intval > SPA_VERSION_BEFORE_FEATURES ||
556 			    has_feature))
557 				error = SET_ERROR(EINVAL);
558 			break;
559 
560 		case ZPOOL_PROP_DELEGATION:
561 		case ZPOOL_PROP_AUTOREPLACE:
562 		case ZPOOL_PROP_LISTSNAPS:
563 		case ZPOOL_PROP_AUTOEXPAND:
564 		case ZPOOL_PROP_AUTOTRIM:
565 			error = nvpair_value_uint64(elem, &intval);
566 			if (!error && intval > 1)
567 				error = SET_ERROR(EINVAL);
568 			break;
569 
570 		case ZPOOL_PROP_MULTIHOST:
571 			error = nvpair_value_uint64(elem, &intval);
572 			if (!error && intval > 1)
573 				error = SET_ERROR(EINVAL);
574 
575 			if (!error && !spa_get_hostid())
576 				error = SET_ERROR(ENOTSUP);
577 
578 			break;
579 
580 		case ZPOOL_PROP_BOOTFS:
581 			/*
582 			 * If the pool version is less than SPA_VERSION_BOOTFS,
583 			 * or the pool is still being created (version == 0),
584 			 * the bootfs property cannot be set.
585 			 */
586 			if (spa_version(spa) < SPA_VERSION_BOOTFS) {
587 				error = SET_ERROR(ENOTSUP);
588 				break;
589 			}
590 
591 			/*
592 			 * Make sure the vdev config is bootable
593 			 */
594 			if (!vdev_is_bootable(spa->spa_root_vdev)) {
595 				error = SET_ERROR(ENOTSUP);
596 				break;
597 			}
598 
599 			reset_bootfs = 1;
600 
601 			error = nvpair_value_string(elem, &strval);
602 
603 			if (!error) {
604 				objset_t *os;
605 				uint64_t propval;
606 
607 				if (strval == NULL || strval[0] == '\0') {
608 					objnum = zpool_prop_default_numeric(
609 					    ZPOOL_PROP_BOOTFS);
610 					break;
611 				}
612 
613 				error = dmu_objset_hold(strval, FTAG, &os);
614 				if (error != 0)
615 					break;
616 
617 				/*
618 				 * Must be ZPL, and its property settings
619 				 * must be supported.
620 				 */
621 
622 				if (dmu_objset_type(os) != DMU_OST_ZFS) {
623 					error = SET_ERROR(ENOTSUP);
624 				} else if ((error =
625 				    dsl_prop_get_int_ds(dmu_objset_ds(os),
626 				    zfs_prop_to_name(ZFS_PROP_COMPRESSION),
627 				    &propval)) == 0 &&
628 				    !BOOTFS_COMPRESS_VALID(propval)) {
629 					error = SET_ERROR(ENOTSUP);
630 				} else {
631 					objnum = dmu_objset_id(os);
632 				}
633 				dmu_objset_rele(os, FTAG);
634 			}
635 			break;
636 
637 		case ZPOOL_PROP_FAILUREMODE:
638 			error = nvpair_value_uint64(elem, &intval);
639 			if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
640 			    intval > ZIO_FAILURE_MODE_PANIC))
641 				error = SET_ERROR(EINVAL);
642 
643 			/*
644 			 * This is a special case which only occurs when
645 			 * the pool has completely failed. This allows
646 			 * the user to change the in-core failmode property
647 			 * without syncing it out to disk (I/Os might
648 			 * currently be blocked). We do this by returning
649 			 * EIO to the caller (spa_prop_set) to trick it
650 			 * into thinking we encountered a property validation
651 			 * error.
652 			 */
653 			if (!error && spa_suspended(spa)) {
654 				spa->spa_failmode = intval;
655 				error = SET_ERROR(EIO);
656 			}
657 			break;
658 
659 		case ZPOOL_PROP_CACHEFILE:
660 			if ((error = nvpair_value_string(elem, &strval)) != 0)
661 				break;
662 
663 			if (strval[0] == '\0')
664 				break;
665 
666 			if (strcmp(strval, "none") == 0)
667 				break;
668 
669 			if (strval[0] != '/') {
670 				error = SET_ERROR(EINVAL);
671 				break;
672 			}
673 
674 			slash = strrchr(strval, '/');
675 			ASSERT(slash != NULL);
676 
677 			if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
678 			    strcmp(slash, "/..") == 0)
679 				error = SET_ERROR(EINVAL);
680 			break;
681 
682 		case ZPOOL_PROP_COMMENT:
683 			if ((error = nvpair_value_string(elem, &strval)) != 0)
684 				break;
685 			for (check = strval; *check != '\0'; check++) {
686 				/*
687 				 * The kernel doesn't have an easy isprint()
688 				 * check.  For this kernel check, we merely
689 				 * check ASCII apart from DEL.  Fix this if
690 				 * there is an easy-to-use kernel isprint().
691 				 */
692 				if (*check >= 0x7f) {
693 					error = SET_ERROR(EINVAL);
694 					break;
695 				}
696 			}
697 			if (strlen(strval) > ZPROP_MAX_COMMENT)
698 				error = E2BIG;
699 			break;
700 
701 		case ZPOOL_PROP_DEDUPDITTO:
702 			if (spa_version(spa) < SPA_VERSION_DEDUP)
703 				error = SET_ERROR(ENOTSUP);
704 			else
705 				error = nvpair_value_uint64(elem, &intval);
706 			if (error == 0 &&
707 			    intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
708 				error = SET_ERROR(EINVAL);
709 			break;
710 		}
711 
712 		if (error)
713 			break;
714 	}
715 
716 	if (!error && reset_bootfs) {
717 		error = nvlist_remove(props,
718 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
719 
720 		if (!error) {
721 			error = nvlist_add_uint64(props,
722 			    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
723 		}
724 	}
725 
726 	return (error);
727 }
728 
729 void
spa_configfile_set(spa_t * spa,nvlist_t * nvp,boolean_t need_sync)730 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
731 {
732 	char *cachefile;
733 	spa_config_dirent_t *dp;
734 
735 	if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
736 	    &cachefile) != 0)
737 		return;
738 
739 	dp = kmem_alloc(sizeof (spa_config_dirent_t),
740 	    KM_SLEEP);
741 
742 	if (cachefile[0] == '\0')
743 		dp->scd_path = spa_strdup(spa_config_path);
744 	else if (strcmp(cachefile, "none") == 0)
745 		dp->scd_path = NULL;
746 	else
747 		dp->scd_path = spa_strdup(cachefile);
748 
749 	list_insert_head(&spa->spa_config_list, dp);
750 	if (need_sync)
751 		spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
752 }
753 
754 int
spa_prop_set(spa_t * spa,nvlist_t * nvp)755 spa_prop_set(spa_t *spa, nvlist_t *nvp)
756 {
757 	int error;
758 	nvpair_t *elem = NULL;
759 	boolean_t need_sync = B_FALSE;
760 
761 	if ((error = spa_prop_validate(spa, nvp)) != 0)
762 		return (error);
763 
764 	while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
765 		zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
766 
767 		if (prop == ZPOOL_PROP_CACHEFILE ||
768 		    prop == ZPOOL_PROP_ALTROOT ||
769 		    prop == ZPOOL_PROP_READONLY)
770 			continue;
771 
772 		if (prop == ZPOOL_PROP_VERSION || prop == ZPOOL_PROP_INVAL) {
773 			uint64_t ver;
774 
775 			if (prop == ZPOOL_PROP_VERSION) {
776 				VERIFY(nvpair_value_uint64(elem, &ver) == 0);
777 			} else {
778 				ASSERT(zpool_prop_feature(nvpair_name(elem)));
779 				ver = SPA_VERSION_FEATURES;
780 				need_sync = B_TRUE;
781 			}
782 
783 			/* Save time if the version is already set. */
784 			if (ver == spa_version(spa))
785 				continue;
786 
787 			/*
788 			 * In addition to the pool directory object, we might
789 			 * create the pool properties object, the features for
790 			 * read object, the features for write object, or the
791 			 * feature descriptions object.
792 			 */
793 			error = dsl_sync_task(spa->spa_name, NULL,
794 			    spa_sync_version, &ver,
795 			    6, ZFS_SPACE_CHECK_RESERVED);
796 			if (error)
797 				return (error);
798 			continue;
799 		}
800 
801 		need_sync = B_TRUE;
802 		break;
803 	}
804 
805 	if (need_sync) {
806 		return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
807 		    nvp, 6, ZFS_SPACE_CHECK_RESERVED));
808 	}
809 
810 	return (0);
811 }
812 
813 /*
814  * If the bootfs property value is dsobj, clear it.
815  */
816 void
spa_prop_clear_bootfs(spa_t * spa,uint64_t dsobj,dmu_tx_t * tx)817 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
818 {
819 	if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
820 		VERIFY(zap_remove(spa->spa_meta_objset,
821 		    spa->spa_pool_props_object,
822 		    zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
823 		spa->spa_bootfs = 0;
824 	}
825 }
826 
827 /*ARGSUSED*/
828 static int
spa_change_guid_check(void * arg,dmu_tx_t * tx)829 spa_change_guid_check(void *arg, dmu_tx_t *tx)
830 {
831 	uint64_t *newguid = arg;
832 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
833 	vdev_t *rvd = spa->spa_root_vdev;
834 	uint64_t vdev_state;
835 
836 	if (spa_feature_is_active(spa, SPA_FEATURE_POOL_CHECKPOINT)) {
837 		int error = (spa_has_checkpoint(spa)) ?
838 		    ZFS_ERR_CHECKPOINT_EXISTS : ZFS_ERR_DISCARDING_CHECKPOINT;
839 		return (SET_ERROR(error));
840 	}
841 
842 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
843 	vdev_state = rvd->vdev_state;
844 	spa_config_exit(spa, SCL_STATE, FTAG);
845 
846 	if (vdev_state != VDEV_STATE_HEALTHY)
847 		return (SET_ERROR(ENXIO));
848 
849 	ASSERT3U(spa_guid(spa), !=, *newguid);
850 
851 	return (0);
852 }
853 
854 static void
spa_change_guid_sync(void * arg,dmu_tx_t * tx)855 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
856 {
857 	uint64_t *newguid = arg;
858 	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
859 	uint64_t oldguid;
860 	vdev_t *rvd = spa->spa_root_vdev;
861 
862 	oldguid = spa_guid(spa);
863 
864 	spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
865 	rvd->vdev_guid = *newguid;
866 	rvd->vdev_guid_sum += (*newguid - oldguid);
867 	vdev_config_dirty(rvd);
868 	spa_config_exit(spa, SCL_STATE, FTAG);
869 
870 	spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
871 	    oldguid, *newguid);
872 }
873 
874 /*
875  * Change the GUID for the pool.  This is done so that we can later
876  * re-import a pool built from a clone of our own vdevs.  We will modify
877  * the root vdev's guid, our own pool guid, and then mark all of our
878  * vdevs dirty.  Note that we must make sure that all our vdevs are
879  * online when we do this, or else any vdevs that weren't present
880  * would be orphaned from our pool.  We are also going to issue a
881  * sysevent to update any watchers.
882  */
883 int
spa_change_guid(spa_t * spa)884 spa_change_guid(spa_t *spa)
885 {
886 	int error;
887 	uint64_t guid;
888 
889 	mutex_enter(&spa->spa_vdev_top_lock);
890 	mutex_enter(&spa_namespace_lock);
891 	guid = spa_generate_guid(NULL);
892 
893 	error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
894 	    spa_change_guid_sync, &guid, 5, ZFS_SPACE_CHECK_RESERVED);
895 
896 	if (error == 0) {
897 		spa_write_cachefile(spa, B_FALSE, B_TRUE);
898 		spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
899 	}
900 
901 	mutex_exit(&spa_namespace_lock);
902 	mutex_exit(&spa->spa_vdev_top_lock);
903 
904 	return (error);
905 }
906 
907 /*
908  * ==========================================================================
909  * SPA state manipulation (open/create/destroy/import/export)
910  * ==========================================================================
911  */
912 
913 static int
spa_error_entry_compare(const void * a,const void * b)914 spa_error_entry_compare(const void *a, const void *b)
915 {
916 	const spa_error_entry_t *sa = (const spa_error_entry_t *)a;
917 	const spa_error_entry_t *sb = (const spa_error_entry_t *)b;
918 	int ret;
919 
920 	ret = memcmp(&sa->se_bookmark, &sb->se_bookmark,
921 	    sizeof (zbookmark_phys_t));
922 
923 	return (TREE_ISIGN(ret));
924 }
925 
926 /*
927  * Utility function which retrieves copies of the current logs and
928  * re-initializes them in the process.
929  */
930 void
spa_get_errlists(spa_t * spa,avl_tree_t * last,avl_tree_t * scrub)931 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
932 {
933 	ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
934 
935 	bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
936 	bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
937 
938 	avl_create(&spa->spa_errlist_scrub,
939 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
940 	    offsetof(spa_error_entry_t, se_avl));
941 	avl_create(&spa->spa_errlist_last,
942 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
943 	    offsetof(spa_error_entry_t, se_avl));
944 }
945 
946 static void
spa_taskqs_init(spa_t * spa,zio_type_t t,zio_taskq_type_t q)947 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
948 {
949 	const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
950 	enum zti_modes mode = ztip->zti_mode;
951 	uint_t value = ztip->zti_value;
952 	uint_t count = ztip->zti_count;
953 	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
954 	char name[32];
955 	uint_t flags = 0;
956 	boolean_t batch = B_FALSE;
957 
958 	if (mode == ZTI_MODE_NULL) {
959 		tqs->stqs_count = 0;
960 		tqs->stqs_taskq = NULL;
961 		return;
962 	}
963 
964 	ASSERT3U(count, >, 0);
965 
966 	tqs->stqs_count = count;
967 	tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
968 
969 	switch (mode) {
970 	case ZTI_MODE_FIXED:
971 		ASSERT3U(value, >=, 1);
972 		value = MAX(value, 1);
973 		break;
974 
975 	case ZTI_MODE_BATCH:
976 		batch = B_TRUE;
977 		flags |= TASKQ_THREADS_CPU_PCT;
978 		value = zio_taskq_batch_pct;
979 		break;
980 
981 	default:
982 		panic("unrecognized mode for %s_%s taskq (%u:%u) in "
983 		    "spa_activate()",
984 		    zio_type_name[t], zio_taskq_types[q], mode, value);
985 		break;
986 	}
987 
988 	for (uint_t i = 0; i < count; i++) {
989 		taskq_t *tq;
990 
991 		if (count > 1) {
992 			(void) snprintf(name, sizeof (name), "%s_%s_%u",
993 			    zio_type_name[t], zio_taskq_types[q], i);
994 		} else {
995 			(void) snprintf(name, sizeof (name), "%s_%s",
996 			    zio_type_name[t], zio_taskq_types[q]);
997 		}
998 
999 		if (zio_taskq_sysdc && spa->spa_proc != &p0) {
1000 			if (batch)
1001 				flags |= TASKQ_DC_BATCH;
1002 
1003 			tq = taskq_create_sysdc(name, value, 50, INT_MAX,
1004 			    spa->spa_proc, zio_taskq_basedc, flags);
1005 		} else {
1006 			pri_t pri = maxclsyspri;
1007 			/*
1008 			 * The write issue taskq can be extremely CPU
1009 			 * intensive.  Run it at slightly lower priority
1010 			 * than the other taskqs.
1011 			 */
1012 			if (t == ZIO_TYPE_WRITE && q == ZIO_TASKQ_ISSUE)
1013 				pri--;
1014 
1015 			tq = taskq_create_proc(name, value, pri, 50,
1016 			    INT_MAX, spa->spa_proc, flags);
1017 		}
1018 
1019 		tqs->stqs_taskq[i] = tq;
1020 	}
1021 }
1022 
1023 static void
spa_taskqs_fini(spa_t * spa,zio_type_t t,zio_taskq_type_t q)1024 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
1025 {
1026 	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1027 
1028 	if (tqs->stqs_taskq == NULL) {
1029 		ASSERT0(tqs->stqs_count);
1030 		return;
1031 	}
1032 
1033 	for (uint_t i = 0; i < tqs->stqs_count; i++) {
1034 		ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
1035 		taskq_destroy(tqs->stqs_taskq[i]);
1036 	}
1037 
1038 	kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
1039 	tqs->stqs_taskq = NULL;
1040 }
1041 
1042 /*
1043  * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
1044  * Note that a type may have multiple discrete taskqs to avoid lock contention
1045  * on the taskq itself. In that case we choose which taskq at random by using
1046  * the low bits of gethrtime().
1047  */
1048 void
spa_taskq_dispatch_ent(spa_t * spa,zio_type_t t,zio_taskq_type_t q,task_func_t * func,void * arg,uint_t flags,taskq_ent_t * ent)1049 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
1050     task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
1051 {
1052 	spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1053 	taskq_t *tq;
1054 
1055 	ASSERT3P(tqs->stqs_taskq, !=, NULL);
1056 	ASSERT3U(tqs->stqs_count, !=, 0);
1057 
1058 	if (tqs->stqs_count == 1) {
1059 		tq = tqs->stqs_taskq[0];
1060 	} else {
1061 		tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
1062 	}
1063 
1064 	taskq_dispatch_ent(tq, func, arg, flags, ent);
1065 }
1066 
1067 static void
spa_create_zio_taskqs(spa_t * spa)1068 spa_create_zio_taskqs(spa_t *spa)
1069 {
1070 	for (int t = 0; t < ZIO_TYPES; t++) {
1071 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1072 			spa_taskqs_init(spa, t, q);
1073 		}
1074 	}
1075 }
1076 
1077 #ifdef _KERNEL
1078 static void
spa_thread(void * arg)1079 spa_thread(void *arg)
1080 {
1081 	callb_cpr_t cprinfo;
1082 	spa_t *spa = arg;
1083 	char spa_id_readable[CB_MAXNAME + 1];
1084 	user_t *pu = PTOU(curproc);
1085 
1086 	(void) snprintf(spa_id_readable, sizeof (spa_id_readable), "SPA:0x%p",
1087 	    spa);
1088 
1089 	CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
1090 	    spa_id_readable);
1091 
1092 	ASSERT(curproc != &p0);
1093 	(void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
1094 	    "zpool-%s", spa->spa_name);
1095 	(void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
1096 
1097 	/* bind this thread to the requested psrset */
1098 	if (zio_taskq_psrset_bind != PS_NONE) {
1099 		pool_lock();
1100 		mutex_enter(&cpu_lock);
1101 		mutex_enter(&pidlock);
1102 		mutex_enter(&curproc->p_lock);
1103 
1104 		if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
1105 		    0, NULL, NULL) == 0)  {
1106 			curthread->t_bind_pset = zio_taskq_psrset_bind;
1107 		} else {
1108 			cmn_err(CE_WARN,
1109 			    "Couldn't bind process for zfs pool \"%s\" to "
1110 			    "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
1111 		}
1112 
1113 		mutex_exit(&curproc->p_lock);
1114 		mutex_exit(&pidlock);
1115 		mutex_exit(&cpu_lock);
1116 		pool_unlock();
1117 	}
1118 
1119 	if (zio_taskq_sysdc) {
1120 		sysdc_thread_enter(curthread, 100, 0);
1121 	}
1122 
1123 	spa->spa_proc = curproc;
1124 	spa->spa_did = curthread->t_did;
1125 
1126 	spa_create_zio_taskqs(spa);
1127 
1128 	mutex_enter(&spa->spa_proc_lock);
1129 	ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
1130 
1131 	spa->spa_proc_state = SPA_PROC_ACTIVE;
1132 	cv_broadcast(&spa->spa_proc_cv);
1133 
1134 	CALLB_CPR_SAFE_BEGIN(&cprinfo);
1135 	while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1136 		cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1137 	CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1138 
1139 	ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1140 	spa->spa_proc_state = SPA_PROC_GONE;
1141 	spa->spa_proc = &p0;
1142 	cv_broadcast(&spa->spa_proc_cv);
1143 	CALLB_CPR_EXIT(&cprinfo);	/* drops spa_proc_lock */
1144 
1145 	mutex_enter(&curproc->p_lock);
1146 	lwp_exit();
1147 }
1148 #endif
1149 
1150 /*
1151  * Activate an uninitialized pool.
1152  */
1153 static void
spa_activate(spa_t * spa,int mode)1154 spa_activate(spa_t *spa, int mode)
1155 {
1156 	ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1157 
1158 	spa->spa_state = POOL_STATE_ACTIVE;
1159 	spa->spa_mode = mode;
1160 
1161 	spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1162 	spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1163 	spa->spa_special_class = metaslab_class_create(spa, zfs_metaslab_ops);
1164 	spa->spa_dedup_class = metaslab_class_create(spa, zfs_metaslab_ops);
1165 
1166 	/* Try to create a covering process */
1167 	mutex_enter(&spa->spa_proc_lock);
1168 	ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1169 	ASSERT(spa->spa_proc == &p0);
1170 	spa->spa_did = 0;
1171 
1172 	/* Only create a process if we're going to be around a while. */
1173 	if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1174 		if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1175 		    NULL, 0) == 0) {
1176 			spa->spa_proc_state = SPA_PROC_CREATED;
1177 			while (spa->spa_proc_state == SPA_PROC_CREATED) {
1178 				cv_wait(&spa->spa_proc_cv,
1179 				    &spa->spa_proc_lock);
1180 			}
1181 			ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1182 			ASSERT(spa->spa_proc != &p0);
1183 			ASSERT(spa->spa_did != 0);
1184 		} else {
1185 #ifdef _KERNEL
1186 			cmn_err(CE_WARN,
1187 			    "Couldn't create process for zfs pool \"%s\"\n",
1188 			    spa->spa_name);
1189 #endif
1190 		}
1191 	}
1192 	mutex_exit(&spa->spa_proc_lock);
1193 
1194 	/* If we didn't create a process, we need to create our taskqs. */
1195 	if (spa->spa_proc == &p0) {
1196 		spa_create_zio_taskqs(spa);
1197 	}
1198 
1199 	for (size_t i = 0; i < TXG_SIZE; i++) {
1200 		spa->spa_txg_zio[i] = zio_root(spa, NULL, NULL,
1201 		    ZIO_FLAG_CANFAIL);
1202 	}
1203 
1204 	list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1205 	    offsetof(vdev_t, vdev_config_dirty_node));
1206 	list_create(&spa->spa_evicting_os_list, sizeof (objset_t),
1207 	    offsetof(objset_t, os_evicting_node));
1208 	list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1209 	    offsetof(vdev_t, vdev_state_dirty_node));
1210 
1211 	txg_list_create(&spa->spa_vdev_txg_list, spa,
1212 	    offsetof(struct vdev, vdev_txg_node));
1213 
1214 	avl_create(&spa->spa_errlist_scrub,
1215 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1216 	    offsetof(spa_error_entry_t, se_avl));
1217 	avl_create(&spa->spa_errlist_last,
1218 	    spa_error_entry_compare, sizeof (spa_error_entry_t),
1219 	    offsetof(spa_error_entry_t, se_avl));
1220 
1221 	spa_keystore_init(&spa->spa_keystore);
1222 
1223 	/*
1224 	 * The taskq to upgrade datasets in this pool. Currently used by
1225 	 * feature SPA_FEATURE_USEROBJ_ACCOUNTING/SPA_FEATURE_PROJECT_QUOTA.
1226 	 */
1227 	spa->spa_upgrade_taskq = taskq_create("z_upgrade", boot_ncpus,
1228 	    minclsyspri, 1, INT_MAX, TASKQ_DYNAMIC);
1229 }
1230 
1231 /*
1232  * Opposite of spa_activate().
1233  */
1234 static void
spa_deactivate(spa_t * spa)1235 spa_deactivate(spa_t *spa)
1236 {
1237 	ASSERT(spa->spa_sync_on == B_FALSE);
1238 	ASSERT(spa->spa_dsl_pool == NULL);
1239 	ASSERT(spa->spa_root_vdev == NULL);
1240 	ASSERT(spa->spa_async_zio_root == NULL);
1241 	ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1242 
1243 	spa_evicting_os_wait(spa);
1244 
1245 	if (spa->spa_upgrade_taskq) {
1246 		taskq_destroy(spa->spa_upgrade_taskq);
1247 		spa->spa_upgrade_taskq = NULL;
1248 	}
1249 
1250 	txg_list_destroy(&spa->spa_vdev_txg_list);
1251 
1252 	list_destroy(&spa->spa_config_dirty_list);
1253 	list_destroy(&spa->spa_evicting_os_list);
1254 	list_destroy(&spa->spa_state_dirty_list);
1255 
1256 	for (int t = 0; t < ZIO_TYPES; t++) {
1257 		for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1258 			spa_taskqs_fini(spa, t, q);
1259 		}
1260 	}
1261 
1262 	for (size_t i = 0; i < TXG_SIZE; i++) {
1263 		ASSERT3P(spa->spa_txg_zio[i], !=, NULL);
1264 		VERIFY0(zio_wait(spa->spa_txg_zio[i]));
1265 		spa->spa_txg_zio[i] = NULL;
1266 	}
1267 
1268 	metaslab_class_destroy(spa->spa_normal_class);
1269 	spa->spa_normal_class = NULL;
1270 
1271 	metaslab_class_destroy(spa->spa_log_class);
1272 	spa->spa_log_class = NULL;
1273 
1274 	metaslab_class_destroy(spa->spa_special_class);
1275 	spa->spa_special_class = NULL;
1276 
1277 	metaslab_class_destroy(spa->spa_dedup_class);
1278 	spa->spa_dedup_class = NULL;
1279 
1280 	/*
1281 	 * If this was part of an import or the open otherwise failed, we may
1282 	 * still have errors left in the queues.  Empty them just in case.
1283 	 */
1284 	spa_errlog_drain(spa);
1285 	avl_destroy(&spa->spa_errlist_scrub);
1286 	avl_destroy(&spa->spa_errlist_last);
1287 
1288 	spa_keystore_fini(&spa->spa_keystore);
1289 
1290 	spa->spa_state = POOL_STATE_UNINITIALIZED;
1291 
1292 	mutex_enter(&spa->spa_proc_lock);
1293 	if (spa->spa_proc_state != SPA_PROC_NONE) {
1294 		ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1295 		spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1296 		cv_broadcast(&spa->spa_proc_cv);
1297 		while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1298 			ASSERT(spa->spa_proc != &p0);
1299 			cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1300 		}
1301 		ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1302 		spa->spa_proc_state = SPA_PROC_NONE;
1303 	}
1304 	ASSERT(spa->spa_proc == &p0);
1305 	mutex_exit(&spa->spa_proc_lock);
1306 
1307 	/*
1308 	 * We want to make sure spa_thread() has actually exited the ZFS
1309 	 * module, so that the module can't be unloaded out from underneath
1310 	 * it.
1311 	 */
1312 	if (spa->spa_did != 0) {
1313 		thread_join(spa->spa_did);
1314 		spa->spa_did = 0;
1315 	}
1316 }
1317 
1318 /*
1319  * Verify a pool configuration, and construct the vdev tree appropriately.  This
1320  * will create all the necessary vdevs in the appropriate layout, with each vdev
1321  * in the CLOSED state.  This will prep the pool before open/creation/import.
1322  * All vdev validation is done by the vdev_alloc() routine.
1323  */
1324 static int
spa_config_parse(spa_t * spa,vdev_t ** vdp,nvlist_t * nv,vdev_t * parent,uint_t id,int atype)1325 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1326     uint_t id, int atype)
1327 {
1328 	nvlist_t **child;
1329 	uint_t children;
1330 	int error;
1331 
1332 	if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1333 		return (error);
1334 
1335 	if ((*vdp)->vdev_ops->vdev_op_leaf)
1336 		return (0);
1337 
1338 	error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1339 	    &child, &children);
1340 
1341 	if (error == ENOENT)
1342 		return (0);
1343 
1344 	if (error) {
1345 		vdev_free(*vdp);
1346 		*vdp = NULL;
1347 		return (SET_ERROR(EINVAL));
1348 	}
1349 
1350 	for (int c = 0; c < children; c++) {
1351 		vdev_t *vd;
1352 		if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1353 		    atype)) != 0) {
1354 			vdev_free(*vdp);
1355 			*vdp = NULL;
1356 			return (error);
1357 		}
1358 	}
1359 
1360 	ASSERT(*vdp != NULL);
1361 
1362 	return (0);
1363 }
1364 
1365 static boolean_t
spa_should_flush_logs_on_unload(spa_t * spa)1366 spa_should_flush_logs_on_unload(spa_t *spa)
1367 {
1368 	if (!spa_feature_is_active(spa, SPA_FEATURE_LOG_SPACEMAP))
1369 		return (B_FALSE);
1370 
1371 	if (!spa_writeable(spa))
1372 		return (B_FALSE);
1373 
1374 	if (!spa->spa_sync_on)
1375 		return (B_FALSE);
1376 
1377 	if (spa_state(spa) != POOL_STATE_EXPORTED)
1378 		return (B_FALSE);
1379 
1380 	if (zfs_keep_log_spacemaps_at_export)
1381 		return (B_FALSE);
1382 
1383 	return (B_TRUE);
1384 }
1385 
1386 /*
1387  * Opens a transaction that will set the flag that will instruct
1388  * spa_sync to attempt to flush all the metaslabs for that txg.
1389  */
1390 static void
spa_unload_log_sm_flush_all(spa_t * spa)1391 spa_unload_log_sm_flush_all(spa_t *spa)
1392 {
1393 	dmu_tx_t *tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
1394 
1395 	VERIFY0(dmu_tx_assign(tx, TXG_WAIT));
1396 
1397 	ASSERT3U(spa->spa_log_flushall_txg, ==, 0);
1398 	spa->spa_log_flushall_txg = dmu_tx_get_txg(tx);
1399 
1400 	dmu_tx_commit(tx);
1401 	txg_wait_synced(spa_get_dsl(spa), spa->spa_log_flushall_txg);
1402 }
1403 
1404 static void
spa_unload_log_sm_metadata(spa_t * spa)1405 spa_unload_log_sm_metadata(spa_t *spa)
1406 {
1407 	void *cookie = NULL;
1408 	spa_log_sm_t *sls;
1409 
1410 	while ((sls = avl_destroy_nodes(&spa->spa_sm_logs_by_txg,
1411 	    &cookie)) != NULL) {
1412 		VERIFY0(sls->sls_mscount);
1413 		kmem_free(sls, sizeof (spa_log_sm_t));
1414 	}
1415 
1416 	for (log_summary_entry_t *e = list_head(&spa->spa_log_summary);
1417 	    e != NULL; e = list_head(&spa->spa_log_summary)) {
1418 		VERIFY0(e->lse_mscount);
1419 		list_remove(&spa->spa_log_summary, e);
1420 		kmem_free(e, sizeof (log_summary_entry_t));
1421 	}
1422 
1423 	spa->spa_unflushed_stats.sus_nblocks = 0;
1424 	spa->spa_unflushed_stats.sus_memused = 0;
1425 	spa->spa_unflushed_stats.sus_blocklimit = 0;
1426 }
1427 
1428 /*
1429  * Opposite of spa_load().
1430  */
1431 static void
spa_unload(spa_t * spa)1432 spa_unload(spa_t *spa)
1433 {
1434 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
1435 	ASSERT(spa_state(spa) != POOL_STATE_UNINITIALIZED);
1436 
1437 	spa_import_progress_remove(spa);
1438 	spa_load_note(spa, "UNLOADING");
1439 
1440 	/*
1441 	 * If the log space map feature is enabled and the pool is getting
1442 	 * exported (but not destroyed), we want to spend some time flushing
1443 	 * as many metaslabs as we can in an attempt to destroy log space
1444 	 * maps and save import time.
1445 	 */
1446 	if (spa_should_flush_logs_on_unload(spa))
1447 		spa_unload_log_sm_flush_all(spa);
1448 
1449 	/*
1450 	 * Stop async tasks.
1451 	 */
1452 	spa_async_suspend(spa);
1453 
1454 	if (spa->spa_root_vdev) {
1455 		vdev_t *root_vdev = spa->spa_root_vdev;
1456 		vdev_initialize_stop_all(root_vdev, VDEV_INITIALIZE_ACTIVE);
1457 		vdev_trim_stop_all(root_vdev, VDEV_TRIM_ACTIVE);
1458 		vdev_autotrim_stop_all(spa);
1459 	}
1460 
1461 	/*
1462 	 * Stop syncing.
1463 	 */
1464 	if (spa->spa_sync_on) {
1465 		txg_sync_stop(spa->spa_dsl_pool);
1466 		spa->spa_sync_on = B_FALSE;
1467 	}
1468 
1469 	/*
1470 	 * This ensures that there is no async metaslab prefetching
1471 	 * while we attempt to unload the spa.
1472 	 */
1473 	if (spa->spa_root_vdev != NULL) {
1474 		for (int c = 0; c < spa->spa_root_vdev->vdev_children; c++) {
1475 			vdev_t *vc = spa->spa_root_vdev->vdev_child[c];
1476 			if (vc->vdev_mg != NULL)
1477 				taskq_wait(vc->vdev_mg->mg_taskq);
1478 		}
1479 	}
1480 
1481 	if (spa->spa_mmp.mmp_thread)
1482 		mmp_thread_stop(spa);
1483 
1484 	/*
1485 	 * Wait for any outstanding async I/O to complete.
1486 	 */
1487 	if (spa->spa_async_zio_root != NULL) {
1488 		for (int i = 0; i < max_ncpus; i++)
1489 			(void) zio_wait(spa->spa_async_zio_root[i]);
1490 		kmem_free(spa->spa_async_zio_root, max_ncpus * sizeof (void *));
1491 		spa->spa_async_zio_root = NULL;
1492 	}
1493 
1494 	if (spa->spa_vdev_removal != NULL) {
1495 		spa_vdev_removal_destroy(spa->spa_vdev_removal);
1496 		spa->spa_vdev_removal = NULL;
1497 	}
1498 
1499 	if (spa->spa_condense_zthr != NULL) {
1500 		zthr_destroy(spa->spa_condense_zthr);
1501 		spa->spa_condense_zthr = NULL;
1502 	}
1503 
1504 	if (spa->spa_checkpoint_discard_zthr != NULL) {
1505 		zthr_destroy(spa->spa_checkpoint_discard_zthr);
1506 		spa->spa_checkpoint_discard_zthr = NULL;
1507 	}
1508 
1509 	spa_condense_fini(spa);
1510 
1511 	bpobj_close(&spa->spa_deferred_bpobj);
1512 
1513 	spa_config_enter(spa, SCL_ALL, spa, RW_WRITER);
1514 
1515 	/*
1516 	 * Close all vdevs.
1517 	 */
1518 	if (spa->spa_root_vdev)
1519 		vdev_free(spa->spa_root_vdev);
1520 	ASSERT(spa->spa_root_vdev == NULL);
1521 
1522 	/*
1523 	 * Close the dsl pool.
1524 	 */
1525 	if (spa->spa_dsl_pool) {
1526 		dsl_pool_close(spa->spa_dsl_pool);
1527 		spa->spa_dsl_pool = NULL;
1528 		spa->spa_meta_objset = NULL;
1529 	}
1530 
1531 	ddt_unload(spa);
1532 	spa_unload_log_sm_metadata(spa);
1533 
1534 	/*
1535 	 * Drop and purge level 2 cache
1536 	 */
1537 	spa_l2cache_drop(spa);
1538 
1539 	for (int i = 0; i < spa->spa_spares.sav_count; i++)
1540 		vdev_free(spa->spa_spares.sav_vdevs[i]);
1541 	if (spa->spa_spares.sav_vdevs) {
1542 		kmem_free(spa->spa_spares.sav_vdevs,
1543 		    spa->spa_spares.sav_count * sizeof (void *));
1544 		spa->spa_spares.sav_vdevs = NULL;
1545 	}
1546 	if (spa->spa_spares.sav_config) {
1547 		nvlist_free(spa->spa_spares.sav_config);
1548 		spa->spa_spares.sav_config = NULL;
1549 	}
1550 	spa->spa_spares.sav_count = 0;
1551 
1552 	for (int i = 0; i < spa->spa_l2cache.sav_count; i++) {
1553 		vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1554 		vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1555 	}
1556 	if (spa->spa_l2cache.sav_vdevs) {
1557 		kmem_free(spa->spa_l2cache.sav_vdevs,
1558 		    spa->spa_l2cache.sav_count * sizeof (void *));
1559 		spa->spa_l2cache.sav_vdevs = NULL;
1560 	}
1561 	if (spa->spa_l2cache.sav_config) {
1562 		nvlist_free(spa->spa_l2cache.sav_config);
1563 		spa->spa_l2cache.sav_config = NULL;
1564 	}
1565 	spa->spa_l2cache.sav_count = 0;
1566 
1567 	spa->spa_async_suspended = 0;
1568 
1569 	spa->spa_indirect_vdevs_loaded = B_FALSE;
1570 
1571 	if (spa->spa_comment != NULL) {
1572 		spa_strfree(spa->spa_comment);
1573 		spa->spa_comment = NULL;
1574 	}
1575 
1576 	spa_config_exit(spa, SCL_ALL, spa);
1577 }
1578 
1579 /*
1580  * Load (or re-load) the current list of vdevs describing the active spares for
1581  * this pool.  When this is called, we have some form of basic information in
1582  * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1583  * then re-generate a more complete list including status information.
1584  */
1585 void
spa_load_spares(spa_t * spa)1586 spa_load_spares(spa_t *spa)
1587 {
1588 	nvlist_t **spares;
1589 	uint_t nspares;
1590 	int i;
1591 	vdev_t *vd, *tvd;
1592 
1593 #ifndef _KERNEL
1594 	/*
1595 	 * zdb opens both the current state of the pool and the
1596 	 * checkpointed state (if present), with a different spa_t.
1597 	 *
1598 	 * As spare vdevs are shared among open pools, we skip loading
1599 	 * them when we load the checkpointed state of the pool.
1600 	 */
1601 	if (!spa_writeable(spa))
1602 		return;
1603 #endif
1604 
1605 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1606 
1607 	/*
1608 	 * First, close and free any existing spare vdevs.
1609 	 */
1610 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1611 		vd = spa->spa_spares.sav_vdevs[i];
1612 
1613 		/* Undo the call to spa_activate() below */
1614 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1615 		    B_FALSE)) != NULL && tvd->vdev_isspare)
1616 			spa_spare_remove(tvd);
1617 		vdev_close(vd);
1618 		vdev_free(vd);
1619 	}
1620 
1621 	if (spa->spa_spares.sav_vdevs)
1622 		kmem_free(spa->spa_spares.sav_vdevs,
1623 		    spa->spa_spares.sav_count * sizeof (void *));
1624 
1625 	if (spa->spa_spares.sav_config == NULL)
1626 		nspares = 0;
1627 	else
1628 		VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1629 		    ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1630 
1631 	spa->spa_spares.sav_count = (int)nspares;
1632 	spa->spa_spares.sav_vdevs = NULL;
1633 
1634 	if (nspares == 0)
1635 		return;
1636 
1637 	/*
1638 	 * Construct the array of vdevs, opening them to get status in the
1639 	 * process.   For each spare, there is potentially two different vdev_t
1640 	 * structures associated with it: one in the list of spares (used only
1641 	 * for basic validation purposes) and one in the active vdev
1642 	 * configuration (if it's spared in).  During this phase we open and
1643 	 * validate each vdev on the spare list.  If the vdev also exists in the
1644 	 * active configuration, then we also mark this vdev as an active spare.
1645 	 */
1646 	spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1647 	    KM_SLEEP);
1648 	for (i = 0; i < spa->spa_spares.sav_count; i++) {
1649 		VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1650 		    VDEV_ALLOC_SPARE) == 0);
1651 		ASSERT(vd != NULL);
1652 
1653 		spa->spa_spares.sav_vdevs[i] = vd;
1654 
1655 		if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1656 		    B_FALSE)) != NULL) {
1657 			if (!tvd->vdev_isspare)
1658 				spa_spare_add(tvd);
1659 
1660 			/*
1661 			 * We only mark the spare active if we were successfully
1662 			 * able to load the vdev.  Otherwise, importing a pool
1663 			 * with a bad active spare would result in strange
1664 			 * behavior, because multiple pool would think the spare
1665 			 * is actively in use.
1666 			 *
1667 			 * There is a vulnerability here to an equally bizarre
1668 			 * circumstance, where a dead active spare is later
1669 			 * brought back to life (onlined or otherwise).  Given
1670 			 * the rarity of this scenario, and the extra complexity
1671 			 * it adds, we ignore the possibility.
1672 			 */
1673 			if (!vdev_is_dead(tvd))
1674 				spa_spare_activate(tvd);
1675 		}
1676 
1677 		vd->vdev_top = vd;
1678 		vd->vdev_aux = &spa->spa_spares;
1679 
1680 		if (vdev_open(vd) != 0)
1681 			continue;
1682 
1683 		if (vdev_validate_aux(vd) == 0)
1684 			spa_spare_add(vd);
1685 	}
1686 
1687 	/*
1688 	 * Recompute the stashed list of spares, with status information
1689 	 * this time.
1690 	 */
1691 	VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1692 	    DATA_TYPE_NVLIST_ARRAY) == 0);
1693 
1694 	spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1695 	    KM_SLEEP);
1696 	for (i = 0; i < spa->spa_spares.sav_count; i++)
1697 		spares[i] = vdev_config_generate(spa,
1698 		    spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1699 	VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1700 	    ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1701 	for (i = 0; i < spa->spa_spares.sav_count; i++)
1702 		nvlist_free(spares[i]);
1703 	kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1704 }
1705 
1706 /*
1707  * Load (or re-load) the current list of vdevs describing the active l2cache for
1708  * this pool.  When this is called, we have some form of basic information in
1709  * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1710  * then re-generate a more complete list including status information.
1711  * Devices which are already active have their details maintained, and are
1712  * not re-opened.
1713  */
1714 void
spa_load_l2cache(spa_t * spa)1715 spa_load_l2cache(spa_t *spa)
1716 {
1717 	nvlist_t **l2cache;
1718 	uint_t nl2cache;
1719 	int i, j, oldnvdevs;
1720 	uint64_t guid;
1721 	vdev_t *vd, **oldvdevs, **newvdevs;
1722 	spa_aux_vdev_t *sav = &spa->spa_l2cache;
1723 
1724 #ifndef _KERNEL
1725 	/*
1726 	 * zdb opens both the current state of the pool and the
1727 	 * checkpointed state (if present), with a different spa_t.
1728 	 *
1729 	 * As L2 caches are part of the ARC which is shared among open
1730 	 * pools, we skip loading them when we load the checkpointed
1731 	 * state of the pool.
1732 	 */
1733 	if (!spa_writeable(spa))
1734 		return;
1735 #endif
1736 
1737 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1738 
1739 	nl2cache = 0;
1740 	newvdevs = NULL;
1741 	if (sav->sav_config != NULL) {
1742 		VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1743 		    ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1744 		if (nl2cache > 0) {
1745 			newvdevs = kmem_alloc(
1746 			    nl2cache * sizeof (void *), KM_SLEEP);
1747 		}
1748 	}
1749 
1750 	oldvdevs = sav->sav_vdevs;
1751 	oldnvdevs = sav->sav_count;
1752 	sav->sav_vdevs = NULL;
1753 	sav->sav_count = 0;
1754 
1755 	/*
1756 	 * Process new nvlist of vdevs.
1757 	 */
1758 	for (i = 0; i < nl2cache; i++) {
1759 		VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1760 		    &guid) == 0);
1761 
1762 		newvdevs[i] = NULL;
1763 		for (j = 0; j < oldnvdevs; j++) {
1764 			vd = oldvdevs[j];
1765 			if (vd != NULL && guid == vd->vdev_guid) {
1766 				/*
1767 				 * Retain previous vdev for add/remove ops.
1768 				 */
1769 				newvdevs[i] = vd;
1770 				oldvdevs[j] = NULL;
1771 				break;
1772 			}
1773 		}
1774 
1775 		if (newvdevs[i] == NULL) {
1776 			/*
1777 			 * Create new vdev
1778 			 */
1779 			VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1780 			    VDEV_ALLOC_L2CACHE) == 0);
1781 			ASSERT(vd != NULL);
1782 			newvdevs[i] = vd;
1783 
1784 			/*
1785 			 * Commit this vdev as an l2cache device,
1786 			 * even if it fails to open.
1787 			 */
1788 			spa_l2cache_add(vd);
1789 
1790 			vd->vdev_top = vd;
1791 			vd->vdev_aux = sav;
1792 
1793 			spa_l2cache_activate(vd);
1794 
1795 			if (vdev_open(vd) != 0)
1796 				continue;
1797 
1798 			(void) vdev_validate_aux(vd);
1799 
1800 			if (!vdev_is_dead(vd))
1801 				l2arc_add_vdev(spa, vd);
1802 		}
1803 	}
1804 
1805 	/*
1806 	 * Purge vdevs that were dropped
1807 	 */
1808 	for (i = 0; i < oldnvdevs; i++) {
1809 		uint64_t pool;
1810 
1811 		vd = oldvdevs[i];
1812 		if (vd != NULL) {
1813 			ASSERT(vd->vdev_isl2cache);
1814 
1815 			if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1816 			    pool != 0ULL && l2arc_vdev_present(vd))
1817 				l2arc_remove_vdev(vd);
1818 			vdev_clear_stats(vd);
1819 			vdev_free(vd);
1820 		}
1821 	}
1822 
1823 	if (oldvdevs)
1824 		kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1825 
1826 	if (sav->sav_config == NULL)
1827 		goto out;
1828 
1829 	sav->sav_vdevs = newvdevs;
1830 	sav->sav_count = (int)nl2cache;
1831 
1832 	/*
1833 	 * Recompute the stashed list of l2cache devices, with status
1834 	 * information this time.
1835 	 */
1836 	VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1837 	    DATA_TYPE_NVLIST_ARRAY) == 0);
1838 
1839 	l2cache = NULL;
1840 	if (sav->sav_count > 0) {
1841 		l2cache = kmem_alloc(
1842 		    sav->sav_count * sizeof (void *), KM_SLEEP);
1843 	}
1844 	for (i = 0; i < sav->sav_count; i++)
1845 		l2cache[i] = vdev_config_generate(spa,
1846 		    sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1847 	VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1848 	    ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1849 out:
1850 	for (i = 0; i < sav->sav_count; i++)
1851 		nvlist_free(l2cache[i]);
1852 	if (sav->sav_count)
1853 		kmem_free(l2cache, sav->sav_count * sizeof (void *));
1854 }
1855 
1856 static int
load_nvlist(spa_t * spa,uint64_t obj,nvlist_t ** value)1857 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1858 {
1859 	dmu_buf_t *db;
1860 	char *packed = NULL;
1861 	size_t nvsize = 0;
1862 	int error;
1863 	*value = NULL;
1864 
1865 	error = dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db);
1866 	if (error != 0)
1867 		return (error);
1868 
1869 	nvsize = *(uint64_t *)db->db_data;
1870 	dmu_buf_rele(db, FTAG);
1871 
1872 	packed = kmem_alloc(nvsize, KM_SLEEP);
1873 	error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1874 	    DMU_READ_PREFETCH);
1875 	if (error == 0)
1876 		error = nvlist_unpack(packed, nvsize, value, 0);
1877 	kmem_free(packed, nvsize);
1878 
1879 	return (error);
1880 }
1881 
1882 /*
1883  * Concrete top-level vdevs that are not missing and are not logs. At every
1884  * spa_sync we write new uberblocks to at least SPA_SYNC_MIN_VDEVS core tvds.
1885  */
1886 static uint64_t
spa_healthy_core_tvds(spa_t * spa)1887 spa_healthy_core_tvds(spa_t *spa)
1888 {
1889 	vdev_t *rvd = spa->spa_root_vdev;
1890 	uint64_t tvds = 0;
1891 
1892 	for (uint64_t i = 0; i < rvd->vdev_children; i++) {
1893 		vdev_t *vd = rvd->vdev_child[i];
1894 		if (vd->vdev_islog)
1895 			continue;
1896 		if (vdev_is_concrete(vd) && !vdev_is_dead(vd))
1897 			tvds++;
1898 	}
1899 
1900 	return (tvds);
1901 }
1902 
1903 /*
1904  * Checks to see if the given vdev could not be opened, in which case we post a
1905  * sysevent to notify the autoreplace code that the device has been removed.
1906  */
1907 static void
spa_check_removed(vdev_t * vd)1908 spa_check_removed(vdev_t *vd)
1909 {
1910 	for (uint64_t c = 0; c < vd->vdev_children; c++)
1911 		spa_check_removed(vd->vdev_child[c]);
1912 
1913 	if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1914 	    vdev_is_concrete(vd)) {
1915 		zfs_post_autoreplace(vd->vdev_spa, vd);
1916 		spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
1917 	}
1918 }
1919 
1920 static int
spa_check_for_missing_logs(spa_t * spa)1921 spa_check_for_missing_logs(spa_t *spa)
1922 {
1923 	vdev_t *rvd = spa->spa_root_vdev;
1924 
1925 	/*
1926 	 * If we're doing a normal import, then build up any additional
1927 	 * diagnostic information about missing log devices.
1928 	 * We'll pass this up to the user for further processing.
1929 	 */
1930 	if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1931 		nvlist_t **child, *nv;
1932 		uint64_t idx = 0;
1933 
1934 		child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1935 		    KM_SLEEP);
1936 		VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1937 
1938 		for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1939 			vdev_t *tvd = rvd->vdev_child[c];
1940 
1941 			/*
1942 			 * We consider a device as missing only if it failed
1943 			 * to open (i.e. offline or faulted is not considered
1944 			 * as missing).
1945 			 */
1946 			if (tvd->vdev_islog &&
1947 			    tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1948 				child[idx++] = vdev_config_generate(spa, tvd,
1949 				    B_FALSE, VDEV_CONFIG_MISSING);
1950 			}
1951 		}
1952 
1953 		if (idx > 0) {
1954 			fnvlist_add_nvlist_array(nv,
1955 			    ZPOOL_CONFIG_CHILDREN, child, idx);
1956 			fnvlist_add_nvlist(spa->spa_load_info,
1957 			    ZPOOL_CONFIG_MISSING_DEVICES, nv);
1958 
1959 			for (uint64_t i = 0; i < idx; i++)
1960 				nvlist_free(child[i]);
1961 		}
1962 		nvlist_free(nv);
1963 		kmem_free(child, rvd->vdev_children * sizeof (char **));
1964 
1965 		if (idx > 0) {
1966 			spa_load_failed(spa, "some log devices are missing");
1967 			vdev_dbgmsg_print_tree(rvd, 2);
1968 			return (SET_ERROR(ENXIO));
1969 		}
1970 	} else {
1971 		for (uint64_t c = 0; c < rvd->vdev_children; c++) {
1972 			vdev_t *tvd = rvd->vdev_child[c];
1973 
1974 			if (tvd->vdev_islog &&
1975 			    tvd->vdev_state == VDEV_STATE_CANT_OPEN) {
1976 				spa_set_log_state(spa, SPA_LOG_CLEAR);
1977 				spa_load_note(spa, "some log devices are "
1978 				    "missing, ZIL is dropped.");
1979 				vdev_dbgmsg_print_tree(rvd, 2);
1980 				break;
1981 			}
1982 		}
1983 	}
1984 
1985 	return (0);
1986 }
1987 
1988 /*
1989  * Check for missing log devices
1990  */
1991 static boolean_t
spa_check_logs(spa_t * spa)1992 spa_check_logs(spa_t *spa)
1993 {
1994 	boolean_t rv = B_FALSE;
1995 	dsl_pool_t *dp = spa_get_dsl(spa);
1996 
1997 	switch (spa->spa_log_state) {
1998 	case SPA_LOG_MISSING:
1999 		/* need to recheck in case slog has been restored */
2000 	case SPA_LOG_UNKNOWN:
2001 		rv = (dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
2002 		    zil_check_log_chain, NULL, DS_FIND_CHILDREN) != 0);
2003 		if (rv)
2004 			spa_set_log_state(spa, SPA_LOG_MISSING);
2005 		break;
2006 	}
2007 	return (rv);
2008 }
2009 
2010 static boolean_t
spa_passivate_log(spa_t * spa)2011 spa_passivate_log(spa_t *spa)
2012 {
2013 	vdev_t *rvd = spa->spa_root_vdev;
2014 	boolean_t slog_found = B_FALSE;
2015 
2016 	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2017 
2018 	if (!spa_has_slogs(spa))
2019 		return (B_FALSE);
2020 
2021 	for (int c = 0; c < rvd->vdev_children; c++) {
2022 		vdev_t *tvd = rvd->vdev_child[c];
2023 		metaslab_group_t *mg = tvd->vdev_mg;
2024 
2025 		if (tvd->vdev_islog) {
2026 			metaslab_group_passivate(mg);
2027 			slog_found = B_TRUE;
2028 		}
2029 	}
2030 
2031 	return (slog_found);
2032 }
2033 
2034 static void
spa_activate_log(spa_t * spa)2035 spa_activate_log(spa_t *spa)
2036 {
2037 	vdev_t *rvd = spa->spa_root_vdev;
2038 
2039 	ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
2040 
2041 	for (int c = 0; c < rvd->vdev_children; c++) {
2042 		vdev_t *tvd = rvd->vdev_child[c];
2043 		metaslab_group_t *mg = tvd->vdev_mg;
2044 
2045 		if (tvd->vdev_islog)
2046 			metaslab_group_activate(mg);
2047 	}
2048 }
2049 
2050 int
spa_reset_logs(spa_t * spa)2051 spa_reset_logs(spa_t *spa)
2052 {
2053 	int error;
2054 
2055 	error = dmu_objset_find(spa_name(spa), zil_reset,
2056 	    NULL, DS_FIND_CHILDREN);
2057 	if (error == 0) {
2058 		/*
2059 		 * We successfully offlined the log device, sync out the
2060 		 * current txg so that the "stubby" block can be removed
2061 		 * by zil_sync().
2062 		 */
2063 		txg_wait_synced(spa->spa_dsl_pool, 0);
2064 	}
2065 	return (error);
2066 }
2067 
2068 static void
spa_aux_check_removed(spa_aux_vdev_t * sav)2069 spa_aux_check_removed(spa_aux_vdev_t *sav)
2070 {
2071 	for (int i = 0; i < sav->sav_count; i++)
2072 		spa_check_removed(sav->sav_vdevs[i]);
2073 }
2074 
2075 void
spa_claim_notify(zio_t * zio)2076 spa_claim_notify(zio_t *zio)
2077 {
2078 	spa_t *spa = zio->io_spa;
2079 
2080 	if (zio->io_error)
2081 		return;
2082 
2083 	mutex_enter(&spa->spa_props_lock);	/* any mutex will do */
2084 	if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
2085 		spa->spa_claim_max_txg = zio->io_bp->blk_birth;
2086 	mutex_exit(&spa->spa_props_lock);
2087 }
2088 
2089 typedef struct spa_load_error {
2090 	uint64_t	sle_meta_count;
2091 	uint64_t	sle_data_count;
2092 } spa_load_error_t;
2093 
2094 static void
spa_load_verify_done(zio_t * zio)2095 spa_load_verify_done(zio_t *zio)
2096 {
2097 	blkptr_t *bp = zio->io_bp;
2098 	spa_load_error_t *sle = zio->io_private;
2099 	dmu_object_type_t type = BP_GET_TYPE(bp);
2100 	int error = zio->io_error;
2101 	spa_t *spa = zio->io_spa;
2102 
2103 	abd_free(zio->io_abd);
2104 	if (error) {
2105 		if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
2106 		    type != DMU_OT_INTENT_LOG)
2107 			atomic_inc_64(&sle->sle_meta_count);
2108 		else
2109 			atomic_inc_64(&sle->sle_data_count);
2110 	}
2111 
2112 	mutex_enter(&spa->spa_scrub_lock);
2113 	spa->spa_load_verify_ios--;
2114 	cv_broadcast(&spa->spa_scrub_io_cv);
2115 	mutex_exit(&spa->spa_scrub_lock);
2116 }
2117 
2118 /*
2119  * Maximum number of concurrent scrub i/os to create while verifying
2120  * a pool while importing it.
2121  */
2122 int spa_load_verify_maxinflight = 10000;
2123 boolean_t spa_load_verify_metadata = B_TRUE;
2124 boolean_t spa_load_verify_data = B_TRUE;
2125 
2126 /*ARGSUSED*/
2127 static int
spa_load_verify_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)2128 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
2129     const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2130 {
2131 	if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
2132 		return (0);
2133 	/*
2134 	 * Note: normally this routine will not be called if
2135 	 * spa_load_verify_metadata is not set.  However, it may be useful
2136 	 * to manually set the flag after the traversal has begun.
2137 	 */
2138 	if (!spa_load_verify_metadata)
2139 		return (0);
2140 	if (!BP_IS_METADATA(bp) && !spa_load_verify_data)
2141 		return (0);
2142 
2143 	zio_t *rio = arg;
2144 	size_t size = BP_GET_PSIZE(bp);
2145 
2146 	mutex_enter(&spa->spa_scrub_lock);
2147 	while (spa->spa_load_verify_ios >= spa_load_verify_maxinflight)
2148 		cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
2149 	spa->spa_load_verify_ios++;
2150 	mutex_exit(&spa->spa_scrub_lock);
2151 
2152 	zio_nowait(zio_read(rio, spa, bp, abd_alloc_for_io(size, B_FALSE), size,
2153 	    spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
2154 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
2155 	    ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
2156 	return (0);
2157 }
2158 
2159 /* ARGSUSED */
2160 int
verify_dataset_name_len(dsl_pool_t * dp,dsl_dataset_t * ds,void * arg)2161 verify_dataset_name_len(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
2162 {
2163 	if (dsl_dataset_namelen(ds) >= ZFS_MAX_DATASET_NAME_LEN)
2164 		return (SET_ERROR(ENAMETOOLONG));
2165 
2166 	return (0);
2167 }
2168 
2169 static int
spa_load_verify(spa_t * spa)2170 spa_load_verify(spa_t *spa)
2171 {
2172 	zio_t *rio;
2173 	spa_load_error_t sle = { 0 };
2174 	zpool_load_policy_t policy;
2175 	boolean_t verify_ok = B_FALSE;
2176 	int error = 0;
2177 
2178 	zpool_get_load_policy(spa->spa_config, &policy);
2179 
2180 	if (policy.zlp_rewind & ZPOOL_NEVER_REWIND)
2181 		return (0);
2182 
2183 	dsl_pool_config_enter(spa->spa_dsl_pool, FTAG);
2184 	error = dmu_objset_find_dp(spa->spa_dsl_pool,
2185 	    spa->spa_dsl_pool->dp_root_dir_obj, verify_dataset_name_len, NULL,
2186 	    DS_FIND_CHILDREN);
2187 	dsl_pool_config_exit(spa->spa_dsl_pool, FTAG);
2188 	if (error != 0)
2189 		return (error);
2190 
2191 	rio = zio_root(spa, NULL, &sle,
2192 	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
2193 
2194 	if (spa_load_verify_metadata) {
2195 		if (spa->spa_extreme_rewind) {
2196 			spa_load_note(spa, "performing a complete scan of the "
2197 			    "pool since extreme rewind is on. This may take "
2198 			    "a very long time.\n  (spa_load_verify_data=%u, "
2199 			    "spa_load_verify_metadata=%u)",
2200 			    spa_load_verify_data, spa_load_verify_metadata);
2201 		}
2202 		error = traverse_pool(spa, spa->spa_verify_min_txg,
2203 		    TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA |
2204 		    TRAVERSE_NO_DECRYPT, spa_load_verify_cb, rio);
2205 	}
2206 
2207 	(void) zio_wait(rio);
2208 
2209 	spa->spa_load_meta_errors = sle.sle_meta_count;
2210 	spa->spa_load_data_errors = sle.sle_data_count;
2211 
2212 	if (sle.sle_meta_count != 0 || sle.sle_data_count != 0) {
2213 		spa_load_note(spa, "spa_load_verify found %llu metadata errors "
2214 		    "and %llu data errors", (u_longlong_t)sle.sle_meta_count,
2215 		    (u_longlong_t)sle.sle_data_count);
2216 	}
2217 
2218 	if (spa_load_verify_dryrun ||
2219 	    (!error && sle.sle_meta_count <= policy.zlp_maxmeta &&
2220 	    sle.sle_data_count <= policy.zlp_maxdata)) {
2221 		int64_t loss = 0;
2222 
2223 		verify_ok = B_TRUE;
2224 		spa->spa_load_txg = spa->spa_uberblock.ub_txg;
2225 		spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
2226 
2227 		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
2228 		VERIFY(nvlist_add_uint64(spa->spa_load_info,
2229 		    ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
2230 		VERIFY(nvlist_add_int64(spa->spa_load_info,
2231 		    ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
2232 		VERIFY(nvlist_add_uint64(spa->spa_load_info,
2233 		    ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
2234 	} else {
2235 		spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
2236 	}
2237 
2238 	if (spa_load_verify_dryrun)
2239 		return (0);
2240 
2241 	if (error) {
2242 		if (error != ENXIO && error != EIO)
2243 			error = SET_ERROR(EIO);
2244 		return (error);
2245 	}
2246 
2247 	return (verify_ok ? 0 : EIO);
2248 }
2249 
2250 /*
2251  * Find a value in the pool props object.
2252  */
2253 static void
spa_prop_find(spa_t * spa,zpool_prop_t prop,uint64_t * val)2254 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
2255 {
2256 	(void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
2257 	    zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
2258 }
2259 
2260 /*
2261  * Find a value in the pool directory object.
2262  */
2263 static int
spa_dir_prop(spa_t * spa,const char * name,uint64_t * val,boolean_t log_enoent)2264 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val, boolean_t log_enoent)
2265 {
2266 	int error = zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
2267 	    name, sizeof (uint64_t), 1, val);
2268 
2269 	if (error != 0 && (error != ENOENT || log_enoent)) {
2270 		spa_load_failed(spa, "couldn't get '%s' value in MOS directory "
2271 		    "[error=%d]", name, error);
2272 	}
2273 
2274 	return (error);
2275 }
2276 
2277 static int
spa_vdev_err(vdev_t * vdev,vdev_aux_t aux,int err)2278 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
2279 {
2280 	vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
2281 	return (SET_ERROR(err));
2282 }
2283 
2284 static void
spa_spawn_aux_threads(spa_t * spa)2285 spa_spawn_aux_threads(spa_t *spa)
2286 {
2287 	ASSERT(spa_writeable(spa));
2288 
2289 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
2290 
2291 	spa_start_indirect_condensing_thread(spa);
2292 
2293 	ASSERT3P(spa->spa_checkpoint_discard_zthr, ==, NULL);
2294 	spa->spa_checkpoint_discard_zthr =
2295 	    zthr_create(spa_checkpoint_discard_thread_check,
2296 	    spa_checkpoint_discard_thread, spa);
2297 }
2298 
2299 /*
2300  * Fix up config after a partly-completed split.  This is done with the
2301  * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
2302  * pool have that entry in their config, but only the splitting one contains
2303  * a list of all the guids of the vdevs that are being split off.
2304  *
2305  * This function determines what to do with that list: either rejoin
2306  * all the disks to the pool, or complete the splitting process.  To attempt
2307  * the rejoin, each disk that is offlined is marked online again, and
2308  * we do a reopen() call.  If the vdev label for every disk that was
2309  * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
2310  * then we call vdev_split() on each disk, and complete the split.
2311  *
2312  * Otherwise we leave the config alone, with all the vdevs in place in
2313  * the original pool.
2314  */
2315 static void
spa_try_repair(spa_t * spa,nvlist_t * config)2316 spa_try_repair(spa_t *spa, nvlist_t *config)
2317 {
2318 	uint_t extracted;
2319 	uint64_t *glist;
2320 	uint_t i, gcount;
2321 	nvlist_t *nvl;
2322 	vdev_t **vd;
2323 	boolean_t attempt_reopen;
2324 
2325 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
2326 		return;
2327 
2328 	/* check that the config is complete */
2329 	if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
2330 	    &glist, &gcount) != 0)
2331 		return;
2332 
2333 	vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
2334 
2335 	/* attempt to online all the vdevs & validate */
2336 	attempt_reopen = B_TRUE;
2337 	for (i = 0; i < gcount; i++) {
2338 		if (glist[i] == 0)	/* vdev is hole */
2339 			continue;
2340 
2341 		vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
2342 		if (vd[i] == NULL) {
2343 			/*
2344 			 * Don't bother attempting to reopen the disks;
2345 			 * just do the split.
2346 			 */
2347 			attempt_reopen = B_FALSE;
2348 		} else {
2349 			/* attempt to re-online it */
2350 			vd[i]->vdev_offline = B_FALSE;
2351 		}
2352 	}
2353 
2354 	if (attempt_reopen) {
2355 		vdev_reopen(spa->spa_root_vdev);
2356 
2357 		/* check each device to see what state it's in */
2358 		for (extracted = 0, i = 0; i < gcount; i++) {
2359 			if (vd[i] != NULL &&
2360 			    vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
2361 				break;
2362 			++extracted;
2363 		}
2364 	}
2365 
2366 	/*
2367 	 * If every disk has been moved to the new pool, or if we never
2368 	 * even attempted to look at them, then we split them off for
2369 	 * good.
2370 	 */
2371 	if (!attempt_reopen || gcount == extracted) {
2372 		for (i = 0; i < gcount; i++)
2373 			if (vd[i] != NULL)
2374 				vdev_split(vd[i]);
2375 		vdev_reopen(spa->spa_root_vdev);
2376 	}
2377 
2378 	kmem_free(vd, gcount * sizeof (vdev_t *));
2379 }
2380 
2381 static int
spa_load(spa_t * spa,spa_load_state_t state,spa_import_type_t type)2382 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type)
2383 {
2384 	char *ereport = FM_EREPORT_ZFS_POOL;
2385 	int error;
2386 
2387 	spa->spa_load_state = state;
2388 	(void) spa_import_progress_set_state(spa, spa_load_state(spa));
2389 
2390 	gethrestime(&spa->spa_loaded_ts);
2391 	error = spa_load_impl(spa, type, &ereport);
2392 
2393 	/*
2394 	 * Don't count references from objsets that are already closed
2395 	 * and are making their way through the eviction process.
2396 	 */
2397 	spa_evicting_os_wait(spa);
2398 	spa->spa_minref = zfs_refcount_count(&spa->spa_refcount);
2399 	if (error) {
2400 		if (error != EEXIST) {
2401 			spa->spa_loaded_ts.tv_sec = 0;
2402 			spa->spa_loaded_ts.tv_nsec = 0;
2403 		}
2404 		if (error != EBADF) {
2405 			(void) zfs_ereport_post(ereport, spa,
2406 			    NULL, NULL, NULL, 0, 0);
2407 		}
2408 	}
2409 	spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2410 	spa->spa_ena = 0;
2411 
2412 	(void) spa_import_progress_set_state(spa, spa_load_state(spa));
2413 
2414 	return (error);
2415 }
2416 
2417 /*
2418  * Count the number of per-vdev ZAPs associated with all of the vdevs in the
2419  * vdev tree rooted in the given vd, and ensure that each ZAP is present in the
2420  * spa's per-vdev ZAP list.
2421  */
2422 static uint64_t
vdev_count_verify_zaps(vdev_t * vd)2423 vdev_count_verify_zaps(vdev_t *vd)
2424 {
2425 	spa_t *spa = vd->vdev_spa;
2426 	uint64_t total = 0;
2427 	if (vd->vdev_top_zap != 0) {
2428 		total++;
2429 		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2430 		    spa->spa_all_vdev_zaps, vd->vdev_top_zap));
2431 	}
2432 	if (vd->vdev_leaf_zap != 0) {
2433 		total++;
2434 		ASSERT0(zap_lookup_int(spa->spa_meta_objset,
2435 		    spa->spa_all_vdev_zaps, vd->vdev_leaf_zap));
2436 	}
2437 
2438 	for (uint64_t i = 0; i < vd->vdev_children; i++) {
2439 		total += vdev_count_verify_zaps(vd->vdev_child[i]);
2440 	}
2441 
2442 	return (total);
2443 }
2444 
2445 /*
2446  * Determine whether the activity check is required.
2447  */
2448 static boolean_t
spa_activity_check_required(spa_t * spa,uberblock_t * ub,nvlist_t * label,nvlist_t * config)2449 spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label,
2450     nvlist_t *config)
2451 {
2452 	uint64_t state = 0;
2453 	uint64_t hostid = 0;
2454 	uint64_t tryconfig_txg = 0;
2455 	uint64_t tryconfig_timestamp = 0;
2456 	uint16_t tryconfig_mmp_seq = 0;
2457 	nvlist_t *nvinfo;
2458 
2459 	if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
2460 		nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
2461 		(void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG,
2462 		    &tryconfig_txg);
2463 		(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
2464 		    &tryconfig_timestamp);
2465 		(void) nvlist_lookup_uint16(nvinfo, ZPOOL_CONFIG_MMP_SEQ,
2466 		    &tryconfig_mmp_seq);
2467 	}
2468 
2469 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state);
2470 
2471 	/*
2472 	 * Disable the MMP activity check - This is used by zdb which
2473 	 * is intended to be used on potentially active pools.
2474 	 */
2475 	if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP)
2476 		return (B_FALSE);
2477 
2478 	/*
2479 	 * Skip the activity check when the MMP feature is disabled.
2480 	 */
2481 	if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0)
2482 		return (B_FALSE);
2483 
2484 	/*
2485 	 * If the tryconfig_ values are nonzero, they are the results of an
2486 	 * earlier tryimport.  If they all match the uberblock we just found,
2487 	 * then the pool has not changed and we return false so we do not test
2488 	 * a second time.
2489 	 */
2490 	if (tryconfig_txg && tryconfig_txg == ub->ub_txg &&
2491 	    tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp &&
2492 	    tryconfig_mmp_seq && tryconfig_mmp_seq ==
2493 	    (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0))
2494 		return (B_FALSE);
2495 
2496 	/*
2497 	 * Allow the activity check to be skipped when importing the pool
2498 	 * on the same host which last imported it.  Since the hostid from
2499 	 * configuration may be stale use the one read from the label.
2500 	 */
2501 	if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID))
2502 		hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID);
2503 
2504 	if (hostid == spa_get_hostid())
2505 		return (B_FALSE);
2506 
2507 	/*
2508 	 * Skip the activity test when the pool was cleanly exported.
2509 	 */
2510 	if (state != POOL_STATE_ACTIVE)
2511 		return (B_FALSE);
2512 
2513 	return (B_TRUE);
2514 }
2515 
2516 /*
2517  * Nanoseconds the activity check must watch for changes on-disk.
2518  */
2519 static uint64_t
spa_activity_check_duration(spa_t * spa,uberblock_t * ub)2520 spa_activity_check_duration(spa_t *spa, uberblock_t *ub)
2521 {
2522 	uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1);
2523 	uint64_t multihost_interval = MSEC2NSEC(
2524 	    MMP_INTERVAL_OK(zfs_multihost_interval));
2525 	uint64_t import_delay = MAX(NANOSEC, import_intervals *
2526 	    multihost_interval);
2527 
2528 	/*
2529 	 * Local tunables determine a minimum duration except for the case
2530 	 * where we know when the remote host will suspend the pool if MMP
2531 	 * writes do not land.
2532 	 *
2533 	 * See Big Theory comment at the top of mmp.c for the reasoning behind
2534 	 * these cases and times.
2535 	 */
2536 
2537 	ASSERT(MMP_IMPORT_SAFETY_FACTOR >= 100);
2538 
2539 	if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
2540 	    MMP_FAIL_INT(ub) > 0) {
2541 
2542 		/* MMP on remote host will suspend pool after failed writes */
2543 		import_delay = MMP_FAIL_INT(ub) * MSEC2NSEC(MMP_INTERVAL(ub)) *
2544 		    MMP_IMPORT_SAFETY_FACTOR / 100;
2545 
2546 		zfs_dbgmsg("fail_intvals>0 import_delay=%llu ub_mmp "
2547 		    "mmp_fails=%llu ub_mmp mmp_interval=%llu "
2548 		    "import_intervals=%u", import_delay, MMP_FAIL_INT(ub),
2549 		    MMP_INTERVAL(ub), import_intervals);
2550 
2551 	} else if (MMP_INTERVAL_VALID(ub) && MMP_FAIL_INT_VALID(ub) &&
2552 	    MMP_FAIL_INT(ub) == 0) {
2553 
2554 		/* MMP on remote host will never suspend pool */
2555 		import_delay = MAX(import_delay, (MSEC2NSEC(MMP_INTERVAL(ub)) +
2556 		    ub->ub_mmp_delay) * import_intervals);
2557 
2558 		zfs_dbgmsg("fail_intvals=0 import_delay=%llu ub_mmp "
2559 		    "mmp_interval=%llu ub_mmp_delay=%llu "
2560 		    "import_intervals=%u", import_delay, MMP_INTERVAL(ub),
2561 		    ub->ub_mmp_delay, import_intervals);
2562 
2563 	} else if (MMP_VALID(ub)) {
2564 		/*
2565 		 * zfs-0.7 compatability case
2566 		 */
2567 
2568 		import_delay = MAX(import_delay, (multihost_interval +
2569 		    ub->ub_mmp_delay) * import_intervals);
2570 
2571 		zfs_dbgmsg("import_delay=%llu ub_mmp_delay=%llu "
2572 		    "import_intervals=%u leaves=%u", import_delay,
2573 		    ub->ub_mmp_delay, import_intervals,
2574 		    vdev_count_leaves(spa));
2575 	} else {
2576 		/* Using local tunings is the only reasonable option */
2577 		zfs_dbgmsg("pool last imported on non-MMP aware "
2578 		    "host using import_delay=%llu multihost_interval=%llu "
2579 		    "import_intervals=%u", import_delay, multihost_interval,
2580 		    import_intervals);
2581 	}
2582 
2583 	return (import_delay);
2584 }
2585 
2586 /*
2587  * Perform the import activity check.  If the user canceled the import or
2588  * we detected activity then fail.
2589  */
2590 static int
spa_activity_check(spa_t * spa,uberblock_t * ub,nvlist_t * config)2591 spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
2592 {
2593 	uint64_t txg = ub->ub_txg;
2594 	uint64_t timestamp = ub->ub_timestamp;
2595 	uint64_t mmp_config = ub->ub_mmp_config;
2596 	uint16_t mmp_seq = MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0;
2597 	uint64_t import_delay;
2598 	hrtime_t import_expire;
2599 	nvlist_t *mmp_label = NULL;
2600 	vdev_t *rvd = spa->spa_root_vdev;
2601 	kcondvar_t cv;
2602 	kmutex_t mtx;
2603 	int error = 0;
2604 
2605 	cv_init(&cv, NULL, CV_DEFAULT, NULL);
2606 	mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL);
2607 	mutex_enter(&mtx);
2608 
2609 	/*
2610 	 * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
2611 	 * during the earlier tryimport.  If the txg recorded there is 0 then
2612 	 * the pool is known to be active on another host.
2613 	 *
2614 	 * Otherwise, the pool might be in use on another host.  Check for
2615 	 * changes in the uberblocks on disk if necessary.
2616 	 */
2617 	if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
2618 		nvlist_t *nvinfo = fnvlist_lookup_nvlist(config,
2619 		    ZPOOL_CONFIG_LOAD_INFO);
2620 
2621 		if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) &&
2622 		    fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) {
2623 			vdev_uberblock_load(rvd, ub, &mmp_label);
2624 			error = SET_ERROR(EREMOTEIO);
2625 			goto out;
2626 		}
2627 	}
2628 
2629 	import_delay = spa_activity_check_duration(spa, ub);
2630 
2631 	/* Add a small random factor in case of simultaneous imports (0-25%) */
2632 	import_delay += import_delay * spa_get_random(250) / 1000;
2633 
2634 	import_expire = gethrtime() + import_delay;
2635 
2636 	while (gethrtime() < import_expire) {
2637 		(void) spa_import_progress_set_mmp_check(spa,
2638 		    NSEC2SEC(import_expire - gethrtime()));
2639 
2640 		vdev_uberblock_load(rvd, ub, &mmp_label);
2641 
2642 		if (txg != ub->ub_txg || timestamp != ub->ub_timestamp ||
2643 		    mmp_seq != (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0)) {
2644 			zfs_dbgmsg("multihost activity detected "
2645 			    "txg %llu ub_txg  %llu "
2646 			    "timestamp %llu ub_timestamp  %llu "
2647 			    "mmp_config %#llx ub_mmp_config %#llx",
2648 			    txg, ub->ub_txg, timestamp, ub->ub_timestamp,
2649 			    mmp_config, ub->ub_mmp_config);
2650 
2651 			error = SET_ERROR(EREMOTEIO);
2652 			break;
2653 		}
2654 
2655 		if (mmp_label) {
2656 			nvlist_free(mmp_label);
2657 			mmp_label = NULL;
2658 		}
2659 
2660 		error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz);
2661 		if (error != -1) {
2662 			error = SET_ERROR(EINTR);
2663 			break;
2664 		}
2665 		error = 0;
2666 	}
2667 
2668 out:
2669 	mutex_exit(&mtx);
2670 	mutex_destroy(&mtx);
2671 	cv_destroy(&cv);
2672 
2673 	/*
2674 	 * If the pool is determined to be active store the status in the
2675 	 * spa->spa_load_info nvlist.  If the remote hostname or hostid are
2676 	 * available from configuration read from disk store them as well.
2677 	 * This allows 'zpool import' to generate a more useful message.
2678 	 *
2679 	 * ZPOOL_CONFIG_MMP_STATE    - observed pool status (mandatory)
2680 	 * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
2681 	 * ZPOOL_CONFIG_MMP_HOSTID   - hostid from the active pool
2682 	 */
2683 	if (error == EREMOTEIO) {
2684 		char *hostname = "<unknown>";
2685 		uint64_t hostid = 0;
2686 
2687 		if (mmp_label) {
2688 			if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) {
2689 				hostname = fnvlist_lookup_string(mmp_label,
2690 				    ZPOOL_CONFIG_HOSTNAME);
2691 				fnvlist_add_string(spa->spa_load_info,
2692 				    ZPOOL_CONFIG_MMP_HOSTNAME, hostname);
2693 			}
2694 
2695 			if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) {
2696 				hostid = fnvlist_lookup_uint64(mmp_label,
2697 				    ZPOOL_CONFIG_HOSTID);
2698 				fnvlist_add_uint64(spa->spa_load_info,
2699 				    ZPOOL_CONFIG_MMP_HOSTID, hostid);
2700 			}
2701 		}
2702 
2703 		fnvlist_add_uint64(spa->spa_load_info,
2704 		    ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE);
2705 		fnvlist_add_uint64(spa->spa_load_info,
2706 		    ZPOOL_CONFIG_MMP_TXG, 0);
2707 
2708 		error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO);
2709 	}
2710 
2711 	if (mmp_label)
2712 		nvlist_free(mmp_label);
2713 
2714 	return (error);
2715 }
2716 
2717 static int
spa_verify_host(spa_t * spa,nvlist_t * mos_config)2718 spa_verify_host(spa_t *spa, nvlist_t *mos_config)
2719 {
2720 	uint64_t hostid;
2721 	char *hostname;
2722 	uint64_t myhostid = 0;
2723 
2724 	if (!spa_is_root(spa) && nvlist_lookup_uint64(mos_config,
2725 	    ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2726 		hostname = fnvlist_lookup_string(mos_config,
2727 		    ZPOOL_CONFIG_HOSTNAME);
2728 
2729 		myhostid = zone_get_hostid(NULL);
2730 
2731 		if (hostid != 0 && myhostid != 0 && hostid != myhostid) {
2732 			cmn_err(CE_WARN, "pool '%s' could not be "
2733 			    "loaded as it was last accessed by "
2734 			    "another system (host: %s hostid: 0x%llx). "
2735 			    "See: http://illumos.org/msg/ZFS-8000-EY",
2736 			    spa_name(spa), hostname, (u_longlong_t)hostid);
2737 			spa_load_failed(spa, "hostid verification failed: pool "
2738 			    "last accessed by host: %s (hostid: 0x%llx)",
2739 			    hostname, (u_longlong_t)hostid);
2740 			return (SET_ERROR(EBADF));
2741 		}
2742 	}
2743 
2744 	return (0);
2745 }
2746 
2747 static int
spa_ld_parse_config(spa_t * spa,spa_import_type_t type)2748 spa_ld_parse_config(spa_t *spa, spa_import_type_t type)
2749 {
2750 	int error = 0;
2751 	nvlist_t *nvtree, *nvl, *config = spa->spa_config;
2752 	int parse;
2753 	vdev_t *rvd;
2754 	uint64_t pool_guid;
2755 	char *comment;
2756 
2757 	/*
2758 	 * Versioning wasn't explicitly added to the label until later, so if
2759 	 * it's not present treat it as the initial version.
2760 	 */
2761 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2762 	    &spa->spa_ubsync.ub_version) != 0)
2763 		spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2764 
2765 	if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid)) {
2766 		spa_load_failed(spa, "invalid config provided: '%s' missing",
2767 		    ZPOOL_CONFIG_POOL_GUID);
2768 		return (SET_ERROR(EINVAL));
2769 	}
2770 
2771 	/*
2772 	 * If we are doing an import, ensure that the pool is not already
2773 	 * imported by checking if its pool guid already exists in the
2774 	 * spa namespace.
2775 	 *
2776 	 * The only case that we allow an already imported pool to be
2777 	 * imported again, is when the pool is checkpointed and we want to
2778 	 * look at its checkpointed state from userland tools like zdb.
2779 	 */
2780 #ifdef _KERNEL
2781 	if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2782 	    spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2783 	    spa_guid_exists(pool_guid, 0)) {
2784 #else
2785 	if ((spa->spa_load_state == SPA_LOAD_IMPORT ||
2786 	    spa->spa_load_state == SPA_LOAD_TRYIMPORT) &&
2787 	    spa_guid_exists(pool_guid, 0) &&
2788 	    !spa_importing_readonly_checkpoint(spa)) {
2789 #endif
2790 		spa_load_failed(spa, "a pool with guid %llu is already open",
2791 		    (u_longlong_t)pool_guid);
2792 		return (SET_ERROR(EEXIST));
2793 	}
2794 
2795 	spa->spa_config_guid = pool_guid;
2796 
2797 	nvlist_free(spa->spa_load_info);
2798 	spa->spa_load_info = fnvlist_alloc();
2799 
2800 	ASSERT(spa->spa_comment == NULL);
2801 	if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2802 		spa->spa_comment = spa_strdup(comment);
2803 
2804 	(void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2805 	    &spa->spa_config_txg);
2806 
2807 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) == 0)
2808 		spa->spa_config_splitting = fnvlist_dup(nvl);
2809 
2810 	if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvtree)) {
2811 		spa_load_failed(spa, "invalid config provided: '%s' missing",
2812 		    ZPOOL_CONFIG_VDEV_TREE);
2813 		return (SET_ERROR(EINVAL));
2814 	}
2815 
2816 	/*
2817 	 * Create "The Godfather" zio to hold all async IOs
2818 	 */
2819 	spa->spa_async_zio_root = kmem_alloc(max_ncpus * sizeof (void *),
2820 	    KM_SLEEP);
2821 	for (int i = 0; i < max_ncpus; i++) {
2822 		spa->spa_async_zio_root[i] = zio_root(spa, NULL, NULL,
2823 		    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
2824 		    ZIO_FLAG_GODFATHER);
2825 	}
2826 
2827 	/*
2828 	 * Parse the configuration into a vdev tree.  We explicitly set the
2829 	 * value that will be returned by spa_version() since parsing the
2830 	 * configuration requires knowing the version number.
2831 	 */
2832 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2833 	parse = (type == SPA_IMPORT_EXISTING ?
2834 	    VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2835 	error = spa_config_parse(spa, &rvd, nvtree, NULL, 0, parse);
2836 	spa_config_exit(spa, SCL_ALL, FTAG);
2837 
2838 	if (error != 0) {
2839 		spa_load_failed(spa, "unable to parse config [error=%d]",
2840 		    error);
2841 		return (error);
2842 	}
2843 
2844 	ASSERT(spa->spa_root_vdev == rvd);
2845 	ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
2846 	ASSERT3U(spa->spa_max_ashift, <=, SPA_MAXBLOCKSHIFT);
2847 
2848 	if (type != SPA_IMPORT_ASSEMBLE) {
2849 		ASSERT(spa_guid(spa) == pool_guid);
2850 	}
2851 
2852 	return (0);
2853 }
2854 
2855 /*
2856  * Recursively open all vdevs in the vdev tree. This function is called twice:
2857  * first with the untrusted config, then with the trusted config.
2858  */
2859 static int
2860 spa_ld_open_vdevs(spa_t *spa)
2861 {
2862 	int error = 0;
2863 
2864 	/*
2865 	 * spa_missing_tvds_allowed defines how many top-level vdevs can be
2866 	 * missing/unopenable for the root vdev to be still considered openable.
2867 	 */
2868 	if (spa->spa_trust_config) {
2869 		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds;
2870 	} else if (spa->spa_config_source == SPA_CONFIG_SRC_CACHEFILE) {
2871 		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_cachefile;
2872 	} else if (spa->spa_config_source == SPA_CONFIG_SRC_SCAN) {
2873 		spa->spa_missing_tvds_allowed = zfs_max_missing_tvds_scan;
2874 	} else {
2875 		spa->spa_missing_tvds_allowed = 0;
2876 	}
2877 
2878 	spa->spa_missing_tvds_allowed =
2879 	    MAX(zfs_max_missing_tvds, spa->spa_missing_tvds_allowed);
2880 
2881 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2882 	error = vdev_open(spa->spa_root_vdev);
2883 	spa_config_exit(spa, SCL_ALL, FTAG);
2884 
2885 	if (spa->spa_missing_tvds != 0) {
2886 		spa_load_note(spa, "vdev tree has %lld missing top-level "
2887 		    "vdevs.", (u_longlong_t)spa->spa_missing_tvds);
2888 		if (spa->spa_trust_config && (spa->spa_mode & FWRITE)) {
2889 			/*
2890 			 * Although theoretically we could allow users to open
2891 			 * incomplete pools in RW mode, we'd need to add a lot
2892 			 * of extra logic (e.g. adjust pool space to account
2893 			 * for missing vdevs).
2894 			 * This limitation also prevents users from accidentally
2895 			 * opening the pool in RW mode during data recovery and
2896 			 * damaging it further.
2897 			 */
2898 			spa_load_note(spa, "pools with missing top-level "
2899 			    "vdevs can only be opened in read-only mode.");
2900 			error = SET_ERROR(ENXIO);
2901 		} else {
2902 			spa_load_note(spa, "current settings allow for maximum "
2903 			    "%lld missing top-level vdevs at this stage.",
2904 			    (u_longlong_t)spa->spa_missing_tvds_allowed);
2905 		}
2906 	}
2907 	if (error != 0) {
2908 		spa_load_failed(spa, "unable to open vdev tree [error=%d]",
2909 		    error);
2910 	}
2911 	if (spa->spa_missing_tvds != 0 || error != 0)
2912 		vdev_dbgmsg_print_tree(spa->spa_root_vdev, 2);
2913 
2914 	return (error);
2915 }
2916 
2917 /*
2918  * We need to validate the vdev labels against the configuration that
2919  * we have in hand. This function is called twice: first with an untrusted
2920  * config, then with a trusted config. The validation is more strict when the
2921  * config is trusted.
2922  */
2923 static int
2924 spa_ld_validate_vdevs(spa_t *spa)
2925 {
2926 	int error = 0;
2927 	vdev_t *rvd = spa->spa_root_vdev;
2928 
2929 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2930 	error = vdev_validate(rvd);
2931 	spa_config_exit(spa, SCL_ALL, FTAG);
2932 
2933 	if (error != 0) {
2934 		spa_load_failed(spa, "vdev_validate failed [error=%d]", error);
2935 		return (error);
2936 	}
2937 
2938 	if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN) {
2939 		spa_load_failed(spa, "cannot open vdev tree after invalidating "
2940 		    "some vdevs");
2941 		vdev_dbgmsg_print_tree(rvd, 2);
2942 		return (SET_ERROR(ENXIO));
2943 	}
2944 
2945 	return (0);
2946 }
2947 
2948 static void
2949 spa_ld_select_uberblock_done(spa_t *spa, uberblock_t *ub)
2950 {
2951 	spa->spa_state = POOL_STATE_ACTIVE;
2952 	spa->spa_ubsync = spa->spa_uberblock;
2953 	spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2954 	    TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2955 	spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2956 	    spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2957 	spa->spa_claim_max_txg = spa->spa_first_txg;
2958 	spa->spa_prev_software_version = ub->ub_software_version;
2959 }
2960 
2961 static int
2962 spa_ld_select_uberblock(spa_t *spa, spa_import_type_t type)
2963 {
2964 	vdev_t *rvd = spa->spa_root_vdev;
2965 	nvlist_t *label;
2966 	uberblock_t *ub = &spa->spa_uberblock;
2967 	boolean_t activity_check = B_FALSE;
2968 
2969 	/*
2970 	 * If we are opening the checkpointed state of the pool by
2971 	 * rewinding to it, at this point we will have written the
2972 	 * checkpointed uberblock to the vdev labels, so searching
2973 	 * the labels will find the right uberblock.  However, if
2974 	 * we are opening the checkpointed state read-only, we have
2975 	 * not modified the labels. Therefore, we must ignore the
2976 	 * labels and continue using the spa_uberblock that was set
2977 	 * by spa_ld_checkpoint_rewind.
2978 	 *
2979 	 * Note that it would be fine to ignore the labels when
2980 	 * rewinding (opening writeable) as well. However, if we
2981 	 * crash just after writing the labels, we will end up
2982 	 * searching the labels. Doing so in the common case means
2983 	 * that this code path gets exercised normally, rather than
2984 	 * just in the edge case.
2985 	 */
2986 	if (ub->ub_checkpoint_txg != 0 &&
2987 	    spa_importing_readonly_checkpoint(spa)) {
2988 		spa_ld_select_uberblock_done(spa, ub);
2989 		return (0);
2990 	}
2991 
2992 	/*
2993 	 * Find the best uberblock.
2994 	 */
2995 	vdev_uberblock_load(rvd, ub, &label);
2996 
2997 	/*
2998 	 * If we weren't able to find a single valid uberblock, return failure.
2999 	 */
3000 	if (ub->ub_txg == 0) {
3001 		nvlist_free(label);
3002 		spa_load_failed(spa, "no valid uberblock found");
3003 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
3004 	}
3005 
3006 	if (spa->spa_load_max_txg != UINT64_MAX) {
3007 		(void) spa_import_progress_set_max_txg(spa,
3008 		    (u_longlong_t)spa->spa_load_max_txg);
3009 	}
3010 	spa_load_note(spa, "using uberblock with txg=%llu",
3011 	    (u_longlong_t)ub->ub_txg);
3012 
3013 	/*
3014 	 * For pools which have the multihost property on determine if the
3015 	 * pool is truly inactive and can be safely imported.  Prevent
3016 	 * hosts which don't have a hostid set from importing the pool.
3017 	 */
3018 	activity_check = spa_activity_check_required(spa, ub, label,
3019 	    spa->spa_config);
3020 	if (activity_check) {
3021 		if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay &&
3022 		    spa_get_hostid() == 0) {
3023 			nvlist_free(label);
3024 			fnvlist_add_uint64(spa->spa_load_info,
3025 			    ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
3026 			return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
3027 		}
3028 
3029 		int error = spa_activity_check(spa, ub, spa->spa_config);
3030 		if (error) {
3031 			nvlist_free(label);
3032 			return (error);
3033 		}
3034 
3035 		fnvlist_add_uint64(spa->spa_load_info,
3036 		    ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE);
3037 		fnvlist_add_uint64(spa->spa_load_info,
3038 		    ZPOOL_CONFIG_MMP_TXG, ub->ub_txg);
3039 		fnvlist_add_uint16(spa->spa_load_info,
3040 		    ZPOOL_CONFIG_MMP_SEQ,
3041 		    (MMP_SEQ_VALID(ub) ? MMP_SEQ(ub) : 0));
3042 	}
3043 
3044 	/*
3045 	 * If the pool has an unsupported version we can't open it.
3046 	 */
3047 	if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
3048 		nvlist_free(label);
3049 		spa_load_failed(spa, "version %llu is not supported",
3050 		    (u_longlong_t)ub->ub_version);
3051 		return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
3052 	}
3053 
3054 	if (ub->ub_version >= SPA_VERSION_FEATURES) {
3055 		nvlist_t *features;
3056 
3057 		/*
3058 		 * If we weren't able to find what's necessary for reading the
3059 		 * MOS in the label, return failure.
3060 		 */
3061 		if (label == NULL) {
3062 			spa_load_failed(spa, "label config unavailable");
3063 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3064 			    ENXIO));
3065 		}
3066 
3067 		if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_FEATURES_FOR_READ,
3068 		    &features) != 0) {
3069 			nvlist_free(label);
3070 			spa_load_failed(spa, "invalid label: '%s' missing",
3071 			    ZPOOL_CONFIG_FEATURES_FOR_READ);
3072 			return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
3073 			    ENXIO));
3074 		}
3075 
3076 		/*
3077 		 * Update our in-core representation with the definitive values
3078 		 * from the label.
3079 		 */
3080 		nvlist_free(spa->spa_label_features);
3081 		VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
3082 	}
3083 
3084 	nvlist_free(label);
3085 
3086 	/*
3087 	 * Look through entries in the label nvlist's features_for_read. If
3088 	 * there is a feature listed there which we don't understand then we
3089 	 * cannot open a pool.
3090 	 */
3091 	if (ub->ub_version >= SPA_VERSION_FEATURES) {
3092 		nvlist_t *unsup_feat;
3093 
3094 		VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
3095 		    0);
3096 
3097 		for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
3098 		    NULL); nvp != NULL;
3099 		    nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
3100 			if (!zfeature_is_supported(nvpair_name(nvp))) {
3101 				VERIFY(nvlist_add_string(unsup_feat,
3102 				    nvpair_name(nvp), "") == 0);
3103 			}
3104 		}
3105 
3106 		if (!nvlist_empty(unsup_feat)) {
3107 			VERIFY(nvlist_add_nvlist(spa->spa_load_info,
3108 			    ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
3109 			nvlist_free(unsup_feat);
3110 			spa_load_failed(spa, "some features are unsupported");
3111 			return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
3112 			    ENOTSUP));
3113 		}
3114 
3115 		nvlist_free(unsup_feat);
3116 	}
3117 
3118 	if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
3119 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3120 		spa_try_repair(spa, spa->spa_config);
3121 		spa_config_exit(spa, SCL_ALL, FTAG);
3122 		nvlist_free(spa->spa_config_splitting);
3123 		spa->spa_config_splitting = NULL;
3124 	}
3125 
3126 	/*
3127 	 * Initialize internal SPA structures.
3128 	 */
3129 	spa_ld_select_uberblock_done(spa, ub);
3130 
3131 	return (0);
3132 }
3133 
3134 static int
3135 spa_ld_open_rootbp(spa_t *spa)
3136 {
3137 	int error = 0;
3138 	vdev_t *rvd = spa->spa_root_vdev;
3139 
3140 	error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
3141 	if (error != 0) {
3142 		spa_load_failed(spa, "unable to open rootbp in dsl_pool_init "
3143 		    "[error=%d]", error);
3144 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3145 	}
3146 	spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
3147 
3148 	return (0);
3149 }
3150 
3151 static int
3152 spa_ld_trusted_config(spa_t *spa, spa_import_type_t type,
3153     boolean_t reloading)
3154 {
3155 	vdev_t *mrvd, *rvd = spa->spa_root_vdev;
3156 	nvlist_t *nv, *mos_config, *policy;
3157 	int error = 0, copy_error;
3158 	uint64_t healthy_tvds, healthy_tvds_mos;
3159 	uint64_t mos_config_txg;
3160 
3161 	if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object, B_TRUE)
3162 	    != 0)
3163 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3164 
3165 	/*
3166 	 * If we're assembling a pool from a split, the config provided is
3167 	 * already trusted so there is nothing to do.
3168 	 */
3169 	if (type == SPA_IMPORT_ASSEMBLE)
3170 		return (0);
3171 
3172 	healthy_tvds = spa_healthy_core_tvds(spa);
3173 
3174 	if (load_nvlist(spa, spa->spa_config_object, &mos_config)
3175 	    != 0) {
3176 		spa_load_failed(spa, "unable to retrieve MOS config");
3177 		return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
3178 	}
3179 
3180 	/*
3181 	 * If we are doing an open, pool owner wasn't verified yet, thus do
3182 	 * the verification here.
3183 	 */
3184 	if (spa->spa_load_state == SPA_LOAD_OPEN) {
3185 		error = spa_verify_host(spa, mos_config);
3186 		if (error != 0) {
3187 			nvlist_free(mos_config);
3188 			return (error);
3189 		}
3190 	}
3191 
3192 	nv = fnvlist_lookup_nvlist(mos_config, ZPOOL_CONFIG_VDEV_TREE);
3193 
3194 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3195 
3196 	/*
3197 	 * Build a new vdev tree from the trusted config
3198 	 */
3199 	VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
3200 
3201 	/*
3202 	 * Vdev paths in the MOS may be obsolete. If the untrusted config was
3203 	 * obtained by scanning /dev/dsk, then it will have the right vdev
3204 	 * paths. We update the trusted MOS config with this information.
3205 	 * We first try to copy the paths with vdev_copy_path_strict, which
3206 	 * succeeds only when both configs have exactly the same vdev tree.
3207 	 * If that fails, we fall back to a more flexible method that has a
3208 	 * best effort policy.
3209 	 */
3210 	copy_error = vdev_copy_path_strict(rvd, mrvd);
3211 	if (copy_error != 0 || spa_load_print_vdev_tree) {
3212 		spa_load_note(spa, "provided vdev tree:");
3213 		vdev_dbgmsg_print_tree(rvd, 2);
3214 		spa_load_note(spa, "MOS vdev tree:");
3215 		vdev_dbgmsg_print_tree(mrvd, 2);
3216 	}
3217 	if (copy_error != 0) {
3218 		spa_load_note(spa, "vdev_copy_path_strict failed, falling "
3219 		    "back to vdev_copy_path_relaxed");
3220 		vdev_copy_path_relaxed(rvd, mrvd);
3221 	}
3222 
3223 	vdev_close(rvd);
3224 	vdev_free(rvd);
3225 	spa->spa_root_vdev = mrvd;
3226 	rvd = mrvd;
3227 	spa_config_exit(spa, SCL_ALL, FTAG);
3228 
3229 	/*
3230 	 * We will use spa_config if we decide to reload the spa or if spa_load
3231 	 * fails and we rewind. We must thus regenerate the config using the
3232 	 * MOS information with the updated paths. ZPOOL_LOAD_POLICY is used to
3233 	 * pass settings on how to load the pool and is not stored in the MOS.
3234 	 * We copy it over to our new, trusted config.
3235 	 */
3236 	mos_config_txg = fnvlist_lookup_uint64(mos_config,
3237 	    ZPOOL_CONFIG_POOL_TXG);
3238 	nvlist_free(mos_config);
3239 	mos_config = spa_config_generate(spa, NULL, mos_config_txg, B_FALSE);
3240 	if (nvlist_lookup_nvlist(spa->spa_config, ZPOOL_LOAD_POLICY,
3241 	    &policy) == 0)
3242 		fnvlist_add_nvlist(mos_config, ZPOOL_LOAD_POLICY, policy);
3243