xref: /titanic_41/usr/src/uts/common/fs/zfs/vdev_label.c (revision d5de16744667e32c8dd538e62d90c3a857910804)
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) 2013 by Delphix. All rights reserved.
25  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
26  */
27 
28 /*
29  * Virtual Device Labels
30  * ---------------------
31  *
32  * The vdev label serves several distinct purposes:
33  *
34  *	1. Uniquely identify this device as part of a ZFS pool and confirm its
35  *	   identity within the pool.
36  *
37  * 	2. Verify that all the devices given in a configuration are present
38  *         within the pool.
39  *
40  * 	3. Determine the uberblock for the pool.
41  *
42  * 	4. In case of an import operation, determine the configuration of the
43  *         toplevel vdev of which it is a part.
44  *
45  * 	5. If an import operation cannot find all the devices in the pool,
46  *         provide enough information to the administrator to determine which
47  *         devices are missing.
48  *
49  * It is important to note that while the kernel is responsible for writing the
50  * label, it only consumes the information in the first three cases.  The
51  * latter information is only consumed in userland when determining the
52  * configuration to import a pool.
53  *
54  *
55  * Label Organization
56  * ------------------
57  *
58  * Before describing the contents of the label, it's important to understand how
59  * the labels are written and updated with respect to the uberblock.
60  *
61  * When the pool configuration is altered, either because it was newly created
62  * or a device was added, we want to update all the labels such that we can deal
63  * with fatal failure at any point.  To this end, each disk has two labels which
64  * are updated before and after the uberblock is synced.  Assuming we have
65  * labels and an uberblock with the following transaction groups:
66  *
67  *              L1          UB          L2
68  *           +------+    +------+    +------+
69  *           |      |    |      |    |      |
70  *           | t10  |    | t10  |    | t10  |
71  *           |      |    |      |    |      |
72  *           +------+    +------+    +------+
73  *
74  * In this stable state, the labels and the uberblock were all updated within
75  * the same transaction group (10).  Each label is mirrored and checksummed, so
76  * that we can detect when we fail partway through writing the label.
77  *
78  * In order to identify which labels are valid, the labels are written in the
79  * following manner:
80  *
81  * 	1. For each vdev, update 'L1' to the new label
82  * 	2. Update the uberblock
83  * 	3. For each vdev, update 'L2' to the new label
84  *
85  * Given arbitrary failure, we can determine the correct label to use based on
86  * the transaction group.  If we fail after updating L1 but before updating the
87  * UB, we will notice that L1's transaction group is greater than the uberblock,
88  * so L2 must be valid.  If we fail after writing the uberblock but before
89  * writing L2, we will notice that L2's transaction group is less than L1, and
90  * therefore L1 is valid.
91  *
92  * Another added complexity is that not every label is updated when the config
93  * is synced.  If we add a single device, we do not want to have to re-write
94  * every label for every device in the pool.  This means that both L1 and L2 may
95  * be older than the pool uberblock, because the necessary information is stored
96  * on another vdev.
97  *
98  *
99  * On-disk Format
100  * --------------
101  *
102  * The vdev label consists of two distinct parts, and is wrapped within the
103  * vdev_label_t structure.  The label includes 8k of padding to permit legacy
104  * VTOC disk labels, but is otherwise ignored.
105  *
106  * The first half of the label is a packed nvlist which contains pool wide
107  * properties, per-vdev properties, and configuration information.  It is
108  * described in more detail below.
109  *
110  * The latter half of the label consists of a redundant array of uberblocks.
111  * These uberblocks are updated whenever a transaction group is committed,
112  * or when the configuration is updated.  When a pool is loaded, we scan each
113  * vdev for the 'best' uberblock.
114  *
115  *
116  * Configuration Information
117  * -------------------------
118  *
119  * The nvlist describing the pool and vdev contains the following elements:
120  *
121  * 	version		ZFS on-disk version
122  * 	name		Pool name
123  * 	state		Pool state
124  * 	txg		Transaction group in which this label was written
125  * 	pool_guid	Unique identifier for this pool
126  * 	vdev_tree	An nvlist describing vdev tree.
127  *	features_for_read
128  *			An nvlist of the features necessary for reading the MOS.
129  *
130  * Each leaf device label also contains the following:
131  *
132  * 	top_guid	Unique ID for top-level vdev in which this is contained
133  * 	guid		Unique ID for the leaf vdev
134  *
135  * The 'vs' configuration follows the format described in 'spa_config.c'.
136  */
137 
138 #include <sys/zfs_context.h>
139 #include <sys/spa.h>
140 #include <sys/spa_impl.h>
141 #include <sys/dmu.h>
142 #include <sys/zap.h>
143 #include <sys/vdev.h>
144 #include <sys/vdev_impl.h>
145 #include <sys/uberblock_impl.h>
146 #include <sys/metaslab.h>
147 #include <sys/zio.h>
148 #include <sys/dsl_scan.h>
149 #include <sys/fs/zfs.h>
150 
151 /*
152  * Basic routines to read and write from a vdev label.
153  * Used throughout the rest of this file.
154  */
155 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)156 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
157 {
158 	ASSERT(offset < sizeof (vdev_label_t));
159 	ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
160 
161 	return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
162 	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
163 }
164 
165 /*
166  * Returns back the vdev label associated with the passed in offset.
167  */
168 int
vdev_label_number(uint64_t psize,uint64_t offset)169 vdev_label_number(uint64_t psize, uint64_t offset)
170 {
171 	int l;
172 
173 	if (offset >= psize - VDEV_LABEL_END_SIZE) {
174 		offset -= psize - VDEV_LABEL_END_SIZE;
175 		offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
176 	}
177 	l = offset / sizeof (vdev_label_t);
178 	return (l < VDEV_LABELS ? l : -1);
179 }
180 
181 static void
vdev_label_read(zio_t * zio,vdev_t * vd,int l,void * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)182 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
183 	uint64_t size, zio_done_func_t *done, void *private, int flags)
184 {
185 	ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
186 	    SCL_STATE_ALL);
187 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
188 
189 	zio_nowait(zio_read_phys(zio, vd,
190 	    vdev_label_offset(vd->vdev_psize, l, offset),
191 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
192 	    ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
193 }
194 
195 static void
vdev_label_write(zio_t * zio,vdev_t * vd,int l,void * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)196 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
197 	uint64_t size, zio_done_func_t *done, void *private, int flags)
198 {
199 	ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
200 	    (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
201 	    (SCL_CONFIG | SCL_STATE) &&
202 	    dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
203 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
204 
205 	zio_nowait(zio_write_phys(zio, vd,
206 	    vdev_label_offset(vd->vdev_psize, l, offset),
207 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
208 	    ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
209 }
210 
211 /*
212  * Generate the nvlist representing this vdev's config.
213  */
214 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)215 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
216     vdev_config_flag_t flags)
217 {
218 	nvlist_t *nv = NULL;
219 
220 	nv = fnvlist_alloc();
221 
222 	fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
223 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
224 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
225 	fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
226 
227 	if (vd->vdev_path != NULL)
228 		fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
229 
230 	if (vd->vdev_devid != NULL)
231 		fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
232 
233 	if (vd->vdev_physpath != NULL)
234 		fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
235 		    vd->vdev_physpath);
236 
237 	if (vd->vdev_fru != NULL)
238 		fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
239 
240 	if (vd->vdev_nparity != 0) {
241 		ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
242 		    VDEV_TYPE_RAIDZ) == 0);
243 
244 		/*
245 		 * Make sure someone hasn't managed to sneak a fancy new vdev
246 		 * into a crufty old storage pool.
247 		 */
248 		ASSERT(vd->vdev_nparity == 1 ||
249 		    (vd->vdev_nparity <= 2 &&
250 		    spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
251 		    (vd->vdev_nparity <= 3 &&
252 		    spa_version(spa) >= SPA_VERSION_RAIDZ3));
253 
254 		/*
255 		 * Note that we'll add the nparity tag even on storage pools
256 		 * that only support a single parity device -- older software
257 		 * will just ignore it.
258 		 */
259 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
260 	}
261 
262 	if (vd->vdev_wholedisk != -1ULL)
263 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
264 		    vd->vdev_wholedisk);
265 
266 	if (vd->vdev_not_present)
267 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
268 
269 	if (vd->vdev_isspare)
270 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
271 
272 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
273 	    vd == vd->vdev_top) {
274 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
275 		    vd->vdev_ms_array);
276 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
277 		    vd->vdev_ms_shift);
278 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
279 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
280 		    vd->vdev_asize);
281 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
282 		if (vd->vdev_removing)
283 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
284 			    vd->vdev_removing);
285 	}
286 
287 	if (flags & VDEV_CONFIG_L2CACHE)
288 		/* indicate that we support L2ARC persistency */
289 		VERIFY(nvlist_add_boolean_value(nv,
290 		    ZPOOL_CONFIG_L2CACHE_PERSISTENT, B_TRUE) == 0);
291 
292 	if (vd->vdev_dtl_sm != NULL) {
293 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
294 		    space_map_object(vd->vdev_dtl_sm));
295 	}
296 
297 	if (vd->vdev_crtxg)
298 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
299 
300 	if (getstats) {
301 		vdev_stat_t vs;
302 		pool_scan_stat_t ps;
303 
304 		vdev_get_stats(vd, &vs);
305 		fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
306 		    (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
307 
308 		/* provide either current or previous scan information */
309 		if (spa_scan_get_stats(spa, &ps) == 0) {
310 			fnvlist_add_uint64_array(nv,
311 			    ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
312 			    sizeof (pool_scan_stat_t) / sizeof (uint64_t));
313 		}
314 	}
315 
316 	if (!vd->vdev_ops->vdev_op_leaf) {
317 		nvlist_t **child;
318 		int c, idx;
319 
320 		ASSERT(!vd->vdev_ishole);
321 
322 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
323 		    KM_SLEEP);
324 
325 		for (c = 0, idx = 0; c < vd->vdev_children; c++) {
326 			vdev_t *cvd = vd->vdev_child[c];
327 
328 			/*
329 			 * If we're generating an nvlist of removing
330 			 * vdevs then skip over any device which is
331 			 * not being removed.
332 			 */
333 			if ((flags & VDEV_CONFIG_REMOVING) &&
334 			    !cvd->vdev_removing)
335 				continue;
336 
337 			child[idx++] = vdev_config_generate(spa, cvd,
338 			    getstats, flags);
339 		}
340 
341 		if (idx) {
342 			fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
343 			    child, idx);
344 		}
345 
346 		for (c = 0; c < idx; c++)
347 			nvlist_free(child[c]);
348 
349 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
350 
351 	} else {
352 		const char *aux = NULL;
353 
354 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
355 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
356 		if (vd->vdev_resilver_txg != 0)
357 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
358 			    vd->vdev_resilver_txg);
359 		if (vd->vdev_faulted)
360 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
361 		if (vd->vdev_degraded)
362 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
363 		if (vd->vdev_removed)
364 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
365 		if (vd->vdev_unspare)
366 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
367 		if (vd->vdev_ishole)
368 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
369 
370 		switch (vd->vdev_stat.vs_aux) {
371 		case VDEV_AUX_ERR_EXCEEDED:
372 			aux = "err_exceeded";
373 			break;
374 
375 		case VDEV_AUX_EXTERNAL:
376 			aux = "external";
377 			break;
378 		}
379 
380 		if (aux != NULL)
381 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
382 
383 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
384 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
385 			    vd->vdev_orig_guid);
386 		}
387 	}
388 
389 	return (nv);
390 }
391 
392 /*
393  * Generate a view of the top-level vdevs.  If we currently have holes
394  * in the namespace, then generate an array which contains a list of holey
395  * vdevs.  Additionally, add the number of top-level children that currently
396  * exist.
397  */
398 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)399 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
400 {
401 	vdev_t *rvd = spa->spa_root_vdev;
402 	uint64_t *array;
403 	uint_t c, idx;
404 
405 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
406 
407 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
408 		vdev_t *tvd = rvd->vdev_child[c];
409 
410 		if (tvd->vdev_ishole)
411 			array[idx++] = c;
412 	}
413 
414 	if (idx) {
415 		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
416 		    array, idx) == 0);
417 	}
418 
419 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
420 	    rvd->vdev_children) == 0);
421 
422 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
423 }
424 
425 /*
426  * Returns the configuration from the label of the given vdev. For vdevs
427  * which don't have a txg value stored on their label (i.e. spares/cache)
428  * or have not been completely initialized (txg = 0) just return
429  * the configuration from the first valid label we find. Otherwise,
430  * find the most up-to-date label that does not exceed the specified
431  * 'txg' value.
432  */
433 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)434 vdev_label_read_config(vdev_t *vd, uint64_t txg)
435 {
436 	spa_t *spa = vd->vdev_spa;
437 	nvlist_t *config = NULL;
438 	vdev_phys_t *vp;
439 	zio_t *zio;
440 	uint64_t best_txg = 0;
441 	int error = 0;
442 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
443 	    ZIO_FLAG_SPECULATIVE;
444 
445 	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
446 
447 	if (!vdev_readable(vd))
448 		return (NULL);
449 
450 	vp = zio_buf_alloc(sizeof (vdev_phys_t));
451 
452 retry:
453 	for (int l = 0; l < VDEV_LABELS; l++) {
454 		nvlist_t *label = NULL;
455 
456 		zio = zio_root(spa, NULL, NULL, flags);
457 
458 		vdev_label_read(zio, vd, l, vp,
459 		    offsetof(vdev_label_t, vl_vdev_phys),
460 		    sizeof (vdev_phys_t), NULL, NULL, flags);
461 
462 		if (zio_wait(zio) == 0 &&
463 		    nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
464 		    &label, 0) == 0) {
465 			uint64_t label_txg = 0;
466 
467 			/*
468 			 * Auxiliary vdevs won't have txg values in their
469 			 * labels and newly added vdevs may not have been
470 			 * completely initialized so just return the
471 			 * configuration from the first valid label we
472 			 * encounter.
473 			 */
474 			error = nvlist_lookup_uint64(label,
475 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
476 			if ((error || label_txg == 0) && !config) {
477 				config = label;
478 				break;
479 			} else if (label_txg <= txg && label_txg > best_txg) {
480 				best_txg = label_txg;
481 				nvlist_free(config);
482 				config = fnvlist_dup(label);
483 			}
484 		}
485 
486 		if (label != NULL) {
487 			nvlist_free(label);
488 			label = NULL;
489 		}
490 	}
491 
492 	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
493 		flags |= ZIO_FLAG_TRYHARD;
494 		goto retry;
495 	}
496 
497 	zio_buf_free(vp, sizeof (vdev_phys_t));
498 
499 	return (config);
500 }
501 
502 /*
503  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
504  * in with the device guid if this spare is active elsewhere on the system.
505  */
506 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)507 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
508     uint64_t *spare_guid, uint64_t *l2cache_guid)
509 {
510 	spa_t *spa = vd->vdev_spa;
511 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
512 	uint64_t vdtxg = 0;
513 	nvlist_t *label;
514 
515 	if (spare_guid)
516 		*spare_guid = 0ULL;
517 	if (l2cache_guid)
518 		*l2cache_guid = 0ULL;
519 
520 	/*
521 	 * Read the label, if any, and perform some basic sanity checks.
522 	 */
523 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
524 		return (B_FALSE);
525 
526 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
527 	    &vdtxg);
528 
529 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
530 	    &state) != 0 ||
531 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
532 	    &device_guid) != 0) {
533 		nvlist_free(label);
534 		return (B_FALSE);
535 	}
536 
537 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
538 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
539 	    &pool_guid) != 0 ||
540 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
541 	    &txg) != 0)) {
542 		nvlist_free(label);
543 		return (B_FALSE);
544 	}
545 
546 	nvlist_free(label);
547 
548 	/*
549 	 * Check to see if this device indeed belongs to the pool it claims to
550 	 * be a part of.  The only way this is allowed is if the device is a hot
551 	 * spare (which we check for later on).
552 	 */
553 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
554 	    !spa_guid_exists(pool_guid, device_guid) &&
555 	    !spa_spare_exists(device_guid, NULL, NULL) &&
556 	    !spa_l2cache_exists(device_guid, NULL))
557 		return (B_FALSE);
558 
559 	/*
560 	 * If the transaction group is zero, then this an initialized (but
561 	 * unused) label.  This is only an error if the create transaction
562 	 * on-disk is the same as the one we're using now, in which case the
563 	 * user has attempted to add the same vdev multiple times in the same
564 	 * transaction.
565 	 */
566 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
567 	    txg == 0 && vdtxg == crtxg)
568 		return (B_TRUE);
569 
570 	/*
571 	 * Check to see if this is a spare device.  We do an explicit check for
572 	 * spa_has_spare() here because it may be on our pending list of spares
573 	 * to add.  We also check if it is an l2cache device.
574 	 */
575 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
576 	    spa_has_spare(spa, device_guid)) {
577 		if (spare_guid)
578 			*spare_guid = device_guid;
579 
580 		switch (reason) {
581 		case VDEV_LABEL_CREATE:
582 		case VDEV_LABEL_L2CACHE:
583 			return (B_TRUE);
584 
585 		case VDEV_LABEL_REPLACE:
586 			return (!spa_has_spare(spa, device_guid) ||
587 			    spare_pool != 0ULL);
588 
589 		case VDEV_LABEL_SPARE:
590 			return (spa_has_spare(spa, device_guid));
591 		}
592 	}
593 
594 	/*
595 	 * Check to see if this is an l2cache device.
596 	 */
597 	if (spa_l2cache_exists(device_guid, NULL))
598 		return (B_TRUE);
599 
600 	/*
601 	 * We can't rely on a pool's state if it's been imported
602 	 * read-only.  Instead we look to see if the pools is marked
603 	 * read-only in the namespace and set the state to active.
604 	 */
605 	if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
606 	    spa_mode(spa) == FREAD)
607 		state = POOL_STATE_ACTIVE;
608 
609 	/*
610 	 * If the device is marked ACTIVE, then this device is in use by another
611 	 * pool on the system.
612 	 */
613 	return (state == POOL_STATE_ACTIVE);
614 }
615 
616 /*
617  * Initialize a vdev label.  We check to make sure each leaf device is not in
618  * use, and writable.  We put down an initial label which we will later
619  * overwrite with a complete label.  Note that it's important to do this
620  * sequentially, not in parallel, so that we catch cases of multiple use of the
621  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
622  * itself.
623  */
624 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)625 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
626 {
627 	spa_t *spa = vd->vdev_spa;
628 	nvlist_t *label;
629 	vdev_phys_t *vp;
630 	char *pad2;
631 	uberblock_t *ub;
632 	zio_t *zio;
633 	char *buf;
634 	size_t buflen;
635 	int error;
636 	uint64_t spare_guid, l2cache_guid;
637 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
638 
639 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
640 
641 	for (int c = 0; c < vd->vdev_children; c++)
642 		if ((error = vdev_label_init(vd->vdev_child[c],
643 		    crtxg, reason)) != 0)
644 			return (error);
645 
646 	/* Track the creation time for this vdev */
647 	vd->vdev_crtxg = crtxg;
648 
649 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
650 		return (0);
651 
652 	/*
653 	 * Dead vdevs cannot be initialized.
654 	 */
655 	if (vdev_is_dead(vd))
656 		return (SET_ERROR(EIO));
657 
658 	/*
659 	 * Determine if the vdev is in use.
660 	 */
661 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
662 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
663 		return (SET_ERROR(EBUSY));
664 
665 	/*
666 	 * If this is a request to add or replace a spare or l2cache device
667 	 * that is in use elsewhere on the system, then we must update the
668 	 * guid (which was initialized to a random value) to reflect the
669 	 * actual GUID (which is shared between multiple pools).
670 	 */
671 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
672 	    spare_guid != 0ULL) {
673 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
674 
675 		vd->vdev_guid += guid_delta;
676 
677 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
678 			pvd->vdev_guid_sum += guid_delta;
679 
680 		/*
681 		 * If this is a replacement, then we want to fallthrough to the
682 		 * rest of the code.  If we're adding a spare, then it's already
683 		 * labeled appropriately and we can just return.
684 		 */
685 		if (reason == VDEV_LABEL_SPARE)
686 			return (0);
687 		ASSERT(reason == VDEV_LABEL_REPLACE ||
688 		    reason == VDEV_LABEL_SPLIT);
689 	}
690 
691 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
692 	    l2cache_guid != 0ULL) {
693 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
694 
695 		vd->vdev_guid += guid_delta;
696 
697 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
698 			pvd->vdev_guid_sum += guid_delta;
699 
700 		/*
701 		 * If this is a replacement, then we want to fallthrough to the
702 		 * rest of the code.  If we're adding an l2cache, then it's
703 		 * already labeled appropriately and we can just return.
704 		 */
705 		if (reason == VDEV_LABEL_L2CACHE)
706 			return (0);
707 		ASSERT(reason == VDEV_LABEL_REPLACE);
708 	}
709 
710 	/*
711 	 * Initialize its label.
712 	 */
713 	vp = zio_buf_alloc(sizeof (vdev_phys_t));
714 	bzero(vp, sizeof (vdev_phys_t));
715 
716 	/*
717 	 * Generate a label describing the pool and our top-level vdev.
718 	 * We mark it as being from txg 0 to indicate that it's not
719 	 * really part of an active pool just yet.  The labels will
720 	 * be written again with a meaningful txg by spa_sync().
721 	 */
722 	if (reason == VDEV_LABEL_SPARE ||
723 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
724 		/*
725 		 * For inactive hot spares, we generate a special label that
726 		 * identifies as a mutually shared hot spare.  We write the
727 		 * label if we are adding a hot spare, or if we are removing an
728 		 * active hot spare (in which case we want to revert the
729 		 * labels).
730 		 */
731 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
732 
733 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
734 		    spa_version(spa)) == 0);
735 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
736 		    POOL_STATE_SPARE) == 0);
737 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
738 		    vd->vdev_guid) == 0);
739 	} else if (reason == VDEV_LABEL_L2CACHE ||
740 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
741 		/*
742 		 * For level 2 ARC devices, add a special label.
743 		 */
744 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
745 
746 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
747 		    spa_version(spa)) == 0);
748 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
749 		    POOL_STATE_L2CACHE) == 0);
750 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
751 		    vd->vdev_guid) == 0);
752 	} else {
753 		uint64_t txg = 0ULL;
754 
755 		if (reason == VDEV_LABEL_SPLIT)
756 			txg = spa->spa_uberblock.ub_txg;
757 		label = spa_config_generate(spa, vd, txg, B_FALSE);
758 
759 		/*
760 		 * Add our creation time.  This allows us to detect multiple
761 		 * vdev uses as described above, and automatically expires if we
762 		 * fail.
763 		 */
764 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
765 		    crtxg) == 0);
766 	}
767 
768 	buf = vp->vp_nvlist;
769 	buflen = sizeof (vp->vp_nvlist);
770 
771 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
772 	if (error != 0) {
773 		nvlist_free(label);
774 		zio_buf_free(vp, sizeof (vdev_phys_t));
775 		/* EFAULT means nvlist_pack ran out of room */
776 		return (error == EFAULT ? ENAMETOOLONG : EINVAL);
777 	}
778 
779 	/*
780 	 * Initialize uberblock template.
781 	 */
782 	ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
783 	bzero(ub, VDEV_UBERBLOCK_RING);
784 	*ub = spa->spa_uberblock;
785 	ub->ub_txg = 0;
786 
787 	/* Initialize the 2nd padding area. */
788 	pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
789 	bzero(pad2, VDEV_PAD_SIZE);
790 
791 	/*
792 	 * Write everything in parallel.
793 	 */
794 retry:
795 	zio = zio_root(spa, NULL, NULL, flags);
796 
797 	for (int l = 0; l < VDEV_LABELS; l++) {
798 
799 		vdev_label_write(zio, vd, l, vp,
800 		    offsetof(vdev_label_t, vl_vdev_phys),
801 		    sizeof (vdev_phys_t), NULL, NULL, flags);
802 
803 		/*
804 		 * Skip the 1st padding area.
805 		 * Zero out the 2nd padding area where it might have
806 		 * left over data from previous filesystem format.
807 		 */
808 		vdev_label_write(zio, vd, l, pad2,
809 		    offsetof(vdev_label_t, vl_pad2),
810 		    VDEV_PAD_SIZE, NULL, NULL, flags);
811 
812 		vdev_label_write(zio, vd, l, ub,
813 		    offsetof(vdev_label_t, vl_uberblock),
814 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
815 	}
816 
817 	error = zio_wait(zio);
818 
819 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
820 		flags |= ZIO_FLAG_TRYHARD;
821 		goto retry;
822 	}
823 
824 	nvlist_free(label);
825 	zio_buf_free(pad2, VDEV_PAD_SIZE);
826 	zio_buf_free(ub, VDEV_UBERBLOCK_RING);
827 	zio_buf_free(vp, sizeof (vdev_phys_t));
828 
829 	/*
830 	 * If this vdev hasn't been previously identified as a spare, then we
831 	 * mark it as such only if a) we are labeling it as a spare, or b) it
832 	 * exists as a spare elsewhere in the system.  Do the same for
833 	 * level 2 ARC devices.
834 	 */
835 	if (error == 0 && !vd->vdev_isspare &&
836 	    (reason == VDEV_LABEL_SPARE ||
837 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
838 		spa_spare_add(vd);
839 
840 	if (error == 0 && !vd->vdev_isl2cache &&
841 	    (reason == VDEV_LABEL_L2CACHE ||
842 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
843 		spa_l2cache_add(vd);
844 
845 	return (error);
846 }
847 
848 /*
849  * ==========================================================================
850  * uberblock load/sync
851  * ==========================================================================
852  */
853 
854 /*
855  * Consider the following situation: txg is safely synced to disk.  We've
856  * written the first uberblock for txg + 1, and then we lose power.  When we
857  * come back up, we fail to see the uberblock for txg + 1 because, say,
858  * it was on a mirrored device and the replica to which we wrote txg + 1
859  * is now offline.  If we then make some changes and sync txg + 1, and then
860  * the missing replica comes back, then for a few seconds we'll have two
861  * conflicting uberblocks on disk with the same txg.  The solution is simple:
862  * among uberblocks with equal txg, choose the one with the latest timestamp.
863  */
864 static int
vdev_uberblock_compare(uberblock_t * ub1,uberblock_t * ub2)865 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
866 {
867 	if (ub1->ub_txg < ub2->ub_txg)
868 		return (-1);
869 	if (ub1->ub_txg > ub2->ub_txg)
870 		return (1);
871 
872 	if (ub1->ub_timestamp < ub2->ub_timestamp)
873 		return (-1);
874 	if (ub1->ub_timestamp > ub2->ub_timestamp)
875 		return (1);
876 
877 	return (0);
878 }
879 
880 struct ubl_cbdata {
881 	uberblock_t	*ubl_ubbest;	/* Best uberblock */
882 	vdev_t		*ubl_vd;	/* vdev associated with the above */
883 };
884 
885 static void
vdev_uberblock_load_done(zio_t * zio)886 vdev_uberblock_load_done(zio_t *zio)
887 {
888 	vdev_t *vd = zio->io_vd;
889 	spa_t *spa = zio->io_spa;
890 	zio_t *rio = zio->io_private;
891 	uberblock_t *ub = zio->io_data;
892 	struct ubl_cbdata *cbp = rio->io_private;
893 
894 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
895 
896 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
897 		mutex_enter(&rio->io_lock);
898 		if (ub->ub_txg <= spa->spa_load_max_txg &&
899 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
900 			/*
901 			 * Keep track of the vdev in which this uberblock
902 			 * was found. We will use this information later
903 			 * to obtain the config nvlist associated with
904 			 * this uberblock.
905 			 */
906 			*cbp->ubl_ubbest = *ub;
907 			cbp->ubl_vd = vd;
908 		}
909 		mutex_exit(&rio->io_lock);
910 	}
911 
912 	zio_buf_free(zio->io_data, zio->io_size);
913 }
914 
915 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)916 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
917     struct ubl_cbdata *cbp)
918 {
919 	for (int c = 0; c < vd->vdev_children; c++)
920 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
921 
922 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
923 		for (int l = 0; l < VDEV_LABELS; l++) {
924 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
925 				vdev_label_read(zio, vd, l,
926 				    zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
927 				    VDEV_UBERBLOCK_OFFSET(vd, n),
928 				    VDEV_UBERBLOCK_SIZE(vd),
929 				    vdev_uberblock_load_done, zio, flags);
930 			}
931 		}
932 	}
933 }
934 
935 /*
936  * Reads the 'best' uberblock from disk along with its associated
937  * configuration. First, we read the uberblock array of each label of each
938  * vdev, keeping track of the uberblock with the highest txg in each array.
939  * Then, we read the configuration from the same vdev as the best uberblock.
940  */
941 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)942 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
943 {
944 	zio_t *zio;
945 	spa_t *spa = rvd->vdev_spa;
946 	struct ubl_cbdata cb;
947 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
948 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
949 
950 	ASSERT(ub);
951 	ASSERT(config);
952 
953 	bzero(ub, sizeof (uberblock_t));
954 	*config = NULL;
955 
956 	cb.ubl_ubbest = ub;
957 	cb.ubl_vd = NULL;
958 
959 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
960 	zio = zio_root(spa, NULL, &cb, flags);
961 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
962 	(void) zio_wait(zio);
963 
964 	/*
965 	 * It's possible that the best uberblock was discovered on a label
966 	 * that has a configuration which was written in a future txg.
967 	 * Search all labels on this vdev to find the configuration that
968 	 * matches the txg for our uberblock.
969 	 */
970 	if (cb.ubl_vd != NULL)
971 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
972 	spa_config_exit(spa, SCL_ALL, FTAG);
973 }
974 
975 /*
976  * On success, increment root zio's count of good writes.
977  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
978  */
979 static void
vdev_uberblock_sync_done(zio_t * zio)980 vdev_uberblock_sync_done(zio_t *zio)
981 {
982 	uint64_t *good_writes = zio->io_private;
983 
984 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
985 		atomic_inc_64(good_writes);
986 }
987 
988 /*
989  * Write the uberblock to all labels of all leaves of the specified vdev.
990  */
991 static void
vdev_uberblock_sync(zio_t * zio,uberblock_t * ub,vdev_t * vd,int flags)992 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
993 {
994 	uberblock_t *ubbuf;
995 	int n;
996 
997 	for (int c = 0; c < vd->vdev_children; c++)
998 		vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
999 
1000 	if (!vd->vdev_ops->vdev_op_leaf)
1001 		return;
1002 
1003 	if (!vdev_writeable(vd))
1004 		return;
1005 
1006 	n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1007 
1008 	ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1009 	bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1010 	*ubbuf = *ub;
1011 
1012 	for (int l = 0; l < VDEV_LABELS; l++)
1013 		vdev_label_write(zio, vd, l, ubbuf,
1014 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1015 		    vdev_uberblock_sync_done, zio->io_private,
1016 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1017 
1018 	zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1019 }
1020 
1021 /* Sync the uberblocks to all vdevs in svd[] */
1022 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1023 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1024 {
1025 	spa_t *spa = svd[0]->vdev_spa;
1026 	zio_t *zio;
1027 	uint64_t good_writes = 0;
1028 
1029 	zio = zio_root(spa, NULL, &good_writes, flags);
1030 
1031 	for (int v = 0; v < svdcount; v++)
1032 		vdev_uberblock_sync(zio, ub, svd[v], flags);
1033 
1034 	(void) zio_wait(zio);
1035 
1036 	/*
1037 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1038 	 * are no longer needed (because the new uberblocks and the even
1039 	 * labels are safely on disk), so it is safe to overwrite them.
1040 	 */
1041 	zio = zio_root(spa, NULL, NULL, flags);
1042 
1043 	for (int v = 0; v < svdcount; v++)
1044 		zio_flush(zio, svd[v]);
1045 
1046 	(void) zio_wait(zio);
1047 
1048 	return (good_writes >= 1 ? 0 : EIO);
1049 }
1050 
1051 /*
1052  * On success, increment the count of good writes for our top-level vdev.
1053  */
1054 static void
vdev_label_sync_done(zio_t * zio)1055 vdev_label_sync_done(zio_t *zio)
1056 {
1057 	uint64_t *good_writes = zio->io_private;
1058 
1059 	if (zio->io_error == 0)
1060 		atomic_inc_64(good_writes);
1061 }
1062 
1063 /*
1064  * If there weren't enough good writes, indicate failure to the parent.
1065  */
1066 static void
vdev_label_sync_top_done(zio_t * zio)1067 vdev_label_sync_top_done(zio_t *zio)
1068 {
1069 	uint64_t *good_writes = zio->io_private;
1070 
1071 	if (*good_writes == 0)
1072 		zio->io_error = SET_ERROR(EIO);
1073 
1074 	kmem_free(good_writes, sizeof (uint64_t));
1075 }
1076 
1077 /*
1078  * We ignore errors for log and cache devices, simply free the private data.
1079  */
1080 static void
vdev_label_sync_ignore_done(zio_t * zio)1081 vdev_label_sync_ignore_done(zio_t *zio)
1082 {
1083 	kmem_free(zio->io_private, sizeof (uint64_t));
1084 }
1085 
1086 /*
1087  * Write all even or odd labels to all leaves of the specified vdev.
1088  */
1089 static void
vdev_label_sync(zio_t * zio,vdev_t * vd,int l,uint64_t txg,int flags)1090 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1091 {
1092 	nvlist_t *label;
1093 	vdev_phys_t *vp;
1094 	char *buf;
1095 	size_t buflen;
1096 
1097 	for (int c = 0; c < vd->vdev_children; c++)
1098 		vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1099 
1100 	if (!vd->vdev_ops->vdev_op_leaf)
1101 		return;
1102 
1103 	if (!vdev_writeable(vd))
1104 		return;
1105 
1106 	/*
1107 	 * Generate a label describing the top-level config to which we belong.
1108 	 */
1109 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1110 
1111 	vp = zio_buf_alloc(sizeof (vdev_phys_t));
1112 	bzero(vp, sizeof (vdev_phys_t));
1113 
1114 	buf = vp->vp_nvlist;
1115 	buflen = sizeof (vp->vp_nvlist);
1116 
1117 	if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1118 		for (; l < VDEV_LABELS; l += 2) {
1119 			vdev_label_write(zio, vd, l, vp,
1120 			    offsetof(vdev_label_t, vl_vdev_phys),
1121 			    sizeof (vdev_phys_t),
1122 			    vdev_label_sync_done, zio->io_private,
1123 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1124 		}
1125 	}
1126 
1127 	zio_buf_free(vp, sizeof (vdev_phys_t));
1128 	nvlist_free(label);
1129 }
1130 
1131 int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1132 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1133 {
1134 	list_t *dl = &spa->spa_config_dirty_list;
1135 	vdev_t *vd;
1136 	zio_t *zio;
1137 	int error;
1138 
1139 	/*
1140 	 * Write the new labels to disk.
1141 	 */
1142 	zio = zio_root(spa, NULL, NULL, flags);
1143 
1144 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1145 		uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1146 		    KM_SLEEP);
1147 
1148 		ASSERT(!vd->vdev_ishole);
1149 
1150 		zio_t *vio = zio_null(zio, spa, NULL,
1151 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1152 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1153 		    good_writes, flags);
1154 		vdev_label_sync(vio, vd, l, txg, flags);
1155 		zio_nowait(vio);
1156 	}
1157 
1158 	error = zio_wait(zio);
1159 
1160 	/*
1161 	 * Flush the new labels to disk.
1162 	 */
1163 	zio = zio_root(spa, NULL, NULL, flags);
1164 
1165 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1166 		zio_flush(zio, vd);
1167 
1168 	(void) zio_wait(zio);
1169 
1170 	return (error);
1171 }
1172 
1173 /*
1174  * Sync the uberblock and any changes to the vdev configuration.
1175  *
1176  * The order of operations is carefully crafted to ensure that
1177  * if the system panics or loses power at any time, the state on disk
1178  * is still transactionally consistent.  The in-line comments below
1179  * describe the failure semantics at each stage.
1180  *
1181  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1182  * at any time, you can just call it again, and it will resume its work.
1183  */
1184 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg,boolean_t tryhard)1185 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1186 {
1187 	spa_t *spa = svd[0]->vdev_spa;
1188 	uberblock_t *ub = &spa->spa_uberblock;
1189 	vdev_t *vd;
1190 	zio_t *zio;
1191 	int error;
1192 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1193 
1194 	/*
1195 	 * Normally, we don't want to try too hard to write every label and
1196 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1197 	 * sync process to block while we retry.  But if we can't write a
1198 	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1199 	 * bailing out and declaring the pool faulted.
1200 	 */
1201 	if (tryhard)
1202 		flags |= ZIO_FLAG_TRYHARD;
1203 
1204 	ASSERT(ub->ub_txg <= txg);
1205 
1206 	/*
1207 	 * If this isn't a resync due to I/O errors,
1208 	 * and nothing changed in this transaction group,
1209 	 * and the vdev configuration hasn't changed,
1210 	 * then there's nothing to do.
1211 	 */
1212 	if (ub->ub_txg < txg &&
1213 	    uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1214 	    list_is_empty(&spa->spa_config_dirty_list))
1215 		return (0);
1216 
1217 	if (txg > spa_freeze_txg(spa))
1218 		return (0);
1219 
1220 	ASSERT(txg <= spa->spa_final_txg);
1221 
1222 	/*
1223 	 * Flush the write cache of every disk that's been written to
1224 	 * in this transaction group.  This ensures that all blocks
1225 	 * written in this txg will be committed to stable storage
1226 	 * before any uberblock that references them.
1227 	 */
1228 	zio = zio_root(spa, NULL, NULL, flags);
1229 
1230 	for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1231 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1232 		zio_flush(zio, vd);
1233 
1234 	(void) zio_wait(zio);
1235 
1236 	/*
1237 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1238 	 * system dies in the middle of this process, that's OK: all of the
1239 	 * even labels that made it to disk will be newer than any uberblock,
1240 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
1241 	 * which have not yet been touched, will still be valid.  We flush
1242 	 * the new labels to disk to ensure that all even-label updates
1243 	 * are committed to stable storage before the uberblock update.
1244 	 */
1245 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1246 		return (error);
1247 
1248 	/*
1249 	 * Sync the uberblocks to all vdevs in svd[].
1250 	 * If the system dies in the middle of this step, there are two cases
1251 	 * to consider, and the on-disk state is consistent either way:
1252 	 *
1253 	 * (1)	If none of the new uberblocks made it to disk, then the
1254 	 *	previous uberblock will be the newest, and the odd labels
1255 	 *	(which had not yet been touched) will be valid with respect
1256 	 *	to that uberblock.
1257 	 *
1258 	 * (2)	If one or more new uberblocks made it to disk, then they
1259 	 *	will be the newest, and the even labels (which had all
1260 	 *	been successfully committed) will be valid with respect
1261 	 *	to the new uberblocks.
1262 	 */
1263 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1264 		return (error);
1265 
1266 	/*
1267 	 * Sync out odd labels for every dirty vdev.  If the system dies
1268 	 * in the middle of this process, the even labels and the new
1269 	 * uberblocks will suffice to open the pool.  The next time
1270 	 * the pool is opened, the first thing we'll do -- before any
1271 	 * user data is modified -- is mark every vdev dirty so that
1272 	 * all labels will be brought up to date.  We flush the new labels
1273 	 * to disk to ensure that all odd-label updates are committed to
1274 	 * stable storage before the next transaction group begins.
1275 	 */
1276 	return (vdev_label_sync_list(spa, 1, txg, flags));
1277 }
1278