xref: /freebsd-13-stable/sys/contrib/openzfs/module/zfs/vdev_label.c (revision b9c2c366db1beb2ed276947056f45938ad8f57ec)
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) 2012, 2020 by Delphix. All rights reserved.
25  * Copyright (c) 2017, Intel Corporation.
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/vdev_draid.h>
146 #include <sys/uberblock_impl.h>
147 #include <sys/metaslab.h>
148 #include <sys/metaslab_impl.h>
149 #include <sys/zio.h>
150 #include <sys/dsl_scan.h>
151 #include <sys/abd.h>
152 #include <sys/fs/zfs.h>
153 #include <sys/byteorder.h>
154 #include <sys/zfs_bootenv.h>
155 
156 /*
157  * Basic routines to read and write from a vdev label.
158  * Used throughout the rest of this file.
159  */
160 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)161 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
162 {
163 	ASSERT(offset < sizeof (vdev_label_t));
164 	ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
165 
166 	return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
167 	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
168 }
169 
170 /*
171  * Returns back the vdev label associated with the passed in offset.
172  */
173 int
vdev_label_number(uint64_t psize,uint64_t offset)174 vdev_label_number(uint64_t psize, uint64_t offset)
175 {
176 	int l;
177 
178 	if (offset >= psize - VDEV_LABEL_END_SIZE) {
179 		offset -= psize - VDEV_LABEL_END_SIZE;
180 		offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
181 	}
182 	l = offset / sizeof (vdev_label_t);
183 	return (l < VDEV_LABELS ? l : -1);
184 }
185 
186 static void
vdev_label_read(zio_t * zio,vdev_t * vd,int l,abd_t * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)187 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
188     uint64_t size, zio_done_func_t *done, void *private, int flags)
189 {
190 	ASSERT(
191 	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
192 	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
193 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
194 
195 	zio_nowait(zio_read_phys(zio, vd,
196 	    vdev_label_offset(vd->vdev_psize, l, offset),
197 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
198 	    ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
199 }
200 
201 void
vdev_label_write(zio_t * zio,vdev_t * vd,int l,abd_t * buf,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,int flags)202 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
203     uint64_t size, zio_done_func_t *done, void *private, int flags)
204 {
205 	ASSERT(
206 	    spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
207 	    spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
208 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
209 
210 	zio_nowait(zio_write_phys(zio, vd,
211 	    vdev_label_offset(vd->vdev_psize, l, offset),
212 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
213 	    ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
214 }
215 
216 /*
217  * Generate the nvlist representing this vdev's stats
218  */
219 void
vdev_config_generate_stats(vdev_t * vd,nvlist_t * nv)220 vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
221 {
222 	nvlist_t *nvx;
223 	vdev_stat_t *vs;
224 	vdev_stat_ex_t *vsx;
225 
226 	vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
227 	vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
228 
229 	vdev_get_stats_ex(vd, vs, vsx);
230 	fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
231 	    (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
232 
233 	/*
234 	 * Add extended stats into a special extended stats nvlist.  This keeps
235 	 * all the extended stats nicely grouped together.  The extended stats
236 	 * nvlist is then added to the main nvlist.
237 	 */
238 	nvx = fnvlist_alloc();
239 
240 	/* ZIOs in flight to disk */
241 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,
242 	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]);
243 
244 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,
245 	    vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]);
246 
247 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,
248 	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]);
249 
250 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,
251 	    vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]);
252 
253 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,
254 	    vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]);
255 
256 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE,
257 	    vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]);
258 
259 	/* ZIOs pending */
260 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
261 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
262 
263 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
264 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
265 
266 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
267 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
268 
269 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
270 	    vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
271 
272 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
273 	    vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
274 
275 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
276 	    vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
277 
278 	/* Histograms */
279 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
280 	    vsx->vsx_total_histo[ZIO_TYPE_READ],
281 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
282 
283 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
284 	    vsx->vsx_total_histo[ZIO_TYPE_WRITE],
285 	    ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
286 
287 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
288 	    vsx->vsx_disk_histo[ZIO_TYPE_READ],
289 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
290 
291 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
292 	    vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
293 	    ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
294 
295 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
296 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
297 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
298 
299 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
300 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
301 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
302 
303 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
304 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
305 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
306 
307 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
308 	    vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
309 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
310 
311 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
312 	    vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
313 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
314 
315 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
316 	    vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
317 	    ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
318 
319 	/* Request sizes */
320 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
321 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
322 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
323 
324 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
325 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
326 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
327 
328 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
329 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
330 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
331 
332 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
333 	    vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
334 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
335 
336 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
337 	    vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
338 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
339 
340 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
341 	    vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
342 	    ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
343 
344 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
345 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
346 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
347 
348 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
349 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
350 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
351 
352 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
353 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
354 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
355 
356 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
357 	    vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
358 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
359 
360 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
361 	    vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
362 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
363 
364 	fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
365 	    vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
366 	    ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
367 
368 	/* IO delays */
369 	fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
370 
371 	/* Add extended stats nvlist to main nvlist */
372 	fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
373 
374 	fnvlist_free(nvx);
375 	kmem_free(vs, sizeof (*vs));
376 	kmem_free(vsx, sizeof (*vsx));
377 }
378 
379 static void
root_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)380 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
381 {
382 	spa_t *spa = vd->vdev_spa;
383 
384 	if (vd != spa->spa_root_vdev)
385 		return;
386 
387 	/* provide either current or previous scan information */
388 	pool_scan_stat_t ps;
389 	if (spa_scan_get_stats(spa, &ps) == 0) {
390 		fnvlist_add_uint64_array(nvl,
391 		    ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
392 		    sizeof (pool_scan_stat_t) / sizeof (uint64_t));
393 	}
394 
395 	pool_removal_stat_t prs;
396 	if (spa_removal_get_stats(spa, &prs) == 0) {
397 		fnvlist_add_uint64_array(nvl,
398 		    ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
399 		    sizeof (prs) / sizeof (uint64_t));
400 	}
401 
402 	pool_checkpoint_stat_t pcs;
403 	if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
404 		fnvlist_add_uint64_array(nvl,
405 		    ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
406 		    sizeof (pcs) / sizeof (uint64_t));
407 	}
408 }
409 
410 static void
top_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)411 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
412 {
413 	if (vd == vd->vdev_top) {
414 		vdev_rebuild_stat_t vrs;
415 		if (vdev_rebuild_get_stats(vd, &vrs) == 0) {
416 			fnvlist_add_uint64_array(nvl,
417 			    ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs,
418 			    sizeof (vrs) / sizeof (uint64_t));
419 		}
420 	}
421 }
422 
423 /*
424  * Generate the nvlist representing this vdev's config.
425  */
426 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)427 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
428     vdev_config_flag_t flags)
429 {
430 	nvlist_t *nv = NULL;
431 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
432 
433 	nv = fnvlist_alloc();
434 
435 	fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
436 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
437 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
438 	fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
439 
440 	if (vd->vdev_path != NULL)
441 		fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
442 
443 	if (vd->vdev_devid != NULL)
444 		fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
445 
446 	if (vd->vdev_physpath != NULL)
447 		fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
448 		    vd->vdev_physpath);
449 
450 	if (vd->vdev_enc_sysfs_path != NULL)
451 		fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
452 		    vd->vdev_enc_sysfs_path);
453 
454 	if (vd->vdev_fru != NULL)
455 		fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
456 
457 	if (vd->vdev_ops->vdev_op_config_generate != NULL)
458 		vd->vdev_ops->vdev_op_config_generate(vd, nv);
459 
460 	if (vd->vdev_wholedisk != -1ULL) {
461 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
462 		    vd->vdev_wholedisk);
463 	}
464 
465 	if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
466 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
467 
468 	if (vd->vdev_isspare)
469 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
470 
471 	if (flags & VDEV_CONFIG_L2CACHE)
472 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
473 
474 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
475 	    vd == vd->vdev_top) {
476 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
477 		    vd->vdev_ms_array);
478 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
479 		    vd->vdev_ms_shift);
480 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
481 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
482 		    vd->vdev_asize);
483 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
484 		if (vd->vdev_removing) {
485 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
486 			    vd->vdev_removing);
487 		}
488 
489 		/* zpool command expects alloc class data */
490 		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
491 			const char *bias = NULL;
492 
493 			switch (vd->vdev_alloc_bias) {
494 			case VDEV_BIAS_LOG:
495 				bias = VDEV_ALLOC_BIAS_LOG;
496 				break;
497 			case VDEV_BIAS_SPECIAL:
498 				bias = VDEV_ALLOC_BIAS_SPECIAL;
499 				break;
500 			case VDEV_BIAS_DEDUP:
501 				bias = VDEV_ALLOC_BIAS_DEDUP;
502 				break;
503 			default:
504 				ASSERT3U(vd->vdev_alloc_bias, ==,
505 				    VDEV_BIAS_NONE);
506 			}
507 			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
508 			    bias);
509 		}
510 	}
511 
512 	if (vd->vdev_dtl_sm != NULL) {
513 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
514 		    space_map_object(vd->vdev_dtl_sm));
515 	}
516 
517 	if (vic->vic_mapping_object != 0) {
518 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
519 		    vic->vic_mapping_object);
520 	}
521 
522 	if (vic->vic_births_object != 0) {
523 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
524 		    vic->vic_births_object);
525 	}
526 
527 	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
528 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
529 		    vic->vic_prev_indirect_vdev);
530 	}
531 
532 	if (vd->vdev_crtxg)
533 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
534 
535 	if (vd->vdev_expansion_time)
536 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
537 		    vd->vdev_expansion_time);
538 
539 	if (flags & VDEV_CONFIG_MOS) {
540 		if (vd->vdev_leaf_zap != 0) {
541 			ASSERT(vd->vdev_ops->vdev_op_leaf);
542 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
543 			    vd->vdev_leaf_zap);
544 		}
545 
546 		if (vd->vdev_top_zap != 0) {
547 			ASSERT(vd == vd->vdev_top);
548 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
549 			    vd->vdev_top_zap);
550 		}
551 
552 		if (vd->vdev_resilver_deferred) {
553 			ASSERT(vd->vdev_ops->vdev_op_leaf);
554 			ASSERT(spa->spa_resilver_deferred);
555 			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
556 		}
557 	}
558 
559 	if (getstats) {
560 		vdev_config_generate_stats(vd, nv);
561 
562 		root_vdev_actions_getprogress(vd, nv);
563 		top_vdev_actions_getprogress(vd, nv);
564 
565 		/*
566 		 * Note: this can be called from open context
567 		 * (spa_get_stats()), so we need the rwlock to prevent
568 		 * the mapping from being changed by condensing.
569 		 */
570 		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
571 		if (vd->vdev_indirect_mapping != NULL) {
572 			ASSERT(vd->vdev_indirect_births != NULL);
573 			vdev_indirect_mapping_t *vim =
574 			    vd->vdev_indirect_mapping;
575 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
576 			    vdev_indirect_mapping_size(vim));
577 		}
578 		rw_exit(&vd->vdev_indirect_rwlock);
579 		if (vd->vdev_mg != NULL &&
580 		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
581 			/*
582 			 * Compute approximately how much memory would be used
583 			 * for the indirect mapping if this device were to
584 			 * be removed.
585 			 *
586 			 * Note: If the frag metric is invalid, then not
587 			 * enough metaslabs have been converted to have
588 			 * histograms.
589 			 */
590 			uint64_t seg_count = 0;
591 			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
592 
593 			/*
594 			 * There are the same number of allocated segments
595 			 * as free segments, so we will have at least one
596 			 * entry per free segment.  However, small free
597 			 * segments (smaller than vdev_removal_max_span)
598 			 * will be combined with adjacent allocated segments
599 			 * as a single mapping.
600 			 */
601 			for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
602 				if (i + 1 < highbit64(vdev_removal_max_span)
603 				    - 1) {
604 					to_alloc +=
605 					    vd->vdev_mg->mg_histogram[i] <<
606 					    (i + 1);
607 				} else {
608 					seg_count +=
609 					    vd->vdev_mg->mg_histogram[i];
610 				}
611 			}
612 
613 			/*
614 			 * The maximum length of a mapping is
615 			 * zfs_remove_max_segment, so we need at least one entry
616 			 * per zfs_remove_max_segment of allocated data.
617 			 */
618 			seg_count += to_alloc / spa_remove_max_segment(spa);
619 
620 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
621 			    seg_count *
622 			    sizeof (vdev_indirect_mapping_entry_phys_t));
623 		}
624 	}
625 
626 	if (!vd->vdev_ops->vdev_op_leaf) {
627 		nvlist_t **child;
628 		int c, idx;
629 
630 		ASSERT(!vd->vdev_ishole);
631 
632 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
633 		    KM_SLEEP);
634 
635 		for (c = 0, idx = 0; c < vd->vdev_children; c++) {
636 			vdev_t *cvd = vd->vdev_child[c];
637 
638 			/*
639 			 * If we're generating an nvlist of removing
640 			 * vdevs then skip over any device which is
641 			 * not being removed.
642 			 */
643 			if ((flags & VDEV_CONFIG_REMOVING) &&
644 			    !cvd->vdev_removing)
645 				continue;
646 
647 			child[idx++] = vdev_config_generate(spa, cvd,
648 			    getstats, flags);
649 		}
650 
651 		if (idx) {
652 			fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
653 			    child, idx);
654 		}
655 
656 		for (c = 0; c < idx; c++)
657 			nvlist_free(child[c]);
658 
659 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
660 
661 	} else {
662 		const char *aux = NULL;
663 
664 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
665 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
666 		if (vd->vdev_resilver_txg != 0)
667 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
668 			    vd->vdev_resilver_txg);
669 		if (vd->vdev_rebuild_txg != 0)
670 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
671 			    vd->vdev_rebuild_txg);
672 		if (vd->vdev_faulted)
673 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
674 		if (vd->vdev_degraded)
675 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
676 		if (vd->vdev_removed)
677 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
678 		if (vd->vdev_unspare)
679 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
680 		if (vd->vdev_ishole)
681 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
682 
683 		/* Set the reason why we're FAULTED/DEGRADED. */
684 		switch (vd->vdev_stat.vs_aux) {
685 		case VDEV_AUX_ERR_EXCEEDED:
686 			aux = "err_exceeded";
687 			break;
688 
689 		case VDEV_AUX_EXTERNAL:
690 			aux = "external";
691 			break;
692 		}
693 
694 		if (aux != NULL && !vd->vdev_tmpoffline) {
695 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
696 		} else {
697 			/*
698 			 * We're healthy - clear any previous AUX_STATE values.
699 			 */
700 			if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
701 				nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
702 		}
703 
704 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
705 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
706 			    vd->vdev_orig_guid);
707 		}
708 	}
709 
710 	return (nv);
711 }
712 
713 /*
714  * Generate a view of the top-level vdevs.  If we currently have holes
715  * in the namespace, then generate an array which contains a list of holey
716  * vdevs.  Additionally, add the number of top-level children that currently
717  * exist.
718  */
719 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)720 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
721 {
722 	vdev_t *rvd = spa->spa_root_vdev;
723 	uint64_t *array;
724 	uint_t c, idx;
725 
726 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
727 
728 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
729 		vdev_t *tvd = rvd->vdev_child[c];
730 
731 		if (tvd->vdev_ishole) {
732 			array[idx++] = c;
733 		}
734 	}
735 
736 	if (idx) {
737 		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
738 		    array, idx) == 0);
739 	}
740 
741 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
742 	    rvd->vdev_children) == 0);
743 
744 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
745 }
746 
747 /*
748  * Returns the configuration from the label of the given vdev. For vdevs
749  * which don't have a txg value stored on their label (i.e. spares/cache)
750  * or have not been completely initialized (txg = 0) just return
751  * the configuration from the first valid label we find. Otherwise,
752  * find the most up-to-date label that does not exceed the specified
753  * 'txg' value.
754  */
755 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)756 vdev_label_read_config(vdev_t *vd, uint64_t txg)
757 {
758 	spa_t *spa = vd->vdev_spa;
759 	nvlist_t *config = NULL;
760 	vdev_phys_t *vp[VDEV_LABELS];
761 	abd_t *vp_abd[VDEV_LABELS];
762 	zio_t *zio[VDEV_LABELS];
763 	uint64_t best_txg = 0;
764 	uint64_t label_txg = 0;
765 	int error = 0;
766 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
767 	    ZIO_FLAG_SPECULATIVE;
768 
769 	ASSERT(vd->vdev_validate_thread == curthread ||
770 	    spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
771 
772 	if (!vdev_readable(vd))
773 		return (NULL);
774 
775 	/*
776 	 * The label for a dRAID distributed spare is not stored on disk.
777 	 * Instead it is generated when needed which allows us to bypass
778 	 * the pipeline when reading the config from the label.
779 	 */
780 	if (vd->vdev_ops == &vdev_draid_spare_ops)
781 		return (vdev_draid_read_config_spare(vd));
782 
783 	for (int l = 0; l < VDEV_LABELS; l++) {
784 		vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
785 		vp[l] = abd_to_buf(vp_abd[l]);
786 	}
787 
788 retry:
789 	for (int l = 0; l < VDEV_LABELS; l++) {
790 		zio[l] = zio_root(spa, NULL, NULL, flags);
791 
792 		vdev_label_read(zio[l], vd, l, vp_abd[l],
793 		    offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
794 		    NULL, NULL, flags);
795 	}
796 	for (int l = 0; l < VDEV_LABELS; l++) {
797 		nvlist_t *label = NULL;
798 
799 		if (zio_wait(zio[l]) == 0 &&
800 		    nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist),
801 		    &label, 0) == 0) {
802 			/*
803 			 * Auxiliary vdevs won't have txg values in their
804 			 * labels and newly added vdevs may not have been
805 			 * completely initialized so just return the
806 			 * configuration from the first valid label we
807 			 * encounter.
808 			 */
809 			error = nvlist_lookup_uint64(label,
810 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
811 			if ((error || label_txg == 0) && !config) {
812 				config = label;
813 				for (l++; l < VDEV_LABELS; l++)
814 					zio_wait(zio[l]);
815 				break;
816 			} else if (label_txg <= txg && label_txg > best_txg) {
817 				best_txg = label_txg;
818 				nvlist_free(config);
819 				config = fnvlist_dup(label);
820 			}
821 		}
822 
823 		if (label != NULL) {
824 			nvlist_free(label);
825 			label = NULL;
826 		}
827 	}
828 
829 	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
830 		flags |= ZIO_FLAG_TRYHARD;
831 		goto retry;
832 	}
833 
834 	/*
835 	 * We found a valid label but it didn't pass txg restrictions.
836 	 */
837 	if (config == NULL && label_txg != 0) {
838 		vdev_dbgmsg(vd, "label discarded as txg is too large "
839 		    "(%llu > %llu)", (u_longlong_t)label_txg,
840 		    (u_longlong_t)txg);
841 	}
842 
843 	for (int l = 0; l < VDEV_LABELS; l++) {
844 		abd_free(vp_abd[l]);
845 	}
846 
847 	return (config);
848 }
849 
850 /*
851  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
852  * in with the device guid if this spare is active elsewhere on the system.
853  */
854 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)855 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
856     uint64_t *spare_guid, uint64_t *l2cache_guid)
857 {
858 	spa_t *spa = vd->vdev_spa;
859 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
860 	uint64_t vdtxg = 0;
861 	nvlist_t *label;
862 
863 	if (spare_guid)
864 		*spare_guid = 0ULL;
865 	if (l2cache_guid)
866 		*l2cache_guid = 0ULL;
867 
868 	/*
869 	 * Read the label, if any, and perform some basic sanity checks.
870 	 */
871 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
872 		return (B_FALSE);
873 
874 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
875 	    &vdtxg);
876 
877 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
878 	    &state) != 0 ||
879 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
880 	    &device_guid) != 0) {
881 		nvlist_free(label);
882 		return (B_FALSE);
883 	}
884 
885 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
886 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
887 	    &pool_guid) != 0 ||
888 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
889 	    &txg) != 0)) {
890 		nvlist_free(label);
891 		return (B_FALSE);
892 	}
893 
894 	nvlist_free(label);
895 
896 	/*
897 	 * Check to see if this device indeed belongs to the pool it claims to
898 	 * be a part of.  The only way this is allowed is if the device is a hot
899 	 * spare (which we check for later on).
900 	 */
901 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
902 	    !spa_guid_exists(pool_guid, device_guid) &&
903 	    !spa_spare_exists(device_guid, NULL, NULL) &&
904 	    !spa_l2cache_exists(device_guid, NULL))
905 		return (B_FALSE);
906 
907 	/*
908 	 * If the transaction group is zero, then this an initialized (but
909 	 * unused) label.  This is only an error if the create transaction
910 	 * on-disk is the same as the one we're using now, in which case the
911 	 * user has attempted to add the same vdev multiple times in the same
912 	 * transaction.
913 	 */
914 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
915 	    txg == 0 && vdtxg == crtxg)
916 		return (B_TRUE);
917 
918 	/*
919 	 * Check to see if this is a spare device.  We do an explicit check for
920 	 * spa_has_spare() here because it may be on our pending list of spares
921 	 * to add.  We also check if it is an l2cache device.
922 	 */
923 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
924 	    spa_has_spare(spa, device_guid)) {
925 		if (spare_guid)
926 			*spare_guid = device_guid;
927 
928 		switch (reason) {
929 		case VDEV_LABEL_CREATE:
930 		case VDEV_LABEL_L2CACHE:
931 			return (B_TRUE);
932 
933 		case VDEV_LABEL_REPLACE:
934 			return (!spa_has_spare(spa, device_guid) ||
935 			    spare_pool != 0ULL);
936 
937 		case VDEV_LABEL_SPARE:
938 			return (spa_has_spare(spa, device_guid));
939 		default:
940 			break;
941 		}
942 	}
943 
944 	/*
945 	 * Check to see if this is an l2cache device.
946 	 */
947 	if (spa_l2cache_exists(device_guid, NULL))
948 		return (B_TRUE);
949 
950 	/*
951 	 * We can't rely on a pool's state if it's been imported
952 	 * read-only.  Instead we look to see if the pools is marked
953 	 * read-only in the namespace and set the state to active.
954 	 */
955 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
956 	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
957 	    spa_mode(spa) == SPA_MODE_READ)
958 		state = POOL_STATE_ACTIVE;
959 
960 	/*
961 	 * If the device is marked ACTIVE, then this device is in use by another
962 	 * pool on the system.
963 	 */
964 	return (state == POOL_STATE_ACTIVE);
965 }
966 
967 /*
968  * Initialize a vdev label.  We check to make sure each leaf device is not in
969  * use, and writable.  We put down an initial label which we will later
970  * overwrite with a complete label.  Note that it's important to do this
971  * sequentially, not in parallel, so that we catch cases of multiple use of the
972  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
973  * itself.
974  */
975 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)976 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
977 {
978 	spa_t *spa = vd->vdev_spa;
979 	nvlist_t *label;
980 	vdev_phys_t *vp;
981 	abd_t *vp_abd;
982 	abd_t *bootenv;
983 	uberblock_t *ub;
984 	abd_t *ub_abd;
985 	zio_t *zio;
986 	char *buf;
987 	size_t buflen;
988 	int error;
989 	uint64_t spare_guid = 0, l2cache_guid = 0;
990 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
991 
992 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
993 
994 	for (int c = 0; c < vd->vdev_children; c++)
995 		if ((error = vdev_label_init(vd->vdev_child[c],
996 		    crtxg, reason)) != 0)
997 			return (error);
998 
999 	/* Track the creation time for this vdev */
1000 	vd->vdev_crtxg = crtxg;
1001 
1002 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1003 		return (0);
1004 
1005 	/*
1006 	 * Dead vdevs cannot be initialized.
1007 	 */
1008 	if (vdev_is_dead(vd))
1009 		return (SET_ERROR(EIO));
1010 
1011 	/*
1012 	 * Determine if the vdev is in use.
1013 	 */
1014 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1015 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1016 		return (SET_ERROR(EBUSY));
1017 
1018 	/*
1019 	 * If this is a request to add or replace a spare or l2cache device
1020 	 * that is in use elsewhere on the system, then we must update the
1021 	 * guid (which was initialized to a random value) to reflect the
1022 	 * actual GUID (which is shared between multiple pools).
1023 	 */
1024 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1025 	    spare_guid != 0ULL) {
1026 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
1027 
1028 		vd->vdev_guid += guid_delta;
1029 
1030 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1031 			pvd->vdev_guid_sum += guid_delta;
1032 
1033 		/*
1034 		 * If this is a replacement, then we want to fallthrough to the
1035 		 * rest of the code.  If we're adding a spare, then it's already
1036 		 * labeled appropriately and we can just return.
1037 		 */
1038 		if (reason == VDEV_LABEL_SPARE)
1039 			return (0);
1040 		ASSERT(reason == VDEV_LABEL_REPLACE ||
1041 		    reason == VDEV_LABEL_SPLIT);
1042 	}
1043 
1044 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1045 	    l2cache_guid != 0ULL) {
1046 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1047 
1048 		vd->vdev_guid += guid_delta;
1049 
1050 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1051 			pvd->vdev_guid_sum += guid_delta;
1052 
1053 		/*
1054 		 * If this is a replacement, then we want to fallthrough to the
1055 		 * rest of the code.  If we're adding an l2cache, then it's
1056 		 * already labeled appropriately and we can just return.
1057 		 */
1058 		if (reason == VDEV_LABEL_L2CACHE)
1059 			return (0);
1060 		ASSERT(reason == VDEV_LABEL_REPLACE);
1061 	}
1062 
1063 	/*
1064 	 * Initialize its label.
1065 	 */
1066 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1067 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1068 	vp = abd_to_buf(vp_abd);
1069 
1070 	/*
1071 	 * Generate a label describing the pool and our top-level vdev.
1072 	 * We mark it as being from txg 0 to indicate that it's not
1073 	 * really part of an active pool just yet.  The labels will
1074 	 * be written again with a meaningful txg by spa_sync().
1075 	 */
1076 	if (reason == VDEV_LABEL_SPARE ||
1077 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1078 		/*
1079 		 * For inactive hot spares, we generate a special label that
1080 		 * identifies as a mutually shared hot spare.  We write the
1081 		 * label if we are adding a hot spare, or if we are removing an
1082 		 * active hot spare (in which case we want to revert the
1083 		 * labels).
1084 		 */
1085 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1086 
1087 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1088 		    spa_version(spa)) == 0);
1089 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1090 		    POOL_STATE_SPARE) == 0);
1091 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1092 		    vd->vdev_guid) == 0);
1093 	} else if (reason == VDEV_LABEL_L2CACHE ||
1094 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1095 		/*
1096 		 * For level 2 ARC devices, add a special label.
1097 		 */
1098 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1099 
1100 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1101 		    spa_version(spa)) == 0);
1102 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1103 		    POOL_STATE_L2CACHE) == 0);
1104 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1105 		    vd->vdev_guid) == 0);
1106 
1107 		/*
1108 		 * This is merely to facilitate reporting the ashift of the
1109 		 * cache device through zdb. The actual retrieval of the
1110 		 * ashift (in vdev_alloc()) uses the nvlist
1111 		 * spa->spa_l2cache->sav_config (populated in
1112 		 * spa_ld_open_aux_vdevs()).
1113 		 */
1114 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_ASHIFT,
1115 		    vd->vdev_ashift) == 0);
1116 	} else {
1117 		uint64_t txg = 0ULL;
1118 
1119 		if (reason == VDEV_LABEL_SPLIT)
1120 			txg = spa->spa_uberblock.ub_txg;
1121 		label = spa_config_generate(spa, vd, txg, B_FALSE);
1122 
1123 		/*
1124 		 * Add our creation time.  This allows us to detect multiple
1125 		 * vdev uses as described above, and automatically expires if we
1126 		 * fail.
1127 		 */
1128 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1129 		    crtxg) == 0);
1130 	}
1131 
1132 	buf = vp->vp_nvlist;
1133 	buflen = sizeof (vp->vp_nvlist);
1134 
1135 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1136 	if (error != 0) {
1137 		nvlist_free(label);
1138 		abd_free(vp_abd);
1139 		/* EFAULT means nvlist_pack ran out of room */
1140 		return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1141 	}
1142 
1143 	/*
1144 	 * Initialize uberblock template.
1145 	 */
1146 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1147 	abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1148 	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1149 	ub = abd_to_buf(ub_abd);
1150 	ub->ub_txg = 0;
1151 
1152 	/* Initialize the 2nd padding area. */
1153 	bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1154 	abd_zero(bootenv, VDEV_PAD_SIZE);
1155 
1156 	/*
1157 	 * Write everything in parallel.
1158 	 */
1159 retry:
1160 	zio = zio_root(spa, NULL, NULL, flags);
1161 
1162 	for (int l = 0; l < VDEV_LABELS; l++) {
1163 
1164 		vdev_label_write(zio, vd, l, vp_abd,
1165 		    offsetof(vdev_label_t, vl_vdev_phys),
1166 		    sizeof (vdev_phys_t), NULL, NULL, flags);
1167 
1168 		/*
1169 		 * Skip the 1st padding area.
1170 		 * Zero out the 2nd padding area where it might have
1171 		 * left over data from previous filesystem format.
1172 		 */
1173 		vdev_label_write(zio, vd, l, bootenv,
1174 		    offsetof(vdev_label_t, vl_be),
1175 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1176 
1177 		vdev_label_write(zio, vd, l, ub_abd,
1178 		    offsetof(vdev_label_t, vl_uberblock),
1179 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1180 	}
1181 
1182 	error = zio_wait(zio);
1183 
1184 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1185 		flags |= ZIO_FLAG_TRYHARD;
1186 		goto retry;
1187 	}
1188 
1189 	nvlist_free(label);
1190 	abd_free(bootenv);
1191 	abd_free(ub_abd);
1192 	abd_free(vp_abd);
1193 
1194 	/*
1195 	 * If this vdev hasn't been previously identified as a spare, then we
1196 	 * mark it as such only if a) we are labeling it as a spare, or b) it
1197 	 * exists as a spare elsewhere in the system.  Do the same for
1198 	 * level 2 ARC devices.
1199 	 */
1200 	if (error == 0 && !vd->vdev_isspare &&
1201 	    (reason == VDEV_LABEL_SPARE ||
1202 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1203 		spa_spare_add(vd);
1204 
1205 	if (error == 0 && !vd->vdev_isl2cache &&
1206 	    (reason == VDEV_LABEL_L2CACHE ||
1207 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1208 		spa_l2cache_add(vd);
1209 
1210 	return (error);
1211 }
1212 
1213 /*
1214  * Done callback for vdev_label_read_bootenv_impl. If this is the first
1215  * callback to finish, store our abd in the callback pointer. Otherwise, we
1216  * just free our abd and return.
1217  */
1218 static void
vdev_label_read_bootenv_done(zio_t * zio)1219 vdev_label_read_bootenv_done(zio_t *zio)
1220 {
1221 	zio_t *rio = zio->io_private;
1222 	abd_t **cbp = rio->io_private;
1223 
1224 	ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1225 
1226 	if (zio->io_error == 0) {
1227 		mutex_enter(&rio->io_lock);
1228 		if (*cbp == NULL) {
1229 			/* Will free this buffer in vdev_label_read_bootenv. */
1230 			*cbp = zio->io_abd;
1231 		} else {
1232 			abd_free(zio->io_abd);
1233 		}
1234 		mutex_exit(&rio->io_lock);
1235 	} else {
1236 		abd_free(zio->io_abd);
1237 	}
1238 }
1239 
1240 static void
vdev_label_read_bootenv_impl(zio_t * zio,vdev_t * vd,int flags)1241 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1242 {
1243 	for (int c = 0; c < vd->vdev_children; c++)
1244 		vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1245 
1246 	/*
1247 	 * We just use the first label that has a correct checksum; the
1248 	 * bootloader should have rewritten them all to be the same on boot,
1249 	 * and any changes we made since boot have been the same across all
1250 	 * labels.
1251 	 */
1252 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1253 		for (int l = 0; l < VDEV_LABELS; l++) {
1254 			vdev_label_read(zio, vd, l,
1255 			    abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1256 			    offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1257 			    vdev_label_read_bootenv_done, zio, flags);
1258 		}
1259 	}
1260 }
1261 
1262 int
vdev_label_read_bootenv(vdev_t * rvd,nvlist_t * bootenv)1263 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1264 {
1265 	nvlist_t *config;
1266 	spa_t *spa = rvd->vdev_spa;
1267 	abd_t *abd = NULL;
1268 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1269 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1270 
1271 	ASSERT(bootenv);
1272 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1273 
1274 	zio_t *zio = zio_root(spa, NULL, &abd, flags);
1275 	vdev_label_read_bootenv_impl(zio, rvd, flags);
1276 	int err = zio_wait(zio);
1277 
1278 	if (abd != NULL) {
1279 		char *buf;
1280 		vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1281 
1282 		vbe->vbe_version = ntohll(vbe->vbe_version);
1283 		switch (vbe->vbe_version) {
1284 		case VB_RAW:
1285 			/*
1286 			 * if we have textual data in vbe_bootenv, create nvlist
1287 			 * with key "envmap".
1288 			 */
1289 			fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1290 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1291 			fnvlist_add_string(bootenv, GRUB_ENVMAP,
1292 			    vbe->vbe_bootenv);
1293 			break;
1294 
1295 		case VB_NVLIST:
1296 			err = nvlist_unpack(vbe->vbe_bootenv,
1297 			    sizeof (vbe->vbe_bootenv), &config, 0);
1298 			if (err == 0) {
1299 				fnvlist_merge(bootenv, config);
1300 				nvlist_free(config);
1301 				break;
1302 			}
1303 			fallthrough;
1304 		default:
1305 			/* Check for FreeBSD zfs bootonce command string */
1306 			buf = abd_to_buf(abd);
1307 			if (*buf == '\0') {
1308 				fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1309 				    VB_NVLIST);
1310 				break;
1311 			}
1312 			fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1313 		}
1314 
1315 		/*
1316 		 * abd was allocated in vdev_label_read_bootenv_impl()
1317 		 */
1318 		abd_free(abd);
1319 		/*
1320 		 * If we managed to read any successfully,
1321 		 * return success.
1322 		 */
1323 		return (0);
1324 	}
1325 	return (err);
1326 }
1327 
1328 int
vdev_label_write_bootenv(vdev_t * vd,nvlist_t * env)1329 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1330 {
1331 	zio_t *zio;
1332 	spa_t *spa = vd->vdev_spa;
1333 	vdev_boot_envblock_t *bootenv;
1334 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1335 	int error;
1336 	size_t nvsize;
1337 	char *nvbuf;
1338 
1339 	error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1340 	if (error != 0)
1341 		return (SET_ERROR(error));
1342 
1343 	if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1344 		return (SET_ERROR(E2BIG));
1345 	}
1346 
1347 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1348 
1349 	error = ENXIO;
1350 	for (int c = 0; c < vd->vdev_children; c++) {
1351 		int child_err;
1352 
1353 		child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1354 		/*
1355 		 * As long as any of the disks managed to write all of their
1356 		 * labels successfully, return success.
1357 		 */
1358 		if (child_err == 0)
1359 			error = child_err;
1360 	}
1361 
1362 	if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1363 	    !vdev_writeable(vd)) {
1364 		return (error);
1365 	}
1366 	ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1367 	abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1368 	abd_zero(abd, VDEV_PAD_SIZE);
1369 
1370 	bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1371 	nvbuf = bootenv->vbe_bootenv;
1372 	nvsize = sizeof (bootenv->vbe_bootenv);
1373 
1374 	bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1375 	switch (bootenv->vbe_version) {
1376 	case VB_RAW:
1377 		if (nvlist_lookup_string(env, GRUB_ENVMAP, &nvbuf) == 0) {
1378 			(void) strlcpy(bootenv->vbe_bootenv, nvbuf, nvsize);
1379 		}
1380 		error = 0;
1381 		break;
1382 
1383 	case VB_NVLIST:
1384 		error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1385 		    KM_SLEEP);
1386 		break;
1387 
1388 	default:
1389 		error = EINVAL;
1390 		break;
1391 	}
1392 
1393 	if (error == 0) {
1394 		bootenv->vbe_version = htonll(bootenv->vbe_version);
1395 		abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1396 	} else {
1397 		abd_free(abd);
1398 		return (SET_ERROR(error));
1399 	}
1400 
1401 retry:
1402 	zio = zio_root(spa, NULL, NULL, flags);
1403 	for (int l = 0; l < VDEV_LABELS; l++) {
1404 		vdev_label_write(zio, vd, l, abd,
1405 		    offsetof(vdev_label_t, vl_be),
1406 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1407 	}
1408 
1409 	error = zio_wait(zio);
1410 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1411 		flags |= ZIO_FLAG_TRYHARD;
1412 		goto retry;
1413 	}
1414 
1415 	abd_free(abd);
1416 	return (error);
1417 }
1418 
1419 /*
1420  * ==========================================================================
1421  * uberblock load/sync
1422  * ==========================================================================
1423  */
1424 
1425 /*
1426  * Consider the following situation: txg is safely synced to disk.  We've
1427  * written the first uberblock for txg + 1, and then we lose power.  When we
1428  * come back up, we fail to see the uberblock for txg + 1 because, say,
1429  * it was on a mirrored device and the replica to which we wrote txg + 1
1430  * is now offline.  If we then make some changes and sync txg + 1, and then
1431  * the missing replica comes back, then for a few seconds we'll have two
1432  * conflicting uberblocks on disk with the same txg.  The solution is simple:
1433  * among uberblocks with equal txg, choose the one with the latest timestamp.
1434  */
1435 static int
vdev_uberblock_compare(const uberblock_t * ub1,const uberblock_t * ub2)1436 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1437 {
1438 	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1439 
1440 	if (likely(cmp))
1441 		return (cmp);
1442 
1443 	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1444 	if (likely(cmp))
1445 		return (cmp);
1446 
1447 	/*
1448 	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1449 	 * ZFS, e.g. OpenZFS >= 0.7.
1450 	 *
1451 	 * If one ub has MMP and the other does not, they were written by
1452 	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1453 	 * a 0 value.
1454 	 *
1455 	 * Since timestamp and txg are the same if we get this far, either is
1456 	 * acceptable for importing the pool.
1457 	 */
1458 	unsigned int seq1 = 0;
1459 	unsigned int seq2 = 0;
1460 
1461 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1462 		seq1 = MMP_SEQ(ub1);
1463 
1464 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1465 		seq2 = MMP_SEQ(ub2);
1466 
1467 	return (TREE_CMP(seq1, seq2));
1468 }
1469 
1470 struct ubl_cbdata {
1471 	uberblock_t	*ubl_ubbest;	/* Best uberblock */
1472 	vdev_t		*ubl_vd;	/* vdev associated with the above */
1473 };
1474 
1475 static void
vdev_uberblock_load_done(zio_t * zio)1476 vdev_uberblock_load_done(zio_t *zio)
1477 {
1478 	vdev_t *vd = zio->io_vd;
1479 	spa_t *spa = zio->io_spa;
1480 	zio_t *rio = zio->io_private;
1481 	uberblock_t *ub = abd_to_buf(zio->io_abd);
1482 	struct ubl_cbdata *cbp = rio->io_private;
1483 
1484 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1485 
1486 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1487 		mutex_enter(&rio->io_lock);
1488 		if (ub->ub_txg <= spa->spa_load_max_txg &&
1489 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1490 			/*
1491 			 * Keep track of the vdev in which this uberblock
1492 			 * was found. We will use this information later
1493 			 * to obtain the config nvlist associated with
1494 			 * this uberblock.
1495 			 */
1496 			*cbp->ubl_ubbest = *ub;
1497 			cbp->ubl_vd = vd;
1498 		}
1499 		mutex_exit(&rio->io_lock);
1500 	}
1501 
1502 	abd_free(zio->io_abd);
1503 }
1504 
1505 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)1506 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1507     struct ubl_cbdata *cbp)
1508 {
1509 	for (int c = 0; c < vd->vdev_children; c++)
1510 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1511 
1512 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1513 	    vd->vdev_ops != &vdev_draid_spare_ops) {
1514 		for (int l = 0; l < VDEV_LABELS; l++) {
1515 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1516 				vdev_label_read(zio, vd, l,
1517 				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1518 				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1519 				    VDEV_UBERBLOCK_SIZE(vd),
1520 				    vdev_uberblock_load_done, zio, flags);
1521 			}
1522 		}
1523 	}
1524 }
1525 
1526 /*
1527  * Reads the 'best' uberblock from disk along with its associated
1528  * configuration. First, we read the uberblock array of each label of each
1529  * vdev, keeping track of the uberblock with the highest txg in each array.
1530  * Then, we read the configuration from the same vdev as the best uberblock.
1531  */
1532 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)1533 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1534 {
1535 	zio_t *zio;
1536 	spa_t *spa = rvd->vdev_spa;
1537 	struct ubl_cbdata cb;
1538 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1539 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1540 
1541 	ASSERT(ub);
1542 	ASSERT(config);
1543 
1544 	bzero(ub, sizeof (uberblock_t));
1545 	*config = NULL;
1546 
1547 	cb.ubl_ubbest = ub;
1548 	cb.ubl_vd = NULL;
1549 
1550 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1551 	zio = zio_root(spa, NULL, &cb, flags);
1552 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1553 	(void) zio_wait(zio);
1554 
1555 	/*
1556 	 * It's possible that the best uberblock was discovered on a label
1557 	 * that has a configuration which was written in a future txg.
1558 	 * Search all labels on this vdev to find the configuration that
1559 	 * matches the txg for our uberblock.
1560 	 */
1561 	if (cb.ubl_vd != NULL) {
1562 		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1563 		    "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1564 
1565 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1566 		if (*config == NULL && spa->spa_extreme_rewind) {
1567 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1568 			    "Trying again without txg restrictions.");
1569 			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1570 		}
1571 		if (*config == NULL) {
1572 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1573 		}
1574 	}
1575 	spa_config_exit(spa, SCL_ALL, FTAG);
1576 }
1577 
1578 /*
1579  * For use when a leaf vdev is expanded.
1580  * The location of labels 2 and 3 changed, and at the new location the
1581  * uberblock rings are either empty or contain garbage.  The sync will write
1582  * new configs there because the vdev is dirty, but expansion also needs the
1583  * uberblock rings copied.  Read them from label 0 which did not move.
1584  *
1585  * Since the point is to populate labels {2,3} with valid uberblocks,
1586  * we zero uberblocks we fail to read or which are not valid.
1587  */
1588 
1589 static void
vdev_copy_uberblocks(vdev_t * vd)1590 vdev_copy_uberblocks(vdev_t *vd)
1591 {
1592 	abd_t *ub_abd;
1593 	zio_t *write_zio;
1594 	int locks = (SCL_L2ARC | SCL_ZIO);
1595 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1596 	    ZIO_FLAG_SPECULATIVE;
1597 
1598 	ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1599 	    SCL_STATE);
1600 	ASSERT(vd->vdev_ops->vdev_op_leaf);
1601 
1602 	/*
1603 	 * No uberblocks are stored on distributed spares, they may be
1604 	 * safely skipped when expanding a leaf vdev.
1605 	 */
1606 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1607 		return;
1608 
1609 	spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1610 
1611 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1612 
1613 	write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1614 	for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1615 		const int src_label = 0;
1616 		zio_t *zio;
1617 
1618 		zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1619 		vdev_label_read(zio, vd, src_label, ub_abd,
1620 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1621 		    NULL, NULL, flags);
1622 
1623 		if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1624 			abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1625 
1626 		for (int l = 2; l < VDEV_LABELS; l++)
1627 			vdev_label_write(write_zio, vd, l, ub_abd,
1628 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1629 			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1630 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1631 	}
1632 	(void) zio_wait(write_zio);
1633 
1634 	spa_config_exit(vd->vdev_spa, locks, FTAG);
1635 
1636 	abd_free(ub_abd);
1637 }
1638 
1639 /*
1640  * On success, increment root zio's count of good writes.
1641  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1642  */
1643 static void
vdev_uberblock_sync_done(zio_t * zio)1644 vdev_uberblock_sync_done(zio_t *zio)
1645 {
1646 	uint64_t *good_writes = zio->io_private;
1647 
1648 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1649 		atomic_inc_64(good_writes);
1650 }
1651 
1652 /*
1653  * Write the uberblock to all labels of all leaves of the specified vdev.
1654  */
1655 static void
vdev_uberblock_sync(zio_t * zio,uint64_t * good_writes,uberblock_t * ub,vdev_t * vd,int flags)1656 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1657     uberblock_t *ub, vdev_t *vd, int flags)
1658 {
1659 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1660 		vdev_uberblock_sync(zio, good_writes,
1661 		    ub, vd->vdev_child[c], flags);
1662 	}
1663 
1664 	if (!vd->vdev_ops->vdev_op_leaf)
1665 		return;
1666 
1667 	if (!vdev_writeable(vd))
1668 		return;
1669 
1670 	/*
1671 	 * There's no need to write uberblocks to a distributed spare, they
1672 	 * are already stored on all the leaves of the parent dRAID.  For
1673 	 * this same reason vdev_uberblock_load_impl() skips distributed
1674 	 * spares when reading uberblocks.
1675 	 */
1676 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1677 		return;
1678 
1679 	/* If the vdev was expanded, need to copy uberblock rings. */
1680 	if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1681 	    vd->vdev_copy_uberblocks == B_TRUE) {
1682 		vdev_copy_uberblocks(vd);
1683 		vd->vdev_copy_uberblocks = B_FALSE;
1684 	}
1685 
1686 	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1687 	int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1688 
1689 	/* Copy the uberblock_t into the ABD */
1690 	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1691 	abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1692 	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1693 
1694 	for (int l = 0; l < VDEV_LABELS; l++)
1695 		vdev_label_write(zio, vd, l, ub_abd,
1696 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1697 		    vdev_uberblock_sync_done, good_writes,
1698 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1699 
1700 	abd_free(ub_abd);
1701 }
1702 
1703 /* Sync the uberblocks to all vdevs in svd[] */
1704 static int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1705 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1706 {
1707 	spa_t *spa = svd[0]->vdev_spa;
1708 	zio_t *zio;
1709 	uint64_t good_writes = 0;
1710 
1711 	zio = zio_root(spa, NULL, NULL, flags);
1712 
1713 	for (int v = 0; v < svdcount; v++)
1714 		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1715 
1716 	(void) zio_wait(zio);
1717 
1718 	/*
1719 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1720 	 * are no longer needed (because the new uberblocks and the even
1721 	 * labels are safely on disk), so it is safe to overwrite them.
1722 	 */
1723 	zio = zio_root(spa, NULL, NULL, flags);
1724 
1725 	for (int v = 0; v < svdcount; v++) {
1726 		if (vdev_writeable(svd[v])) {
1727 			zio_flush(zio, svd[v]);
1728 		}
1729 	}
1730 
1731 	(void) zio_wait(zio);
1732 
1733 	return (good_writes >= 1 ? 0 : EIO);
1734 }
1735 
1736 /*
1737  * On success, increment the count of good writes for our top-level vdev.
1738  */
1739 static void
vdev_label_sync_done(zio_t * zio)1740 vdev_label_sync_done(zio_t *zio)
1741 {
1742 	uint64_t *good_writes = zio->io_private;
1743 
1744 	if (zio->io_error == 0)
1745 		atomic_inc_64(good_writes);
1746 }
1747 
1748 /*
1749  * If there weren't enough good writes, indicate failure to the parent.
1750  */
1751 static void
vdev_label_sync_top_done(zio_t * zio)1752 vdev_label_sync_top_done(zio_t *zio)
1753 {
1754 	uint64_t *good_writes = zio->io_private;
1755 
1756 	if (*good_writes == 0)
1757 		zio->io_error = SET_ERROR(EIO);
1758 
1759 	kmem_free(good_writes, sizeof (uint64_t));
1760 }
1761 
1762 /*
1763  * We ignore errors for log and cache devices, simply free the private data.
1764  */
1765 static void
vdev_label_sync_ignore_done(zio_t * zio)1766 vdev_label_sync_ignore_done(zio_t *zio)
1767 {
1768 	kmem_free(zio->io_private, sizeof (uint64_t));
1769 }
1770 
1771 /*
1772  * Write all even or odd labels to all leaves of the specified vdev.
1773  */
1774 static void
vdev_label_sync(zio_t * zio,uint64_t * good_writes,vdev_t * vd,int l,uint64_t txg,int flags)1775 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1776     vdev_t *vd, int l, uint64_t txg, int flags)
1777 {
1778 	nvlist_t *label;
1779 	vdev_phys_t *vp;
1780 	abd_t *vp_abd;
1781 	char *buf;
1782 	size_t buflen;
1783 
1784 	for (int c = 0; c < vd->vdev_children; c++) {
1785 		vdev_label_sync(zio, good_writes,
1786 		    vd->vdev_child[c], l, txg, flags);
1787 	}
1788 
1789 	if (!vd->vdev_ops->vdev_op_leaf)
1790 		return;
1791 
1792 	if (!vdev_writeable(vd))
1793 		return;
1794 
1795 	/*
1796 	 * The top-level config never needs to be written to a distributed
1797 	 * spare.  When read vdev_dspare_label_read_config() will generate
1798 	 * the config for the vdev_label_read_config().
1799 	 */
1800 	if (vd->vdev_ops == &vdev_draid_spare_ops)
1801 		return;
1802 
1803 	/*
1804 	 * Generate a label describing the top-level config to which we belong.
1805 	 */
1806 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1807 
1808 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1809 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1810 	vp = abd_to_buf(vp_abd);
1811 
1812 	buf = vp->vp_nvlist;
1813 	buflen = sizeof (vp->vp_nvlist);
1814 
1815 	if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1816 		for (; l < VDEV_LABELS; l += 2) {
1817 			vdev_label_write(zio, vd, l, vp_abd,
1818 			    offsetof(vdev_label_t, vl_vdev_phys),
1819 			    sizeof (vdev_phys_t),
1820 			    vdev_label_sync_done, good_writes,
1821 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1822 		}
1823 	}
1824 
1825 	abd_free(vp_abd);
1826 	nvlist_free(label);
1827 }
1828 
1829 static int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1830 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1831 {
1832 	list_t *dl = &spa->spa_config_dirty_list;
1833 	vdev_t *vd;
1834 	zio_t *zio;
1835 	int error;
1836 
1837 	/*
1838 	 * Write the new labels to disk.
1839 	 */
1840 	zio = zio_root(spa, NULL, NULL, flags);
1841 
1842 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1843 		uint64_t *good_writes;
1844 
1845 		ASSERT(!vd->vdev_ishole);
1846 
1847 		good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1848 		zio_t *vio = zio_null(zio, spa, NULL,
1849 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1850 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1851 		    good_writes, flags);
1852 		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1853 		zio_nowait(vio);
1854 	}
1855 
1856 	error = zio_wait(zio);
1857 
1858 	/*
1859 	 * Flush the new labels to disk.
1860 	 */
1861 	zio = zio_root(spa, NULL, NULL, flags);
1862 
1863 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1864 		zio_flush(zio, vd);
1865 
1866 	(void) zio_wait(zio);
1867 
1868 	return (error);
1869 }
1870 
1871 /*
1872  * Sync the uberblock and any changes to the vdev configuration.
1873  *
1874  * The order of operations is carefully crafted to ensure that
1875  * if the system panics or loses power at any time, the state on disk
1876  * is still transactionally consistent.  The in-line comments below
1877  * describe the failure semantics at each stage.
1878  *
1879  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1880  * at any time, you can just call it again, and it will resume its work.
1881  */
1882 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg)1883 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1884 {
1885 	spa_t *spa = svd[0]->vdev_spa;
1886 	uberblock_t *ub = &spa->spa_uberblock;
1887 	int error = 0;
1888 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1889 
1890 	ASSERT(svdcount != 0);
1891 retry:
1892 	/*
1893 	 * Normally, we don't want to try too hard to write every label and
1894 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1895 	 * sync process to block while we retry.  But if we can't write a
1896 	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1897 	 * bailing out and declaring the pool faulted.
1898 	 */
1899 	if (error != 0) {
1900 		if ((flags & ZIO_FLAG_TRYHARD) != 0)
1901 			return (error);
1902 		flags |= ZIO_FLAG_TRYHARD;
1903 	}
1904 
1905 	ASSERT(ub->ub_txg <= txg);
1906 
1907 	/*
1908 	 * If this isn't a resync due to I/O errors,
1909 	 * and nothing changed in this transaction group,
1910 	 * and the vdev configuration hasn't changed,
1911 	 * then there's nothing to do.
1912 	 */
1913 	if (ub->ub_txg < txg) {
1914 		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1915 		    txg, spa->spa_mmp.mmp_delay);
1916 
1917 		if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1918 			return (0);
1919 	}
1920 
1921 	if (txg > spa_freeze_txg(spa))
1922 		return (0);
1923 
1924 	ASSERT(txg <= spa->spa_final_txg);
1925 
1926 	/*
1927 	 * Flush the write cache of every disk that's been written to
1928 	 * in this transaction group.  This ensures that all blocks
1929 	 * written in this txg will be committed to stable storage
1930 	 * before any uberblock that references them.
1931 	 */
1932 	zio_t *zio = zio_root(spa, NULL, NULL, flags);
1933 
1934 	for (vdev_t *vd =
1935 	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1936 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1937 		zio_flush(zio, vd);
1938 
1939 	(void) zio_wait(zio);
1940 
1941 	/*
1942 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1943 	 * system dies in the middle of this process, that's OK: all of the
1944 	 * even labels that made it to disk will be newer than any uberblock,
1945 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
1946 	 * which have not yet been touched, will still be valid.  We flush
1947 	 * the new labels to disk to ensure that all even-label updates
1948 	 * are committed to stable storage before the uberblock update.
1949 	 */
1950 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1951 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1952 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1953 			    "for pool '%s' when syncing out the even labels "
1954 			    "of dirty vdevs", error, spa_name(spa));
1955 		}
1956 		goto retry;
1957 	}
1958 
1959 	/*
1960 	 * Sync the uberblocks to all vdevs in svd[].
1961 	 * If the system dies in the middle of this step, there are two cases
1962 	 * to consider, and the on-disk state is consistent either way:
1963 	 *
1964 	 * (1)	If none of the new uberblocks made it to disk, then the
1965 	 *	previous uberblock will be the newest, and the odd labels
1966 	 *	(which had not yet been touched) will be valid with respect
1967 	 *	to that uberblock.
1968 	 *
1969 	 * (2)	If one or more new uberblocks made it to disk, then they
1970 	 *	will be the newest, and the even labels (which had all
1971 	 *	been successfully committed) will be valid with respect
1972 	 *	to the new uberblocks.
1973 	 */
1974 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1975 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1976 			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1977 			    "%d for pool '%s'", error, spa_name(spa));
1978 		}
1979 		goto retry;
1980 	}
1981 
1982 	if (spa_multihost(spa))
1983 		mmp_update_uberblock(spa, ub);
1984 
1985 	/*
1986 	 * Sync out odd labels for every dirty vdev.  If the system dies
1987 	 * in the middle of this process, the even labels and the new
1988 	 * uberblocks will suffice to open the pool.  The next time
1989 	 * the pool is opened, the first thing we'll do -- before any
1990 	 * user data is modified -- is mark every vdev dirty so that
1991 	 * all labels will be brought up to date.  We flush the new labels
1992 	 * to disk to ensure that all odd-label updates are committed to
1993 	 * stable storage before the next transaction group begins.
1994 	 */
1995 	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
1996 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1997 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1998 			    "for pool '%s' when syncing out the odd labels of "
1999 			    "dirty vdevs", error, spa_name(spa));
2000 		}
2001 		goto retry;
2002 	}
2003 
2004 	return (0);
2005 }
2006