xref: /illumos-gate/usr/src/uts/common/fs/zfs/vdev_label.c (revision 74009d0f)
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  * Copyright 2020 Joyent, Inc.
27  */
28 
29 /*
30  * Virtual Device Labels
31  * ---------------------
32  *
33  * The vdev label serves several distinct purposes:
34  *
35  *	1. Uniquely identify this device as part of a ZFS pool and confirm its
36  *	   identity within the pool.
37  *
38  *	2. Verify that all the devices given in a configuration are present
39  *         within the pool.
40  *
41  *	3. Determine the uberblock for the pool.
42  *
43  *	4. In case of an import operation, determine the configuration of the
44  *         toplevel vdev of which it is a part.
45  *
46  *	5. If an import operation cannot find all the devices in the pool,
47  *         provide enough information to the administrator to determine which
48  *         devices are missing.
49  *
50  * It is important to note that while the kernel is responsible for writing the
51  * label, it only consumes the information in the first three cases.  The
52  * latter information is only consumed in userland when determining the
53  * configuration to import a pool.
54  *
55  *
56  * Label Organization
57  * ------------------
58  *
59  * Before describing the contents of the label, it's important to understand how
60  * the labels are written and updated with respect to the uberblock.
61  *
62  * When the pool configuration is altered, either because it was newly created
63  * or a device was added, we want to update all the labels such that we can deal
64  * with fatal failure at any point.  To this end, each disk has two labels which
65  * are updated before and after the uberblock is synced.  Assuming we have
66  * labels and an uberblock with the following transaction groups:
67  *
68  *              L1          UB          L2
69  *           +------+    +------+    +------+
70  *           |      |    |      |    |      |
71  *           | t10  |    | t10  |    | t10  |
72  *           |      |    |      |    |      |
73  *           +------+    +------+    +------+
74  *
75  * In this stable state, the labels and the uberblock were all updated within
76  * the same transaction group (10).  Each label is mirrored and checksummed, so
77  * that we can detect when we fail partway through writing the label.
78  *
79  * In order to identify which labels are valid, the labels are written in the
80  * following manner:
81  *
82  *	1. For each vdev, update 'L1' to the new label
83  *	2. Update the uberblock
84  *	3. For each vdev, update 'L2' to the new label
85  *
86  * Given arbitrary failure, we can determine the correct label to use based on
87  * the transaction group.  If we fail after updating L1 but before updating the
88  * UB, we will notice that L1's transaction group is greater than the uberblock,
89  * so L2 must be valid.  If we fail after writing the uberblock but before
90  * writing L2, we will notice that L2's transaction group is less than L1, and
91  * therefore L1 is valid.
92  *
93  * Another added complexity is that not every label is updated when the config
94  * is synced.  If we add a single device, we do not want to have to re-write
95  * every label for every device in the pool.  This means that both L1 and L2 may
96  * be older than the pool uberblock, because the necessary information is stored
97  * on another vdev.
98  *
99  *
100  * On-disk Format
101  * --------------
102  *
103  * The vdev label consists of two distinct parts, and is wrapped within the
104  * vdev_label_t structure.  The label includes 8k of padding to permit legacy
105  * VTOC disk labels, but is otherwise ignored.
106  *
107  * The first half of the label is a packed nvlist which contains pool wide
108  * properties, per-vdev properties, and configuration information.  It is
109  * described in more detail below.
110  *
111  * The latter half of the label consists of a redundant array of uberblocks.
112  * These uberblocks are updated whenever a transaction group is committed,
113  * or when the configuration is updated.  When a pool is loaded, we scan each
114  * vdev for the 'best' uberblock.
115  *
116  *
117  * Configuration Information
118  * -------------------------
119  *
120  * The nvlist describing the pool and vdev contains the following elements:
121  *
122  *	version		ZFS on-disk version
123  *	name		Pool name
124  *	state		Pool state
125  *	txg		Transaction group in which this label was written
126  *	pool_guid	Unique identifier for this pool
127  *	vdev_tree	An nvlist describing vdev tree.
128  *	features_for_read
129  *			An nvlist of the features necessary for reading the MOS.
130  *
131  * Each leaf device label also contains the following:
132  *
133  *	top_guid	Unique ID for top-level vdev in which this is contained
134  *	guid		Unique ID for the leaf vdev
135  *
136  * The 'vs' configuration follows the format described in 'spa_config.c'.
137  */
138 
139 #include <sys/zfs_context.h>
140 #include <sys/spa.h>
141 #include <sys/spa_impl.h>
142 #include <sys/dmu.h>
143 #include <sys/zap.h>
144 #include <sys/vdev.h>
145 #include <sys/vdev_impl.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 	nvlist_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 /*
411  * Generate the nvlist representing this vdev's config.
412  */
413 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)414 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
415     vdev_config_flag_t flags)
416 {
417 	nvlist_t *nv = NULL;
418 	vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
419 
420 	nv = fnvlist_alloc();
421 
422 	fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
423 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
424 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
425 	fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
426 
427 	if (vd->vdev_path != NULL)
428 		fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
429 
430 	if (vd->vdev_devid != NULL)
431 		fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
432 
433 	if (vd->vdev_physpath != NULL)
434 		fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
435 		    vd->vdev_physpath);
436 
437 	if (vd->vdev_fru != NULL)
438 		fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
439 
440 	if (vd->vdev_nparity != 0) {
441 		ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
442 		    VDEV_TYPE_RAIDZ) == 0);
443 
444 		/*
445 		 * Make sure someone hasn't managed to sneak a fancy new vdev
446 		 * into a crufty old storage pool.
447 		 */
448 		ASSERT(vd->vdev_nparity == 1 ||
449 		    (vd->vdev_nparity <= 2 &&
450 		    spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
451 		    (vd->vdev_nparity <= 3 &&
452 		    spa_version(spa) >= SPA_VERSION_RAIDZ3));
453 
454 		/*
455 		 * Note that we'll add the nparity tag even on storage pools
456 		 * that only support a single parity device -- older software
457 		 * will just ignore it.
458 		 */
459 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
460 	}
461 
462 	if (vd->vdev_wholedisk != -1ULL)
463 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
464 		    vd->vdev_wholedisk);
465 
466 	if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
467 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
468 
469 	if (vd->vdev_isspare)
470 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
471 
472 	if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
473 	    vd == vd->vdev_top) {
474 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
475 		    vd->vdev_ms_array);
476 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
477 		    vd->vdev_ms_shift);
478 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
479 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
480 		    vd->vdev_asize);
481 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
482 		if (vd->vdev_removing) {
483 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
484 			    vd->vdev_removing);
485 		}
486 
487 		/* zpool command expects alloc class data */
488 		if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
489 			const char *bias = NULL;
490 
491 			switch (vd->vdev_alloc_bias) {
492 			case VDEV_BIAS_LOG:
493 				bias = VDEV_ALLOC_BIAS_LOG;
494 				break;
495 			case VDEV_BIAS_SPECIAL:
496 				bias = VDEV_ALLOC_BIAS_SPECIAL;
497 				break;
498 			case VDEV_BIAS_DEDUP:
499 				bias = VDEV_ALLOC_BIAS_DEDUP;
500 				break;
501 			default:
502 				ASSERT3U(vd->vdev_alloc_bias, ==,
503 				    VDEV_BIAS_NONE);
504 			}
505 			fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
506 			    bias);
507 		}
508 	}
509 
510 	if (vd->vdev_dtl_sm != NULL) {
511 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
512 		    space_map_object(vd->vdev_dtl_sm));
513 	}
514 
515 	if (vic->vic_mapping_object != 0) {
516 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
517 		    vic->vic_mapping_object);
518 	}
519 
520 	if (vic->vic_births_object != 0) {
521 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
522 		    vic->vic_births_object);
523 	}
524 
525 	if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
526 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
527 		    vic->vic_prev_indirect_vdev);
528 	}
529 
530 	if (vd->vdev_crtxg)
531 		fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
532 
533 	if (flags & VDEV_CONFIG_MOS) {
534 		if (vd->vdev_leaf_zap != 0) {
535 			ASSERT(vd->vdev_ops->vdev_op_leaf);
536 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
537 			    vd->vdev_leaf_zap);
538 		}
539 
540 		if (vd->vdev_top_zap != 0) {
541 			ASSERT(vd == vd->vdev_top);
542 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
543 			    vd->vdev_top_zap);
544 		}
545 
546 		if (vd->vdev_resilver_deferred) {
547 			ASSERT(vd->vdev_ops->vdev_op_leaf);
548 			ASSERT(spa->spa_resilver_deferred);
549 			fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
550 		}
551 	}
552 
553 	if (getstats) {
554 		vdev_config_generate_stats(vd, nv);
555 
556 		root_vdev_actions_getprogress(vd, nv);
557 
558 		/*
559 		 * Note: this can be called from open context
560 		 * (spa_get_stats()), so we need the rwlock to prevent
561 		 * the mapping from being changed by condensing.
562 		 */
563 		rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
564 		if (vd->vdev_indirect_mapping != NULL) {
565 			ASSERT(vd->vdev_indirect_births != NULL);
566 			vdev_indirect_mapping_t *vim =
567 			    vd->vdev_indirect_mapping;
568 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
569 			    vdev_indirect_mapping_size(vim));
570 		}
571 		rw_exit(&vd->vdev_indirect_rwlock);
572 		if (vd->vdev_mg != NULL &&
573 		    vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
574 			/*
575 			 * Compute approximately how much memory would be used
576 			 * for the indirect mapping if this device were to
577 			 * be removed.
578 			 *
579 			 * Note: If the frag metric is invalid, then not
580 			 * enough metaslabs have been converted to have
581 			 * histograms.
582 			 */
583 			uint64_t seg_count = 0;
584 			uint64_t to_alloc = vd->vdev_stat.vs_alloc;
585 
586 			/*
587 			 * There are the same number of allocated segments
588 			 * as free segments, so we will have at least one
589 			 * entry per free segment.  However, small free
590 			 * segments (smaller than vdev_removal_max_span)
591 			 * will be combined with adjacent allocated segments
592 			 * as a single mapping.
593 			 */
594 			for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) {
595 				if (1ULL << (i + 1) < vdev_removal_max_span) {
596 					to_alloc +=
597 					    vd->vdev_mg->mg_histogram[i] <<
598 					    i + 1;
599 				} else {
600 					seg_count +=
601 					    vd->vdev_mg->mg_histogram[i];
602 				}
603 			}
604 
605 			/*
606 			 * The maximum length of a mapping is
607 			 * zfs_remove_max_segment, so we need at least one entry
608 			 * per zfs_remove_max_segment of allocated data.
609 			 */
610 			seg_count += to_alloc / zfs_remove_max_segment;
611 
612 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
613 			    seg_count *
614 			    sizeof (vdev_indirect_mapping_entry_phys_t));
615 		}
616 	}
617 
618 	if (!vd->vdev_ops->vdev_op_leaf) {
619 		nvlist_t **child;
620 		int c, idx;
621 
622 		ASSERT(!vd->vdev_ishole);
623 
624 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
625 		    KM_SLEEP);
626 
627 		for (c = 0, idx = 0; c < vd->vdev_children; c++) {
628 			vdev_t *cvd = vd->vdev_child[c];
629 
630 			/*
631 			 * If we're generating an nvlist of removing
632 			 * vdevs then skip over any device which is
633 			 * not being removed.
634 			 */
635 			if ((flags & VDEV_CONFIG_REMOVING) &&
636 			    !cvd->vdev_removing)
637 				continue;
638 
639 			child[idx++] = vdev_config_generate(spa, cvd,
640 			    getstats, flags);
641 		}
642 
643 		if (idx) {
644 			fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
645 			    child, idx);
646 		}
647 
648 		for (c = 0; c < idx; c++)
649 			nvlist_free(child[c]);
650 
651 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
652 
653 	} else {
654 		const char *aux = NULL;
655 
656 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
657 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
658 		if (vd->vdev_resilver_txg != 0)
659 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
660 			    vd->vdev_resilver_txg);
661 		if (vd->vdev_faulted)
662 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
663 		if (vd->vdev_degraded)
664 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
665 		if (vd->vdev_removed)
666 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
667 		if (vd->vdev_unspare)
668 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
669 		if (vd->vdev_ishole)
670 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
671 
672 		switch (vd->vdev_stat.vs_aux) {
673 		case VDEV_AUX_ERR_EXCEEDED:
674 			aux = "err_exceeded";
675 			break;
676 
677 		case VDEV_AUX_EXTERNAL:
678 			aux = "external";
679 			break;
680 		}
681 
682 		if (aux != NULL)
683 			fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
684 
685 		if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
686 			fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
687 			    vd->vdev_orig_guid);
688 		}
689 	}
690 
691 	return (nv);
692 }
693 
694 /*
695  * Generate a view of the top-level vdevs.  If we currently have holes
696  * in the namespace, then generate an array which contains a list of holey
697  * vdevs.  Additionally, add the number of top-level children that currently
698  * exist.
699  */
700 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)701 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
702 {
703 	vdev_t *rvd = spa->spa_root_vdev;
704 	uint64_t *array;
705 	uint_t c, idx;
706 
707 	array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
708 
709 	for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
710 		vdev_t *tvd = rvd->vdev_child[c];
711 
712 		if (tvd->vdev_ishole) {
713 			array[idx++] = c;
714 		}
715 	}
716 
717 	if (idx) {
718 		VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
719 		    array, idx) == 0);
720 	}
721 
722 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
723 	    rvd->vdev_children) == 0);
724 
725 	kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
726 }
727 
728 /*
729  * Returns the configuration from the label of the given vdev. For vdevs
730  * which don't have a txg value stored on their label (i.e. spares/cache)
731  * or have not been completely initialized (txg = 0) just return
732  * the configuration from the first valid label we find. Otherwise,
733  * find the most up-to-date label that does not exceed the specified
734  * 'txg' value.
735  */
736 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)737 vdev_label_read_config(vdev_t *vd, uint64_t txg)
738 {
739 	spa_t *spa = vd->vdev_spa;
740 	nvlist_t *config = NULL;
741 	vdev_phys_t *vp;
742 	abd_t *vp_abd;
743 	zio_t *zio;
744 	uint64_t best_txg = 0;
745 	uint64_t label_txg = 0;
746 	int error = 0;
747 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
748 	    ZIO_FLAG_SPECULATIVE;
749 
750 	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
751 
752 	if (!vdev_readable(vd))
753 		return (NULL);
754 
755 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
756 	vp = abd_to_buf(vp_abd);
757 
758 retry:
759 	for (int l = 0; l < VDEV_LABELS; l++) {
760 		nvlist_t *label = NULL;
761 
762 		zio = zio_root(spa, NULL, NULL, flags);
763 
764 		vdev_label_read(zio, vd, l, vp_abd,
765 		    offsetof(vdev_label_t, vl_vdev_phys),
766 		    sizeof (vdev_phys_t), NULL, NULL, flags);
767 
768 		if (zio_wait(zio) == 0 &&
769 		    nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
770 		    &label, 0) == 0) {
771 			/*
772 			 * Auxiliary vdevs won't have txg values in their
773 			 * labels and newly added vdevs may not have been
774 			 * completely initialized so just return the
775 			 * configuration from the first valid label we
776 			 * encounter.
777 			 */
778 			error = nvlist_lookup_uint64(label,
779 			    ZPOOL_CONFIG_POOL_TXG, &label_txg);
780 			if ((error || label_txg == 0) && !config) {
781 				config = label;
782 				break;
783 			} else if (label_txg <= txg && label_txg > best_txg) {
784 				best_txg = label_txg;
785 				nvlist_free(config);
786 				config = fnvlist_dup(label);
787 			}
788 		}
789 
790 		if (label != NULL) {
791 			nvlist_free(label);
792 			label = NULL;
793 		}
794 	}
795 
796 	if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
797 		flags |= ZIO_FLAG_TRYHARD;
798 		goto retry;
799 	}
800 
801 	/*
802 	 * We found a valid label but it didn't pass txg restrictions.
803 	 */
804 	if (config == NULL && label_txg != 0) {
805 		vdev_dbgmsg(vd, "label discarded as txg is too large "
806 		    "(%llu > %llu)", (u_longlong_t)label_txg,
807 		    (u_longlong_t)txg);
808 	}
809 
810 	abd_free(vp_abd);
811 
812 	return (config);
813 }
814 
815 /*
816  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
817  * in with the device guid if this spare is active elsewhere on the system.
818  */
819 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)820 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
821     uint64_t *spare_guid, uint64_t *l2cache_guid)
822 {
823 	spa_t *spa = vd->vdev_spa;
824 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
825 	uint64_t vdtxg = 0;
826 	nvlist_t *label;
827 
828 	if (spare_guid)
829 		*spare_guid = 0ULL;
830 	if (l2cache_guid)
831 		*l2cache_guid = 0ULL;
832 
833 	/*
834 	 * Read the label, if any, and perform some basic sanity checks.
835 	 */
836 	if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
837 		return (B_FALSE);
838 
839 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
840 	    &vdtxg);
841 
842 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
843 	    &state) != 0 ||
844 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
845 	    &device_guid) != 0) {
846 		nvlist_free(label);
847 		return (B_FALSE);
848 	}
849 
850 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
851 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
852 	    &pool_guid) != 0 ||
853 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
854 	    &txg) != 0)) {
855 		nvlist_free(label);
856 		return (B_FALSE);
857 	}
858 
859 	nvlist_free(label);
860 
861 	/*
862 	 * Check to see if this device indeed belongs to the pool it claims to
863 	 * be a part of.  The only way this is allowed is if the device is a hot
864 	 * spare (which we check for later on).
865 	 */
866 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
867 	    !spa_guid_exists(pool_guid, device_guid) &&
868 	    !spa_spare_exists(device_guid, NULL, NULL) &&
869 	    !spa_l2cache_exists(device_guid, NULL))
870 		return (B_FALSE);
871 
872 	/*
873 	 * If the transaction group is zero, then this an initialized (but
874 	 * unused) label.  This is only an error if the create transaction
875 	 * on-disk is the same as the one we're using now, in which case the
876 	 * user has attempted to add the same vdev multiple times in the same
877 	 * transaction.
878 	 */
879 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
880 	    txg == 0 && vdtxg == crtxg)
881 		return (B_TRUE);
882 
883 	/*
884 	 * Check to see if this is a spare device.  We do an explicit check for
885 	 * spa_has_spare() here because it may be on our pending list of spares
886 	 * to add.  We also check if it is an l2cache device.
887 	 */
888 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
889 	    spa_has_spare(spa, device_guid)) {
890 		if (spare_guid)
891 			*spare_guid = device_guid;
892 
893 		switch (reason) {
894 		case VDEV_LABEL_CREATE:
895 		case VDEV_LABEL_L2CACHE:
896 			return (B_TRUE);
897 
898 		case VDEV_LABEL_REPLACE:
899 			return (!spa_has_spare(spa, device_guid) ||
900 			    spare_pool != 0ULL);
901 
902 		case VDEV_LABEL_SPARE:
903 			return (spa_has_spare(spa, device_guid));
904 		}
905 	}
906 
907 	/*
908 	 * Check to see if this is an l2cache device.
909 	 */
910 	if (spa_l2cache_exists(device_guid, NULL))
911 		return (B_TRUE);
912 
913 	/*
914 	 * We can't rely on a pool's state if it's been imported
915 	 * read-only.  Instead we look to see if the pools is marked
916 	 * read-only in the namespace and set the state to active.
917 	 */
918 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
919 	    (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
920 	    spa_mode(spa) == FREAD)
921 		state = POOL_STATE_ACTIVE;
922 
923 	/*
924 	 * If the device is marked ACTIVE, then this device is in use by another
925 	 * pool on the system.
926 	 */
927 	return (state == POOL_STATE_ACTIVE);
928 }
929 
930 /*
931  * Initialize a vdev label.  We check to make sure each leaf device is not in
932  * use, and writable.  We put down an initial label which we will later
933  * overwrite with a complete label.  Note that it's important to do this
934  * sequentially, not in parallel, so that we catch cases of multiple use of the
935  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
936  * itself.
937  */
938 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)939 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
940 {
941 	spa_t *spa = vd->vdev_spa;
942 	nvlist_t *label;
943 	vdev_phys_t *vp;
944 	abd_t *vp_abd;
945 	abd_t *bootenv;
946 	uberblock_t *ub;
947 	abd_t *ub_abd;
948 	zio_t *zio;
949 	char *buf;
950 	size_t buflen;
951 	int error;
952 	uint64_t spare_guid = 0, l2cache_guid;
953 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
954 
955 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
956 
957 	for (int c = 0; c < vd->vdev_children; c++)
958 		if ((error = vdev_label_init(vd->vdev_child[c],
959 		    crtxg, reason)) != 0)
960 			return (error);
961 
962 	/* Track the creation time for this vdev */
963 	vd->vdev_crtxg = crtxg;
964 
965 	if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
966 		return (0);
967 
968 	/*
969 	 * Dead vdevs cannot be initialized.
970 	 */
971 	if (vdev_is_dead(vd))
972 		return (SET_ERROR(EIO));
973 
974 	/*
975 	 * Determine if the vdev is in use.
976 	 */
977 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
978 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
979 		return (SET_ERROR(EBUSY));
980 
981 	/*
982 	 * If this is a request to add or replace a spare or l2cache device
983 	 * that is in use elsewhere on the system, then we must update the
984 	 * guid (which was initialized to a random value) to reflect the
985 	 * actual GUID (which is shared between multiple pools).
986 	 */
987 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
988 	    spare_guid != 0ULL) {
989 		uint64_t guid_delta = spare_guid - vd->vdev_guid;
990 
991 		vd->vdev_guid += guid_delta;
992 
993 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
994 			pvd->vdev_guid_sum += guid_delta;
995 
996 		/*
997 		 * If this is a replacement, then we want to fallthrough to the
998 		 * rest of the code.  If we're adding a spare, then it's already
999 		 * labeled appropriately and we can just return.
1000 		 */
1001 		if (reason == VDEV_LABEL_SPARE)
1002 			return (0);
1003 		ASSERT(reason == VDEV_LABEL_REPLACE ||
1004 		    reason == VDEV_LABEL_SPLIT);
1005 	}
1006 
1007 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1008 	    l2cache_guid != 0ULL) {
1009 		uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1010 
1011 		vd->vdev_guid += guid_delta;
1012 
1013 		for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1014 			pvd->vdev_guid_sum += guid_delta;
1015 
1016 		/*
1017 		 * If this is a replacement, then we want to fallthrough to the
1018 		 * rest of the code.  If we're adding an l2cache, then it's
1019 		 * already labeled appropriately and we can just return.
1020 		 */
1021 		if (reason == VDEV_LABEL_L2CACHE)
1022 			return (0);
1023 		ASSERT(reason == VDEV_LABEL_REPLACE);
1024 	}
1025 
1026 	/*
1027 	 * Initialize its label.
1028 	 */
1029 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1030 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1031 	vp = abd_to_buf(vp_abd);
1032 
1033 	/*
1034 	 * Generate a label describing the pool and our top-level vdev.
1035 	 * We mark it as being from txg 0 to indicate that it's not
1036 	 * really part of an active pool just yet.  The labels will
1037 	 * be written again with a meaningful txg by spa_sync().
1038 	 */
1039 	if (reason == VDEV_LABEL_SPARE ||
1040 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
1041 		/*
1042 		 * For inactive hot spares, we generate a special label that
1043 		 * identifies as a mutually shared hot spare.  We write the
1044 		 * label if we are adding a hot spare, or if we are removing an
1045 		 * active hot spare (in which case we want to revert the
1046 		 * labels).
1047 		 */
1048 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1049 
1050 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1051 		    spa_version(spa)) == 0);
1052 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1053 		    POOL_STATE_SPARE) == 0);
1054 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1055 		    vd->vdev_guid) == 0);
1056 	} else if (reason == VDEV_LABEL_L2CACHE ||
1057 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
1058 		/*
1059 		 * For level 2 ARC devices, add a special label.
1060 		 */
1061 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1062 
1063 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1064 		    spa_version(spa)) == 0);
1065 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1066 		    POOL_STATE_L2CACHE) == 0);
1067 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
1068 		    vd->vdev_guid) == 0);
1069 	} else {
1070 		uint64_t txg = 0ULL;
1071 
1072 		if (reason == VDEV_LABEL_SPLIT)
1073 			txg = spa->spa_uberblock.ub_txg;
1074 		label = spa_config_generate(spa, vd, txg, B_FALSE);
1075 
1076 		/*
1077 		 * Add our creation time.  This allows us to detect multiple
1078 		 * vdev uses as described above, and automatically expires if we
1079 		 * fail.
1080 		 */
1081 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1082 		    crtxg) == 0);
1083 	}
1084 
1085 	buf = vp->vp_nvlist;
1086 	buflen = sizeof (vp->vp_nvlist);
1087 
1088 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1089 	if (error != 0) {
1090 		nvlist_free(label);
1091 		abd_free(vp_abd);
1092 		/* EFAULT means nvlist_pack ran out of room */
1093 		return (error == EFAULT ? ENAMETOOLONG : EINVAL);
1094 	}
1095 
1096 	/*
1097 	 * Initialize uberblock template.
1098 	 */
1099 	ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1100 	abd_zero(ub_abd, VDEV_UBERBLOCK_RING);
1101 	abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1102 	ub = abd_to_buf(ub_abd);
1103 	ub->ub_txg = 0;
1104 
1105 	/* Initialize the 2nd padding area. */
1106 	bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1107 	abd_zero(bootenv, VDEV_PAD_SIZE);
1108 
1109 	/*
1110 	 * Write everything in parallel.
1111 	 */
1112 retry:
1113 	zio = zio_root(spa, NULL, NULL, flags);
1114 
1115 	for (int l = 0; l < VDEV_LABELS; l++) {
1116 
1117 		vdev_label_write(zio, vd, l, vp_abd,
1118 		    offsetof(vdev_label_t, vl_vdev_phys),
1119 		    sizeof (vdev_phys_t), NULL, NULL, flags);
1120 
1121 		/*
1122 		 * Skip the 1st padding area.
1123 		 * Zero out the 2nd padding area where it might have
1124 		 * left over data from previous filesystem format.
1125 		 */
1126 		vdev_label_write(zio, vd, l, bootenv,
1127 		    offsetof(vdev_label_t, vl_be),
1128 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1129 
1130 		vdev_label_write(zio, vd, l, ub_abd,
1131 		    offsetof(vdev_label_t, vl_uberblock),
1132 		    VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1133 	}
1134 
1135 	error = zio_wait(zio);
1136 
1137 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1138 		flags |= ZIO_FLAG_TRYHARD;
1139 		goto retry;
1140 	}
1141 
1142 	nvlist_free(label);
1143 	abd_free(bootenv);
1144 	abd_free(ub_abd);
1145 	abd_free(vp_abd);
1146 
1147 	/*
1148 	 * If this vdev hasn't been previously identified as a spare, then we
1149 	 * mark it as such only if a) we are labeling it as a spare, or b) it
1150 	 * exists as a spare elsewhere in the system.  Do the same for
1151 	 * level 2 ARC devices.
1152 	 */
1153 	if (error == 0 && !vd->vdev_isspare &&
1154 	    (reason == VDEV_LABEL_SPARE ||
1155 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1156 		spa_spare_add(vd);
1157 
1158 	if (error == 0 && !vd->vdev_isl2cache &&
1159 	    (reason == VDEV_LABEL_L2CACHE ||
1160 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
1161 		spa_l2cache_add(vd);
1162 
1163 	return (error);
1164 }
1165 
1166 /*
1167  * Done callback for vdev_label_read_bootenv_impl. If this is the first
1168  * callback to finish, store our abd in the callback pointer. Otherwise, we
1169  * just free our abd and return.
1170  */
1171 static void
vdev_label_read_bootenv_done(zio_t * zio)1172 vdev_label_read_bootenv_done(zio_t *zio)
1173 {
1174 	zio_t *rio = zio->io_private;
1175 	abd_t **cbp = rio->io_private;
1176 
1177 	ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1178 
1179 	if (zio->io_error == 0) {
1180 		mutex_enter(&rio->io_lock);
1181 		if (*cbp == NULL) {
1182 			/* Will free this buffer in vdev_label_read_bootenv. */
1183 			*cbp = zio->io_abd;
1184 		} else {
1185 			abd_free(zio->io_abd);
1186 		}
1187 		mutex_exit(&rio->io_lock);
1188 	} else {
1189 		abd_free(zio->io_abd);
1190 	}
1191 }
1192 
1193 static void
vdev_label_read_bootenv_impl(zio_t * zio,vdev_t * vd,int flags)1194 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1195 {
1196 	for (int c = 0; c < vd->vdev_children; c++)
1197 		vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1198 
1199 	/*
1200 	 * We just use the first label that has a correct checksum; the
1201 	 * bootloader should have rewritten them all to be the same on boot,
1202 	 * and any changes we made since boot have been the same across all
1203 	 * labels.
1204 	 */
1205 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1206 		for (int l = 0; l < VDEV_LABELS; l++) {
1207 			vdev_label_read(zio, vd, l,
1208 			    abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1209 			    offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1210 			    vdev_label_read_bootenv_done, zio, flags);
1211 		}
1212 	}
1213 }
1214 
1215 int
vdev_label_read_bootenv(vdev_t * rvd,nvlist_t * bootenv)1216 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1217 {
1218 	nvlist_t *config;
1219 	spa_t *spa = rvd->vdev_spa;
1220 	abd_t *abd = NULL;
1221 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1222 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1223 
1224 	ASSERT(bootenv);
1225 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1226 
1227 	zio_t *zio = zio_root(spa, NULL, &abd, flags);
1228 	vdev_label_read_bootenv_impl(zio, rvd, flags);
1229 	int err = zio_wait(zio);
1230 
1231 	if (abd != NULL) {
1232 		char *buf;
1233 		vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1234 
1235 		vbe->vbe_version = ntohll(vbe->vbe_version);
1236 		switch (vbe->vbe_version) {
1237 		case VB_RAW:
1238 			/*
1239 			 * if we have textual data in vbe_bootenv, create nvlist
1240 			 * with key "envmap".
1241 			 */
1242 			fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1243 			vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1244 			fnvlist_add_string(bootenv, GRUB_ENVMAP,
1245 			    vbe->vbe_bootenv);
1246 			break;
1247 
1248 		case VB_NVLIST:
1249 			err = nvlist_unpack(vbe->vbe_bootenv,
1250 			    sizeof (vbe->vbe_bootenv), &config, 0);
1251 			if (err == 0) {
1252 				fnvlist_merge(bootenv, config);
1253 				nvlist_free(config);
1254 				break;
1255 			}
1256 			/* FALLTHROUGH */
1257 		default:
1258 			/* Check for FreeBSD zfs bootonce command string */
1259 			buf = abd_to_buf(abd);
1260 			if (*buf == '\0') {
1261 				fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1262 				    VB_NVLIST);
1263 				break;
1264 			}
1265 			fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1266 		}
1267 
1268 		/*
1269 		 * abd was allocated in vdev_label_read_bootenv_impl()
1270 		 */
1271 		abd_free(abd);
1272 		/*
1273 		 * If we managed to read any successfully,
1274 		 * return success.
1275 		 */
1276 		return (0);
1277 	}
1278 	return (err);
1279 }
1280 
1281 int
vdev_label_write_bootenv(vdev_t * vd,nvlist_t * env)1282 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1283 {
1284 	zio_t *zio;
1285 	spa_t *spa = vd->vdev_spa;
1286 	vdev_boot_envblock_t *bootenv;
1287 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1288 	int error;
1289 	size_t nvsize;
1290 	char *nvbuf;
1291 
1292 	error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1293 	if (error != 0)
1294 		return (SET_ERROR(error));
1295 
1296 	if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1297 		return (SET_ERROR(E2BIG));
1298 	}
1299 
1300 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1301 
1302 	error = ENXIO;
1303 	for (int c = 0; c < vd->vdev_children; c++) {
1304 		int child_err;
1305 
1306 		child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1307 		/*
1308 		 * As long as any of the disks managed to write all of their
1309 		 * labels successfully, return success.
1310 		 */
1311 		if (child_err == 0)
1312 			error = child_err;
1313 	}
1314 
1315 	if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1316 	    !vdev_writeable(vd)) {
1317 		return (error);
1318 	}
1319 	ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1320 	abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1321 	abd_zero(abd, VDEV_PAD_SIZE);
1322 
1323 	bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1324 	nvbuf = bootenv->vbe_bootenv;
1325 	nvsize = sizeof (bootenv->vbe_bootenv);
1326 
1327 	bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1328 	switch (bootenv->vbe_version) {
1329 	case VB_RAW:
1330 		if (nvlist_lookup_string(env, GRUB_ENVMAP, &nvbuf) == 0) {
1331 			(void) strlcpy(bootenv->vbe_bootenv, nvbuf, nvsize);
1332 		}
1333 		error = 0;
1334 		break;
1335 
1336 	case VB_NVLIST:
1337 		error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1338 		    KM_SLEEP);
1339 		break;
1340 
1341 	default:
1342 		error = EINVAL;
1343 		break;
1344 	}
1345 
1346 	if (error == 0) {
1347 		bootenv->vbe_version = htonll(bootenv->vbe_version);
1348 		abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1349 	} else {
1350 		abd_free(abd);
1351 		return (SET_ERROR(error));
1352 	}
1353 
1354 retry:
1355 	zio = zio_root(spa, NULL, NULL, flags);
1356 	for (int l = 0; l < VDEV_LABELS; l++) {
1357 		vdev_label_write(zio, vd, l, abd,
1358 		    offsetof(vdev_label_t, vl_be),
1359 		    VDEV_PAD_SIZE, NULL, NULL, flags);
1360 	}
1361 
1362 	error = zio_wait(zio);
1363 	if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1364 		flags |= ZIO_FLAG_TRYHARD;
1365 		goto retry;
1366 	}
1367 
1368 	abd_free(abd);
1369 	return (error);
1370 }
1371 
1372 /*
1373  * ==========================================================================
1374  * uberblock load/sync
1375  * ==========================================================================
1376  */
1377 
1378 /*
1379  * Consider the following situation: txg is safely synced to disk.  We've
1380  * written the first uberblock for txg + 1, and then we lose power.  When we
1381  * come back up, we fail to see the uberblock for txg + 1 because, say,
1382  * it was on a mirrored device and the replica to which we wrote txg + 1
1383  * is now offline.  If we then make some changes and sync txg + 1, and then
1384  * the missing replica comes back, then for a few seconds we'll have two
1385  * conflicting uberblocks on disk with the same txg.  The solution is simple:
1386  * among uberblocks with equal txg, choose the one with the latest timestamp.
1387  */
1388 static int
vdev_uberblock_compare(const uberblock_t * ub1,const uberblock_t * ub2)1389 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1390 {
1391 	int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1392 
1393 	if (likely(cmp))
1394 		return (cmp);
1395 
1396 	cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1397 	if (likely(cmp))
1398 		return (cmp);
1399 
1400 	/*
1401 	 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1402 	 * ZFS, e.g. zfsonlinux >= 0.7.
1403 	 *
1404 	 * If one ub has MMP and the other does not, they were written by
1405 	 * different hosts, which matters for MMP.  So we treat no MMP/no SEQ as
1406 	 * a 0 value.
1407 	 *
1408 	 * Since timestamp and txg are the same if we get this far, either is
1409 	 * acceptable for importing the pool.
1410 	 */
1411 	unsigned int seq1 = 0;
1412 	unsigned int seq2 = 0;
1413 
1414 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1415 		seq1 = MMP_SEQ(ub1);
1416 
1417 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1418 		seq2 = MMP_SEQ(ub2);
1419 
1420 	return (TREE_CMP(seq1, seq2));
1421 }
1422 
1423 struct ubl_cbdata {
1424 	uberblock_t	*ubl_ubbest;	/* Best uberblock */
1425 	vdev_t		*ubl_vd;	/* vdev associated with the above */
1426 };
1427 
1428 static void
vdev_uberblock_load_done(zio_t * zio)1429 vdev_uberblock_load_done(zio_t *zio)
1430 {
1431 	vdev_t *vd = zio->io_vd;
1432 	spa_t *spa = zio->io_spa;
1433 	zio_t *rio = zio->io_private;
1434 	uberblock_t *ub = abd_to_buf(zio->io_abd);
1435 	struct ubl_cbdata *cbp = rio->io_private;
1436 
1437 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1438 
1439 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1440 		mutex_enter(&rio->io_lock);
1441 		if (ub->ub_txg <= spa->spa_load_max_txg &&
1442 		    vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1443 			/*
1444 			 * Keep track of the vdev in which this uberblock
1445 			 * was found. We will use this information later
1446 			 * to obtain the config nvlist associated with
1447 			 * this uberblock.
1448 			 */
1449 			*cbp->ubl_ubbest = *ub;
1450 			cbp->ubl_vd = vd;
1451 		}
1452 		mutex_exit(&rio->io_lock);
1453 	}
1454 
1455 	abd_free(zio->io_abd);
1456 }
1457 
1458 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)1459 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1460     struct ubl_cbdata *cbp)
1461 {
1462 	for (int c = 0; c < vd->vdev_children; c++)
1463 		vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1464 
1465 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1466 		for (int l = 0; l < VDEV_LABELS; l++) {
1467 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1468 				vdev_label_read(zio, vd, l,
1469 				    abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1470 				    B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1471 				    VDEV_UBERBLOCK_SIZE(vd),
1472 				    vdev_uberblock_load_done, zio, flags);
1473 			}
1474 		}
1475 	}
1476 }
1477 
1478 /*
1479  * Reads the 'best' uberblock from disk along with its associated
1480  * configuration. First, we read the uberblock array of each label of each
1481  * vdev, keeping track of the uberblock with the highest txg in each array.
1482  * Then, we read the configuration from the same vdev as the best uberblock.
1483  */
1484 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)1485 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1486 {
1487 	zio_t *zio;
1488 	spa_t *spa = rvd->vdev_spa;
1489 	struct ubl_cbdata cb;
1490 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1491 	    ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1492 
1493 	ASSERT(ub);
1494 	ASSERT(config);
1495 
1496 	bzero(ub, sizeof (uberblock_t));
1497 	*config = NULL;
1498 
1499 	cb.ubl_ubbest = ub;
1500 	cb.ubl_vd = NULL;
1501 
1502 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1503 	zio = zio_root(spa, NULL, &cb, flags);
1504 	vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1505 	(void) zio_wait(zio);
1506 
1507 	/*
1508 	 * It's possible that the best uberblock was discovered on a label
1509 	 * that has a configuration which was written in a future txg.
1510 	 * Search all labels on this vdev to find the configuration that
1511 	 * matches the txg for our uberblock.
1512 	 */
1513 	if (cb.ubl_vd != NULL) {
1514 		vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1515 		    "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1516 
1517 		*config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1518 		if (*config == NULL && spa->spa_extreme_rewind) {
1519 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1520 			    "Trying again without txg restrictions.");
1521 			*config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1522 		}
1523 		if (*config == NULL) {
1524 			vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1525 		}
1526 	}
1527 	spa_config_exit(spa, SCL_ALL, FTAG);
1528 }
1529 
1530 /*
1531  * On success, increment root zio's count of good writes.
1532  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1533  */
1534 static void
vdev_uberblock_sync_done(zio_t * zio)1535 vdev_uberblock_sync_done(zio_t *zio)
1536 {
1537 	uint64_t *good_writes = zio->io_private;
1538 
1539 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1540 		atomic_inc_64(good_writes);
1541 }
1542 
1543 /*
1544  * Write the uberblock to all labels of all leaves of the specified vdev.
1545  */
1546 static void
vdev_uberblock_sync(zio_t * zio,uint64_t * good_writes,uberblock_t * ub,vdev_t * vd,int flags)1547 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1548     uberblock_t *ub, vdev_t *vd, int flags)
1549 {
1550 	for (uint64_t c = 0; c < vd->vdev_children; c++) {
1551 		vdev_uberblock_sync(zio, good_writes,
1552 		    ub, vd->vdev_child[c], flags);
1553 	}
1554 
1555 	if (!vd->vdev_ops->vdev_op_leaf)
1556 		return;
1557 
1558 	if (!vdev_writeable(vd))
1559 		return;
1560 
1561 	int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1562 	int n = ub->ub_txg % (VDEV_UBERBLOCK_COUNT(vd) - m);
1563 
1564 	/* Copy the uberblock_t into the ABD */
1565 	abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1566 	abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1567 	abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1568 
1569 	for (int l = 0; l < VDEV_LABELS; l++)
1570 		vdev_label_write(zio, vd, l, ub_abd,
1571 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1572 		    vdev_uberblock_sync_done, good_writes,
1573 		    flags | ZIO_FLAG_DONT_PROPAGATE);
1574 
1575 	abd_free(ub_abd);
1576 }
1577 
1578 /* Sync the uberblocks to all vdevs in svd[] */
1579 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1580 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1581 {
1582 	spa_t *spa = svd[0]->vdev_spa;
1583 	zio_t *zio;
1584 	uint64_t good_writes = 0;
1585 
1586 	zio = zio_root(spa, NULL, NULL, flags);
1587 
1588 	for (int v = 0; v < svdcount; v++)
1589 		vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1590 
1591 	(void) zio_wait(zio);
1592 
1593 	/*
1594 	 * Flush the uberblocks to disk.  This ensures that the odd labels
1595 	 * are no longer needed (because the new uberblocks and the even
1596 	 * labels are safely on disk), so it is safe to overwrite them.
1597 	 */
1598 	zio = zio_root(spa, NULL, NULL, flags);
1599 
1600 	for (int v = 0; v < svdcount; v++) {
1601 		if (vdev_writeable(svd[v])) {
1602 			zio_flush(zio, svd[v]);
1603 		}
1604 	}
1605 
1606 	(void) zio_wait(zio);
1607 
1608 	return (good_writes >= 1 ? 0 : EIO);
1609 }
1610 
1611 /*
1612  * On success, increment the count of good writes for our top-level vdev.
1613  */
1614 static void
vdev_label_sync_done(zio_t * zio)1615 vdev_label_sync_done(zio_t *zio)
1616 {
1617 	uint64_t *good_writes = zio->io_private;
1618 
1619 	if (zio->io_error == 0)
1620 		atomic_inc_64(good_writes);
1621 }
1622 
1623 /*
1624  * If there weren't enough good writes, indicate failure to the parent.
1625  */
1626 static void
vdev_label_sync_top_done(zio_t * zio)1627 vdev_label_sync_top_done(zio_t *zio)
1628 {
1629 	uint64_t *good_writes = zio->io_private;
1630 
1631 	if (*good_writes == 0)
1632 		zio->io_error = SET_ERROR(EIO);
1633 
1634 	kmem_free(good_writes, sizeof (uint64_t));
1635 }
1636 
1637 /*
1638  * We ignore errors for log and cache devices, simply free the private data.
1639  */
1640 static void
vdev_label_sync_ignore_done(zio_t * zio)1641 vdev_label_sync_ignore_done(zio_t *zio)
1642 {
1643 	kmem_free(zio->io_private, sizeof (uint64_t));
1644 }
1645 
1646 /*
1647  * Write all even or odd labels to all leaves of the specified vdev.
1648  */
1649 static void
vdev_label_sync(zio_t * zio,uint64_t * good_writes,vdev_t * vd,int l,uint64_t txg,int flags)1650 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1651     vdev_t *vd, int l, uint64_t txg, int flags)
1652 {
1653 	nvlist_t *label;
1654 	vdev_phys_t *vp;
1655 	abd_t *vp_abd;
1656 	char *buf;
1657 	size_t buflen;
1658 
1659 	for (int c = 0; c < vd->vdev_children; c++) {
1660 		vdev_label_sync(zio, good_writes,
1661 		    vd->vdev_child[c], l, txg, flags);
1662 	}
1663 
1664 	if (!vd->vdev_ops->vdev_op_leaf)
1665 		return;
1666 
1667 	if (!vdev_writeable(vd))
1668 		return;
1669 
1670 	/*
1671 	 * Generate a label describing the top-level config to which we belong.
1672 	 */
1673 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1674 
1675 	vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1676 	abd_zero(vp_abd, sizeof (vdev_phys_t));
1677 	vp = abd_to_buf(vp_abd);
1678 
1679 	buf = vp->vp_nvlist;
1680 	buflen = sizeof (vp->vp_nvlist);
1681 
1682 	if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1683 		for (; l < VDEV_LABELS; l += 2) {
1684 			vdev_label_write(zio, vd, l, vp_abd,
1685 			    offsetof(vdev_label_t, vl_vdev_phys),
1686 			    sizeof (vdev_phys_t),
1687 			    vdev_label_sync_done, good_writes,
1688 			    flags | ZIO_FLAG_DONT_PROPAGATE);
1689 		}
1690 	}
1691 
1692 	abd_free(vp_abd);
1693 	nvlist_free(label);
1694 }
1695 
1696 int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1697 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1698 {
1699 	list_t *dl = &spa->spa_config_dirty_list;
1700 	vdev_t *vd;
1701 	zio_t *zio;
1702 	int error;
1703 
1704 	/*
1705 	 * Write the new labels to disk.
1706 	 */
1707 	zio = zio_root(spa, NULL, NULL, flags);
1708 
1709 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1710 		uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1711 		    KM_SLEEP);
1712 
1713 		ASSERT(!vd->vdev_ishole);
1714 
1715 		zio_t *vio = zio_null(zio, spa, NULL,
1716 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
1717 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1718 		    good_writes, flags);
1719 		vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1720 		zio_nowait(vio);
1721 	}
1722 
1723 	error = zio_wait(zio);
1724 
1725 	/*
1726 	 * Flush the new labels to disk.
1727 	 */
1728 	zio = zio_root(spa, NULL, NULL, flags);
1729 
1730 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1731 		zio_flush(zio, vd);
1732 
1733 	(void) zio_wait(zio);
1734 
1735 	return (error);
1736 }
1737 
1738 /*
1739  * Sync the uberblock and any changes to the vdev configuration.
1740  *
1741  * The order of operations is carefully crafted to ensure that
1742  * if the system panics or loses power at any time, the state on disk
1743  * is still transactionally consistent.  The in-line comments below
1744  * describe the failure semantics at each stage.
1745  *
1746  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1747  * at any time, you can just call it again, and it will resume its work.
1748  */
1749 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg)1750 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
1751 {
1752 	spa_t *spa = svd[0]->vdev_spa;
1753 	uberblock_t *ub = &spa->spa_uberblock;
1754 	int error = 0;
1755 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1756 
1757 	ASSERT(svdcount != 0);
1758 retry:
1759 	/*
1760 	 * Normally, we don't want to try too hard to write every label and
1761 	 * uberblock.  If there is a flaky disk, we don't want the rest of the
1762 	 * sync process to block while we retry.  But if we can't write a
1763 	 * single label out, we should retry with ZIO_FLAG_TRYHARD before
1764 	 * bailing out and declaring the pool faulted.
1765 	 */
1766 	if (error != 0) {
1767 		if ((flags & ZIO_FLAG_TRYHARD) != 0)
1768 			return (error);
1769 		flags |= ZIO_FLAG_TRYHARD;
1770 	}
1771 
1772 	ASSERT(ub->ub_txg <= txg);
1773 
1774 	/*
1775 	 * If this isn't a resync due to I/O errors,
1776 	 * and nothing changed in this transaction group,
1777 	 * and the vdev configuration hasn't changed,
1778 	 * then there's nothing to do.
1779 	 */
1780 	if (ub->ub_txg < txg) {
1781 		boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
1782 		    txg, spa->spa_mmp.mmp_delay);
1783 
1784 		if (!changed && list_is_empty(&spa->spa_config_dirty_list))
1785 			return (0);
1786 	}
1787 
1788 	if (txg > spa_freeze_txg(spa))
1789 		return (0);
1790 
1791 	ASSERT(txg <= spa->spa_final_txg);
1792 
1793 	/*
1794 	 * Flush the write cache of every disk that's been written to
1795 	 * in this transaction group.  This ensures that all blocks
1796 	 * written in this txg will be committed to stable storage
1797 	 * before any uberblock that references them.
1798 	 */
1799 	zio_t *zio = zio_root(spa, NULL, NULL, flags);
1800 
1801 	for (vdev_t *vd =
1802 	    txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
1803 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1804 		zio_flush(zio, vd);
1805 
1806 	(void) zio_wait(zio);
1807 
1808 	/*
1809 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1810 	 * system dies in the middle of this process, that's OK: all of the
1811 	 * even labels that made it to disk will be newer than any uberblock,
1812 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
1813 	 * which have not yet been touched, will still be valid.  We flush
1814 	 * the new labels to disk to ensure that all even-label updates
1815 	 * are committed to stable storage before the uberblock update.
1816 	 */
1817 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
1818 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1819 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1820 			    "for pool '%s' when syncing out the even labels "
1821 			    "of dirty vdevs", error, spa_name(spa));
1822 		}
1823 		goto retry;
1824 	}
1825 
1826 	/*
1827 	 * Sync the uberblocks to all vdevs in svd[].
1828 	 * If the system dies in the middle of this step, there are two cases
1829 	 * to consider, and the on-disk state is consistent either way:
1830 	 *
1831 	 * (1)	If none of the new uberblocks made it to disk, then the
1832 	 *	previous uberblock will be the newest, and the odd labels
1833 	 *	(which had not yet been touched) will be valid with respect
1834 	 *	to that uberblock.
1835 	 *
1836 	 * (2)	If one or more new uberblocks made it to disk, then they
1837 	 *	will be the newest, and the even labels (which had all
1838 	 *	been successfully committed) will be valid with respect
1839 	 *	to the new uberblocks.
1840 	 */
1841 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
1842 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1843 			zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
1844 			    "%d for pool '%s'", error, spa_name(spa));
1845 		}
1846 		goto retry;
1847 	}
1848 
1849 	if (spa_multihost(spa))
1850 		mmp_update_uberblock(spa, ub);
1851 
1852 	/*
1853 	 * Sync out odd labels for every dirty vdev.  If the system dies
1854 	 * in the middle of this process, the even labels and the new
1855 	 * uberblocks will suffice to open the pool.  The next time
1856 	 * the pool is opened, the first thing we'll do -- before any
1857 	 * user data is modified -- is mark every vdev dirty so that
1858 	 * all labels will be brought up to date.  We flush the new labels
1859 	 * to disk to ensure that all odd-label updates are committed to
1860 	 * stable storage before the next transaction group begins.
1861 	 */
1862 	if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
1863 		if ((flags & ZIO_FLAG_TRYHARD) != 0) {
1864 			zfs_dbgmsg("vdev_label_sync_list() returned error %d "
1865 			    "for pool '%s' when syncing out the odd labels of "
1866 			    "dirty vdevs", error, spa_name(spa));
1867 		}
1868 		goto retry;
1869 	}
1870 
1871 	return (0);
1872 }
1873