1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22
23 /*
24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
26 * Copyright (c) 2017, Intel Corporation.
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/vdev_raidz.h>
147 #include <sys/vdev_draid.h>
148 #include <sys/uberblock_impl.h>
149 #include <sys/metaslab.h>
150 #include <sys/metaslab_impl.h>
151 #include <sys/zio.h>
152 #include <sys/dsl_scan.h>
153 #include <sys/abd.h>
154 #include <sys/fs/zfs.h>
155 #include <sys/byteorder.h>
156 #include <sys/zfs_bootenv.h>
157
158 /*
159 * Basic routines to read and write from a vdev label.
160 * Used throughout the rest of this file.
161 */
162 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)163 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
164 {
165 ASSERT(offset < sizeof (vdev_label_t));
166 ASSERT0(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t));
167
168 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
169 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
170 }
171
172 /*
173 * Returns back the vdev label associated with the passed in offset.
174 */
175 int
vdev_label_number(uint64_t psize,uint64_t offset)176 vdev_label_number(uint64_t psize, uint64_t offset)
177 {
178 int l;
179
180 if (offset >= psize - VDEV_LABEL_END_SIZE) {
181 offset -= psize - VDEV_LABEL_END_SIZE;
182 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
183 }
184 l = offset / sizeof (vdev_label_t);
185 return (l < VDEV_LABELS ? l : -1);
186 }
187
188 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)189 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
190 uint64_t size, zio_done_func_t *done, void *private, int flags)
191 {
192 ASSERT(
193 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
194 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
195 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
196
197 zio_nowait(zio_read_phys(zio, vd,
198 vdev_label_offset(vd->vdev_psize, l, offset),
199 size, buf, ZIO_CHECKSUM_LABEL, done, private,
200 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
201 }
202
203 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)204 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
205 uint64_t size, zio_done_func_t *done, void *private, int flags)
206 {
207 ASSERT(
208 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
209 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
210 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
211
212 zio_nowait(zio_write_phys(zio, vd,
213 vdev_label_offset(vd->vdev_psize, l, offset),
214 size, buf, ZIO_CHECKSUM_LABEL, done, private,
215 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
216 }
217
218 /*
219 * Generate the nvlist representing this vdev's stats
220 */
221 void
vdev_config_generate_stats(vdev_t * vd,nvlist_t * nv)222 vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
223 {
224 nvlist_t *nvx;
225 vdev_stat_t *vs;
226 vdev_stat_ex_t *vsx;
227
228 vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
229 vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
230
231 vdev_get_stats_ex(vd, vs, vsx);
232 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
233 (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
234
235 /*
236 * Add extended stats into a special extended stats nvlist. This keeps
237 * all the extended stats nicely grouped together. The extended stats
238 * nvlist is then added to the main nvlist.
239 */
240 nvx = fnvlist_alloc();
241
242 /* ZIOs in flight to disk */
243 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,
244 vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]);
245
246 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,
247 vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]);
248
249 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,
250 vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]);
251
252 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,
253 vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]);
254
255 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,
256 vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]);
257
258 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE,
259 vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]);
260
261 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE,
262 vsx->vsx_active_queue[ZIO_PRIORITY_REBUILD]);
263
264 /* ZIOs pending */
265 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
266 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
267
268 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
269 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
270
271 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
272 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
273
274 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
275 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
276
277 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
278 vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
279
280 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
281 vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
282
283 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE,
284 vsx->vsx_pend_queue[ZIO_PRIORITY_REBUILD]);
285
286 /* Histograms */
287 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
288 vsx->vsx_total_histo[ZIO_TYPE_READ],
289 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
290
291 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
292 vsx->vsx_total_histo[ZIO_TYPE_WRITE],
293 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
294
295 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
296 vsx->vsx_disk_histo[ZIO_TYPE_READ],
297 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
298
299 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
300 vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
301 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
302
303 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
304 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
305 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
306
307 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
308 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
309 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
310
311 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
312 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
313 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
314
315 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
316 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
317 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
318
319 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
320 vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
321 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
322
323 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
324 vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
325 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
326
327 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_REBUILD_LAT_HISTO,
328 vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD],
329 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD]));
330
331 /* Request sizes */
332 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
333 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
334 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
335
336 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
337 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
338 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
339
340 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
341 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
342 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
343
344 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
345 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
346 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
347
348 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
349 vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
350 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
351
352 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
353 vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
354 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
355
356 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_REBUILD_HISTO,
357 vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD],
358 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD]));
359
360 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
361 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
362 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
363
364 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
365 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
366 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
367
368 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
369 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
370 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
371
372 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
373 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
374 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
375
376 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
377 vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
378 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
379
380 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
381 vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
382 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
383
384 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_REBUILD_HISTO,
385 vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD],
386 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD]));
387
388 /* IO delays */
389 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
390
391 /* Direct I/O write verify errors */
392 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_DIO_VERIFY_ERRORS,
393 vs->vs_dio_verify_errors);
394
395 /* Add extended stats nvlist to main nvlist */
396 fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
397
398 fnvlist_free(nvx);
399 kmem_free(vs, sizeof (*vs));
400 kmem_free(vsx, sizeof (*vsx));
401 }
402
403 static void
root_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)404 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
405 {
406 spa_t *spa = vd->vdev_spa;
407
408 if (vd != spa->spa_root_vdev)
409 return;
410
411 /* provide either current or previous scan information */
412 pool_scan_stat_t ps;
413 if (spa_scan_get_stats(spa, &ps) == 0) {
414 fnvlist_add_uint64_array(nvl,
415 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
416 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
417 }
418
419 pool_removal_stat_t prs;
420 if (spa_removal_get_stats(spa, &prs) == 0) {
421 fnvlist_add_uint64_array(nvl,
422 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
423 sizeof (prs) / sizeof (uint64_t));
424 }
425
426 pool_checkpoint_stat_t pcs;
427 if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
428 fnvlist_add_uint64_array(nvl,
429 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
430 sizeof (pcs) / sizeof (uint64_t));
431 }
432
433 pool_raidz_expand_stat_t pres;
434 if (spa_raidz_expand_get_stats(spa, &pres) == 0) {
435 fnvlist_add_uint64_array(nvl,
436 ZPOOL_CONFIG_RAIDZ_EXPAND_STATS, (uint64_t *)&pres,
437 sizeof (pres) / sizeof (uint64_t));
438 }
439 }
440
441 static void
top_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)442 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
443 {
444 if (vd == vd->vdev_top) {
445 vdev_rebuild_stat_t vrs;
446 if (vdev_rebuild_get_stats(vd, &vrs) == 0) {
447 fnvlist_add_uint64_array(nvl,
448 ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs,
449 sizeof (vrs) / sizeof (uint64_t));
450 }
451 }
452 }
453
454 /*
455 * Generate the nvlist representing this vdev's config.
456 */
457 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)458 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
459 vdev_config_flag_t flags)
460 {
461 nvlist_t *nv = NULL;
462 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
463
464 nv = fnvlist_alloc();
465
466 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
467 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
468 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
469 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
470
471 if (vd->vdev_path != NULL)
472 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
473
474 if (vd->vdev_devid != NULL)
475 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
476
477 if (vd->vdev_physpath != NULL)
478 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
479 vd->vdev_physpath);
480
481 if (vd->vdev_enc_sysfs_path != NULL)
482 fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
483 vd->vdev_enc_sysfs_path);
484
485 if (vd->vdev_fru != NULL)
486 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
487
488 if (vd->vdev_ops->vdev_op_config_generate != NULL)
489 vd->vdev_ops->vdev_op_config_generate(vd, nv);
490
491 if (vd->vdev_wholedisk != -1ULL) {
492 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
493 vd->vdev_wholedisk);
494 }
495
496 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
497 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
498
499 if (vd->vdev_isspare)
500 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
501
502 if (flags & VDEV_CONFIG_L2CACHE)
503 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
504
505 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
506 vd == vd->vdev_top) {
507 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
508 vd->vdev_ms_array);
509 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
510 vd->vdev_ms_shift);
511 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
512 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
513 vd->vdev_asize);
514 fnvlist_add_uint64(nv, ZPOOL_CONFIG_MIN_ALLOC,
515 vdev_get_min_alloc(vd));
516 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
517 if (vd->vdev_noalloc) {
518 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NONALLOCATING,
519 vd->vdev_noalloc);
520 }
521
522 /*
523 * Slog devices are removed synchronously so don't
524 * persist the vdev_removing flag to the label.
525 */
526 if (vd->vdev_removing && !vd->vdev_islog) {
527 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
528 vd->vdev_removing);
529 }
530
531 /* zpool command expects alloc class data */
532 if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
533 const char *bias = NULL;
534
535 switch (vd->vdev_alloc_bias) {
536 case VDEV_BIAS_LOG:
537 bias = VDEV_ALLOC_BIAS_LOG;
538 break;
539 case VDEV_BIAS_SPECIAL:
540 bias = VDEV_ALLOC_BIAS_SPECIAL;
541 break;
542 case VDEV_BIAS_DEDUP:
543 bias = VDEV_ALLOC_BIAS_DEDUP;
544 break;
545 default:
546 ASSERT3U(vd->vdev_alloc_bias, ==,
547 VDEV_BIAS_NONE);
548 }
549 fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
550 bias);
551 }
552 }
553
554 if (vd->vdev_dtl_sm != NULL) {
555 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
556 space_map_object(vd->vdev_dtl_sm));
557 }
558
559 if (vic->vic_mapping_object != 0) {
560 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
561 vic->vic_mapping_object);
562 }
563
564 if (vic->vic_births_object != 0) {
565 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
566 vic->vic_births_object);
567 }
568
569 if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
570 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
571 vic->vic_prev_indirect_vdev);
572 }
573
574 if (vd->vdev_crtxg)
575 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
576
577 if (vd->vdev_expansion_time)
578 fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
579 vd->vdev_expansion_time);
580
581 if (flags & VDEV_CONFIG_MOS) {
582 if (vd->vdev_leaf_zap != 0) {
583 ASSERT(vd->vdev_ops->vdev_op_leaf);
584 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
585 vd->vdev_leaf_zap);
586 }
587
588 if (vd->vdev_top_zap != 0) {
589 ASSERT(vd == vd->vdev_top);
590 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
591 vd->vdev_top_zap);
592 }
593
594 if (vd->vdev_ops == &vdev_root_ops && vd->vdev_root_zap != 0 &&
595 spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2)) {
596 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_ROOT_ZAP,
597 vd->vdev_root_zap);
598 }
599
600 if (vd->vdev_resilver_deferred) {
601 ASSERT(vd->vdev_ops->vdev_op_leaf);
602 ASSERT(spa->spa_resilver_deferred);
603 fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
604 }
605 }
606
607 if (getstats) {
608 vdev_config_generate_stats(vd, nv);
609
610 root_vdev_actions_getprogress(vd, nv);
611 top_vdev_actions_getprogress(vd, nv);
612
613 /*
614 * Note: this can be called from open context
615 * (spa_get_stats()), so we need the rwlock to prevent
616 * the mapping from being changed by condensing.
617 */
618 rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
619 if (vd->vdev_indirect_mapping != NULL) {
620 ASSERT(vd->vdev_indirect_births != NULL);
621 vdev_indirect_mapping_t *vim =
622 vd->vdev_indirect_mapping;
623 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
624 vdev_indirect_mapping_size(vim));
625 }
626 rw_exit(&vd->vdev_indirect_rwlock);
627 if (vd->vdev_mg != NULL &&
628 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
629 /*
630 * Compute approximately how much memory would be used
631 * for the indirect mapping if this device were to
632 * be removed.
633 *
634 * Note: If the frag metric is invalid, then not
635 * enough metaslabs have been converted to have
636 * histograms.
637 */
638 uint64_t seg_count = 0;
639 uint64_t to_alloc = vd->vdev_stat.vs_alloc;
640
641 /*
642 * There are the same number of allocated segments
643 * as free segments, so we will have at least one
644 * entry per free segment. However, small free
645 * segments (smaller than vdev_removal_max_span)
646 * will be combined with adjacent allocated segments
647 * as a single mapping.
648 */
649 for (int i = 0; i < ZFS_RANGE_TREE_HISTOGRAM_SIZE;
650 i++) {
651 if (i + 1 < highbit64(vdev_removal_max_span)
652 - 1) {
653 to_alloc +=
654 vd->vdev_mg->mg_histogram[i] <<
655 (i + 1);
656 } else {
657 seg_count +=
658 vd->vdev_mg->mg_histogram[i];
659 }
660 }
661
662 /*
663 * The maximum length of a mapping is
664 * zfs_remove_max_segment, so we need at least one entry
665 * per zfs_remove_max_segment of allocated data.
666 */
667 seg_count += to_alloc / spa_remove_max_segment(spa);
668
669 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
670 seg_count *
671 sizeof (vdev_indirect_mapping_entry_phys_t));
672 }
673 }
674
675 if (!vd->vdev_ops->vdev_op_leaf) {
676 nvlist_t **child;
677 uint64_t c;
678
679 ASSERT(!vd->vdev_ishole);
680
681 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
682 KM_SLEEP);
683
684 for (c = 0; c < vd->vdev_children; c++) {
685 child[c] = vdev_config_generate(spa, vd->vdev_child[c],
686 getstats, flags);
687 }
688
689 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
690 (const nvlist_t * const *)child, vd->vdev_children);
691
692 for (c = 0; c < vd->vdev_children; c++)
693 nvlist_free(child[c]);
694
695 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
696
697 } else {
698 const char *aux = NULL;
699
700 if (vd->vdev_offline && !vd->vdev_tmpoffline)
701 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
702 if (vd->vdev_resilver_txg != 0)
703 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
704 vd->vdev_resilver_txg);
705 if (vd->vdev_rebuild_txg != 0)
706 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
707 vd->vdev_rebuild_txg);
708 if (vd->vdev_faulted)
709 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
710 if (vd->vdev_degraded)
711 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
712 if (vd->vdev_removed)
713 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
714 if (vd->vdev_unspare)
715 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
716 if (vd->vdev_ishole)
717 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
718
719 /* Set the reason why we're FAULTED/DEGRADED. */
720 switch (vd->vdev_stat.vs_aux) {
721 case VDEV_AUX_ERR_EXCEEDED:
722 aux = "err_exceeded";
723 break;
724
725 case VDEV_AUX_EXTERNAL:
726 aux = "external";
727 break;
728 }
729
730 if (aux != NULL && !vd->vdev_tmpoffline) {
731 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
732 } else {
733 /*
734 * We're healthy - clear any previous AUX_STATE values.
735 */
736 if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
737 nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
738 }
739
740 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
741 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
742 vd->vdev_orig_guid);
743 }
744 }
745
746 return (nv);
747 }
748
749 /*
750 * Generate a view of the top-level vdevs. If we currently have holes
751 * in the namespace, then generate an array which contains a list of holey
752 * vdevs. Additionally, add the number of top-level children that currently
753 * exist.
754 */
755 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)756 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
757 {
758 vdev_t *rvd = spa->spa_root_vdev;
759 uint64_t *array;
760 uint_t c, idx;
761
762 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
763
764 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
765 vdev_t *tvd = rvd->vdev_child[c];
766
767 if (tvd->vdev_ishole) {
768 array[idx++] = c;
769 }
770 }
771
772 if (idx) {
773 VERIFY0(nvlist_add_uint64_array(config,
774 ZPOOL_CONFIG_HOLE_ARRAY, array, idx));
775 }
776
777 VERIFY0(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
778 rvd->vdev_children));
779
780 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
781 }
782
783 /*
784 * Returns the configuration from the label of the given vdev. For vdevs
785 * which don't have a txg value stored on their label (i.e. spares/cache)
786 * or have not been completely initialized (txg = 0) just return
787 * the configuration from the first valid label we find. Otherwise,
788 * find the most up-to-date label that does not exceed the specified
789 * 'txg' value.
790 */
791 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)792 vdev_label_read_config(vdev_t *vd, uint64_t txg)
793 {
794 spa_t *spa = vd->vdev_spa;
795 nvlist_t *config = NULL;
796 vdev_phys_t *vp[VDEV_LABELS];
797 abd_t *vp_abd[VDEV_LABELS];
798 zio_t *zio[VDEV_LABELS];
799 uint64_t best_txg = 0;
800 uint64_t label_txg = 0;
801 int error = 0;
802 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
803 ZIO_FLAG_SPECULATIVE;
804
805 ASSERT(vd->vdev_validate_thread == curthread ||
806 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
807
808 if (!vdev_readable(vd))
809 return (NULL);
810
811 /*
812 * The label for a dRAID distributed spare is not stored on disk.
813 * Instead it is generated when needed which allows us to bypass
814 * the pipeline when reading the config from the label.
815 */
816 if (vd->vdev_ops == &vdev_draid_spare_ops)
817 return (vdev_draid_read_config_spare(vd));
818
819 for (int l = 0; l < VDEV_LABELS; l++) {
820 vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
821 vp[l] = abd_to_buf(vp_abd[l]);
822 }
823
824 retry:
825 for (int l = 0; l < VDEV_LABELS; l++) {
826 zio[l] = zio_root(spa, NULL, NULL, flags);
827
828 vdev_label_read(zio[l], vd, l, vp_abd[l],
829 offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
830 NULL, NULL, flags);
831 }
832 for (int l = 0; l < VDEV_LABELS; l++) {
833 nvlist_t *label = NULL;
834
835 if (zio_wait(zio[l]) == 0 &&
836 nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist),
837 &label, 0) == 0) {
838 /*
839 * Auxiliary vdevs won't have txg values in their
840 * labels and newly added vdevs may not have been
841 * completely initialized so just return the
842 * configuration from the first valid label we
843 * encounter.
844 */
845 error = nvlist_lookup_uint64(label,
846 ZPOOL_CONFIG_POOL_TXG, &label_txg);
847 if ((error || label_txg == 0) && !config) {
848 config = label;
849 for (l++; l < VDEV_LABELS; l++)
850 zio_wait(zio[l]);
851 break;
852 } else if (label_txg <= txg && label_txg > best_txg) {
853 best_txg = label_txg;
854 nvlist_free(config);
855 config = fnvlist_dup(label);
856 }
857 }
858
859 if (label != NULL) {
860 nvlist_free(label);
861 label = NULL;
862 }
863 }
864
865 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
866 flags |= ZIO_FLAG_TRYHARD;
867 goto retry;
868 }
869
870 /*
871 * We found a valid label but it didn't pass txg restrictions.
872 */
873 if (config == NULL && label_txg != 0) {
874 vdev_dbgmsg(vd, "label discarded as txg is too large "
875 "(%llu > %llu)", (u_longlong_t)label_txg,
876 (u_longlong_t)txg);
877 }
878
879 for (int l = 0; l < VDEV_LABELS; l++) {
880 abd_free(vp_abd[l]);
881 }
882
883 return (config);
884 }
885
886 /*
887 * Determine if a device is in use. The 'spare_guid' parameter will be filled
888 * in with the device guid if this spare is active elsewhere on the system.
889 */
890 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)891 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
892 uint64_t *spare_guid, uint64_t *l2cache_guid)
893 {
894 spa_t *spa = vd->vdev_spa;
895 uint64_t state, pool_guid, device_guid, txg, spare_pool;
896 uint64_t vdtxg = 0;
897 nvlist_t *label;
898
899 if (spare_guid)
900 *spare_guid = 0ULL;
901 if (l2cache_guid)
902 *l2cache_guid = 0ULL;
903
904 /*
905 * Read the label, if any, and perform some basic sanity checks.
906 */
907 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
908 return (B_FALSE);
909
910 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
911 &vdtxg);
912
913 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
914 &state) != 0 ||
915 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
916 &device_guid) != 0) {
917 nvlist_free(label);
918 return (B_FALSE);
919 }
920
921 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
922 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
923 &pool_guid) != 0 ||
924 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
925 &txg) != 0)) {
926 nvlist_free(label);
927 return (B_FALSE);
928 }
929
930 nvlist_free(label);
931
932 /*
933 * Check to see if this device indeed belongs to the pool it claims to
934 * be a part of. The only way this is allowed is if the device is a hot
935 * spare (which we check for later on).
936 */
937 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
938 !spa_guid_exists(pool_guid, device_guid) &&
939 !spa_spare_exists(device_guid, NULL, NULL) &&
940 !spa_l2cache_exists(device_guid, NULL))
941 return (B_FALSE);
942
943 /*
944 * If the transaction group is zero, then this an initialized (but
945 * unused) label. This is only an error if the create transaction
946 * on-disk is the same as the one we're using now, in which case the
947 * user has attempted to add the same vdev multiple times in the same
948 * transaction.
949 */
950 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
951 txg == 0 && vdtxg == crtxg)
952 return (B_TRUE);
953
954 /*
955 * Check to see if this is a spare device. We do an explicit check for
956 * spa_has_spare() here because it may be on our pending list of spares
957 * to add.
958 */
959 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
960 spa_has_spare(spa, device_guid)) {
961 if (spare_guid)
962 *spare_guid = device_guid;
963
964 switch (reason) {
965 case VDEV_LABEL_CREATE:
966 return (B_TRUE);
967
968 case VDEV_LABEL_REPLACE:
969 return (!spa_has_spare(spa, device_guid) ||
970 spare_pool != 0ULL);
971
972 case VDEV_LABEL_SPARE:
973 return (spa_has_spare(spa, device_guid));
974 default:
975 break;
976 }
977 }
978
979 /*
980 * Check to see if this is an l2cache device.
981 */
982 if (spa_l2cache_exists(device_guid, NULL) ||
983 spa_has_l2cache(spa, device_guid)) {
984 if (l2cache_guid)
985 *l2cache_guid = device_guid;
986
987 switch (reason) {
988 case VDEV_LABEL_CREATE:
989 return (B_TRUE);
990
991 case VDEV_LABEL_REPLACE:
992 return (!spa_has_l2cache(spa, device_guid));
993
994 case VDEV_LABEL_L2CACHE:
995 return (spa_has_l2cache(spa, device_guid));
996 default:
997 break;
998 }
999 }
1000
1001 /*
1002 * We can't rely on a pool's state if it's been imported
1003 * read-only. Instead we look to see if the pools is marked
1004 * read-only in the namespace and set the state to active.
1005 */
1006 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
1007 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
1008 spa_mode(spa) == SPA_MODE_READ)
1009 state = POOL_STATE_ACTIVE;
1010
1011 /*
1012 * If the device is marked ACTIVE, then this device is in use by another
1013 * pool on the system.
1014 */
1015 return (state == POOL_STATE_ACTIVE);
1016 }
1017
1018 static nvlist_t *
vdev_aux_label_generate(vdev_t * vd,boolean_t reason_spare)1019 vdev_aux_label_generate(vdev_t *vd, boolean_t reason_spare)
1020 {
1021 /*
1022 * For inactive hot spares and level 2 ARC devices, we generate
1023 * a special label that identifies as a mutually shared hot
1024 * spare or l2cache device. We write the label in case of
1025 * addition or removal of hot spare or l2cache vdev (in which
1026 * case we want to revert the labels).
1027 */
1028 nvlist_t *label = fnvlist_alloc();
1029 fnvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1030 spa_version(vd->vdev_spa));
1031 fnvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, reason_spare ?
1032 POOL_STATE_SPARE : POOL_STATE_L2CACHE);
1033 fnvlist_add_uint64(label, ZPOOL_CONFIG_GUID, vd->vdev_guid);
1034
1035 /*
1036 * This is merely to facilitate reporting the ashift of the
1037 * cache device through zdb. The actual retrieval of the
1038 * ashift (in vdev_alloc()) uses the nvlist
1039 * spa->spa_l2cache->sav_config (populated in
1040 * spa_ld_open_aux_vdevs()).
1041 */
1042 if (!reason_spare)
1043 fnvlist_add_uint64(label, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
1044
1045 /*
1046 * Add path information to help find it during pool import
1047 */
1048 if (vd->vdev_path != NULL)
1049 fnvlist_add_string(label, ZPOOL_CONFIG_PATH, vd->vdev_path);
1050 if (vd->vdev_devid != NULL)
1051 fnvlist_add_string(label, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
1052 if (vd->vdev_physpath != NULL) {
1053 fnvlist_add_string(label, ZPOOL_CONFIG_PHYS_PATH,
1054 vd->vdev_physpath);
1055 }
1056 return (label);
1057 }
1058
1059 /*
1060 * Initialize a vdev label. We check to make sure each leaf device is not in
1061 * use, and writable. We put down an initial label which we will later
1062 * overwrite with a complete label. Note that it's important to do this
1063 * sequentially, not in parallel, so that we catch cases of multiple use of the
1064 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
1065 * itself.
1066 */
1067 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)1068 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
1069 {
1070 spa_t *spa = vd->vdev_spa;
1071 nvlist_t *label;
1072 vdev_phys_t *vp;
1073 abd_t *vp_abd;
1074 abd_t *bootenv;
1075 uberblock_t *ub;
1076 abd_t *ub_abd;
1077 zio_t *zio;
1078 char *buf;
1079 size_t buflen;
1080 int error;
1081 uint64_t spare_guid = 0, l2cache_guid = 0;
1082 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1083 boolean_t reason_spare = (reason == VDEV_LABEL_SPARE || (reason ==
1084 VDEV_LABEL_REMOVE && vd->vdev_isspare));
1085 boolean_t reason_l2cache = (reason == VDEV_LABEL_L2CACHE || (reason ==
1086 VDEV_LABEL_REMOVE && vd->vdev_isl2cache));
1087
1088 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1089
1090 for (int c = 0; c < vd->vdev_children; c++)
1091 if ((error = vdev_label_init(vd->vdev_child[c],
1092 crtxg, reason)) != 0)
1093 return (error);
1094
1095 /* Track the creation time for this vdev */
1096 vd->vdev_crtxg = crtxg;
1097
1098 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1099 return (0);
1100
1101 /*
1102 * Dead vdevs cannot be initialized.
1103 */
1104 if (vdev_is_dead(vd))
1105 return (SET_ERROR(EIO));
1106
1107 /*
1108 * Determine if the vdev is in use.
1109 */
1110 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1111 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1112 return (SET_ERROR(EBUSY));
1113
1114 /*
1115 * If this is a request to add or replace a spare or l2cache device
1116 * that is in use elsewhere on the system, then we must update the
1117 * guid (which was initialized to a random value) to reflect the
1118 * actual GUID (which is shared between multiple pools).
1119 */
1120 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1121 spare_guid != 0ULL) {
1122 uint64_t guid_delta = spare_guid - vd->vdev_guid;
1123
1124 vd->vdev_guid += guid_delta;
1125
1126 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1127 pvd->vdev_guid_sum += guid_delta;
1128
1129 /*
1130 * If this is a replacement, then we want to fallthrough to the
1131 * rest of the code. If we're adding a spare, then it's already
1132 * labeled appropriately and we can just return.
1133 */
1134 if (reason == VDEV_LABEL_SPARE)
1135 return (0);
1136 ASSERT(reason == VDEV_LABEL_REPLACE ||
1137 reason == VDEV_LABEL_SPLIT);
1138 }
1139
1140 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1141 l2cache_guid != 0ULL) {
1142 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1143
1144 vd->vdev_guid += guid_delta;
1145
1146 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1147 pvd->vdev_guid_sum += guid_delta;
1148
1149 /*
1150 * If this is a replacement, then we want to fallthrough to the
1151 * rest of the code. If we're adding an l2cache, then it's
1152 * already labeled appropriately and we can just return.
1153 */
1154 if (reason == VDEV_LABEL_L2CACHE)
1155 return (0);
1156 ASSERT(reason == VDEV_LABEL_REPLACE);
1157 }
1158
1159 /*
1160 * Initialize its label.
1161 */
1162 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1163 abd_zero(vp_abd, sizeof (vdev_phys_t));
1164 vp = abd_to_buf(vp_abd);
1165
1166 /*
1167 * Generate a label describing the pool and our top-level vdev.
1168 * We mark it as being from txg 0 to indicate that it's not
1169 * really part of an active pool just yet. The labels will
1170 * be written again with a meaningful txg by spa_sync().
1171 */
1172 if (reason_spare || reason_l2cache) {
1173 label = vdev_aux_label_generate(vd, reason_spare);
1174
1175 /*
1176 * When spare or l2cache (aux) vdev is added during pool
1177 * creation, spa->spa_uberblock is not written until this
1178 * point. Write it on next config sync.
1179 */
1180 if (uberblock_verify(&spa->spa_uberblock))
1181 spa->spa_aux_sync_uber = B_TRUE;
1182 } else {
1183 uint64_t txg = 0ULL;
1184
1185 if (reason == VDEV_LABEL_SPLIT)
1186 txg = spa->spa_uberblock.ub_txg;
1187 label = spa_config_generate(spa, vd, txg, B_FALSE);
1188
1189 /*
1190 * Add our creation time. This allows us to detect multiple
1191 * vdev uses as described above, and automatically expires if we
1192 * fail.
1193 */
1194 VERIFY0(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1195 crtxg));
1196 }
1197
1198 buf = vp->vp_nvlist;
1199 buflen = sizeof (vp->vp_nvlist);
1200
1201 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1202 if (error != 0) {
1203 nvlist_free(label);
1204 abd_free(vp_abd);
1205 /* EFAULT means nvlist_pack ran out of room */
1206 return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1207 }
1208
1209 /*
1210 * Initialize uberblock template.
1211 */
1212 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1213 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1214 abd_zero_off(ub_abd, sizeof (uberblock_t),
1215 VDEV_UBERBLOCK_RING - sizeof (uberblock_t));
1216 ub = abd_to_buf(ub_abd);
1217 ub->ub_txg = 0;
1218
1219 /* Initialize the 2nd padding area. */
1220 bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1221 abd_zero(bootenv, VDEV_PAD_SIZE);
1222
1223 /*
1224 * Write everything in parallel.
1225 */
1226 retry:
1227 zio = zio_root(spa, NULL, NULL, flags);
1228
1229 for (int l = 0; l < VDEV_LABELS; l++) {
1230
1231 vdev_label_write(zio, vd, l, vp_abd,
1232 offsetof(vdev_label_t, vl_vdev_phys),
1233 sizeof (vdev_phys_t), NULL, NULL, flags);
1234
1235 /*
1236 * Skip the 1st padding area.
1237 * Zero out the 2nd padding area where it might have
1238 * left over data from previous filesystem format.
1239 */
1240 vdev_label_write(zio, vd, l, bootenv,
1241 offsetof(vdev_label_t, vl_be),
1242 VDEV_PAD_SIZE, NULL, NULL, flags);
1243
1244 vdev_label_write(zio, vd, l, ub_abd,
1245 offsetof(vdev_label_t, vl_uberblock),
1246 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1247 }
1248
1249 error = zio_wait(zio);
1250
1251 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1252 flags |= ZIO_FLAG_TRYHARD;
1253 goto retry;
1254 }
1255
1256 nvlist_free(label);
1257 abd_free(bootenv);
1258 abd_free(ub_abd);
1259 abd_free(vp_abd);
1260
1261 /*
1262 * If this vdev hasn't been previously identified as a spare, then we
1263 * mark it as such only if a) we are labeling it as a spare, or b) it
1264 * exists as a spare elsewhere in the system. Do the same for
1265 * level 2 ARC devices.
1266 */
1267 if (error == 0 && !vd->vdev_isspare &&
1268 (reason == VDEV_LABEL_SPARE ||
1269 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1270 spa_spare_add(vd);
1271
1272 if (error == 0 && !vd->vdev_isl2cache &&
1273 (reason == VDEV_LABEL_L2CACHE ||
1274 spa_l2cache_exists(vd->vdev_guid, NULL)))
1275 spa_l2cache_add(vd);
1276
1277 return (error);
1278 }
1279
1280 /*
1281 * Done callback for vdev_label_read_bootenv_impl. If this is the first
1282 * callback to finish, store our abd in the callback pointer. Otherwise, we
1283 * just free our abd and return.
1284 */
1285 static void
vdev_label_read_bootenv_done(zio_t * zio)1286 vdev_label_read_bootenv_done(zio_t *zio)
1287 {
1288 zio_t *rio = zio->io_private;
1289 abd_t **cbp = rio->io_private;
1290
1291 ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1292
1293 if (zio->io_error == 0) {
1294 mutex_enter(&rio->io_lock);
1295 if (*cbp == NULL) {
1296 /* Will free this buffer in vdev_label_read_bootenv. */
1297 *cbp = zio->io_abd;
1298 } else {
1299 abd_free(zio->io_abd);
1300 }
1301 mutex_exit(&rio->io_lock);
1302 } else {
1303 abd_free(zio->io_abd);
1304 }
1305 }
1306
1307 static void
vdev_label_read_bootenv_impl(zio_t * zio,vdev_t * vd,int flags)1308 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1309 {
1310 for (int c = 0; c < vd->vdev_children; c++)
1311 vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1312
1313 /*
1314 * We just use the first label that has a correct checksum; the
1315 * bootloader should have rewritten them all to be the same on boot,
1316 * and any changes we made since boot have been the same across all
1317 * labels.
1318 */
1319 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1320 for (int l = 0; l < VDEV_LABELS; l++) {
1321 vdev_label_read(zio, vd, l,
1322 abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1323 offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1324 vdev_label_read_bootenv_done, zio, flags);
1325 }
1326 }
1327 }
1328
1329 int
vdev_label_read_bootenv(vdev_t * rvd,nvlist_t * bootenv)1330 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1331 {
1332 nvlist_t *config;
1333 spa_t *spa = rvd->vdev_spa;
1334 abd_t *abd = NULL;
1335 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1336 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1337
1338 ASSERT(bootenv);
1339 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1340
1341 zio_t *zio = zio_root(spa, NULL, &abd, flags);
1342 vdev_label_read_bootenv_impl(zio, rvd, flags);
1343 int err = zio_wait(zio);
1344
1345 if (abd != NULL) {
1346 char *buf;
1347 vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1348
1349 vbe->vbe_version = ntohll(vbe->vbe_version);
1350 switch (vbe->vbe_version) {
1351 case VB_RAW:
1352 /*
1353 * if we have textual data in vbe_bootenv, create nvlist
1354 * with key "envmap".
1355 */
1356 fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1357 vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1358 fnvlist_add_string(bootenv, GRUB_ENVMAP,
1359 vbe->vbe_bootenv);
1360 break;
1361
1362 case VB_NVLIST:
1363 err = nvlist_unpack(vbe->vbe_bootenv,
1364 sizeof (vbe->vbe_bootenv), &config, 0);
1365 if (err == 0) {
1366 fnvlist_merge(bootenv, config);
1367 nvlist_free(config);
1368 break;
1369 }
1370 zfs_fallthrough;
1371 default:
1372 /* Check for FreeBSD zfs bootonce command string */
1373 buf = abd_to_buf(abd);
1374 if (*buf == '\0') {
1375 fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1376 VB_NVLIST);
1377 break;
1378 }
1379 fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1380 }
1381
1382 /*
1383 * abd was allocated in vdev_label_read_bootenv_impl()
1384 */
1385 abd_free(abd);
1386 /*
1387 * If we managed to read any successfully,
1388 * return success.
1389 */
1390 return (0);
1391 }
1392 return (err);
1393 }
1394
1395 int
vdev_label_write_bootenv(vdev_t * vd,nvlist_t * env)1396 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1397 {
1398 zio_t *zio;
1399 spa_t *spa = vd->vdev_spa;
1400 vdev_boot_envblock_t *bootenv;
1401 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1402 int error;
1403 size_t nvsize;
1404 char *nvbuf;
1405 const char *tmp;
1406
1407 error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1408 if (error != 0)
1409 return (SET_ERROR(error));
1410
1411 if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1412 return (SET_ERROR(E2BIG));
1413 }
1414
1415 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1416
1417 error = ENXIO;
1418 for (int c = 0; c < vd->vdev_children; c++) {
1419 int child_err;
1420
1421 child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1422 /*
1423 * As long as any of the disks managed to write all of their
1424 * labels successfully, return success.
1425 */
1426 if (child_err == 0)
1427 error = child_err;
1428 }
1429
1430 if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1431 !vdev_writeable(vd)) {
1432 return (error);
1433 }
1434 ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1435 abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1436 abd_zero(abd, VDEV_PAD_SIZE);
1437
1438 bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1439 nvbuf = bootenv->vbe_bootenv;
1440 nvsize = sizeof (bootenv->vbe_bootenv);
1441
1442 bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1443 switch (bootenv->vbe_version) {
1444 case VB_RAW:
1445 if (nvlist_lookup_string(env, GRUB_ENVMAP, &tmp) == 0) {
1446 (void) strlcpy(bootenv->vbe_bootenv, tmp, nvsize);
1447 }
1448 error = 0;
1449 break;
1450
1451 case VB_NVLIST:
1452 error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1453 KM_SLEEP);
1454 break;
1455
1456 default:
1457 error = EINVAL;
1458 break;
1459 }
1460
1461 if (error == 0) {
1462 bootenv->vbe_version = htonll(bootenv->vbe_version);
1463 abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1464 } else {
1465 abd_free(abd);
1466 return (SET_ERROR(error));
1467 }
1468
1469 retry:
1470 zio = zio_root(spa, NULL, NULL, flags);
1471 for (int l = 0; l < VDEV_LABELS; l++) {
1472 vdev_label_write(zio, vd, l, abd,
1473 offsetof(vdev_label_t, vl_be),
1474 VDEV_PAD_SIZE, NULL, NULL, flags);
1475 }
1476
1477 error = zio_wait(zio);
1478 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1479 flags |= ZIO_FLAG_TRYHARD;
1480 goto retry;
1481 }
1482
1483 abd_free(abd);
1484 return (error);
1485 }
1486
1487 /*
1488 * ==========================================================================
1489 * uberblock load/sync
1490 * ==========================================================================
1491 */
1492
1493 /*
1494 * Consider the following situation: txg is safely synced to disk. We've
1495 * written the first uberblock for txg + 1, and then we lose power. When we
1496 * come back up, we fail to see the uberblock for txg + 1 because, say,
1497 * it was on a mirrored device and the replica to which we wrote txg + 1
1498 * is now offline. If we then make some changes and sync txg + 1, and then
1499 * the missing replica comes back, then for a few seconds we'll have two
1500 * conflicting uberblocks on disk with the same txg. The solution is simple:
1501 * among uberblocks with equal txg, choose the one with the latest timestamp.
1502 */
1503 static int
vdev_uberblock_compare(const uberblock_t * ub1,const uberblock_t * ub2)1504 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1505 {
1506 int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1507
1508 if (likely(cmp))
1509 return (cmp);
1510
1511 cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1512 if (likely(cmp))
1513 return (cmp);
1514
1515 /*
1516 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1517 * ZFS, e.g. OpenZFS >= 0.7.
1518 *
1519 * If one ub has MMP and the other does not, they were written by
1520 * different hosts, which matters for MMP. So we treat no MMP/no SEQ as
1521 * a 0 value.
1522 *
1523 * Since timestamp and txg are the same if we get this far, either is
1524 * acceptable for importing the pool.
1525 */
1526 unsigned int seq1 = 0;
1527 unsigned int seq2 = 0;
1528
1529 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1530 seq1 = MMP_SEQ(ub1);
1531
1532 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1533 seq2 = MMP_SEQ(ub2);
1534
1535 return (TREE_CMP(seq1, seq2));
1536 }
1537
1538 struct ubl_cbdata {
1539 uberblock_t ubl_latest; /* Most recent uberblock */
1540 uberblock_t *ubl_ubbest; /* Best uberblock (w/r/t max_txg) */
1541 vdev_t *ubl_vd; /* vdev associated with the above */
1542 };
1543
1544 static void
vdev_uberblock_load_done(zio_t * zio)1545 vdev_uberblock_load_done(zio_t *zio)
1546 {
1547 vdev_t *vd = zio->io_vd;
1548 spa_t *spa = zio->io_spa;
1549 zio_t *rio = zio->io_private;
1550 uberblock_t *ub = abd_to_buf(zio->io_abd);
1551 struct ubl_cbdata *cbp = rio->io_private;
1552
1553 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1554
1555 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1556 mutex_enter(&rio->io_lock);
1557 if (vdev_uberblock_compare(ub, &cbp->ubl_latest) > 0) {
1558 cbp->ubl_latest = *ub;
1559 }
1560 if (ub->ub_txg <= spa->spa_load_max_txg &&
1561 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1562 /*
1563 * Keep track of the vdev in which this uberblock
1564 * was found. We will use this information later
1565 * to obtain the config nvlist associated with
1566 * this uberblock.
1567 */
1568 *cbp->ubl_ubbest = *ub;
1569 cbp->ubl_vd = vd;
1570 }
1571 mutex_exit(&rio->io_lock);
1572 }
1573
1574 abd_free(zio->io_abd);
1575 }
1576
1577 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)1578 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1579 struct ubl_cbdata *cbp)
1580 {
1581 for (int c = 0; c < vd->vdev_children; c++)
1582 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1583
1584 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1585 vd->vdev_ops != &vdev_draid_spare_ops) {
1586 for (int l = 0; l < VDEV_LABELS; l++) {
1587 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1588 vdev_label_read(zio, vd, l,
1589 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1590 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1591 VDEV_UBERBLOCK_SIZE(vd),
1592 vdev_uberblock_load_done, zio, flags);
1593 }
1594 }
1595 }
1596 }
1597
1598 /*
1599 * Reads the 'best' uberblock from disk along with its associated
1600 * configuration. First, we read the uberblock array of each label of each
1601 * vdev, keeping track of the uberblock with the highest txg in each array.
1602 * Then, we read the configuration from the same vdev as the best uberblock.
1603 */
1604 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)1605 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1606 {
1607 zio_t *zio;
1608 spa_t *spa = rvd->vdev_spa;
1609 struct ubl_cbdata cb;
1610 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1611 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1612
1613 ASSERT(ub);
1614 ASSERT(config);
1615
1616 memset(ub, 0, sizeof (uberblock_t));
1617 memset(&cb, 0, sizeof (cb));
1618 *config = NULL;
1619
1620 cb.ubl_ubbest = ub;
1621
1622 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1623 zio = zio_root(spa, NULL, &cb, flags);
1624 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1625 (void) zio_wait(zio);
1626
1627 /*
1628 * It's possible that the best uberblock was discovered on a label
1629 * that has a configuration which was written in a future txg.
1630 * Search all labels on this vdev to find the configuration that
1631 * matches the txg for our uberblock.
1632 */
1633 if (cb.ubl_vd != NULL) {
1634 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1635 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1636
1637 if (ub->ub_raidz_reflow_info !=
1638 cb.ubl_latest.ub_raidz_reflow_info) {
1639 vdev_dbgmsg(cb.ubl_vd,
1640 "spa=%s best uberblock (txg=%llu info=0x%llx) "
1641 "has different raidz_reflow_info than latest "
1642 "uberblock (txg=%llu info=0x%llx)",
1643 spa->spa_name,
1644 (u_longlong_t)ub->ub_txg,
1645 (u_longlong_t)ub->ub_raidz_reflow_info,
1646 (u_longlong_t)cb.ubl_latest.ub_txg,
1647 (u_longlong_t)cb.ubl_latest.ub_raidz_reflow_info);
1648 memset(ub, 0, sizeof (uberblock_t));
1649 spa_config_exit(spa, SCL_ALL, FTAG);
1650 return;
1651 }
1652
1653 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1654 if (*config == NULL && spa->spa_extreme_rewind) {
1655 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1656 "Trying again without txg restrictions.");
1657 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1658 }
1659 if (*config == NULL) {
1660 vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1661 }
1662 }
1663 spa_config_exit(spa, SCL_ALL, FTAG);
1664 }
1665
1666 /*
1667 * For use when a leaf vdev is expanded.
1668 * The location of labels 2 and 3 changed, and at the new location the
1669 * uberblock rings are either empty or contain garbage. The sync will write
1670 * new configs there because the vdev is dirty, but expansion also needs the
1671 * uberblock rings copied. Read them from label 0 which did not move.
1672 *
1673 * Since the point is to populate labels {2,3} with valid uberblocks,
1674 * we zero uberblocks we fail to read or which are not valid.
1675 */
1676
1677 static void
vdev_copy_uberblocks(vdev_t * vd)1678 vdev_copy_uberblocks(vdev_t *vd)
1679 {
1680 abd_t *ub_abd;
1681 zio_t *write_zio;
1682 int locks = (SCL_L2ARC | SCL_ZIO);
1683 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1684 ZIO_FLAG_SPECULATIVE;
1685
1686 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1687 SCL_STATE);
1688 ASSERT(vd->vdev_ops->vdev_op_leaf);
1689
1690 /*
1691 * No uberblocks are stored on distributed spares, they may be
1692 * safely skipped when expanding a leaf vdev.
1693 */
1694 if (vd->vdev_ops == &vdev_draid_spare_ops)
1695 return;
1696
1697 spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1698
1699 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1700
1701 write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1702 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1703 const int src_label = 0;
1704 zio_t *zio;
1705
1706 zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1707 vdev_label_read(zio, vd, src_label, ub_abd,
1708 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1709 NULL, NULL, flags);
1710
1711 if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1712 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1713
1714 for (int l = 2; l < VDEV_LABELS; l++)
1715 vdev_label_write(write_zio, vd, l, ub_abd,
1716 VDEV_UBERBLOCK_OFFSET(vd, n),
1717 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1718 flags | ZIO_FLAG_DONT_PROPAGATE);
1719 }
1720 (void) zio_wait(write_zio);
1721
1722 spa_config_exit(vd->vdev_spa, locks, FTAG);
1723
1724 abd_free(ub_abd);
1725 }
1726
1727 /*
1728 * On success, increment root zio's count of good writes.
1729 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1730 */
1731 static void
vdev_uberblock_sync_done(zio_t * zio)1732 vdev_uberblock_sync_done(zio_t *zio)
1733 {
1734 uint64_t *good_writes = zio->io_private;
1735
1736 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1737 atomic_inc_64(good_writes);
1738 }
1739
1740 /*
1741 * Write the uberblock to all labels of all leaves of the specified vdev.
1742 */
1743 static void
vdev_uberblock_sync(zio_t * zio,uint64_t * good_writes,uberblock_t * ub,vdev_t * vd,int flags)1744 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1745 uberblock_t *ub, vdev_t *vd, int flags)
1746 {
1747 for (uint64_t c = 0; c < vd->vdev_children; c++) {
1748 vdev_uberblock_sync(zio, good_writes,
1749 ub, vd->vdev_child[c], flags);
1750 }
1751
1752 if (!vd->vdev_ops->vdev_op_leaf)
1753 return;
1754
1755 if (!vdev_writeable(vd))
1756 return;
1757
1758 /*
1759 * There's no need to write uberblocks to a distributed spare, they
1760 * are already stored on all the leaves of the parent dRAID. For
1761 * this same reason vdev_uberblock_load_impl() skips distributed
1762 * spares when reading uberblocks.
1763 */
1764 if (vd->vdev_ops == &vdev_draid_spare_ops)
1765 return;
1766
1767 /* If the vdev was expanded, need to copy uberblock rings. */
1768 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1769 vd->vdev_copy_uberblocks == B_TRUE) {
1770 vdev_copy_uberblocks(vd);
1771 vd->vdev_copy_uberblocks = B_FALSE;
1772 }
1773
1774 /*
1775 * We chose a slot based on the txg. If this uberblock has a special
1776 * RAIDZ expansion state, then it is essentially an update of the
1777 * current uberblock (it has the same txg). However, the current
1778 * state is committed, so we want to write it to a different slot. If
1779 * we overwrote the same slot, and we lose power during the uberblock
1780 * write, and the disk does not do single-sector overwrites
1781 * atomically (even though it is required to - i.e. we should see
1782 * either the old or the new uberblock), then we could lose this
1783 * txg's uberblock. Rewinding to the previous txg's uberblock may not
1784 * be possible because RAIDZ expansion may have already overwritten
1785 * some of the data, so we need the progress indicator in the
1786 * uberblock.
1787 */
1788 int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1789 int n = (ub->ub_txg - (RRSS_GET_STATE(ub) == RRSS_SCRATCH_VALID)) %
1790 (VDEV_UBERBLOCK_COUNT(vd) - m);
1791
1792 /* Copy the uberblock_t into the ABD */
1793 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1794 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1795 abd_zero_off(ub_abd, sizeof (uberblock_t),
1796 VDEV_UBERBLOCK_SIZE(vd) - sizeof (uberblock_t));
1797
1798 for (int l = 0; l < VDEV_LABELS; l++)
1799 vdev_label_write(zio, vd, l, ub_abd,
1800 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1801 vdev_uberblock_sync_done, good_writes,
1802 flags | ZIO_FLAG_DONT_PROPAGATE);
1803
1804 abd_free(ub_abd);
1805 }
1806
1807 /* Sync the uberblocks to all vdevs in svd[] */
1808 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1809 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1810 {
1811 spa_t *spa = svd[0]->vdev_spa;
1812 zio_t *zio;
1813 uint64_t good_writes = 0;
1814
1815 zio = zio_root(spa, NULL, NULL, flags);
1816
1817 for (int v = 0; v < svdcount; v++)
1818 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1819
1820 if (spa->spa_aux_sync_uber) {
1821 for (int v = 0; v < spa->spa_spares.sav_count; v++) {
1822 vdev_uberblock_sync(zio, &good_writes, ub,
1823 spa->spa_spares.sav_vdevs[v], flags);
1824 }
1825 for (int v = 0; v < spa->spa_l2cache.sav_count; v++) {
1826 vdev_uberblock_sync(zio, &good_writes, ub,
1827 spa->spa_l2cache.sav_vdevs[v], flags);
1828 }
1829 }
1830 (void) zio_wait(zio);
1831
1832 /*
1833 * Flush the uberblocks to disk. This ensures that the odd labels
1834 * are no longer needed (because the new uberblocks and the even
1835 * labels are safely on disk), so it is safe to overwrite them.
1836 */
1837 zio = zio_root(spa, NULL, NULL, flags);
1838
1839 for (int v = 0; v < svdcount; v++) {
1840 if (vdev_writeable(svd[v])) {
1841 zio_flush(zio, svd[v]);
1842 }
1843 }
1844 if (spa->spa_aux_sync_uber) {
1845 spa->spa_aux_sync_uber = B_FALSE;
1846 for (int v = 0; v < spa->spa_spares.sav_count; v++) {
1847 if (vdev_writeable(spa->spa_spares.sav_vdevs[v])) {
1848 zio_flush(zio, spa->spa_spares.sav_vdevs[v]);
1849 }
1850 }
1851 for (int v = 0; v < spa->spa_l2cache.sav_count; v++) {
1852 if (vdev_writeable(spa->spa_l2cache.sav_vdevs[v])) {
1853 zio_flush(zio, spa->spa_l2cache.sav_vdevs[v]);
1854 }
1855 }
1856 }
1857
1858 (void) zio_wait(zio);
1859
1860 return (good_writes >= 1 ? 0 : EIO);
1861 }
1862
1863 /*
1864 * On success, increment the count of good writes for our top-level vdev.
1865 */
1866 static void
vdev_label_sync_done(zio_t * zio)1867 vdev_label_sync_done(zio_t *zio)
1868 {
1869 uint64_t *good_writes = zio->io_private;
1870
1871 if (zio->io_error == 0)
1872 atomic_inc_64(good_writes);
1873 }
1874
1875 /*
1876 * If there weren't enough good writes, indicate failure to the parent.
1877 */
1878 static void
vdev_label_sync_top_done(zio_t * zio)1879 vdev_label_sync_top_done(zio_t *zio)
1880 {
1881 uint64_t *good_writes = zio->io_private;
1882
1883 if (*good_writes == 0)
1884 zio->io_error = SET_ERROR(EIO);
1885
1886 kmem_free(good_writes, sizeof (uint64_t));
1887 }
1888
1889 /*
1890 * We ignore errors for log and cache devices, simply free the private data.
1891 */
1892 static void
vdev_label_sync_ignore_done(zio_t * zio)1893 vdev_label_sync_ignore_done(zio_t *zio)
1894 {
1895 kmem_free(zio->io_private, sizeof (uint64_t));
1896 }
1897
1898 /*
1899 * Write all even or odd labels to all leaves of the specified vdev.
1900 */
1901 static void
vdev_label_sync(zio_t * zio,uint64_t * good_writes,vdev_t * vd,int l,uint64_t txg,int flags)1902 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1903 vdev_t *vd, int l, uint64_t txg, int flags)
1904 {
1905 nvlist_t *label;
1906 vdev_phys_t *vp;
1907 abd_t *vp_abd;
1908 char *buf;
1909 size_t buflen;
1910 vdev_t *pvd = vd->vdev_parent;
1911 boolean_t spare_in_use = B_FALSE;
1912
1913 for (int c = 0; c < vd->vdev_children; c++) {
1914 vdev_label_sync(zio, good_writes,
1915 vd->vdev_child[c], l, txg, flags);
1916 }
1917
1918 if (!vd->vdev_ops->vdev_op_leaf)
1919 return;
1920
1921 if (!vdev_writeable(vd))
1922 return;
1923
1924 /*
1925 * The top-level config never needs to be written to a distributed
1926 * spare. When read vdev_dspare_label_read_config() will generate
1927 * the config for the vdev_label_read_config().
1928 */
1929 if (vd->vdev_ops == &vdev_draid_spare_ops)
1930 return;
1931
1932 if (pvd && pvd->vdev_ops == &vdev_spare_ops)
1933 spare_in_use = B_TRUE;
1934
1935 /*
1936 * Generate a label describing the top-level config to which we belong.
1937 */
1938 if ((vd->vdev_isspare && !spare_in_use) || vd->vdev_isl2cache) {
1939 label = vdev_aux_label_generate(vd, vd->vdev_isspare);
1940 } else {
1941 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1942 }
1943
1944 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1945 abd_zero(vp_abd, sizeof (vdev_phys_t));
1946 vp = abd_to_buf(vp_abd);
1947
1948 buf = vp->vp_nvlist;
1949 buflen = sizeof (vp->vp_nvlist);
1950
1951 if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1952 for (; l < VDEV_LABELS; l += 2) {
1953 vdev_label_write(zio, vd, l, vp_abd,
1954 offsetof(vdev_label_t, vl_vdev_phys),
1955 sizeof (vdev_phys_t),
1956 vdev_label_sync_done, good_writes,
1957 flags | ZIO_FLAG_DONT_PROPAGATE);
1958 }
1959 }
1960
1961 abd_free(vp_abd);
1962 nvlist_free(label);
1963 }
1964
1965 static int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1966 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1967 {
1968 list_t *dl = &spa->spa_config_dirty_list;
1969 vdev_t *vd;
1970 zio_t *zio;
1971 int error;
1972
1973 /*
1974 * Write the new labels to disk.
1975 */
1976 zio = zio_root(spa, NULL, NULL, flags);
1977
1978 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1979 uint64_t *good_writes;
1980
1981 ASSERT(!vd->vdev_ishole);
1982
1983 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1984 zio_t *vio = zio_null(zio, spa, NULL,
1985 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1986 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1987 good_writes, flags);
1988 vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1989 zio_nowait(vio);
1990 }
1991
1992 /*
1993 * AUX path may have changed during import
1994 */
1995 spa_aux_vdev_t *sav[2] = {&spa->spa_spares, &spa->spa_l2cache};
1996 for (int i = 0; i < 2; i++) {
1997 for (int v = 0; v < sav[i]->sav_count; v++) {
1998 uint64_t *good_writes;
1999 if (!sav[i]->sav_label_sync)
2000 continue;
2001 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
2002 zio_t *vio = zio_null(zio, spa, NULL,
2003 vdev_label_sync_ignore_done, good_writes, flags);
2004 vdev_label_sync(vio, good_writes, sav[i]->sav_vdevs[v],
2005 l, txg, flags);
2006 zio_nowait(vio);
2007 }
2008 }
2009
2010 error = zio_wait(zio);
2011
2012 /*
2013 * Flush the new labels to disk.
2014 */
2015 zio = zio_root(spa, NULL, NULL, flags);
2016
2017 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
2018 zio_flush(zio, vd);
2019
2020 for (int i = 0; i < 2; i++) {
2021 if (!sav[i]->sav_label_sync)
2022 continue;
2023 for (int v = 0; v < sav[i]->sav_count; v++)
2024 zio_flush(zio, sav[i]->sav_vdevs[v]);
2025 if (l == 1)
2026 sav[i]->sav_label_sync = B_FALSE;
2027 }
2028
2029 (void) zio_wait(zio);
2030
2031 return (error);
2032 }
2033
2034 /*
2035 * Sync the uberblock and any changes to the vdev configuration.
2036 *
2037 * The order of operations is carefully crafted to ensure that
2038 * if the system panics or loses power at any time, the state on disk
2039 * is still transactionally consistent. The in-line comments below
2040 * describe the failure semantics at each stage.
2041 *
2042 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
2043 * at any time, you can just call it again, and it will resume its work.
2044 */
2045 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg)2046 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
2047 {
2048 spa_t *spa = svd[0]->vdev_spa;
2049 uberblock_t *ub = &spa->spa_uberblock;
2050 int error = 0;
2051 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
2052
2053 ASSERT(svdcount != 0);
2054 retry:
2055 /*
2056 * Normally, we don't want to try too hard to write every label and
2057 * uberblock. If there is a flaky disk, we don't want the rest of the
2058 * sync process to block while we retry. But if we can't write a
2059 * single label out, we should retry with ZIO_FLAG_TRYHARD before
2060 * bailing out and declaring the pool faulted.
2061 */
2062 if (error != 0) {
2063 if ((flags & ZIO_FLAG_TRYHARD) != 0)
2064 return (error);
2065 flags |= ZIO_FLAG_TRYHARD;
2066 }
2067
2068 ASSERT(ub->ub_txg <= txg);
2069
2070 /*
2071 * If this isn't a resync due to I/O errors,
2072 * and nothing changed in this transaction group,
2073 * and multihost protection isn't enabled,
2074 * and the vdev configuration hasn't changed,
2075 * then there's nothing to do.
2076 */
2077 if (ub->ub_txg < txg) {
2078 boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
2079 txg, spa->spa_mmp.mmp_delay);
2080
2081 if (!changed && list_is_empty(&spa->spa_config_dirty_list) &&
2082 !spa_multihost(spa))
2083 return (0);
2084 }
2085
2086 if (txg > spa_freeze_txg(spa))
2087 return (0);
2088
2089 ASSERT(txg <= spa->spa_final_txg);
2090
2091 /*
2092 * Flush the write cache of every disk that's been written to
2093 * in this transaction group. This ensures that all blocks
2094 * written in this txg will be committed to stable storage
2095 * before any uberblock that references them.
2096 */
2097 zio_t *zio = zio_root(spa, NULL, NULL, flags);
2098
2099 for (vdev_t *vd =
2100 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
2101 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
2102 zio_flush(zio, vd);
2103
2104 (void) zio_wait(zio);
2105
2106 /*
2107 * Sync out the even labels (L0, L2) for every dirty vdev. If the
2108 * system dies in the middle of this process, that's OK: all of the
2109 * even labels that made it to disk will be newer than any uberblock,
2110 * and will therefore be considered invalid. The odd labels (L1, L3),
2111 * which have not yet been touched, will still be valid. We flush
2112 * the new labels to disk to ensure that all even-label updates
2113 * are committed to stable storage before the uberblock update.
2114 */
2115 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
2116 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2117 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2118 "for pool '%s' when syncing out the even labels "
2119 "of dirty vdevs", error, spa_name(spa));
2120 }
2121 goto retry;
2122 }
2123
2124 /*
2125 * Sync the uberblocks to all vdevs in svd[].
2126 * If the system dies in the middle of this step, there are two cases
2127 * to consider, and the on-disk state is consistent either way:
2128 *
2129 * (1) If none of the new uberblocks made it to disk, then the
2130 * previous uberblock will be the newest, and the odd labels
2131 * (which had not yet been touched) will be valid with respect
2132 * to that uberblock.
2133 *
2134 * (2) If one or more new uberblocks made it to disk, then they
2135 * will be the newest, and the even labels (which had all
2136 * been successfully committed) will be valid with respect
2137 * to the new uberblocks.
2138 */
2139 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
2140 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2141 zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
2142 "%d for pool '%s'", error, spa_name(spa));
2143 }
2144 goto retry;
2145 }
2146
2147 if (spa_multihost(spa))
2148 mmp_update_uberblock(spa, ub);
2149
2150 /*
2151 * Sync out odd labels for every dirty vdev. If the system dies
2152 * in the middle of this process, the even labels and the new
2153 * uberblocks will suffice to open the pool. The next time
2154 * the pool is opened, the first thing we'll do -- before any
2155 * user data is modified -- is mark every vdev dirty so that
2156 * all labels will be brought up to date. We flush the new labels
2157 * to disk to ensure that all odd-label updates are committed to
2158 * stable storage before the next transaction group begins.
2159 */
2160 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
2161 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2162 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2163 "for pool '%s' when syncing out the odd labels of "
2164 "dirty vdevs", error, spa_name(spa));
2165 }
2166 goto retry;
2167 }
2168
2169 return (0);
2170 }
2171