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 https://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2017, Intel Corporation.
26 */
27
28 /*
29 * Virtual Device Labels
30 * ---------------------
31 *
32 * The vdev label serves several distinct purposes:
33 *
34 * 1. Uniquely identify this device as part of a ZFS pool and confirm its
35 * identity within the pool.
36 *
37 * 2. Verify that all the devices given in a configuration are present
38 * within the pool.
39 *
40 * 3. Determine the uberblock for the pool.
41 *
42 * 4. In case of an import operation, determine the configuration of the
43 * toplevel vdev of which it is a part.
44 *
45 * 5. If an import operation cannot find all the devices in the pool,
46 * provide enough information to the administrator to determine which
47 * devices are missing.
48 *
49 * It is important to note that while the kernel is responsible for writing the
50 * label, it only consumes the information in the first three cases. The
51 * latter information is only consumed in userland when determining the
52 * configuration to import a pool.
53 *
54 *
55 * Label Organization
56 * ------------------
57 *
58 * Before describing the contents of the label, it's important to understand how
59 * the labels are written and updated with respect to the uberblock.
60 *
61 * When the pool configuration is altered, either because it was newly created
62 * or a device was added, we want to update all the labels such that we can deal
63 * with fatal failure at any point. To this end, each disk has two labels which
64 * are updated before and after the uberblock is synced. Assuming we have
65 * labels and an uberblock with the following transaction groups:
66 *
67 * L1 UB L2
68 * +------+ +------+ +------+
69 * | | | | | |
70 * | t10 | | t10 | | t10 |
71 * | | | | | |
72 * +------+ +------+ +------+
73 *
74 * In this stable state, the labels and the uberblock were all updated within
75 * the same transaction group (10). Each label is mirrored and checksummed, so
76 * that we can detect when we fail partway through writing the label.
77 *
78 * In order to identify which labels are valid, the labels are written in the
79 * following manner:
80 *
81 * 1. For each vdev, update 'L1' to the new label
82 * 2. Update the uberblock
83 * 3. For each vdev, update 'L2' to the new label
84 *
85 * Given arbitrary failure, we can determine the correct label to use based on
86 * the transaction group. If we fail after updating L1 but before updating the
87 * UB, we will notice that L1's transaction group is greater than the uberblock,
88 * so L2 must be valid. If we fail after writing the uberblock but before
89 * writing L2, we will notice that L2's transaction group is less than L1, and
90 * therefore L1 is valid.
91 *
92 * Another added complexity is that not every label is updated when the config
93 * is synced. If we add a single device, we do not want to have to re-write
94 * every label for every device in the pool. This means that both L1 and L2 may
95 * be older than the pool uberblock, because the necessary information is stored
96 * on another vdev.
97 *
98 *
99 * On-disk Format
100 * --------------
101 *
102 * The vdev label consists of two distinct parts, and is wrapped within the
103 * vdev_label_t structure. The label includes 8k of padding to permit legacy
104 * VTOC disk labels, but is otherwise ignored.
105 *
106 * The first half of the label is a packed nvlist which contains pool wide
107 * properties, per-vdev properties, and configuration information. It is
108 * described in more detail below.
109 *
110 * The latter half of the label consists of a redundant array of uberblocks.
111 * These uberblocks are updated whenever a transaction group is committed,
112 * or when the configuration is updated. When a pool is loaded, we scan each
113 * vdev for the 'best' uberblock.
114 *
115 *
116 * Configuration Information
117 * -------------------------
118 *
119 * The nvlist describing the pool and vdev contains the following elements:
120 *
121 * version ZFS on-disk version
122 * name Pool name
123 * state Pool state
124 * txg Transaction group in which this label was written
125 * pool_guid Unique identifier for this pool
126 * vdev_tree An nvlist describing vdev tree.
127 * features_for_read
128 * An nvlist of the features necessary for reading the MOS.
129 *
130 * Each leaf device label also contains the following:
131 *
132 * top_guid Unique ID for top-level vdev in which this is contained
133 * guid Unique ID for the leaf vdev
134 *
135 * The 'vs' configuration follows the format described in 'spa_config.c'.
136 */
137
138 #include <sys/zfs_context.h>
139 #include <sys/spa.h>
140 #include <sys/spa_impl.h>
141 #include <sys/dmu.h>
142 #include <sys/zap.h>
143 #include <sys/vdev.h>
144 #include <sys/vdev_impl.h>
145 #include <sys/vdev_raidz.h>
146 #include <sys/vdev_draid.h>
147 #include <sys/uberblock_impl.h>
148 #include <sys/metaslab.h>
149 #include <sys/metaslab_impl.h>
150 #include <sys/zio.h>
151 #include <sys/dsl_scan.h>
152 #include <sys/abd.h>
153 #include <sys/fs/zfs.h>
154 #include <sys/byteorder.h>
155 #include <sys/zfs_bootenv.h>
156
157 /*
158 * Basic routines to read and write from a vdev label.
159 * Used throughout the rest of this file.
160 */
161 uint64_t
vdev_label_offset(uint64_t psize,int l,uint64_t offset)162 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
163 {
164 ASSERT(offset < sizeof (vdev_label_t));
165 ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
166
167 return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
168 0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
169 }
170
171 /*
172 * Returns back the vdev label associated with the passed in offset.
173 */
174 int
vdev_label_number(uint64_t psize,uint64_t offset)175 vdev_label_number(uint64_t psize, uint64_t offset)
176 {
177 int l;
178
179 if (offset >= psize - VDEV_LABEL_END_SIZE) {
180 offset -= psize - VDEV_LABEL_END_SIZE;
181 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
182 }
183 l = offset / sizeof (vdev_label_t);
184 return (l < VDEV_LABELS ? l : -1);
185 }
186
187 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)188 vdev_label_read(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
189 uint64_t size, zio_done_func_t *done, void *private, int flags)
190 {
191 ASSERT(
192 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
193 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
194 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
195
196 zio_nowait(zio_read_phys(zio, vd,
197 vdev_label_offset(vd->vdev_psize, l, offset),
198 size, buf, ZIO_CHECKSUM_LABEL, done, private,
199 ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
200 }
201
202 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)203 vdev_label_write(zio_t *zio, vdev_t *vd, int l, abd_t *buf, uint64_t offset,
204 uint64_t size, zio_done_func_t *done, void *private, int flags)
205 {
206 ASSERT(
207 spa_config_held(zio->io_spa, SCL_STATE, RW_READER) == SCL_STATE ||
208 spa_config_held(zio->io_spa, SCL_STATE, RW_WRITER) == SCL_STATE);
209 ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
210
211 zio_nowait(zio_write_phys(zio, vd,
212 vdev_label_offset(vd->vdev_psize, l, offset),
213 size, buf, ZIO_CHECKSUM_LABEL, done, private,
214 ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
215 }
216
217 /*
218 * Generate the nvlist representing this vdev's stats
219 */
220 void
vdev_config_generate_stats(vdev_t * vd,nvlist_t * nv)221 vdev_config_generate_stats(vdev_t *vd, nvlist_t *nv)
222 {
223 nvlist_t *nvx;
224 vdev_stat_t *vs;
225 vdev_stat_ex_t *vsx;
226
227 vs = kmem_alloc(sizeof (*vs), KM_SLEEP);
228 vsx = kmem_alloc(sizeof (*vsx), KM_SLEEP);
229
230 vdev_get_stats_ex(vd, vs, vsx);
231 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
232 (uint64_t *)vs, sizeof (*vs) / sizeof (uint64_t));
233
234 /*
235 * Add extended stats into a special extended stats nvlist. This keeps
236 * all the extended stats nicely grouped together. The extended stats
237 * nvlist is then added to the main nvlist.
238 */
239 nvx = fnvlist_alloc();
240
241 /* ZIOs in flight to disk */
242 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_ACTIVE_QUEUE,
243 vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_READ]);
244
245 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_ACTIVE_QUEUE,
246 vsx->vsx_active_queue[ZIO_PRIORITY_SYNC_WRITE]);
247
248 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_ACTIVE_QUEUE,
249 vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_READ]);
250
251 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_ACTIVE_QUEUE,
252 vsx->vsx_active_queue[ZIO_PRIORITY_ASYNC_WRITE]);
253
254 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_ACTIVE_QUEUE,
255 vsx->vsx_active_queue[ZIO_PRIORITY_SCRUB]);
256
257 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_ACTIVE_QUEUE,
258 vsx->vsx_active_queue[ZIO_PRIORITY_TRIM]);
259
260 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_ACTIVE_QUEUE,
261 vsx->vsx_active_queue[ZIO_PRIORITY_REBUILD]);
262
263 /* ZIOs pending */
264 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_PEND_QUEUE,
265 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_READ]);
266
267 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_PEND_QUEUE,
268 vsx->vsx_pend_queue[ZIO_PRIORITY_SYNC_WRITE]);
269
270 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_PEND_QUEUE,
271 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_READ]);
272
273 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_PEND_QUEUE,
274 vsx->vsx_pend_queue[ZIO_PRIORITY_ASYNC_WRITE]);
275
276 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SCRUB_PEND_QUEUE,
277 vsx->vsx_pend_queue[ZIO_PRIORITY_SCRUB]);
278
279 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_TRIM_PEND_QUEUE,
280 vsx->vsx_pend_queue[ZIO_PRIORITY_TRIM]);
281
282 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_REBUILD_PEND_QUEUE,
283 vsx->vsx_pend_queue[ZIO_PRIORITY_REBUILD]);
284
285 /* Histograms */
286 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_R_LAT_HISTO,
287 vsx->vsx_total_histo[ZIO_TYPE_READ],
288 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_READ]));
289
290 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TOT_W_LAT_HISTO,
291 vsx->vsx_total_histo[ZIO_TYPE_WRITE],
292 ARRAY_SIZE(vsx->vsx_total_histo[ZIO_TYPE_WRITE]));
293
294 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_R_LAT_HISTO,
295 vsx->vsx_disk_histo[ZIO_TYPE_READ],
296 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_READ]));
297
298 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_DISK_W_LAT_HISTO,
299 vsx->vsx_disk_histo[ZIO_TYPE_WRITE],
300 ARRAY_SIZE(vsx->vsx_disk_histo[ZIO_TYPE_WRITE]));
301
302 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_R_LAT_HISTO,
303 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ],
304 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_READ]));
305
306 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_W_LAT_HISTO,
307 vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE],
308 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SYNC_WRITE]));
309
310 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_R_LAT_HISTO,
311 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ],
312 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_READ]));
313
314 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_W_LAT_HISTO,
315 vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE],
316 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_ASYNC_WRITE]));
317
318 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SCRUB_LAT_HISTO,
319 vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB],
320 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_SCRUB]));
321
322 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_TRIM_LAT_HISTO,
323 vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM],
324 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_TRIM]));
325
326 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_REBUILD_LAT_HISTO,
327 vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD],
328 ARRAY_SIZE(vsx->vsx_queue_histo[ZIO_PRIORITY_REBUILD]));
329
330 /* Request sizes */
331 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_R_HISTO,
332 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ],
333 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_READ]));
334
335 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_IND_W_HISTO,
336 vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE],
337 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SYNC_WRITE]));
338
339 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_R_HISTO,
340 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ],
341 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_READ]));
342
343 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_IND_W_HISTO,
344 vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE],
345 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_ASYNC_WRITE]));
346
347 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_SCRUB_HISTO,
348 vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB],
349 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_SCRUB]));
350
351 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_TRIM_HISTO,
352 vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM],
353 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_TRIM]));
354
355 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_IND_REBUILD_HISTO,
356 vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD],
357 ARRAY_SIZE(vsx->vsx_ind_histo[ZIO_PRIORITY_REBUILD]));
358
359 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_R_HISTO,
360 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ],
361 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_READ]));
362
363 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_SYNC_AGG_W_HISTO,
364 vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE],
365 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SYNC_WRITE]));
366
367 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_R_HISTO,
368 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ],
369 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_READ]));
370
371 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_ASYNC_AGG_W_HISTO,
372 vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE],
373 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_ASYNC_WRITE]));
374
375 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_SCRUB_HISTO,
376 vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB],
377 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_SCRUB]));
378
379 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_TRIM_HISTO,
380 vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM],
381 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_TRIM]));
382
383 fnvlist_add_uint64_array(nvx, ZPOOL_CONFIG_VDEV_AGG_REBUILD_HISTO,
384 vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD],
385 ARRAY_SIZE(vsx->vsx_agg_histo[ZIO_PRIORITY_REBUILD]));
386
387 /* IO delays */
388 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_SLOW_IOS, vs->vs_slow_ios);
389
390 /* Direct I/O write verify errors */
391 fnvlist_add_uint64(nvx, ZPOOL_CONFIG_VDEV_DIO_VERIFY_ERRORS,
392 vs->vs_dio_verify_errors);
393
394 /* Add extended stats nvlist to main nvlist */
395 fnvlist_add_nvlist(nv, ZPOOL_CONFIG_VDEV_STATS_EX, nvx);
396
397 fnvlist_free(nvx);
398 kmem_free(vs, sizeof (*vs));
399 kmem_free(vsx, sizeof (*vsx));
400 }
401
402 static void
root_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)403 root_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
404 {
405 spa_t *spa = vd->vdev_spa;
406
407 if (vd != spa->spa_root_vdev)
408 return;
409
410 /* provide either current or previous scan information */
411 pool_scan_stat_t ps;
412 if (spa_scan_get_stats(spa, &ps) == 0) {
413 fnvlist_add_uint64_array(nvl,
414 ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
415 sizeof (pool_scan_stat_t) / sizeof (uint64_t));
416 }
417
418 pool_removal_stat_t prs;
419 if (spa_removal_get_stats(spa, &prs) == 0) {
420 fnvlist_add_uint64_array(nvl,
421 ZPOOL_CONFIG_REMOVAL_STATS, (uint64_t *)&prs,
422 sizeof (prs) / sizeof (uint64_t));
423 }
424
425 pool_checkpoint_stat_t pcs;
426 if (spa_checkpoint_get_stats(spa, &pcs) == 0) {
427 fnvlist_add_uint64_array(nvl,
428 ZPOOL_CONFIG_CHECKPOINT_STATS, (uint64_t *)&pcs,
429 sizeof (pcs) / sizeof (uint64_t));
430 }
431
432 pool_raidz_expand_stat_t pres;
433 if (spa_raidz_expand_get_stats(spa, &pres) == 0) {
434 fnvlist_add_uint64_array(nvl,
435 ZPOOL_CONFIG_RAIDZ_EXPAND_STATS, (uint64_t *)&pres,
436 sizeof (pres) / sizeof (uint64_t));
437 }
438 }
439
440 static void
top_vdev_actions_getprogress(vdev_t * vd,nvlist_t * nvl)441 top_vdev_actions_getprogress(vdev_t *vd, nvlist_t *nvl)
442 {
443 if (vd == vd->vdev_top) {
444 vdev_rebuild_stat_t vrs;
445 if (vdev_rebuild_get_stats(vd, &vrs) == 0) {
446 fnvlist_add_uint64_array(nvl,
447 ZPOOL_CONFIG_REBUILD_STATS, (uint64_t *)&vrs,
448 sizeof (vrs) / sizeof (uint64_t));
449 }
450 }
451 }
452
453 /*
454 * Generate the nvlist representing this vdev's config.
455 */
456 nvlist_t *
vdev_config_generate(spa_t * spa,vdev_t * vd,boolean_t getstats,vdev_config_flag_t flags)457 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
458 vdev_config_flag_t flags)
459 {
460 nvlist_t *nv = NULL;
461 vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
462
463 nv = fnvlist_alloc();
464
465 fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
466 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
467 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
468 fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
469
470 if (vd->vdev_path != NULL)
471 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
472
473 if (vd->vdev_devid != NULL)
474 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
475
476 if (vd->vdev_physpath != NULL)
477 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
478 vd->vdev_physpath);
479
480 if (vd->vdev_enc_sysfs_path != NULL)
481 fnvlist_add_string(nv, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH,
482 vd->vdev_enc_sysfs_path);
483
484 if (vd->vdev_fru != NULL)
485 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
486
487 if (vd->vdev_ops->vdev_op_config_generate != NULL)
488 vd->vdev_ops->vdev_op_config_generate(vd, nv);
489
490 if (vd->vdev_wholedisk != -1ULL) {
491 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
492 vd->vdev_wholedisk);
493 }
494
495 if (vd->vdev_not_present && !(flags & VDEV_CONFIG_MISSING))
496 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
497
498 if (vd->vdev_isspare)
499 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
500
501 if (flags & VDEV_CONFIG_L2CACHE)
502 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
503
504 if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
505 vd == vd->vdev_top) {
506 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
507 vd->vdev_ms_array);
508 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
509 vd->vdev_ms_shift);
510 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
511 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
512 vd->vdev_asize);
513 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
514 if (vd->vdev_noalloc) {
515 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NONALLOCATING,
516 vd->vdev_noalloc);
517 }
518
519 /*
520 * Slog devices are removed synchronously so don't
521 * persist the vdev_removing flag to the label.
522 */
523 if (vd->vdev_removing && !vd->vdev_islog) {
524 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
525 vd->vdev_removing);
526 }
527
528 /* zpool command expects alloc class data */
529 if (getstats && vd->vdev_alloc_bias != VDEV_BIAS_NONE) {
530 const char *bias = NULL;
531
532 switch (vd->vdev_alloc_bias) {
533 case VDEV_BIAS_LOG:
534 bias = VDEV_ALLOC_BIAS_LOG;
535 break;
536 case VDEV_BIAS_SPECIAL:
537 bias = VDEV_ALLOC_BIAS_SPECIAL;
538 break;
539 case VDEV_BIAS_DEDUP:
540 bias = VDEV_ALLOC_BIAS_DEDUP;
541 break;
542 default:
543 ASSERT3U(vd->vdev_alloc_bias, ==,
544 VDEV_BIAS_NONE);
545 }
546 fnvlist_add_string(nv, ZPOOL_CONFIG_ALLOCATION_BIAS,
547 bias);
548 }
549 }
550
551 if (vd->vdev_dtl_sm != NULL) {
552 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
553 space_map_object(vd->vdev_dtl_sm));
554 }
555
556 if (vic->vic_mapping_object != 0) {
557 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT,
558 vic->vic_mapping_object);
559 }
560
561 if (vic->vic_births_object != 0) {
562 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS,
563 vic->vic_births_object);
564 }
565
566 if (vic->vic_prev_indirect_vdev != UINT64_MAX) {
567 fnvlist_add_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
568 vic->vic_prev_indirect_vdev);
569 }
570
571 if (vd->vdev_crtxg)
572 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
573
574 if (vd->vdev_expansion_time)
575 fnvlist_add_uint64(nv, ZPOOL_CONFIG_EXPANSION_TIME,
576 vd->vdev_expansion_time);
577
578 if (flags & VDEV_CONFIG_MOS) {
579 if (vd->vdev_leaf_zap != 0) {
580 ASSERT(vd->vdev_ops->vdev_op_leaf);
581 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_LEAF_ZAP,
582 vd->vdev_leaf_zap);
583 }
584
585 if (vd->vdev_top_zap != 0) {
586 ASSERT(vd == vd->vdev_top);
587 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP,
588 vd->vdev_top_zap);
589 }
590
591 if (vd->vdev_ops == &vdev_root_ops && vd->vdev_root_zap != 0 &&
592 spa_feature_is_active(vd->vdev_spa, SPA_FEATURE_AVZ_V2)) {
593 fnvlist_add_uint64(nv, ZPOOL_CONFIG_VDEV_ROOT_ZAP,
594 vd->vdev_root_zap);
595 }
596
597 if (vd->vdev_resilver_deferred) {
598 ASSERT(vd->vdev_ops->vdev_op_leaf);
599 ASSERT(spa->spa_resilver_deferred);
600 fnvlist_add_boolean(nv, ZPOOL_CONFIG_RESILVER_DEFER);
601 }
602 }
603
604 if (getstats) {
605 vdev_config_generate_stats(vd, nv);
606
607 root_vdev_actions_getprogress(vd, nv);
608 top_vdev_actions_getprogress(vd, nv);
609
610 /*
611 * Note: this can be called from open context
612 * (spa_get_stats()), so we need the rwlock to prevent
613 * the mapping from being changed by condensing.
614 */
615 rw_enter(&vd->vdev_indirect_rwlock, RW_READER);
616 if (vd->vdev_indirect_mapping != NULL) {
617 ASSERT(vd->vdev_indirect_births != NULL);
618 vdev_indirect_mapping_t *vim =
619 vd->vdev_indirect_mapping;
620 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
621 vdev_indirect_mapping_size(vim));
622 }
623 rw_exit(&vd->vdev_indirect_rwlock);
624 if (vd->vdev_mg != NULL &&
625 vd->vdev_mg->mg_fragmentation != ZFS_FRAG_INVALID) {
626 /*
627 * Compute approximately how much memory would be used
628 * for the indirect mapping if this device were to
629 * be removed.
630 *
631 * Note: If the frag metric is invalid, then not
632 * enough metaslabs have been converted to have
633 * histograms.
634 */
635 uint64_t seg_count = 0;
636 uint64_t to_alloc = vd->vdev_stat.vs_alloc;
637
638 /*
639 * There are the same number of allocated segments
640 * as free segments, so we will have at least one
641 * entry per free segment. However, small free
642 * segments (smaller than vdev_removal_max_span)
643 * will be combined with adjacent allocated segments
644 * as a single mapping.
645 */
646 for (int i = 0; i < ZFS_RANGE_TREE_HISTOGRAM_SIZE;
647 i++) {
648 if (i + 1 < highbit64(vdev_removal_max_span)
649 - 1) {
650 to_alloc +=
651 vd->vdev_mg->mg_histogram[i] <<
652 (i + 1);
653 } else {
654 seg_count +=
655 vd->vdev_mg->mg_histogram[i];
656 }
657 }
658
659 /*
660 * The maximum length of a mapping is
661 * zfs_remove_max_segment, so we need at least one entry
662 * per zfs_remove_max_segment of allocated data.
663 */
664 seg_count += to_alloc / spa_remove_max_segment(spa);
665
666 fnvlist_add_uint64(nv, ZPOOL_CONFIG_INDIRECT_SIZE,
667 seg_count *
668 sizeof (vdev_indirect_mapping_entry_phys_t));
669 }
670 }
671
672 if (!vd->vdev_ops->vdev_op_leaf) {
673 nvlist_t **child;
674 uint64_t c;
675
676 ASSERT(!vd->vdev_ishole);
677
678 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
679 KM_SLEEP);
680
681 for (c = 0; c < vd->vdev_children; c++) {
682 child[c] = vdev_config_generate(spa, vd->vdev_child[c],
683 getstats, flags);
684 }
685
686 fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
687 (const nvlist_t * const *)child, vd->vdev_children);
688
689 for (c = 0; c < vd->vdev_children; c++)
690 nvlist_free(child[c]);
691
692 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
693
694 } else {
695 const char *aux = NULL;
696
697 if (vd->vdev_offline && !vd->vdev_tmpoffline)
698 fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
699 if (vd->vdev_resilver_txg != 0)
700 fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
701 vd->vdev_resilver_txg);
702 if (vd->vdev_rebuild_txg != 0)
703 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REBUILD_TXG,
704 vd->vdev_rebuild_txg);
705 if (vd->vdev_faulted)
706 fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
707 if (vd->vdev_degraded)
708 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
709 if (vd->vdev_removed)
710 fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
711 if (vd->vdev_unspare)
712 fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
713 if (vd->vdev_ishole)
714 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
715
716 /* Set the reason why we're FAULTED/DEGRADED. */
717 switch (vd->vdev_stat.vs_aux) {
718 case VDEV_AUX_ERR_EXCEEDED:
719 aux = "err_exceeded";
720 break;
721
722 case VDEV_AUX_EXTERNAL:
723 aux = "external";
724 break;
725 }
726
727 if (aux != NULL && !vd->vdev_tmpoffline) {
728 fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
729 } else {
730 /*
731 * We're healthy - clear any previous AUX_STATE values.
732 */
733 if (nvlist_exists(nv, ZPOOL_CONFIG_AUX_STATE))
734 nvlist_remove_all(nv, ZPOOL_CONFIG_AUX_STATE);
735 }
736
737 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
738 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
739 vd->vdev_orig_guid);
740 }
741 }
742
743 return (nv);
744 }
745
746 /*
747 * Generate a view of the top-level vdevs. If we currently have holes
748 * in the namespace, then generate an array which contains a list of holey
749 * vdevs. Additionally, add the number of top-level children that currently
750 * exist.
751 */
752 void
vdev_top_config_generate(spa_t * spa,nvlist_t * config)753 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
754 {
755 vdev_t *rvd = spa->spa_root_vdev;
756 uint64_t *array;
757 uint_t c, idx;
758
759 array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
760
761 for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
762 vdev_t *tvd = rvd->vdev_child[c];
763
764 if (tvd->vdev_ishole) {
765 array[idx++] = c;
766 }
767 }
768
769 if (idx) {
770 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
771 array, idx) == 0);
772 }
773
774 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
775 rvd->vdev_children) == 0);
776
777 kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
778 }
779
780 /*
781 * Returns the configuration from the label of the given vdev. For vdevs
782 * which don't have a txg value stored on their label (i.e. spares/cache)
783 * or have not been completely initialized (txg = 0) just return
784 * the configuration from the first valid label we find. Otherwise,
785 * find the most up-to-date label that does not exceed the specified
786 * 'txg' value.
787 */
788 nvlist_t *
vdev_label_read_config(vdev_t * vd,uint64_t txg)789 vdev_label_read_config(vdev_t *vd, uint64_t txg)
790 {
791 spa_t *spa = vd->vdev_spa;
792 nvlist_t *config = NULL;
793 vdev_phys_t *vp[VDEV_LABELS];
794 abd_t *vp_abd[VDEV_LABELS];
795 zio_t *zio[VDEV_LABELS];
796 uint64_t best_txg = 0;
797 uint64_t label_txg = 0;
798 int error = 0;
799 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
800 ZIO_FLAG_SPECULATIVE;
801
802 ASSERT(vd->vdev_validate_thread == curthread ||
803 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
804
805 if (!vdev_readable(vd))
806 return (NULL);
807
808 /*
809 * The label for a dRAID distributed spare is not stored on disk.
810 * Instead it is generated when needed which allows us to bypass
811 * the pipeline when reading the config from the label.
812 */
813 if (vd->vdev_ops == &vdev_draid_spare_ops)
814 return (vdev_draid_read_config_spare(vd));
815
816 for (int l = 0; l < VDEV_LABELS; l++) {
817 vp_abd[l] = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
818 vp[l] = abd_to_buf(vp_abd[l]);
819 }
820
821 retry:
822 for (int l = 0; l < VDEV_LABELS; l++) {
823 zio[l] = zio_root(spa, NULL, NULL, flags);
824
825 vdev_label_read(zio[l], vd, l, vp_abd[l],
826 offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
827 NULL, NULL, flags);
828 }
829 for (int l = 0; l < VDEV_LABELS; l++) {
830 nvlist_t *label = NULL;
831
832 if (zio_wait(zio[l]) == 0 &&
833 nvlist_unpack(vp[l]->vp_nvlist, sizeof (vp[l]->vp_nvlist),
834 &label, 0) == 0) {
835 /*
836 * Auxiliary vdevs won't have txg values in their
837 * labels and newly added vdevs may not have been
838 * completely initialized so just return the
839 * configuration from the first valid label we
840 * encounter.
841 */
842 error = nvlist_lookup_uint64(label,
843 ZPOOL_CONFIG_POOL_TXG, &label_txg);
844 if ((error || label_txg == 0) && !config) {
845 config = label;
846 for (l++; l < VDEV_LABELS; l++)
847 zio_wait(zio[l]);
848 break;
849 } else if (label_txg <= txg && label_txg > best_txg) {
850 best_txg = label_txg;
851 nvlist_free(config);
852 config = fnvlist_dup(label);
853 }
854 }
855
856 if (label != NULL) {
857 nvlist_free(label);
858 label = NULL;
859 }
860 }
861
862 if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
863 flags |= ZIO_FLAG_TRYHARD;
864 goto retry;
865 }
866
867 /*
868 * We found a valid label but it didn't pass txg restrictions.
869 */
870 if (config == NULL && label_txg != 0) {
871 vdev_dbgmsg(vd, "label discarded as txg is too large "
872 "(%llu > %llu)", (u_longlong_t)label_txg,
873 (u_longlong_t)txg);
874 }
875
876 for (int l = 0; l < VDEV_LABELS; l++) {
877 abd_free(vp_abd[l]);
878 }
879
880 return (config);
881 }
882
883 /*
884 * Determine if a device is in use. The 'spare_guid' parameter will be filled
885 * in with the device guid if this spare is active elsewhere on the system.
886 */
887 static boolean_t
vdev_inuse(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason,uint64_t * spare_guid,uint64_t * l2cache_guid)888 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
889 uint64_t *spare_guid, uint64_t *l2cache_guid)
890 {
891 spa_t *spa = vd->vdev_spa;
892 uint64_t state, pool_guid, device_guid, txg, spare_pool;
893 uint64_t vdtxg = 0;
894 nvlist_t *label;
895
896 if (spare_guid)
897 *spare_guid = 0ULL;
898 if (l2cache_guid)
899 *l2cache_guid = 0ULL;
900
901 /*
902 * Read the label, if any, and perform some basic sanity checks.
903 */
904 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
905 return (B_FALSE);
906
907 (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
908 &vdtxg);
909
910 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
911 &state) != 0 ||
912 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
913 &device_guid) != 0) {
914 nvlist_free(label);
915 return (B_FALSE);
916 }
917
918 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
919 (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
920 &pool_guid) != 0 ||
921 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
922 &txg) != 0)) {
923 nvlist_free(label);
924 return (B_FALSE);
925 }
926
927 nvlist_free(label);
928
929 /*
930 * Check to see if this device indeed belongs to the pool it claims to
931 * be a part of. The only way this is allowed is if the device is a hot
932 * spare (which we check for later on).
933 */
934 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
935 !spa_guid_exists(pool_guid, device_guid) &&
936 !spa_spare_exists(device_guid, NULL, NULL) &&
937 !spa_l2cache_exists(device_guid, NULL))
938 return (B_FALSE);
939
940 /*
941 * If the transaction group is zero, then this an initialized (but
942 * unused) label. This is only an error if the create transaction
943 * on-disk is the same as the one we're using now, in which case the
944 * user has attempted to add the same vdev multiple times in the same
945 * transaction.
946 */
947 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
948 txg == 0 && vdtxg == crtxg)
949 return (B_TRUE);
950
951 /*
952 * Check to see if this is a spare device. We do an explicit check for
953 * spa_has_spare() here because it may be on our pending list of spares
954 * to add.
955 */
956 if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
957 spa_has_spare(spa, device_guid)) {
958 if (spare_guid)
959 *spare_guid = device_guid;
960
961 switch (reason) {
962 case VDEV_LABEL_CREATE:
963 return (B_TRUE);
964
965 case VDEV_LABEL_REPLACE:
966 return (!spa_has_spare(spa, device_guid) ||
967 spare_pool != 0ULL);
968
969 case VDEV_LABEL_SPARE:
970 return (spa_has_spare(spa, device_guid));
971 default:
972 break;
973 }
974 }
975
976 /*
977 * Check to see if this is an l2cache device.
978 */
979 if (spa_l2cache_exists(device_guid, NULL) ||
980 spa_has_l2cache(spa, device_guid)) {
981 if (l2cache_guid)
982 *l2cache_guid = device_guid;
983
984 switch (reason) {
985 case VDEV_LABEL_CREATE:
986 return (B_TRUE);
987
988 case VDEV_LABEL_REPLACE:
989 return (!spa_has_l2cache(spa, device_guid));
990
991 case VDEV_LABEL_L2CACHE:
992 return (spa_has_l2cache(spa, device_guid));
993 default:
994 break;
995 }
996 }
997
998 /*
999 * We can't rely on a pool's state if it's been imported
1000 * read-only. Instead we look to see if the pools is marked
1001 * read-only in the namespace and set the state to active.
1002 */
1003 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
1004 (spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
1005 spa_mode(spa) == SPA_MODE_READ)
1006 state = POOL_STATE_ACTIVE;
1007
1008 /*
1009 * If the device is marked ACTIVE, then this device is in use by another
1010 * pool on the system.
1011 */
1012 return (state == POOL_STATE_ACTIVE);
1013 }
1014
1015 static nvlist_t *
vdev_aux_label_generate(vdev_t * vd,boolean_t reason_spare)1016 vdev_aux_label_generate(vdev_t *vd, boolean_t reason_spare)
1017 {
1018 /*
1019 * For inactive hot spares and level 2 ARC devices, we generate
1020 * a special label that identifies as a mutually shared hot
1021 * spare or l2cache device. We write the label in case of
1022 * addition or removal of hot spare or l2cache vdev (in which
1023 * case we want to revert the labels).
1024 */
1025 nvlist_t *label = fnvlist_alloc();
1026 fnvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
1027 spa_version(vd->vdev_spa));
1028 fnvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE, reason_spare ?
1029 POOL_STATE_SPARE : POOL_STATE_L2CACHE);
1030 fnvlist_add_uint64(label, ZPOOL_CONFIG_GUID, vd->vdev_guid);
1031
1032 /*
1033 * This is merely to facilitate reporting the ashift of the
1034 * cache device through zdb. The actual retrieval of the
1035 * ashift (in vdev_alloc()) uses the nvlist
1036 * spa->spa_l2cache->sav_config (populated in
1037 * spa_ld_open_aux_vdevs()).
1038 */
1039 if (!reason_spare)
1040 fnvlist_add_uint64(label, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
1041
1042 /*
1043 * Add path information to help find it during pool import
1044 */
1045 if (vd->vdev_path != NULL)
1046 fnvlist_add_string(label, ZPOOL_CONFIG_PATH, vd->vdev_path);
1047 if (vd->vdev_devid != NULL)
1048 fnvlist_add_string(label, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
1049 if (vd->vdev_physpath != NULL) {
1050 fnvlist_add_string(label, ZPOOL_CONFIG_PHYS_PATH,
1051 vd->vdev_physpath);
1052 }
1053 return (label);
1054 }
1055
1056 /*
1057 * Initialize a vdev label. We check to make sure each leaf device is not in
1058 * use, and writable. We put down an initial label which we will later
1059 * overwrite with a complete label. Note that it's important to do this
1060 * sequentially, not in parallel, so that we catch cases of multiple use of the
1061 * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
1062 * itself.
1063 */
1064 int
vdev_label_init(vdev_t * vd,uint64_t crtxg,vdev_labeltype_t reason)1065 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
1066 {
1067 spa_t *spa = vd->vdev_spa;
1068 nvlist_t *label;
1069 vdev_phys_t *vp;
1070 abd_t *vp_abd;
1071 abd_t *bootenv;
1072 uberblock_t *ub;
1073 abd_t *ub_abd;
1074 zio_t *zio;
1075 char *buf;
1076 size_t buflen;
1077 int error;
1078 uint64_t spare_guid = 0, l2cache_guid = 0;
1079 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1080 boolean_t reason_spare = (reason == VDEV_LABEL_SPARE || (reason ==
1081 VDEV_LABEL_REMOVE && vd->vdev_isspare));
1082 boolean_t reason_l2cache = (reason == VDEV_LABEL_L2CACHE || (reason ==
1083 VDEV_LABEL_REMOVE && vd->vdev_isl2cache));
1084
1085 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1086
1087 for (int c = 0; c < vd->vdev_children; c++)
1088 if ((error = vdev_label_init(vd->vdev_child[c],
1089 crtxg, reason)) != 0)
1090 return (error);
1091
1092 /* Track the creation time for this vdev */
1093 vd->vdev_crtxg = crtxg;
1094
1095 if (!vd->vdev_ops->vdev_op_leaf || !spa_writeable(spa))
1096 return (0);
1097
1098 /*
1099 * Dead vdevs cannot be initialized.
1100 */
1101 if (vdev_is_dead(vd))
1102 return (SET_ERROR(EIO));
1103
1104 /*
1105 * Determine if the vdev is in use.
1106 */
1107 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
1108 vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
1109 return (SET_ERROR(EBUSY));
1110
1111 /*
1112 * If this is a request to add or replace a spare or l2cache device
1113 * that is in use elsewhere on the system, then we must update the
1114 * guid (which was initialized to a random value) to reflect the
1115 * actual GUID (which is shared between multiple pools).
1116 */
1117 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
1118 spare_guid != 0ULL) {
1119 uint64_t guid_delta = spare_guid - vd->vdev_guid;
1120
1121 vd->vdev_guid += guid_delta;
1122
1123 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1124 pvd->vdev_guid_sum += guid_delta;
1125
1126 /*
1127 * If this is a replacement, then we want to fallthrough to the
1128 * rest of the code. If we're adding a spare, then it's already
1129 * labeled appropriately and we can just return.
1130 */
1131 if (reason == VDEV_LABEL_SPARE)
1132 return (0);
1133 ASSERT(reason == VDEV_LABEL_REPLACE ||
1134 reason == VDEV_LABEL_SPLIT);
1135 }
1136
1137 if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
1138 l2cache_guid != 0ULL) {
1139 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
1140
1141 vd->vdev_guid += guid_delta;
1142
1143 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
1144 pvd->vdev_guid_sum += guid_delta;
1145
1146 /*
1147 * If this is a replacement, then we want to fallthrough to the
1148 * rest of the code. If we're adding an l2cache, then it's
1149 * already labeled appropriately and we can just return.
1150 */
1151 if (reason == VDEV_LABEL_L2CACHE)
1152 return (0);
1153 ASSERT(reason == VDEV_LABEL_REPLACE);
1154 }
1155
1156 /*
1157 * Initialize its label.
1158 */
1159 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1160 abd_zero(vp_abd, sizeof (vdev_phys_t));
1161 vp = abd_to_buf(vp_abd);
1162
1163 /*
1164 * Generate a label describing the pool and our top-level vdev.
1165 * We mark it as being from txg 0 to indicate that it's not
1166 * really part of an active pool just yet. The labels will
1167 * be written again with a meaningful txg by spa_sync().
1168 */
1169 if (reason_spare || reason_l2cache) {
1170 label = vdev_aux_label_generate(vd, reason_spare);
1171
1172 /*
1173 * When spare or l2cache (aux) vdev is added during pool
1174 * creation, spa->spa_uberblock is not written until this
1175 * point. Write it on next config sync.
1176 */
1177 if (uberblock_verify(&spa->spa_uberblock))
1178 spa->spa_aux_sync_uber = B_TRUE;
1179 } else {
1180 uint64_t txg = 0ULL;
1181
1182 if (reason == VDEV_LABEL_SPLIT)
1183 txg = spa->spa_uberblock.ub_txg;
1184 label = spa_config_generate(spa, vd, txg, B_FALSE);
1185
1186 /*
1187 * Add our creation time. This allows us to detect multiple
1188 * vdev uses as described above, and automatically expires if we
1189 * fail.
1190 */
1191 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
1192 crtxg) == 0);
1193 }
1194
1195 buf = vp->vp_nvlist;
1196 buflen = sizeof (vp->vp_nvlist);
1197
1198 error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
1199 if (error != 0) {
1200 nvlist_free(label);
1201 abd_free(vp_abd);
1202 /* EFAULT means nvlist_pack ran out of room */
1203 return (SET_ERROR(error == EFAULT ? ENAMETOOLONG : EINVAL));
1204 }
1205
1206 /*
1207 * Initialize uberblock template.
1208 */
1209 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_RING, B_TRUE);
1210 abd_copy_from_buf(ub_abd, &spa->spa_uberblock, sizeof (uberblock_t));
1211 abd_zero_off(ub_abd, sizeof (uberblock_t),
1212 VDEV_UBERBLOCK_RING - sizeof (uberblock_t));
1213 ub = abd_to_buf(ub_abd);
1214 ub->ub_txg = 0;
1215
1216 /* Initialize the 2nd padding area. */
1217 bootenv = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1218 abd_zero(bootenv, VDEV_PAD_SIZE);
1219
1220 /*
1221 * Write everything in parallel.
1222 */
1223 retry:
1224 zio = zio_root(spa, NULL, NULL, flags);
1225
1226 for (int l = 0; l < VDEV_LABELS; l++) {
1227
1228 vdev_label_write(zio, vd, l, vp_abd,
1229 offsetof(vdev_label_t, vl_vdev_phys),
1230 sizeof (vdev_phys_t), NULL, NULL, flags);
1231
1232 /*
1233 * Skip the 1st padding area.
1234 * Zero out the 2nd padding area where it might have
1235 * left over data from previous filesystem format.
1236 */
1237 vdev_label_write(zio, vd, l, bootenv,
1238 offsetof(vdev_label_t, vl_be),
1239 VDEV_PAD_SIZE, NULL, NULL, flags);
1240
1241 vdev_label_write(zio, vd, l, ub_abd,
1242 offsetof(vdev_label_t, vl_uberblock),
1243 VDEV_UBERBLOCK_RING, NULL, NULL, flags);
1244 }
1245
1246 error = zio_wait(zio);
1247
1248 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1249 flags |= ZIO_FLAG_TRYHARD;
1250 goto retry;
1251 }
1252
1253 nvlist_free(label);
1254 abd_free(bootenv);
1255 abd_free(ub_abd);
1256 abd_free(vp_abd);
1257
1258 /*
1259 * If this vdev hasn't been previously identified as a spare, then we
1260 * mark it as such only if a) we are labeling it as a spare, or b) it
1261 * exists as a spare elsewhere in the system. Do the same for
1262 * level 2 ARC devices.
1263 */
1264 if (error == 0 && !vd->vdev_isspare &&
1265 (reason == VDEV_LABEL_SPARE ||
1266 spa_spare_exists(vd->vdev_guid, NULL, NULL)))
1267 spa_spare_add(vd);
1268
1269 if (error == 0 && !vd->vdev_isl2cache &&
1270 (reason == VDEV_LABEL_L2CACHE ||
1271 spa_l2cache_exists(vd->vdev_guid, NULL)))
1272 spa_l2cache_add(vd);
1273
1274 return (error);
1275 }
1276
1277 /*
1278 * Done callback for vdev_label_read_bootenv_impl. If this is the first
1279 * callback to finish, store our abd in the callback pointer. Otherwise, we
1280 * just free our abd and return.
1281 */
1282 static void
vdev_label_read_bootenv_done(zio_t * zio)1283 vdev_label_read_bootenv_done(zio_t *zio)
1284 {
1285 zio_t *rio = zio->io_private;
1286 abd_t **cbp = rio->io_private;
1287
1288 ASSERT3U(zio->io_size, ==, VDEV_PAD_SIZE);
1289
1290 if (zio->io_error == 0) {
1291 mutex_enter(&rio->io_lock);
1292 if (*cbp == NULL) {
1293 /* Will free this buffer in vdev_label_read_bootenv. */
1294 *cbp = zio->io_abd;
1295 } else {
1296 abd_free(zio->io_abd);
1297 }
1298 mutex_exit(&rio->io_lock);
1299 } else {
1300 abd_free(zio->io_abd);
1301 }
1302 }
1303
1304 static void
vdev_label_read_bootenv_impl(zio_t * zio,vdev_t * vd,int flags)1305 vdev_label_read_bootenv_impl(zio_t *zio, vdev_t *vd, int flags)
1306 {
1307 for (int c = 0; c < vd->vdev_children; c++)
1308 vdev_label_read_bootenv_impl(zio, vd->vdev_child[c], flags);
1309
1310 /*
1311 * We just use the first label that has a correct checksum; the
1312 * bootloader should have rewritten them all to be the same on boot,
1313 * and any changes we made since boot have been the same across all
1314 * labels.
1315 */
1316 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1317 for (int l = 0; l < VDEV_LABELS; l++) {
1318 vdev_label_read(zio, vd, l,
1319 abd_alloc_linear(VDEV_PAD_SIZE, B_FALSE),
1320 offsetof(vdev_label_t, vl_be), VDEV_PAD_SIZE,
1321 vdev_label_read_bootenv_done, zio, flags);
1322 }
1323 }
1324 }
1325
1326 int
vdev_label_read_bootenv(vdev_t * rvd,nvlist_t * bootenv)1327 vdev_label_read_bootenv(vdev_t *rvd, nvlist_t *bootenv)
1328 {
1329 nvlist_t *config;
1330 spa_t *spa = rvd->vdev_spa;
1331 abd_t *abd = NULL;
1332 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1333 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1334
1335 ASSERT(bootenv);
1336 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1337
1338 zio_t *zio = zio_root(spa, NULL, &abd, flags);
1339 vdev_label_read_bootenv_impl(zio, rvd, flags);
1340 int err = zio_wait(zio);
1341
1342 if (abd != NULL) {
1343 char *buf;
1344 vdev_boot_envblock_t *vbe = abd_to_buf(abd);
1345
1346 vbe->vbe_version = ntohll(vbe->vbe_version);
1347 switch (vbe->vbe_version) {
1348 case VB_RAW:
1349 /*
1350 * if we have textual data in vbe_bootenv, create nvlist
1351 * with key "envmap".
1352 */
1353 fnvlist_add_uint64(bootenv, BOOTENV_VERSION, VB_RAW);
1354 vbe->vbe_bootenv[sizeof (vbe->vbe_bootenv) - 1] = '\0';
1355 fnvlist_add_string(bootenv, GRUB_ENVMAP,
1356 vbe->vbe_bootenv);
1357 break;
1358
1359 case VB_NVLIST:
1360 err = nvlist_unpack(vbe->vbe_bootenv,
1361 sizeof (vbe->vbe_bootenv), &config, 0);
1362 if (err == 0) {
1363 fnvlist_merge(bootenv, config);
1364 nvlist_free(config);
1365 break;
1366 }
1367 zfs_fallthrough;
1368 default:
1369 /* Check for FreeBSD zfs bootonce command string */
1370 buf = abd_to_buf(abd);
1371 if (*buf == '\0') {
1372 fnvlist_add_uint64(bootenv, BOOTENV_VERSION,
1373 VB_NVLIST);
1374 break;
1375 }
1376 fnvlist_add_string(bootenv, FREEBSD_BOOTONCE, buf);
1377 }
1378
1379 /*
1380 * abd was allocated in vdev_label_read_bootenv_impl()
1381 */
1382 abd_free(abd);
1383 /*
1384 * If we managed to read any successfully,
1385 * return success.
1386 */
1387 return (0);
1388 }
1389 return (err);
1390 }
1391
1392 int
vdev_label_write_bootenv(vdev_t * vd,nvlist_t * env)1393 vdev_label_write_bootenv(vdev_t *vd, nvlist_t *env)
1394 {
1395 zio_t *zio;
1396 spa_t *spa = vd->vdev_spa;
1397 vdev_boot_envblock_t *bootenv;
1398 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1399 int error;
1400 size_t nvsize;
1401 char *nvbuf;
1402 const char *tmp;
1403
1404 error = nvlist_size(env, &nvsize, NV_ENCODE_XDR);
1405 if (error != 0)
1406 return (SET_ERROR(error));
1407
1408 if (nvsize >= sizeof (bootenv->vbe_bootenv)) {
1409 return (SET_ERROR(E2BIG));
1410 }
1411
1412 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1413
1414 error = ENXIO;
1415 for (int c = 0; c < vd->vdev_children; c++) {
1416 int child_err;
1417
1418 child_err = vdev_label_write_bootenv(vd->vdev_child[c], env);
1419 /*
1420 * As long as any of the disks managed to write all of their
1421 * labels successfully, return success.
1422 */
1423 if (child_err == 0)
1424 error = child_err;
1425 }
1426
1427 if (!vd->vdev_ops->vdev_op_leaf || vdev_is_dead(vd) ||
1428 !vdev_writeable(vd)) {
1429 return (error);
1430 }
1431 ASSERT3U(sizeof (*bootenv), ==, VDEV_PAD_SIZE);
1432 abd_t *abd = abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE);
1433 abd_zero(abd, VDEV_PAD_SIZE);
1434
1435 bootenv = abd_borrow_buf_copy(abd, VDEV_PAD_SIZE);
1436 nvbuf = bootenv->vbe_bootenv;
1437 nvsize = sizeof (bootenv->vbe_bootenv);
1438
1439 bootenv->vbe_version = fnvlist_lookup_uint64(env, BOOTENV_VERSION);
1440 switch (bootenv->vbe_version) {
1441 case VB_RAW:
1442 if (nvlist_lookup_string(env, GRUB_ENVMAP, &tmp) == 0) {
1443 (void) strlcpy(bootenv->vbe_bootenv, tmp, nvsize);
1444 }
1445 error = 0;
1446 break;
1447
1448 case VB_NVLIST:
1449 error = nvlist_pack(env, &nvbuf, &nvsize, NV_ENCODE_XDR,
1450 KM_SLEEP);
1451 break;
1452
1453 default:
1454 error = EINVAL;
1455 break;
1456 }
1457
1458 if (error == 0) {
1459 bootenv->vbe_version = htonll(bootenv->vbe_version);
1460 abd_return_buf_copy(abd, bootenv, VDEV_PAD_SIZE);
1461 } else {
1462 abd_free(abd);
1463 return (SET_ERROR(error));
1464 }
1465
1466 retry:
1467 zio = zio_root(spa, NULL, NULL, flags);
1468 for (int l = 0; l < VDEV_LABELS; l++) {
1469 vdev_label_write(zio, vd, l, abd,
1470 offsetof(vdev_label_t, vl_be),
1471 VDEV_PAD_SIZE, NULL, NULL, flags);
1472 }
1473
1474 error = zio_wait(zio);
1475 if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
1476 flags |= ZIO_FLAG_TRYHARD;
1477 goto retry;
1478 }
1479
1480 abd_free(abd);
1481 return (error);
1482 }
1483
1484 /*
1485 * ==========================================================================
1486 * uberblock load/sync
1487 * ==========================================================================
1488 */
1489
1490 /*
1491 * Consider the following situation: txg is safely synced to disk. We've
1492 * written the first uberblock for txg + 1, and then we lose power. When we
1493 * come back up, we fail to see the uberblock for txg + 1 because, say,
1494 * it was on a mirrored device and the replica to which we wrote txg + 1
1495 * is now offline. If we then make some changes and sync txg + 1, and then
1496 * the missing replica comes back, then for a few seconds we'll have two
1497 * conflicting uberblocks on disk with the same txg. The solution is simple:
1498 * among uberblocks with equal txg, choose the one with the latest timestamp.
1499 */
1500 static int
vdev_uberblock_compare(const uberblock_t * ub1,const uberblock_t * ub2)1501 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1502 {
1503 int cmp = TREE_CMP(ub1->ub_txg, ub2->ub_txg);
1504
1505 if (likely(cmp))
1506 return (cmp);
1507
1508 cmp = TREE_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1509 if (likely(cmp))
1510 return (cmp);
1511
1512 /*
1513 * If MMP_VALID(ub) && MMP_SEQ_VALID(ub) then the host has an MMP-aware
1514 * ZFS, e.g. OpenZFS >= 0.7.
1515 *
1516 * If one ub has MMP and the other does not, they were written by
1517 * different hosts, which matters for MMP. So we treat no MMP/no SEQ as
1518 * a 0 value.
1519 *
1520 * Since timestamp and txg are the same if we get this far, either is
1521 * acceptable for importing the pool.
1522 */
1523 unsigned int seq1 = 0;
1524 unsigned int seq2 = 0;
1525
1526 if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1527 seq1 = MMP_SEQ(ub1);
1528
1529 if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1530 seq2 = MMP_SEQ(ub2);
1531
1532 return (TREE_CMP(seq1, seq2));
1533 }
1534
1535 struct ubl_cbdata {
1536 uberblock_t ubl_latest; /* Most recent uberblock */
1537 uberblock_t *ubl_ubbest; /* Best uberblock (w/r/t max_txg) */
1538 vdev_t *ubl_vd; /* vdev associated with the above */
1539 };
1540
1541 static void
vdev_uberblock_load_done(zio_t * zio)1542 vdev_uberblock_load_done(zio_t *zio)
1543 {
1544 vdev_t *vd = zio->io_vd;
1545 spa_t *spa = zio->io_spa;
1546 zio_t *rio = zio->io_private;
1547 uberblock_t *ub = abd_to_buf(zio->io_abd);
1548 struct ubl_cbdata *cbp = rio->io_private;
1549
1550 ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
1551
1552 if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
1553 mutex_enter(&rio->io_lock);
1554 if (vdev_uberblock_compare(ub, &cbp->ubl_latest) > 0) {
1555 cbp->ubl_latest = *ub;
1556 }
1557 if (ub->ub_txg <= spa->spa_load_max_txg &&
1558 vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
1559 /*
1560 * Keep track of the vdev in which this uberblock
1561 * was found. We will use this information later
1562 * to obtain the config nvlist associated with
1563 * this uberblock.
1564 */
1565 *cbp->ubl_ubbest = *ub;
1566 cbp->ubl_vd = vd;
1567 }
1568 mutex_exit(&rio->io_lock);
1569 }
1570
1571 abd_free(zio->io_abd);
1572 }
1573
1574 static void
vdev_uberblock_load_impl(zio_t * zio,vdev_t * vd,int flags,struct ubl_cbdata * cbp)1575 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
1576 struct ubl_cbdata *cbp)
1577 {
1578 for (int c = 0; c < vd->vdev_children; c++)
1579 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
1580
1581 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd) &&
1582 vd->vdev_ops != &vdev_draid_spare_ops) {
1583 for (int l = 0; l < VDEV_LABELS; l++) {
1584 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1585 vdev_label_read(zio, vd, l,
1586 abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd),
1587 B_TRUE), VDEV_UBERBLOCK_OFFSET(vd, n),
1588 VDEV_UBERBLOCK_SIZE(vd),
1589 vdev_uberblock_load_done, zio, flags);
1590 }
1591 }
1592 }
1593 }
1594
1595 /*
1596 * Reads the 'best' uberblock from disk along with its associated
1597 * configuration. First, we read the uberblock array of each label of each
1598 * vdev, keeping track of the uberblock with the highest txg in each array.
1599 * Then, we read the configuration from the same vdev as the best uberblock.
1600 */
1601 void
vdev_uberblock_load(vdev_t * rvd,uberblock_t * ub,nvlist_t ** config)1602 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
1603 {
1604 zio_t *zio;
1605 spa_t *spa = rvd->vdev_spa;
1606 struct ubl_cbdata cb;
1607 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1608 ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
1609
1610 ASSERT(ub);
1611 ASSERT(config);
1612
1613 memset(ub, 0, sizeof (uberblock_t));
1614 memset(&cb, 0, sizeof (cb));
1615 *config = NULL;
1616
1617 cb.ubl_ubbest = ub;
1618
1619 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1620 zio = zio_root(spa, NULL, &cb, flags);
1621 vdev_uberblock_load_impl(zio, rvd, flags, &cb);
1622 (void) zio_wait(zio);
1623
1624 /*
1625 * It's possible that the best uberblock was discovered on a label
1626 * that has a configuration which was written in a future txg.
1627 * Search all labels on this vdev to find the configuration that
1628 * matches the txg for our uberblock.
1629 */
1630 if (cb.ubl_vd != NULL) {
1631 vdev_dbgmsg(cb.ubl_vd, "best uberblock found for spa %s. "
1632 "txg %llu", spa->spa_name, (u_longlong_t)ub->ub_txg);
1633
1634 if (ub->ub_raidz_reflow_info !=
1635 cb.ubl_latest.ub_raidz_reflow_info) {
1636 vdev_dbgmsg(cb.ubl_vd,
1637 "spa=%s best uberblock (txg=%llu info=0x%llx) "
1638 "has different raidz_reflow_info than latest "
1639 "uberblock (txg=%llu info=0x%llx)",
1640 spa->spa_name,
1641 (u_longlong_t)ub->ub_txg,
1642 (u_longlong_t)ub->ub_raidz_reflow_info,
1643 (u_longlong_t)cb.ubl_latest.ub_txg,
1644 (u_longlong_t)cb.ubl_latest.ub_raidz_reflow_info);
1645 memset(ub, 0, sizeof (uberblock_t));
1646 spa_config_exit(spa, SCL_ALL, FTAG);
1647 return;
1648 }
1649
1650 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
1651 if (*config == NULL && spa->spa_extreme_rewind) {
1652 vdev_dbgmsg(cb.ubl_vd, "failed to read label config. "
1653 "Trying again without txg restrictions.");
1654 *config = vdev_label_read_config(cb.ubl_vd, UINT64_MAX);
1655 }
1656 if (*config == NULL) {
1657 vdev_dbgmsg(cb.ubl_vd, "failed to read label config");
1658 }
1659 }
1660 spa_config_exit(spa, SCL_ALL, FTAG);
1661 }
1662
1663 /*
1664 * For use when a leaf vdev is expanded.
1665 * The location of labels 2 and 3 changed, and at the new location the
1666 * uberblock rings are either empty or contain garbage. The sync will write
1667 * new configs there because the vdev is dirty, but expansion also needs the
1668 * uberblock rings copied. Read them from label 0 which did not move.
1669 *
1670 * Since the point is to populate labels {2,3} with valid uberblocks,
1671 * we zero uberblocks we fail to read or which are not valid.
1672 */
1673
1674 static void
vdev_copy_uberblocks(vdev_t * vd)1675 vdev_copy_uberblocks(vdev_t *vd)
1676 {
1677 abd_t *ub_abd;
1678 zio_t *write_zio;
1679 int locks = (SCL_L2ARC | SCL_ZIO);
1680 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
1681 ZIO_FLAG_SPECULATIVE;
1682
1683 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_READER) ==
1684 SCL_STATE);
1685 ASSERT(vd->vdev_ops->vdev_op_leaf);
1686
1687 /*
1688 * No uberblocks are stored on distributed spares, they may be
1689 * safely skipped when expanding a leaf vdev.
1690 */
1691 if (vd->vdev_ops == &vdev_draid_spare_ops)
1692 return;
1693
1694 spa_config_enter(vd->vdev_spa, locks, FTAG, RW_READER);
1695
1696 ub_abd = abd_alloc_linear(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1697
1698 write_zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1699 for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1700 const int src_label = 0;
1701 zio_t *zio;
1702
1703 zio = zio_root(vd->vdev_spa, NULL, NULL, flags);
1704 vdev_label_read(zio, vd, src_label, ub_abd,
1705 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1706 NULL, NULL, flags);
1707
1708 if (zio_wait(zio) || uberblock_verify(abd_to_buf(ub_abd)))
1709 abd_zero(ub_abd, VDEV_UBERBLOCK_SIZE(vd));
1710
1711 for (int l = 2; l < VDEV_LABELS; l++)
1712 vdev_label_write(write_zio, vd, l, ub_abd,
1713 VDEV_UBERBLOCK_OFFSET(vd, n),
1714 VDEV_UBERBLOCK_SIZE(vd), NULL, NULL,
1715 flags | ZIO_FLAG_DONT_PROPAGATE);
1716 }
1717 (void) zio_wait(write_zio);
1718
1719 spa_config_exit(vd->vdev_spa, locks, FTAG);
1720
1721 abd_free(ub_abd);
1722 }
1723
1724 /*
1725 * On success, increment root zio's count of good writes.
1726 * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
1727 */
1728 static void
vdev_uberblock_sync_done(zio_t * zio)1729 vdev_uberblock_sync_done(zio_t *zio)
1730 {
1731 uint64_t *good_writes = zio->io_private;
1732
1733 if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1734 atomic_inc_64(good_writes);
1735 }
1736
1737 /*
1738 * Write the uberblock to all labels of all leaves of the specified vdev.
1739 */
1740 static void
vdev_uberblock_sync(zio_t * zio,uint64_t * good_writes,uberblock_t * ub,vdev_t * vd,int flags)1741 vdev_uberblock_sync(zio_t *zio, uint64_t *good_writes,
1742 uberblock_t *ub, vdev_t *vd, int flags)
1743 {
1744 for (uint64_t c = 0; c < vd->vdev_children; c++) {
1745 vdev_uberblock_sync(zio, good_writes,
1746 ub, vd->vdev_child[c], flags);
1747 }
1748
1749 if (!vd->vdev_ops->vdev_op_leaf)
1750 return;
1751
1752 if (!vdev_writeable(vd))
1753 return;
1754
1755 /*
1756 * There's no need to write uberblocks to a distributed spare, they
1757 * are already stored on all the leaves of the parent dRAID. For
1758 * this same reason vdev_uberblock_load_impl() skips distributed
1759 * spares when reading uberblocks.
1760 */
1761 if (vd->vdev_ops == &vdev_draid_spare_ops)
1762 return;
1763
1764 /* If the vdev was expanded, need to copy uberblock rings. */
1765 if (vd->vdev_state == VDEV_STATE_HEALTHY &&
1766 vd->vdev_copy_uberblocks == B_TRUE) {
1767 vdev_copy_uberblocks(vd);
1768 vd->vdev_copy_uberblocks = B_FALSE;
1769 }
1770
1771 /*
1772 * We chose a slot based on the txg. If this uberblock has a special
1773 * RAIDZ expansion state, then it is essentially an update of the
1774 * current uberblock (it has the same txg). However, the current
1775 * state is committed, so we want to write it to a different slot. If
1776 * we overwrote the same slot, and we lose power during the uberblock
1777 * write, and the disk does not do single-sector overwrites
1778 * atomically (even though it is required to - i.e. we should see
1779 * either the old or the new uberblock), then we could lose this
1780 * txg's uberblock. Rewinding to the previous txg's uberblock may not
1781 * be possible because RAIDZ expansion may have already overwritten
1782 * some of the data, so we need the progress indicator in the
1783 * uberblock.
1784 */
1785 int m = spa_multihost(vd->vdev_spa) ? MMP_BLOCKS_PER_LABEL : 0;
1786 int n = (ub->ub_txg - (RRSS_GET_STATE(ub) == RRSS_SCRATCH_VALID)) %
1787 (VDEV_UBERBLOCK_COUNT(vd) - m);
1788
1789 /* Copy the uberblock_t into the ABD */
1790 abd_t *ub_abd = abd_alloc_for_io(VDEV_UBERBLOCK_SIZE(vd), B_TRUE);
1791 abd_copy_from_buf(ub_abd, ub, sizeof (uberblock_t));
1792 abd_zero_off(ub_abd, sizeof (uberblock_t),
1793 VDEV_UBERBLOCK_SIZE(vd) - sizeof (uberblock_t));
1794
1795 for (int l = 0; l < VDEV_LABELS; l++)
1796 vdev_label_write(zio, vd, l, ub_abd,
1797 VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1798 vdev_uberblock_sync_done, good_writes,
1799 flags | ZIO_FLAG_DONT_PROPAGATE);
1800
1801 abd_free(ub_abd);
1802 }
1803
1804 /* Sync the uberblocks to all vdevs in svd[] */
1805 int
vdev_uberblock_sync_list(vdev_t ** svd,int svdcount,uberblock_t * ub,int flags)1806 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1807 {
1808 spa_t *spa = svd[0]->vdev_spa;
1809 zio_t *zio;
1810 uint64_t good_writes = 0;
1811
1812 zio = zio_root(spa, NULL, NULL, flags);
1813
1814 for (int v = 0; v < svdcount; v++)
1815 vdev_uberblock_sync(zio, &good_writes, ub, svd[v], flags);
1816
1817 if (spa->spa_aux_sync_uber) {
1818 for (int v = 0; v < spa->spa_spares.sav_count; v++) {
1819 vdev_uberblock_sync(zio, &good_writes, ub,
1820 spa->spa_spares.sav_vdevs[v], flags);
1821 }
1822 for (int v = 0; v < spa->spa_l2cache.sav_count; v++) {
1823 vdev_uberblock_sync(zio, &good_writes, ub,
1824 spa->spa_l2cache.sav_vdevs[v], flags);
1825 }
1826 }
1827 (void) zio_wait(zio);
1828
1829 /*
1830 * Flush the uberblocks to disk. This ensures that the odd labels
1831 * are no longer needed (because the new uberblocks and the even
1832 * labels are safely on disk), so it is safe to overwrite them.
1833 */
1834 zio = zio_root(spa, NULL, NULL, flags);
1835
1836 for (int v = 0; v < svdcount; v++) {
1837 if (vdev_writeable(svd[v])) {
1838 zio_flush(zio, svd[v]);
1839 }
1840 }
1841 if (spa->spa_aux_sync_uber) {
1842 spa->spa_aux_sync_uber = B_FALSE;
1843 for (int v = 0; v < spa->spa_spares.sav_count; v++) {
1844 if (vdev_writeable(spa->spa_spares.sav_vdevs[v])) {
1845 zio_flush(zio, spa->spa_spares.sav_vdevs[v]);
1846 }
1847 }
1848 for (int v = 0; v < spa->spa_l2cache.sav_count; v++) {
1849 if (vdev_writeable(spa->spa_l2cache.sav_vdevs[v])) {
1850 zio_flush(zio, spa->spa_l2cache.sav_vdevs[v]);
1851 }
1852 }
1853 }
1854
1855 (void) zio_wait(zio);
1856
1857 return (good_writes >= 1 ? 0 : EIO);
1858 }
1859
1860 /*
1861 * On success, increment the count of good writes for our top-level vdev.
1862 */
1863 static void
vdev_label_sync_done(zio_t * zio)1864 vdev_label_sync_done(zio_t *zio)
1865 {
1866 uint64_t *good_writes = zio->io_private;
1867
1868 if (zio->io_error == 0)
1869 atomic_inc_64(good_writes);
1870 }
1871
1872 /*
1873 * If there weren't enough good writes, indicate failure to the parent.
1874 */
1875 static void
vdev_label_sync_top_done(zio_t * zio)1876 vdev_label_sync_top_done(zio_t *zio)
1877 {
1878 uint64_t *good_writes = zio->io_private;
1879
1880 if (*good_writes == 0)
1881 zio->io_error = SET_ERROR(EIO);
1882
1883 kmem_free(good_writes, sizeof (uint64_t));
1884 }
1885
1886 /*
1887 * We ignore errors for log and cache devices, simply free the private data.
1888 */
1889 static void
vdev_label_sync_ignore_done(zio_t * zio)1890 vdev_label_sync_ignore_done(zio_t *zio)
1891 {
1892 kmem_free(zio->io_private, sizeof (uint64_t));
1893 }
1894
1895 /*
1896 * Write all even or odd labels to all leaves of the specified vdev.
1897 */
1898 static void
vdev_label_sync(zio_t * zio,uint64_t * good_writes,vdev_t * vd,int l,uint64_t txg,int flags)1899 vdev_label_sync(zio_t *zio, uint64_t *good_writes,
1900 vdev_t *vd, int l, uint64_t txg, int flags)
1901 {
1902 nvlist_t *label;
1903 vdev_phys_t *vp;
1904 abd_t *vp_abd;
1905 char *buf;
1906 size_t buflen;
1907 vdev_t *pvd = vd->vdev_parent;
1908 boolean_t spare_in_use = B_FALSE;
1909
1910 for (int c = 0; c < vd->vdev_children; c++) {
1911 vdev_label_sync(zio, good_writes,
1912 vd->vdev_child[c], l, txg, flags);
1913 }
1914
1915 if (!vd->vdev_ops->vdev_op_leaf)
1916 return;
1917
1918 if (!vdev_writeable(vd))
1919 return;
1920
1921 /*
1922 * The top-level config never needs to be written to a distributed
1923 * spare. When read vdev_dspare_label_read_config() will generate
1924 * the config for the vdev_label_read_config().
1925 */
1926 if (vd->vdev_ops == &vdev_draid_spare_ops)
1927 return;
1928
1929 if (pvd && pvd->vdev_ops == &vdev_spare_ops)
1930 spare_in_use = B_TRUE;
1931
1932 /*
1933 * Generate a label describing the top-level config to which we belong.
1934 */
1935 if ((vd->vdev_isspare && !spare_in_use) || vd->vdev_isl2cache) {
1936 label = vdev_aux_label_generate(vd, vd->vdev_isspare);
1937 } else {
1938 label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1939 }
1940
1941 vp_abd = abd_alloc_linear(sizeof (vdev_phys_t), B_TRUE);
1942 abd_zero(vp_abd, sizeof (vdev_phys_t));
1943 vp = abd_to_buf(vp_abd);
1944
1945 buf = vp->vp_nvlist;
1946 buflen = sizeof (vp->vp_nvlist);
1947
1948 if (!nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP)) {
1949 for (; l < VDEV_LABELS; l += 2) {
1950 vdev_label_write(zio, vd, l, vp_abd,
1951 offsetof(vdev_label_t, vl_vdev_phys),
1952 sizeof (vdev_phys_t),
1953 vdev_label_sync_done, good_writes,
1954 flags | ZIO_FLAG_DONT_PROPAGATE);
1955 }
1956 }
1957
1958 abd_free(vp_abd);
1959 nvlist_free(label);
1960 }
1961
1962 static int
vdev_label_sync_list(spa_t * spa,int l,uint64_t txg,int flags)1963 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1964 {
1965 list_t *dl = &spa->spa_config_dirty_list;
1966 vdev_t *vd;
1967 zio_t *zio;
1968 int error;
1969
1970 /*
1971 * Write the new labels to disk.
1972 */
1973 zio = zio_root(spa, NULL, NULL, flags);
1974
1975 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1976 uint64_t *good_writes;
1977
1978 ASSERT(!vd->vdev_ishole);
1979
1980 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1981 zio_t *vio = zio_null(zio, spa, NULL,
1982 (vd->vdev_islog || vd->vdev_aux != NULL) ?
1983 vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1984 good_writes, flags);
1985 vdev_label_sync(vio, good_writes, vd, l, txg, flags);
1986 zio_nowait(vio);
1987 }
1988
1989 /*
1990 * AUX path may have changed during import
1991 */
1992 spa_aux_vdev_t *sav[2] = {&spa->spa_spares, &spa->spa_l2cache};
1993 for (int i = 0; i < 2; i++) {
1994 for (int v = 0; v < sav[i]->sav_count; v++) {
1995 uint64_t *good_writes;
1996 if (!sav[i]->sav_label_sync)
1997 continue;
1998 good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
1999 zio_t *vio = zio_null(zio, spa, NULL,
2000 vdev_label_sync_ignore_done, good_writes, flags);
2001 vdev_label_sync(vio, good_writes, sav[i]->sav_vdevs[v],
2002 l, txg, flags);
2003 zio_nowait(vio);
2004 }
2005 }
2006
2007 error = zio_wait(zio);
2008
2009 /*
2010 * Flush the new labels to disk.
2011 */
2012 zio = zio_root(spa, NULL, NULL, flags);
2013
2014 for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
2015 zio_flush(zio, vd);
2016
2017 for (int i = 0; i < 2; i++) {
2018 if (!sav[i]->sav_label_sync)
2019 continue;
2020 for (int v = 0; v < sav[i]->sav_count; v++)
2021 zio_flush(zio, sav[i]->sav_vdevs[v]);
2022 if (l == 1)
2023 sav[i]->sav_label_sync = B_FALSE;
2024 }
2025
2026 (void) zio_wait(zio);
2027
2028 return (error);
2029 }
2030
2031 /*
2032 * Sync the uberblock and any changes to the vdev configuration.
2033 *
2034 * The order of operations is carefully crafted to ensure that
2035 * if the system panics or loses power at any time, the state on disk
2036 * is still transactionally consistent. The in-line comments below
2037 * describe the failure semantics at each stage.
2038 *
2039 * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
2040 * at any time, you can just call it again, and it will resume its work.
2041 */
2042 int
vdev_config_sync(vdev_t ** svd,int svdcount,uint64_t txg)2043 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
2044 {
2045 spa_t *spa = svd[0]->vdev_spa;
2046 uberblock_t *ub = &spa->spa_uberblock;
2047 int error = 0;
2048 int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
2049
2050 ASSERT(svdcount != 0);
2051 retry:
2052 /*
2053 * Normally, we don't want to try too hard to write every label and
2054 * uberblock. If there is a flaky disk, we don't want the rest of the
2055 * sync process to block while we retry. But if we can't write a
2056 * single label out, we should retry with ZIO_FLAG_TRYHARD before
2057 * bailing out and declaring the pool faulted.
2058 */
2059 if (error != 0) {
2060 if ((flags & ZIO_FLAG_TRYHARD) != 0)
2061 return (error);
2062 flags |= ZIO_FLAG_TRYHARD;
2063 }
2064
2065 ASSERT(ub->ub_txg <= txg);
2066
2067 /*
2068 * If this isn't a resync due to I/O errors,
2069 * and nothing changed in this transaction group,
2070 * and multihost protection isn't enabled,
2071 * and the vdev configuration hasn't changed,
2072 * then there's nothing to do.
2073 */
2074 if (ub->ub_txg < txg) {
2075 boolean_t changed = uberblock_update(ub, spa->spa_root_vdev,
2076 txg, spa->spa_mmp.mmp_delay);
2077
2078 if (!changed && list_is_empty(&spa->spa_config_dirty_list) &&
2079 !spa_multihost(spa))
2080 return (0);
2081 }
2082
2083 if (txg > spa_freeze_txg(spa))
2084 return (0);
2085
2086 ASSERT(txg <= spa->spa_final_txg);
2087
2088 /*
2089 * Flush the write cache of every disk that's been written to
2090 * in this transaction group. This ensures that all blocks
2091 * written in this txg will be committed to stable storage
2092 * before any uberblock that references them.
2093 */
2094 zio_t *zio = zio_root(spa, NULL, NULL, flags);
2095
2096 for (vdev_t *vd =
2097 txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd != NULL;
2098 vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
2099 zio_flush(zio, vd);
2100
2101 (void) zio_wait(zio);
2102
2103 /*
2104 * Sync out the even labels (L0, L2) for every dirty vdev. If the
2105 * system dies in the middle of this process, that's OK: all of the
2106 * even labels that made it to disk will be newer than any uberblock,
2107 * and will therefore be considered invalid. The odd labels (L1, L3),
2108 * which have not yet been touched, will still be valid. We flush
2109 * the new labels to disk to ensure that all even-label updates
2110 * are committed to stable storage before the uberblock update.
2111 */
2112 if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0) {
2113 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2114 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2115 "for pool '%s' when syncing out the even labels "
2116 "of dirty vdevs", error, spa_name(spa));
2117 }
2118 goto retry;
2119 }
2120
2121 /*
2122 * Sync the uberblocks to all vdevs in svd[].
2123 * If the system dies in the middle of this step, there are two cases
2124 * to consider, and the on-disk state is consistent either way:
2125 *
2126 * (1) If none of the new uberblocks made it to disk, then the
2127 * previous uberblock will be the newest, and the odd labels
2128 * (which had not yet been touched) will be valid with respect
2129 * to that uberblock.
2130 *
2131 * (2) If one or more new uberblocks made it to disk, then they
2132 * will be the newest, and the even labels (which had all
2133 * been successfully committed) will be valid with respect
2134 * to the new uberblocks.
2135 */
2136 if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0) {
2137 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2138 zfs_dbgmsg("vdev_uberblock_sync_list() returned error "
2139 "%d for pool '%s'", error, spa_name(spa));
2140 }
2141 goto retry;
2142 }
2143
2144 if (spa_multihost(spa))
2145 mmp_update_uberblock(spa, ub);
2146
2147 /*
2148 * Sync out odd labels for every dirty vdev. If the system dies
2149 * in the middle of this process, the even labels and the new
2150 * uberblocks will suffice to open the pool. The next time
2151 * the pool is opened, the first thing we'll do -- before any
2152 * user data is modified -- is mark every vdev dirty so that
2153 * all labels will be brought up to date. We flush the new labels
2154 * to disk to ensure that all odd-label updates are committed to
2155 * stable storage before the next transaction group begins.
2156 */
2157 if ((error = vdev_label_sync_list(spa, 1, txg, flags)) != 0) {
2158 if ((flags & ZIO_FLAG_TRYHARD) != 0) {
2159 zfs_dbgmsg("vdev_label_sync_list() returned error %d "
2160 "for pool '%s' when syncing out the odd labels of "
2161 "dirty vdevs", error, spa_name(spa));
2162 }
2163 goto retry;
2164 }
2165
2166 return (0);
2167 }
2168