789
|
1 |
/*
|
|
2 |
* CDDL HEADER START
|
|
3 |
*
|
|
4 |
* The contents of this file are subject to the terms of the
|
1485
|
5 |
* Common Development and Distribution License (the "License").
|
|
6 |
* You may not use this file except in compliance with the License.
|
789
|
7 |
*
|
|
8 |
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
|
|
9 |
* or http://www.opensolaris.org/os/licensing.
|
|
10 |
* See the License for the specific language governing permissions
|
|
11 |
* and limitations under the License.
|
|
12 |
*
|
|
13 |
* When distributing Covered Code, include this CDDL HEADER in each
|
|
14 |
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
|
|
15 |
* If applicable, add the following below this CDDL HEADER, with the
|
|
16 |
* fields enclosed by brackets "[]" replaced with your own identifying
|
|
17 |
* information: Portions Copyright [yyyy] [name of copyright owner]
|
|
18 |
*
|
|
19 |
* CDDL HEADER END
|
|
20 |
*/
|
|
21 |
/*
|
1485
|
22 |
* Copyright 2006 Sun Microsystems, Inc. All rights reserved.
|
789
|
23 |
* Use is subject to license terms.
|
|
24 |
*/
|
|
25 |
|
|
26 |
#pragma ident "%Z%%M% %I% %E% SMI"
|
|
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 transacation 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 |
*
|
|
128 |
* Each leaf device label also contains the following:
|
|
129 |
*
|
|
130 |
* top_guid Unique ID for top-level vdev in which this is contained
|
|
131 |
* guid Unique ID for the leaf vdev
|
|
132 |
*
|
|
133 |
* The 'vs' configuration follows the format described in 'spa_config.c'.
|
|
134 |
*/
|
|
135 |
|
|
136 |
#include <sys/zfs_context.h>
|
|
137 |
#include <sys/spa.h>
|
|
138 |
#include <sys/spa_impl.h>
|
|
139 |
#include <sys/dmu.h>
|
|
140 |
#include <sys/zap.h>
|
|
141 |
#include <sys/vdev.h>
|
|
142 |
#include <sys/vdev_impl.h>
|
|
143 |
#include <sys/uberblock_impl.h>
|
|
144 |
#include <sys/metaslab.h>
|
|
145 |
#include <sys/zio.h>
|
|
146 |
#include <sys/fs/zfs.h>
|
|
147 |
|
|
148 |
/*
|
|
149 |
* Basic routines to read and write from a vdev label.
|
|
150 |
* Used throughout the rest of this file.
|
|
151 |
*/
|
|
152 |
uint64_t
|
|
153 |
vdev_label_offset(uint64_t psize, int l, uint64_t offset)
|
|
154 |
{
|
|
155 |
return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
|
|
156 |
0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
|
|
157 |
}
|
|
158 |
|
|
159 |
static void
|
|
160 |
vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
|
|
161 |
uint64_t size, zio_done_func_t *done, void *private)
|
|
162 |
{
|
|
163 |
ASSERT(vd->vdev_children == 0);
|
|
164 |
|
|
165 |
zio_nowait(zio_read_phys(zio, vd,
|
|
166 |
vdev_label_offset(vd->vdev_psize, l, offset),
|
|
167 |
size, buf, ZIO_CHECKSUM_LABEL, done, private,
|
1544
|
168 |
ZIO_PRIORITY_SYNC_READ,
|
|
169 |
ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE));
|
789
|
170 |
}
|
|
171 |
|
|
172 |
static void
|
|
173 |
vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
|
|
174 |
uint64_t size, zio_done_func_t *done, void *private)
|
|
175 |
{
|
|
176 |
ASSERT(vd->vdev_children == 0);
|
|
177 |
|
|
178 |
zio_nowait(zio_write_phys(zio, vd,
|
|
179 |
vdev_label_offset(vd->vdev_psize, l, offset),
|
|
180 |
size, buf, ZIO_CHECKSUM_LABEL, done, private,
|
1544
|
181 |
ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL));
|
789
|
182 |
}
|
|
183 |
|
|
184 |
/*
|
|
185 |
* Generate the nvlist representing this vdev's config.
|
|
186 |
*/
|
|
187 |
nvlist_t *
|
|
188 |
vdev_config_generate(vdev_t *vd, int getstats)
|
|
189 |
{
|
|
190 |
nvlist_t *nv = NULL;
|
|
191 |
|
1544
|
192 |
VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
|
789
|
193 |
|
|
194 |
VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
|
|
195 |
vd->vdev_ops->vdev_op_type) == 0);
|
|
196 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id) == 0);
|
|
197 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
|
|
198 |
|
|
199 |
if (vd->vdev_path != NULL)
|
|
200 |
VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
|
|
201 |
vd->vdev_path) == 0);
|
|
202 |
|
|
203 |
if (vd->vdev_devid != NULL)
|
|
204 |
VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
|
|
205 |
vd->vdev_devid) == 0);
|
|
206 |
|
1171
|
207 |
if (vd->vdev_wholedisk != -1ULL)
|
|
208 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
|
|
209 |
vd->vdev_wholedisk) == 0);
|
|
210 |
|
1544
|
211 |
if (vd->vdev_not_present)
|
|
212 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
|
|
213 |
|
789
|
214 |
if (vd == vd->vdev_top) {
|
|
215 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
|
|
216 |
vd->vdev_ms_array) == 0);
|
|
217 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
|
|
218 |
vd->vdev_ms_shift) == 0);
|
|
219 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
|
|
220 |
vd->vdev_ashift) == 0);
|
|
221 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
|
|
222 |
vd->vdev_asize) == 0);
|
|
223 |
}
|
|
224 |
|
|
225 |
if (vd->vdev_dtl.smo_object != 0)
|
|
226 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
|
|
227 |
vd->vdev_dtl.smo_object) == 0);
|
|
228 |
|
|
229 |
if (getstats) {
|
|
230 |
vdev_stat_t vs;
|
|
231 |
vdev_get_stats(vd, &vs);
|
|
232 |
VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS,
|
|
233 |
(uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
|
|
234 |
}
|
|
235 |
|
|
236 |
if (!vd->vdev_ops->vdev_op_leaf) {
|
|
237 |
nvlist_t **child;
|
|
238 |
int c;
|
|
239 |
|
|
240 |
child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
|
|
241 |
KM_SLEEP);
|
|
242 |
|
|
243 |
for (c = 0; c < vd->vdev_children; c++)
|
|
244 |
child[c] = vdev_config_generate(vd->vdev_child[c],
|
|
245 |
getstats);
|
|
246 |
|
|
247 |
VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
|
|
248 |
child, vd->vdev_children) == 0);
|
|
249 |
|
|
250 |
for (c = 0; c < vd->vdev_children; c++)
|
|
251 |
nvlist_free(child[c]);
|
|
252 |
|
|
253 |
kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
|
1485
|
254 |
|
|
255 |
} else {
|
|
256 |
if (!vd->vdev_tmpoffline) {
|
|
257 |
if (vd->vdev_offline)
|
|
258 |
VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
|
|
259 |
B_TRUE) == 0);
|
|
260 |
else
|
|
261 |
(void) nvlist_remove(nv, ZPOOL_CONFIG_OFFLINE,
|
|
262 |
DATA_TYPE_UINT64);
|
|
263 |
}
|
789
|
264 |
}
|
|
265 |
|
|
266 |
return (nv);
|
|
267 |
}
|
|
268 |
|
|
269 |
nvlist_t *
|
|
270 |
vdev_label_read_config(vdev_t *vd)
|
|
271 |
{
|
|
272 |
nvlist_t *config = NULL;
|
|
273 |
vdev_phys_t *vp;
|
|
274 |
zio_t *zio;
|
|
275 |
int l;
|
|
276 |
|
|
277 |
if (vdev_is_dead(vd))
|
|
278 |
return (NULL);
|
|
279 |
|
|
280 |
vp = zio_buf_alloc(sizeof (vdev_phys_t));
|
|
281 |
|
|
282 |
for (l = 0; l < VDEV_LABELS; l++) {
|
|
283 |
|
1544
|
284 |
zio = zio_root(vd->vdev_spa, NULL, NULL, ZIO_FLAG_CANFAIL |
|
|
285 |
ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CONFIG_HELD);
|
789
|
286 |
|
|
287 |
vdev_label_read(zio, vd, l, vp,
|
|
288 |
offsetof(vdev_label_t, vl_vdev_phys),
|
|
289 |
sizeof (vdev_phys_t), NULL, NULL);
|
|
290 |
|
|
291 |
if (zio_wait(zio) == 0 &&
|
|
292 |
nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
|
1544
|
293 |
&config, 0) == 0)
|
789
|
294 |
break;
|
|
295 |
|
|
296 |
if (config != NULL) {
|
|
297 |
nvlist_free(config);
|
|
298 |
config = NULL;
|
|
299 |
}
|
|
300 |
}
|
|
301 |
|
|
302 |
zio_buf_free(vp, sizeof (vdev_phys_t));
|
|
303 |
|
|
304 |
return (config);
|
|
305 |
}
|
|
306 |
|
|
307 |
int
|
|
308 |
vdev_label_init(vdev_t *vd, uint64_t crtxg)
|
|
309 |
{
|
|
310 |
spa_t *spa = vd->vdev_spa;
|
|
311 |
nvlist_t *label;
|
|
312 |
vdev_phys_t *vp;
|
|
313 |
vdev_boot_header_t *vb;
|
|
314 |
uberblock_phys_t *ubphys;
|
|
315 |
zio_t *zio;
|
|
316 |
int l, c, n;
|
|
317 |
char *buf;
|
|
318 |
size_t buflen;
|
|
319 |
int error;
|
|
320 |
|
|
321 |
for (c = 0; c < vd->vdev_children; c++)
|
|
322 |
if ((error = vdev_label_init(vd->vdev_child[c], crtxg)) != 0)
|
|
323 |
return (error);
|
|
324 |
|
|
325 |
if (!vd->vdev_ops->vdev_op_leaf)
|
|
326 |
return (0);
|
|
327 |
|
|
328 |
/*
|
|
329 |
* Make sure each leaf device is writable, and zero its initial content.
|
|
330 |
* Along the way, also make sure that no leaf is already in use.
|
|
331 |
* Note that it's important to do this sequentially, not in parallel,
|
|
332 |
* so that we catch cases of multiple use of the same leaf vdev in
|
|
333 |
* the vdev we're creating -- e.g. mirroring a disk with itself.
|
|
334 |
*/
|
|
335 |
if (vdev_is_dead(vd))
|
|
336 |
return (EIO);
|
|
337 |
|
|
338 |
/*
|
|
339 |
* Check whether this device is already in use.
|
|
340 |
* Ignore the check if crtxg == 0, which we use for device removal.
|
|
341 |
*/
|
1544
|
342 |
if (crtxg != 0 &&
|
|
343 |
(label = vdev_label_read_config(vd)) != NULL) {
|
|
344 |
uint64_t state, pool_guid, device_guid, txg;
|
789
|
345 |
uint64_t mycrtxg = 0;
|
|
346 |
|
|
347 |
(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
|
|
348 |
&mycrtxg);
|
|
349 |
|
1544
|
350 |
if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
|
789
|
351 |
&state) == 0 && state == POOL_STATE_ACTIVE &&
|
|
352 |
nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
|
|
353 |
&pool_guid) == 0 &&
|
|
354 |
nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
|
|
355 |
&device_guid) == 0 &&
|
|
356 |
spa_guid_exists(pool_guid, device_guid) &&
|
|
357 |
nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
|
|
358 |
&txg) == 0 && (txg != 0 || mycrtxg == crtxg)) {
|
|
359 |
dprintf("vdev %s in use, pool_state %d\n",
|
|
360 |
vdev_description(vd), state);
|
|
361 |
nvlist_free(label);
|
|
362 |
return (EBUSY);
|
|
363 |
}
|
|
364 |
nvlist_free(label);
|
|
365 |
}
|
|
366 |
|
|
367 |
/*
|
|
368 |
* The device isn't in use, so initialize its label.
|
|
369 |
*/
|
|
370 |
vp = zio_buf_alloc(sizeof (vdev_phys_t));
|
|
371 |
bzero(vp, sizeof (vdev_phys_t));
|
|
372 |
|
|
373 |
/*
|
|
374 |
* Generate a label describing the pool and our top-level vdev.
|
|
375 |
* We mark it as being from txg 0 to indicate that it's not
|
|
376 |
* really part of an active pool just yet. The labels will
|
|
377 |
* be written again with a meaningful txg by spa_sync().
|
|
378 |
*/
|
|
379 |
label = spa_config_generate(spa, vd, 0ULL, 0);
|
|
380 |
|
|
381 |
/*
|
|
382 |
* Add our creation time. This allows us to detect multiple vdev
|
|
383 |
* uses as described above, and automatically expires if we fail.
|
|
384 |
*/
|
|
385 |
VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG, crtxg) == 0);
|
|
386 |
|
|
387 |
buf = vp->vp_nvlist;
|
|
388 |
buflen = sizeof (vp->vp_nvlist);
|
|
389 |
|
1544
|
390 |
if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) != 0) {
|
789
|
391 |
nvlist_free(label);
|
|
392 |
zio_buf_free(vp, sizeof (vdev_phys_t));
|
|
393 |
return (EINVAL);
|
|
394 |
}
|
|
395 |
|
|
396 |
/*
|
|
397 |
* Initialize boot block header.
|
|
398 |
*/
|
|
399 |
vb = zio_buf_alloc(sizeof (vdev_boot_header_t));
|
|
400 |
bzero(vb, sizeof (vdev_boot_header_t));
|
|
401 |
vb->vb_magic = VDEV_BOOT_MAGIC;
|
|
402 |
vb->vb_version = VDEV_BOOT_VERSION;
|
|
403 |
vb->vb_offset = VDEV_BOOT_OFFSET;
|
|
404 |
vb->vb_size = VDEV_BOOT_SIZE;
|
|
405 |
|
|
406 |
/*
|
|
407 |
* Initialize uberblock template.
|
|
408 |
*/
|
|
409 |
ubphys = zio_buf_alloc(sizeof (uberblock_phys_t));
|
|
410 |
bzero(ubphys, sizeof (uberblock_phys_t));
|
|
411 |
ubphys->ubp_uberblock = spa->spa_uberblock;
|
|
412 |
ubphys->ubp_uberblock.ub_txg = 0;
|
|
413 |
|
|
414 |
/*
|
|
415 |
* Write everything in parallel.
|
|
416 |
*/
|
|
417 |
zio = zio_root(spa, NULL, NULL,
|
|
418 |
ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
|
|
419 |
|
|
420 |
for (l = 0; l < VDEV_LABELS; l++) {
|
|
421 |
|
|
422 |
vdev_label_write(zio, vd, l, vp,
|
|
423 |
offsetof(vdev_label_t, vl_vdev_phys),
|
|
424 |
sizeof (vdev_phys_t), NULL, NULL);
|
|
425 |
|
|
426 |
vdev_label_write(zio, vd, l, vb,
|
|
427 |
offsetof(vdev_label_t, vl_boot_header),
|
|
428 |
sizeof (vdev_boot_header_t), NULL, NULL);
|
|
429 |
|
|
430 |
for (n = 0; n < VDEV_UBERBLOCKS; n++) {
|
|
431 |
|
|
432 |
vdev_label_write(zio, vd, l, ubphys,
|
|
433 |
offsetof(vdev_label_t, vl_uberblock[n]),
|
|
434 |
sizeof (uberblock_phys_t), NULL, NULL);
|
|
435 |
|
|
436 |
}
|
|
437 |
}
|
|
438 |
|
|
439 |
error = zio_wait(zio);
|
|
440 |
|
|
441 |
nvlist_free(label);
|
|
442 |
zio_buf_free(ubphys, sizeof (uberblock_phys_t));
|
|
443 |
zio_buf_free(vb, sizeof (vdev_boot_header_t));
|
|
444 |
zio_buf_free(vp, sizeof (vdev_phys_t));
|
|
445 |
|
|
446 |
return (error);
|
|
447 |
}
|
|
448 |
|
|
449 |
/*
|
|
450 |
* ==========================================================================
|
|
451 |
* uberblock load/sync
|
|
452 |
* ==========================================================================
|
|
453 |
*/
|
|
454 |
|
|
455 |
/*
|
|
456 |
* Consider the following situation: txg is safely synced to disk. We've
|
|
457 |
* written the first uberblock for txg + 1, and then we lose power. When we
|
|
458 |
* come back up, we fail to see the uberblock for txg + 1 because, say,
|
|
459 |
* it was on a mirrored device and the replica to which we wrote txg + 1
|
|
460 |
* is now offline. If we then make some changes and sync txg + 1, and then
|
|
461 |
* the missing replica comes back, then for a new seconds we'll have two
|
|
462 |
* conflicting uberblocks on disk with the same txg. The solution is simple:
|
|
463 |
* among uberblocks with equal txg, choose the one with the latest timestamp.
|
|
464 |
*/
|
|
465 |
static int
|
|
466 |
vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
|
|
467 |
{
|
|
468 |
if (ub1->ub_txg < ub2->ub_txg)
|
|
469 |
return (-1);
|
|
470 |
if (ub1->ub_txg > ub2->ub_txg)
|
|
471 |
return (1);
|
|
472 |
|
|
473 |
if (ub1->ub_timestamp < ub2->ub_timestamp)
|
|
474 |
return (-1);
|
|
475 |
if (ub1->ub_timestamp > ub2->ub_timestamp)
|
|
476 |
return (1);
|
|
477 |
|
|
478 |
return (0);
|
|
479 |
}
|
|
480 |
|
|
481 |
static void
|
|
482 |
vdev_uberblock_load_done(zio_t *zio)
|
|
483 |
{
|
|
484 |
uberblock_phys_t *ubphys = zio->io_data;
|
|
485 |
uberblock_t *ub = &ubphys->ubp_uberblock;
|
|
486 |
uberblock_t *ubbest = zio->io_private;
|
|
487 |
spa_t *spa = zio->io_spa;
|
|
488 |
|
|
489 |
ASSERT3U(zio->io_size, ==, sizeof (uberblock_phys_t));
|
|
490 |
|
1544
|
491 |
if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
|
789
|
492 |
mutex_enter(&spa->spa_uberblock_lock);
|
|
493 |
if (vdev_uberblock_compare(ub, ubbest) > 0)
|
|
494 |
*ubbest = *ub;
|
|
495 |
mutex_exit(&spa->spa_uberblock_lock);
|
|
496 |
}
|
|
497 |
|
|
498 |
zio_buf_free(zio->io_data, zio->io_size);
|
|
499 |
}
|
|
500 |
|
|
501 |
void
|
|
502 |
vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
|
|
503 |
{
|
|
504 |
int l, c, n;
|
|
505 |
|
|
506 |
for (c = 0; c < vd->vdev_children; c++)
|
|
507 |
vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
|
|
508 |
|
|
509 |
if (!vd->vdev_ops->vdev_op_leaf)
|
|
510 |
return;
|
|
511 |
|
|
512 |
if (vdev_is_dead(vd))
|
|
513 |
return;
|
|
514 |
|
|
515 |
for (l = 0; l < VDEV_LABELS; l++) {
|
|
516 |
for (n = 0; n < VDEV_UBERBLOCKS; n++) {
|
|
517 |
vdev_label_read(zio, vd, l,
|
|
518 |
zio_buf_alloc(sizeof (uberblock_phys_t)),
|
|
519 |
offsetof(vdev_label_t, vl_uberblock[n]),
|
|
520 |
sizeof (uberblock_phys_t),
|
|
521 |
vdev_uberblock_load_done, ubbest);
|
|
522 |
}
|
|
523 |
}
|
|
524 |
}
|
|
525 |
|
|
526 |
/*
|
|
527 |
* Write the uberblock to both labels of all leaves of the specified vdev.
|
|
528 |
*/
|
|
529 |
static void
|
|
530 |
vdev_uberblock_sync_done(zio_t *zio)
|
|
531 |
{
|
|
532 |
uint64_t *good_writes = zio->io_root->io_private;
|
|
533 |
|
|
534 |
if (zio->io_error == 0)
|
|
535 |
atomic_add_64(good_writes, 1);
|
|
536 |
}
|
|
537 |
|
|
538 |
static void
|
|
539 |
vdev_uberblock_sync(zio_t *zio, uberblock_phys_t *ubphys, vdev_t *vd,
|
|
540 |
uint64_t txg)
|
|
541 |
{
|
|
542 |
int l, c, n;
|
|
543 |
|
|
544 |
for (c = 0; c < vd->vdev_children; c++)
|
|
545 |
vdev_uberblock_sync(zio, ubphys, vd->vdev_child[c], txg);
|
|
546 |
|
|
547 |
if (!vd->vdev_ops->vdev_op_leaf)
|
|
548 |
return;
|
|
549 |
|
|
550 |
if (vdev_is_dead(vd))
|
|
551 |
return;
|
|
552 |
|
|
553 |
n = txg & (VDEV_UBERBLOCKS - 1);
|
|
554 |
|
|
555 |
ASSERT(ubphys->ubp_uberblock.ub_txg == txg);
|
|
556 |
|
|
557 |
for (l = 0; l < VDEV_LABELS; l++)
|
|
558 |
vdev_label_write(zio, vd, l, ubphys,
|
|
559 |
offsetof(vdev_label_t, vl_uberblock[n]),
|
|
560 |
sizeof (uberblock_phys_t), vdev_uberblock_sync_done, NULL);
|
|
561 |
|
|
562 |
dprintf("vdev %s in txg %llu\n", vdev_description(vd), txg);
|
|
563 |
}
|
|
564 |
|
|
565 |
static int
|
|
566 |
vdev_uberblock_sync_tree(spa_t *spa, uberblock_t *ub, vdev_t *uvd, uint64_t txg)
|
|
567 |
{
|
|
568 |
uberblock_phys_t *ubphys;
|
|
569 |
uint64_t *good_writes;
|
|
570 |
zio_t *zio;
|
|
571 |
int error;
|
|
572 |
|
|
573 |
ubphys = zio_buf_alloc(sizeof (uberblock_phys_t));
|
|
574 |
bzero(ubphys, sizeof (uberblock_phys_t));
|
|
575 |
ubphys->ubp_uberblock = *ub;
|
|
576 |
|
|
577 |
good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
|
|
578 |
|
|
579 |
zio = zio_root(spa, NULL, good_writes,
|
|
580 |
ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
|
|
581 |
|
|
582 |
vdev_uberblock_sync(zio, ubphys, uvd, txg);
|
|
583 |
|
|
584 |
error = zio_wait(zio);
|
|
585 |
|
|
586 |
if (error && *good_writes != 0) {
|
|
587 |
dprintf("partial success: good_writes = %llu\n", *good_writes);
|
|
588 |
error = 0;
|
|
589 |
}
|
|
590 |
|
|
591 |
/*
|
|
592 |
* It's possible to have no good writes and no error if every vdev is in
|
|
593 |
* the CANT_OPEN state.
|
|
594 |
*/
|
|
595 |
if (*good_writes == 0 && error == 0)
|
|
596 |
error = EIO;
|
|
597 |
|
|
598 |
kmem_free(good_writes, sizeof (uint64_t));
|
|
599 |
zio_buf_free(ubphys, sizeof (uberblock_phys_t));
|
|
600 |
|
|
601 |
return (error);
|
|
602 |
}
|
|
603 |
|
|
604 |
/*
|
|
605 |
* Sync out an individual vdev.
|
|
606 |
*/
|
|
607 |
static void
|
|
608 |
vdev_sync_label_done(zio_t *zio)
|
|
609 |
{
|
|
610 |
uint64_t *good_writes = zio->io_root->io_private;
|
|
611 |
|
|
612 |
if (zio->io_error == 0)
|
|
613 |
atomic_add_64(good_writes, 1);
|
|
614 |
}
|
|
615 |
|
|
616 |
static void
|
|
617 |
vdev_sync_label(zio_t *zio, vdev_t *vd, int l, uint64_t txg)
|
|
618 |
{
|
|
619 |
nvlist_t *label;
|
|
620 |
vdev_phys_t *vp;
|
|
621 |
char *buf;
|
|
622 |
size_t buflen;
|
|
623 |
int c;
|
|
624 |
|
|
625 |
for (c = 0; c < vd->vdev_children; c++)
|
|
626 |
vdev_sync_label(zio, vd->vdev_child[c], l, txg);
|
|
627 |
|
|
628 |
if (!vd->vdev_ops->vdev_op_leaf)
|
|
629 |
return;
|
|
630 |
|
|
631 |
if (vdev_is_dead(vd))
|
|
632 |
return;
|
|
633 |
|
|
634 |
/*
|
|
635 |
* Generate a label describing the top-level config to which we belong.
|
|
636 |
*/
|
|
637 |
label = spa_config_generate(vd->vdev_spa, vd, txg, 0);
|
|
638 |
|
|
639 |
vp = zio_buf_alloc(sizeof (vdev_phys_t));
|
|
640 |
bzero(vp, sizeof (vdev_phys_t));
|
|
641 |
|
|
642 |
buf = vp->vp_nvlist;
|
|
643 |
buflen = sizeof (vp->vp_nvlist);
|
|
644 |
|
1544
|
645 |
if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0)
|
789
|
646 |
vdev_label_write(zio, vd, l, vp,
|
|
647 |
offsetof(vdev_label_t, vl_vdev_phys), sizeof (vdev_phys_t),
|
|
648 |
vdev_sync_label_done, NULL);
|
|
649 |
|
|
650 |
zio_buf_free(vp, sizeof (vdev_phys_t));
|
|
651 |
nvlist_free(label);
|
|
652 |
|
|
653 |
dprintf("%s label %d txg %llu\n", vdev_description(vd), l, txg);
|
|
654 |
}
|
|
655 |
|
|
656 |
static int
|
|
657 |
vdev_sync_labels(vdev_t *vd, int l, uint64_t txg)
|
|
658 |
{
|
|
659 |
uint64_t *good_writes;
|
|
660 |
zio_t *zio;
|
|
661 |
int error;
|
|
662 |
|
|
663 |
ASSERT(vd == vd->vdev_top);
|
|
664 |
|
|
665 |
good_writes = kmem_zalloc(sizeof (uint64_t), KM_SLEEP);
|
|
666 |
|
|
667 |
zio = zio_root(vd->vdev_spa, NULL, good_writes,
|
|
668 |
ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
|
|
669 |
|
|
670 |
/*
|
|
671 |
* Recursively kick off writes to all labels.
|
|
672 |
*/
|
|
673 |
vdev_sync_label(zio, vd, l, txg);
|
|
674 |
|
|
675 |
error = zio_wait(zio);
|
|
676 |
|
|
677 |
if (error && *good_writes != 0) {
|
|
678 |
dprintf("partial success: good_writes = %llu\n", *good_writes);
|
|
679 |
error = 0;
|
|
680 |
}
|
|
681 |
|
|
682 |
if (*good_writes == 0 && error == 0)
|
|
683 |
error = ENODEV;
|
|
684 |
|
|
685 |
kmem_free(good_writes, sizeof (uint64_t));
|
|
686 |
|
|
687 |
return (error);
|
|
688 |
}
|
|
689 |
|
|
690 |
/*
|
|
691 |
* Sync the entire vdev configuration.
|
|
692 |
*
|
|
693 |
* The order of operations is carefully crafted to ensure that
|
|
694 |
* if the system panics or loses power at any time, the state on disk
|
|
695 |
* is still transactionally consistent. The in-line comments below
|
|
696 |
* describe the failure semantics at each stage.
|
|
697 |
*
|
|
698 |
* Moreover, it is designed to be idempotent: if spa_sync_labels() fails
|
|
699 |
* at any time, you can just call it again, and it will resume its work.
|
|
700 |
*/
|
|
701 |
int
|
|
702 |
spa_sync_labels(spa_t *spa, uint64_t txg)
|
|
703 |
{
|
|
704 |
uberblock_t *ub = &spa->spa_uberblock;
|
|
705 |
vdev_t *rvd = spa->spa_root_vdev;
|
|
706 |
vdev_t *vd, *uvd;
|
|
707 |
zio_t *zio;
|
|
708 |
int c, l, error;
|
|
709 |
|
|
710 |
ASSERT(ub->ub_txg <= txg);
|
|
711 |
|
|
712 |
/*
|
|
713 |
* If this isn't a resync due to I/O errors, and nothing changed
|
|
714 |
* in this transaction group, and the vdev configuration hasn't changed,
|
|
715 |
* and this isn't an explicit sync-all, then there's nothing to do.
|
|
716 |
*/
|
|
717 |
if (ub->ub_txg < txg && uberblock_update(ub, rvd, txg) == B_FALSE &&
|
|
718 |
list_is_empty(&spa->spa_dirty_list)) {
|
|
719 |
dprintf("nothing to sync in %s in txg %llu\n",
|
|
720 |
spa_name(spa), txg);
|
|
721 |
return (0);
|
|
722 |
}
|
|
723 |
|
|
724 |
if (txg > spa_freeze_txg(spa))
|
|
725 |
return (0);
|
|
726 |
|
|
727 |
dprintf("syncing %s txg %llu\n", spa_name(spa), txg);
|
|
728 |
|
|
729 |
/*
|
|
730 |
* Flush the write cache of every disk that's been written to
|
|
731 |
* in this transaction group. This ensures that all blocks
|
|
732 |
* written in this txg will be committed to stable storage
|
|
733 |
* before any uberblock that references them.
|
|
734 |
*/
|
|
735 |
zio = zio_root(spa, NULL, NULL,
|
|
736 |
ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
|
|
737 |
for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
|
|
738 |
vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg))) {
|
|
739 |
zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE,
|
|
740 |
NULL, NULL, ZIO_PRIORITY_NOW,
|
|
741 |
ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY));
|
|
742 |
}
|
|
743 |
(void) zio_wait(zio);
|
|
744 |
|
|
745 |
/*
|
|
746 |
* Sync out the even labels (L0, L2) for every dirty vdev. If the
|
|
747 |
* system dies in the middle of this process, that's OK: all of the
|
|
748 |
* even labels that made it to disk will be newer than any uberblock,
|
|
749 |
* and will therefore be considered invalid. The odd labels (L1, L3),
|
|
750 |
* which have not yet been touched, will still be valid.
|
|
751 |
*/
|
|
752 |
for (vd = list_head(&spa->spa_dirty_list); vd != NULL;
|
|
753 |
vd = list_next(&spa->spa_dirty_list, vd)) {
|
|
754 |
for (l = 0; l < VDEV_LABELS; l++) {
|
|
755 |
if (l & 1)
|
|
756 |
continue;
|
|
757 |
if ((error = vdev_sync_labels(vd, l, txg)) != 0)
|
|
758 |
return (error);
|
|
759 |
}
|
|
760 |
}
|
|
761 |
|
|
762 |
/*
|
|
763 |
* Flush the new labels to disk. This ensures that all even-label
|
|
764 |
* updates are committed to stable storage before the uberblock update.
|
|
765 |
*/
|
|
766 |
zio = zio_root(spa, NULL, NULL,
|
|
767 |
ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
|
|
768 |
for (vd = list_head(&spa->spa_dirty_list); vd != NULL;
|
|
769 |
vd = list_next(&spa->spa_dirty_list, vd)) {
|
|
770 |
zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE,
|
|
771 |
NULL, NULL, ZIO_PRIORITY_NOW,
|
|
772 |
ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY));
|
|
773 |
}
|
|
774 |
(void) zio_wait(zio);
|
|
775 |
|
|
776 |
/*
|
|
777 |
* If there are any dirty vdevs, sync the uberblock to all vdevs.
|
|
778 |
* Otherwise, pick one top-level vdev at random.
|
|
779 |
*/
|
|
780 |
if (!list_is_empty(&spa->spa_dirty_list))
|
|
781 |
uvd = rvd;
|
|
782 |
else
|
|
783 |
uvd = rvd->vdev_child[spa_get_random(rvd->vdev_children)];
|
|
784 |
|
|
785 |
/*
|
|
786 |
* Sync the uberblocks. If the system dies in the middle of this
|
|
787 |
* step, there are two cases to consider, and the on-disk state
|
|
788 |
* is consistent either way:
|
|
789 |
*
|
|
790 |
* (1) If none of the new uberblocks made it to disk, then the
|
|
791 |
* previous uberblock will be the newest, and the odd labels
|
|
792 |
* (which had not yet been touched) will be valid with respect
|
|
793 |
* to that uberblock.
|
|
794 |
*
|
|
795 |
* (2) If one or more new uberblocks made it to disk, then they
|
|
796 |
* will be the newest, and the even labels (which had all
|
|
797 |
* been successfully committed) will be valid with respect
|
|
798 |
* to the new uberblocks.
|
|
799 |
*/
|
|
800 |
if ((error = vdev_uberblock_sync_tree(spa, ub, uvd, txg)) != 0)
|
|
801 |
return (error);
|
|
802 |
|
|
803 |
/*
|
|
804 |
* Flush the uberblocks to disk. This ensures that the odd labels
|
|
805 |
* are no longer needed (because the new uberblocks and the even
|
|
806 |
* labels are safely on disk), so it is safe to overwrite them.
|
|
807 |
*/
|
|
808 |
(void) zio_wait(zio_ioctl(NULL, spa, uvd, DKIOCFLUSHWRITECACHE,
|
|
809 |
NULL, NULL, ZIO_PRIORITY_NOW,
|
|
810 |
ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY));
|
|
811 |
|
|
812 |
/*
|
|
813 |
* Sync out odd labels for every dirty vdev. If the system dies
|
|
814 |
* in the middle of this process, the even labels and the new
|
|
815 |
* uberblocks will suffice to open the pool. The next time
|
|
816 |
* the pool is opened, the first thing we'll do -- before any
|
|
817 |
* user data is modified -- is mark every vdev dirty so that
|
|
818 |
* all labels will be brought up to date.
|
|
819 |
*/
|
|
820 |
for (vd = list_head(&spa->spa_dirty_list); vd != NULL;
|
|
821 |
vd = list_next(&spa->spa_dirty_list, vd)) {
|
|
822 |
for (l = 0; l < VDEV_LABELS; l++) {
|
|
823 |
if ((l & 1) == 0)
|
|
824 |
continue;
|
|
825 |
if ((error = vdev_sync_labels(vd, l, txg)) != 0)
|
|
826 |
return (error);
|
|
827 |
}
|
|
828 |
}
|
|
829 |
|
|
830 |
/*
|
|
831 |
* Flush the new labels to disk. This ensures that all odd-label
|
|
832 |
* updates are committed to stable storage before the next
|
|
833 |
* transaction group begins.
|
|
834 |
*/
|
|
835 |
zio = zio_root(spa, NULL, NULL,
|
|
836 |
ZIO_FLAG_CONFIG_HELD | ZIO_FLAG_CANFAIL);
|
|
837 |
for (vd = list_head(&spa->spa_dirty_list); vd != NULL;
|
|
838 |
vd = list_next(&spa->spa_dirty_list, vd)) {
|
|
839 |
zio_nowait(zio_ioctl(zio, spa, vd, DKIOCFLUSHWRITECACHE,
|
|
840 |
NULL, NULL, ZIO_PRIORITY_NOW,
|
|
841 |
ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY));
|
|
842 |
}
|
|
843 |
(void) zio_wait(zio);
|
|
844 |
|
|
845 |
/*
|
|
846 |
* Clear the dirty list.
|
|
847 |
*/
|
|
848 |
while (!list_is_empty(&spa->spa_dirty_list))
|
|
849 |
vdev_config_clean(list_head(&spa->spa_dirty_list));
|
|
850 |
|
|
851 |
#ifdef DEBUG
|
|
852 |
for (c = 0; c < rvd->vdev_children; c++) {
|
|
853 |
ASSERT(rvd->vdev_child[c]->vdev_is_dirty == 0);
|
|
854 |
}
|
|
855 |
#endif
|
|
856 |
|
|
857 |
return (0);
|
|
858 |
}
|