1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
26 */
27
28#include <sys/zfs_context.h>
29#include <sys/spa.h>
| 1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
26 */
27
28#include <sys/zfs_context.h>
29#include <sys/spa.h>
|
| 30#include <sys/fm/fs/zfs.h>
|
30#include <sys/spa_impl.h>
31#include <sys/nvpair.h>
32#include <sys/uio.h>
33#include <sys/fs/zfs.h>
34#include <sys/vdev_impl.h>
35#include <sys/zfs_ioctl.h>
36#include <sys/utsname.h>
37#include <sys/sunddi.h>
38#include <sys/zfeature.h>
39#ifdef _KERNEL
40#include <sys/kobj.h>
41#include <sys/zone.h>
42#endif
43
44/*
45 * Pool configuration repository.
46 *
47 * Pool configuration is stored as a packed nvlist on the filesystem. By
48 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
49 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
50 * property set that allows them to be stored in an alternate location until
51 * the control of external software.
52 *
53 * For each cache file, we have a single nvlist which holds all the
54 * configuration information. When the module loads, we read this information
55 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
56 * maintained independently in spa.c. Whenever the namespace is modified, or
57 * the configuration of a pool is changed, we call spa_config_sync(), which
58 * walks through all the active pools and writes the configuration to disk.
59 */
60
61static uint64_t spa_config_generation = 1;
62
63/*
64 * This can be overridden in userland to preserve an alternate namespace for
65 * userland pools when doing testing.
66 */
67const char *spa_config_path = ZPOOL_CACHE;
68
69/*
70 * Called when the module is first loaded, this routine loads the configuration
71 * file into the SPA namespace. It does not actually open or load the pools; it
72 * only populates the namespace.
73 */
74void
75spa_config_load(void)
76{
77 void *buf = NULL;
78 nvlist_t *nvlist, *child;
79 nvpair_t *nvpair;
80 char *pathname;
81 struct _buf *file;
82 uint64_t fsize;
83
84 /*
85 * Open the configuration file.
86 */
87 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
88
89 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path);
90
91 file = kobj_open_file(pathname);
92
93 kmem_free(pathname, MAXPATHLEN);
94
95 if (file == (struct _buf *)-1)
96 return;
97
98 if (kobj_get_filesize(file, &fsize) != 0)
99 goto out;
100
101 buf = kmem_alloc(fsize, KM_SLEEP);
102
103 /*
104 * Read the nvlist from the file.
105 */
106 if (kobj_read_file(file, buf, fsize, 0) < 0)
107 goto out;
108
109 /*
110 * Unpack the nvlist.
111 */
112 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
113 goto out;
114
115 /*
116 * Iterate over all elements in the nvlist, creating a new spa_t for
117 * each one with the specified configuration.
118 */
119 mutex_enter(&spa_namespace_lock);
120 nvpair = NULL;
121 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
122 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
123 continue;
124
125 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
126
127 if (spa_lookup(nvpair_name(nvpair)) != NULL)
128 continue;
129 (void) spa_add(nvpair_name(nvpair), child, NULL);
130 }
131 mutex_exit(&spa_namespace_lock);
132
133 nvlist_free(nvlist);
134
135out:
136 if (buf != NULL)
137 kmem_free(buf, fsize);
138
139 kobj_close_file(file);
140}
141
| 31#include <sys/spa_impl.h>
32#include <sys/nvpair.h>
33#include <sys/uio.h>
34#include <sys/fs/zfs.h>
35#include <sys/vdev_impl.h>
36#include <sys/zfs_ioctl.h>
37#include <sys/utsname.h>
38#include <sys/sunddi.h>
39#include <sys/zfeature.h>
40#ifdef _KERNEL
41#include <sys/kobj.h>
42#include <sys/zone.h>
43#endif
44
45/*
46 * Pool configuration repository.
47 *
48 * Pool configuration is stored as a packed nvlist on the filesystem. By
49 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
50 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
51 * property set that allows them to be stored in an alternate location until
52 * the control of external software.
53 *
54 * For each cache file, we have a single nvlist which holds all the
55 * configuration information. When the module loads, we read this information
56 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
57 * maintained independently in spa.c. Whenever the namespace is modified, or
58 * the configuration of a pool is changed, we call spa_config_sync(), which
59 * walks through all the active pools and writes the configuration to disk.
60 */
61
62static uint64_t spa_config_generation = 1;
63
64/*
65 * This can be overridden in userland to preserve an alternate namespace for
66 * userland pools when doing testing.
67 */
68const char *spa_config_path = ZPOOL_CACHE;
69
70/*
71 * Called when the module is first loaded, this routine loads the configuration
72 * file into the SPA namespace. It does not actually open or load the pools; it
73 * only populates the namespace.
74 */
75void
76spa_config_load(void)
77{
78 void *buf = NULL;
79 nvlist_t *nvlist, *child;
80 nvpair_t *nvpair;
81 char *pathname;
82 struct _buf *file;
83 uint64_t fsize;
84
85 /*
86 * Open the configuration file.
87 */
88 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
89
90 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path);
91
92 file = kobj_open_file(pathname);
93
94 kmem_free(pathname, MAXPATHLEN);
95
96 if (file == (struct _buf *)-1)
97 return;
98
99 if (kobj_get_filesize(file, &fsize) != 0)
100 goto out;
101
102 buf = kmem_alloc(fsize, KM_SLEEP);
103
104 /*
105 * Read the nvlist from the file.
106 */
107 if (kobj_read_file(file, buf, fsize, 0) < 0)
108 goto out;
109
110 /*
111 * Unpack the nvlist.
112 */
113 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
114 goto out;
115
116 /*
117 * Iterate over all elements in the nvlist, creating a new spa_t for
118 * each one with the specified configuration.
119 */
120 mutex_enter(&spa_namespace_lock);
121 nvpair = NULL;
122 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
123 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
124 continue;
125
126 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
127
128 if (spa_lookup(nvpair_name(nvpair)) != NULL)
129 continue;
130 (void) spa_add(nvpair_name(nvpair), child, NULL);
131 }
132 mutex_exit(&spa_namespace_lock);
133
134 nvlist_free(nvlist);
135
136out:
137 if (buf != NULL)
138 kmem_free(buf, fsize);
139
140 kobj_close_file(file);
141}
142
|
142static void
| 143static int
|
143spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
144{
145 size_t buflen;
146 char *buf;
147 vnode_t *vp;
148 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
149 char *temp;
| 144spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
145{
146 size_t buflen;
147 char *buf;
148 vnode_t *vp;
149 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
150 char *temp;
|
| 151 int err;
|
150
151 /*
152 * If the nvlist is empty (NULL), then remove the old cachefile.
153 */
154 if (nvl == NULL) {
| 152
153 /*
154 * If the nvlist is empty (NULL), then remove the old cachefile.
155 */
156 if (nvl == NULL) {
|
155 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
156 return;
| 157 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
158 return (err);
|
157 }
158
159 /*
160 * Pack the configuration into a buffer.
161 */
162 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
163
164 buf = kmem_alloc(buflen, KM_SLEEP);
165 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
166
167 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
168 KM_SLEEP) == 0);
169
170 /*
171 * Write the configuration to disk. We need to do the traditional
172 * 'write to temporary file, sync, move over original' to make sure we
173 * always have a consistent view of the data.
174 */
175 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
176
| 159 }
160
161 /*
162 * Pack the configuration into a buffer.
163 */
164 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
165
166 buf = kmem_alloc(buflen, KM_SLEEP);
167 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
168
169 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
170 KM_SLEEP) == 0);
171
172 /*
173 * Write the configuration to disk. We need to do the traditional
174 * 'write to temporary file, sync, move over original' to make sure we
175 * always have a consistent view of the data.
176 */
177 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
178
|
177 if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
178 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
179 0, RLIM64_INFINITY, kcred, NULL) == 0 &&
180 VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
181 (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
182 }
| 179 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
180 if (err == 0) {
181 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
182 0, RLIM64_INFINITY, kcred, NULL);
183 if (err == 0)
184 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
185 if (err == 0)
186 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
|
183 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
184 }
185
186 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
187
188 kmem_free(buf, buflen);
189 kmem_free(temp, MAXPATHLEN);
| 187 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
188 }
189
190 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
191
192 kmem_free(buf, buflen);
193 kmem_free(temp, MAXPATHLEN);
|
| 194 return (err);
|
190}
191
192/*
193 * Synchronize pool configuration to disk. This must be called with the
194 * namespace lock held.
195 */
196void
197spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
198{
199 spa_config_dirent_t *dp, *tdp;
200 nvlist_t *nvl;
| 195}
196
197/*
198 * Synchronize pool configuration to disk. This must be called with the
199 * namespace lock held.
200 */
201void
202spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
203{
204 spa_config_dirent_t *dp, *tdp;
205 nvlist_t *nvl;
|
| 206 boolean_t ccw_failure;
207 int error;
|
201
202 ASSERT(MUTEX_HELD(&spa_namespace_lock));
203
204 if (rootdir == NULL || !(spa_mode_global & FWRITE))
205 return;
206
207 /*
208 * Iterate over all cachefiles for the pool, past or present. When the
209 * cachefile is changed, the new one is pushed onto this list, allowing
210 * us to update previous cachefiles that no longer contain this pool.
211 */
| 208
209 ASSERT(MUTEX_HELD(&spa_namespace_lock));
210
211 if (rootdir == NULL || !(spa_mode_global & FWRITE))
212 return;
213
214 /*
215 * Iterate over all cachefiles for the pool, past or present. When the
216 * cachefile is changed, the new one is pushed onto this list, allowing
217 * us to update previous cachefiles that no longer contain this pool.
218 */
|
| 219 ccw_failure = B_FALSE;
|
212 for (dp = list_head(&target->spa_config_list); dp != NULL;
213 dp = list_next(&target->spa_config_list, dp)) {
214 spa_t *spa = NULL;
215 if (dp->scd_path == NULL)
216 continue;
217
218 /*
219 * Iterate over all pools, adding any matching pools to 'nvl'.
220 */
221 nvl = NULL;
222 while ((spa = spa_next(spa)) != NULL) {
223 /*
224 * Skip over our own pool if we're about to remove
225 * ourselves from the spa namespace or any pool that
226 * is readonly. Since we cannot guarantee that a
227 * readonly pool would successfully import upon reboot,
228 * we don't allow them to be written to the cache file.
229 */
230 if ((spa == target && removing) ||
231 !spa_writeable(spa))
232 continue;
233
234 mutex_enter(&spa->spa_props_lock);
235 tdp = list_head(&spa->spa_config_list);
236 if (spa->spa_config == NULL ||
237 tdp->scd_path == NULL ||
238 strcmp(tdp->scd_path, dp->scd_path) != 0) {
239 mutex_exit(&spa->spa_props_lock);
240 continue;
241 }
242
243 if (nvl == NULL)
244 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
245 KM_SLEEP) == 0);
246
247 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
248 spa->spa_config) == 0);
249 mutex_exit(&spa->spa_props_lock);
250 }
251
| 220 for (dp = list_head(&target->spa_config_list); dp != NULL;
221 dp = list_next(&target->spa_config_list, dp)) {
222 spa_t *spa = NULL;
223 if (dp->scd_path == NULL)
224 continue;
225
226 /*
227 * Iterate over all pools, adding any matching pools to 'nvl'.
228 */
229 nvl = NULL;
230 while ((spa = spa_next(spa)) != NULL) {
231 /*
232 * Skip over our own pool if we're about to remove
233 * ourselves from the spa namespace or any pool that
234 * is readonly. Since we cannot guarantee that a
235 * readonly pool would successfully import upon reboot,
236 * we don't allow them to be written to the cache file.
237 */
238 if ((spa == target && removing) ||
239 !spa_writeable(spa))
240 continue;
241
242 mutex_enter(&spa->spa_props_lock);
243 tdp = list_head(&spa->spa_config_list);
244 if (spa->spa_config == NULL ||
245 tdp->scd_path == NULL ||
246 strcmp(tdp->scd_path, dp->scd_path) != 0) {
247 mutex_exit(&spa->spa_props_lock);
248 continue;
249 }
250
251 if (nvl == NULL)
252 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
253 KM_SLEEP) == 0);
254
255 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
256 spa->spa_config) == 0);
257 mutex_exit(&spa->spa_props_lock);
258 }
259
|
252 spa_config_write(dp, nvl);
| 260 error = spa_config_write(dp, nvl);
261 if (error != 0)
262 ccw_failure = B_TRUE;
|
253 nvlist_free(nvl);
254 }
255
| 263 nvlist_free(nvl);
264 }
265
|
| 266 if (ccw_failure) {
267 /*
268 * Keep trying so that configuration data is
269 * written if/when any temporary filesystem
270 * resource issues are resolved.
271 */
272 if (target->spa_ccw_fail_time == 0) {
273 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
274 target, NULL, NULL, 0, 0);
275 }
276 target->spa_ccw_fail_time = gethrtime();
277 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
278 } else {
279 /*
280 * Do not rate limit future attempts to update
281 * the config cache.
282 */
283 target->spa_ccw_fail_time = 0;
284 }
285
|
256 /*
257 * Remove any config entries older than the current one.
258 */
259 dp = list_head(&target->spa_config_list);
260 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
261 list_remove(&target->spa_config_list, tdp);
262 if (tdp->scd_path != NULL)
263 spa_strfree(tdp->scd_path);
264 kmem_free(tdp, sizeof (spa_config_dirent_t));
265 }
266
267 spa_config_generation++;
268
269 if (postsysevent)
270 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
271}
272
273/*
274 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
275 * and we don't want to allow the local zone to see all the pools anyway.
276 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
277 * information for all pool visible within the zone.
278 */
279nvlist_t *
280spa_all_configs(uint64_t *generation)
281{
282 nvlist_t *pools;
283 spa_t *spa = NULL;
284
285 if (*generation == spa_config_generation)
286 return (NULL);
287
288 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
289
290 mutex_enter(&spa_namespace_lock);
291 while ((spa = spa_next(spa)) != NULL) {
292 if (INGLOBALZONE(curthread) ||
293 zone_dataset_visible(spa_name(spa), NULL)) {
294 mutex_enter(&spa->spa_props_lock);
295 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
296 spa->spa_config) == 0);
297 mutex_exit(&spa->spa_props_lock);
298 }
299 }
300 *generation = spa_config_generation;
301 mutex_exit(&spa_namespace_lock);
302
303 return (pools);
304}
305
306void
307spa_config_set(spa_t *spa, nvlist_t *config)
308{
309 mutex_enter(&spa->spa_props_lock);
310 if (spa->spa_config != NULL)
311 nvlist_free(spa->spa_config);
312 spa->spa_config = config;
313 mutex_exit(&spa->spa_props_lock);
314}
315
316/*
317 * Generate the pool's configuration based on the current in-core state.
318 *
319 * We infer whether to generate a complete config or just one top-level config
320 * based on whether vd is the root vdev.
321 */
322nvlist_t *
323spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
324{
325 nvlist_t *config, *nvroot;
326 vdev_t *rvd = spa->spa_root_vdev;
327 unsigned long hostid = 0;
328 boolean_t locked = B_FALSE;
329 uint64_t split_guid;
330
331 if (vd == NULL) {
332 vd = rvd;
333 locked = B_TRUE;
334 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
335 }
336
337 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
338 (SCL_CONFIG | SCL_STATE));
339
340 /*
341 * If txg is -1, report the current value of spa->spa_config_txg.
342 */
343 if (txg == -1ULL)
344 txg = spa->spa_config_txg;
345
346 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
347
348 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
349 spa_version(spa)) == 0);
350 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
351 spa_name(spa)) == 0);
352 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
353 spa_state(spa)) == 0);
354 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
355 txg) == 0);
356 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
357 spa_guid(spa)) == 0);
358 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
359 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
360
361
362#ifdef _KERNEL
363 hostid = zone_get_hostid(NULL);
364#else /* _KERNEL */
365 /*
366 * We're emulating the system's hostid in userland, so we can't use
367 * zone_get_hostid().
368 */
369 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
370#endif /* _KERNEL */
371 if (hostid != 0) {
372 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
373 hostid) == 0);
374 }
375 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
376 utsname.nodename) == 0);
377
378 if (vd != rvd) {
379 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
380 vd->vdev_top->vdev_guid) == 0);
381 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
382 vd->vdev_guid) == 0);
383 if (vd->vdev_isspare)
384 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
385 1ULL) == 0);
386 if (vd->vdev_islog)
387 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
388 1ULL) == 0);
389 vd = vd->vdev_top; /* label contains top config */
390 } else {
391 /*
392 * Only add the (potentially large) split information
393 * in the mos config, and not in the vdev labels
394 */
395 if (spa->spa_config_splitting != NULL)
396 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
397 spa->spa_config_splitting) == 0);
398 }
399
400 /*
401 * Add the top-level config. We even add this on pools which
402 * don't support holes in the namespace.
403 */
404 vdev_top_config_generate(spa, config);
405
406 /*
407 * If we're splitting, record the original pool's guid.
408 */
409 if (spa->spa_config_splitting != NULL &&
410 nvlist_lookup_uint64(spa->spa_config_splitting,
411 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
413 split_guid) == 0);
414 }
415
416 nvroot = vdev_config_generate(spa, vd, getstats, 0);
417 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
418 nvlist_free(nvroot);
419
420 /*
421 * Store what's necessary for reading the MOS in the label.
422 */
423 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
424 spa->spa_label_features) == 0);
425
426 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
427 ddt_histogram_t *ddh;
428 ddt_stat_t *dds;
429 ddt_object_t *ddo;
430
431 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
432 ddt_get_dedup_histogram(spa, ddh);
433 VERIFY(nvlist_add_uint64_array(config,
434 ZPOOL_CONFIG_DDT_HISTOGRAM,
435 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
436 kmem_free(ddh, sizeof (ddt_histogram_t));
437
438 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
439 ddt_get_dedup_object_stats(spa, ddo);
440 VERIFY(nvlist_add_uint64_array(config,
441 ZPOOL_CONFIG_DDT_OBJ_STATS,
442 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
443 kmem_free(ddo, sizeof (ddt_object_t));
444
445 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
446 ddt_get_dedup_stats(spa, dds);
447 VERIFY(nvlist_add_uint64_array(config,
448 ZPOOL_CONFIG_DDT_STATS,
449 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
450 kmem_free(dds, sizeof (ddt_stat_t));
451 }
452
453 if (locked)
454 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
455
456 return (config);
457}
458
459/*
460 * Update all disk labels, generate a fresh config based on the current
461 * in-core state, and sync the global config cache (do not sync the config
462 * cache if this is a booting rootpool).
463 */
464void
465spa_config_update(spa_t *spa, int what)
466{
467 vdev_t *rvd = spa->spa_root_vdev;
468 uint64_t txg;
469 int c;
470
471 ASSERT(MUTEX_HELD(&spa_namespace_lock));
472
473 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
474 txg = spa_last_synced_txg(spa) + 1;
475 if (what == SPA_CONFIG_UPDATE_POOL) {
476 vdev_config_dirty(rvd);
477 } else {
478 /*
479 * If we have top-level vdevs that were added but have
480 * not yet been prepared for allocation, do that now.
481 * (It's safe now because the config cache is up to date,
482 * so it will be able to translate the new DVAs.)
483 * See comments in spa_vdev_add() for full details.
484 */
485 for (c = 0; c < rvd->vdev_children; c++) {
486 vdev_t *tvd = rvd->vdev_child[c];
487 if (tvd->vdev_ms_array == 0)
488 vdev_metaslab_set_size(tvd);
489 vdev_expand(tvd, txg);
490 }
491 }
492 spa_config_exit(spa, SCL_ALL, FTAG);
493
494 /*
495 * Wait for the mosconfig to be regenerated and synced.
496 */
497 txg_wait_synced(spa->spa_dsl_pool, txg);
498
499 /*
500 * Update the global config cache to reflect the new mosconfig.
501 */
502 if (!spa->spa_is_root)
503 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
504
505 if (what == SPA_CONFIG_UPDATE_POOL)
506 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
507}
| 286 /*
287 * Remove any config entries older than the current one.
288 */
289 dp = list_head(&target->spa_config_list);
290 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
291 list_remove(&target->spa_config_list, tdp);
292 if (tdp->scd_path != NULL)
293 spa_strfree(tdp->scd_path);
294 kmem_free(tdp, sizeof (spa_config_dirent_t));
295 }
296
297 spa_config_generation++;
298
299 if (postsysevent)
300 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
301}
302
303/*
304 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
305 * and we don't want to allow the local zone to see all the pools anyway.
306 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
307 * information for all pool visible within the zone.
308 */
309nvlist_t *
310spa_all_configs(uint64_t *generation)
311{
312 nvlist_t *pools;
313 spa_t *spa = NULL;
314
315 if (*generation == spa_config_generation)
316 return (NULL);
317
318 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
319
320 mutex_enter(&spa_namespace_lock);
321 while ((spa = spa_next(spa)) != NULL) {
322 if (INGLOBALZONE(curthread) ||
323 zone_dataset_visible(spa_name(spa), NULL)) {
324 mutex_enter(&spa->spa_props_lock);
325 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
326 spa->spa_config) == 0);
327 mutex_exit(&spa->spa_props_lock);
328 }
329 }
330 *generation = spa_config_generation;
331 mutex_exit(&spa_namespace_lock);
332
333 return (pools);
334}
335
336void
337spa_config_set(spa_t *spa, nvlist_t *config)
338{
339 mutex_enter(&spa->spa_props_lock);
340 if (spa->spa_config != NULL)
341 nvlist_free(spa->spa_config);
342 spa->spa_config = config;
343 mutex_exit(&spa->spa_props_lock);
344}
345
346/*
347 * Generate the pool's configuration based on the current in-core state.
348 *
349 * We infer whether to generate a complete config or just one top-level config
350 * based on whether vd is the root vdev.
351 */
352nvlist_t *
353spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
354{
355 nvlist_t *config, *nvroot;
356 vdev_t *rvd = spa->spa_root_vdev;
357 unsigned long hostid = 0;
358 boolean_t locked = B_FALSE;
359 uint64_t split_guid;
360
361 if (vd == NULL) {
362 vd = rvd;
363 locked = B_TRUE;
364 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
365 }
366
367 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
368 (SCL_CONFIG | SCL_STATE));
369
370 /*
371 * If txg is -1, report the current value of spa->spa_config_txg.
372 */
373 if (txg == -1ULL)
374 txg = spa->spa_config_txg;
375
376 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
377
378 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
379 spa_version(spa)) == 0);
380 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
381 spa_name(spa)) == 0);
382 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
383 spa_state(spa)) == 0);
384 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
385 txg) == 0);
386 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
387 spa_guid(spa)) == 0);
388 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
389 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
390
391
392#ifdef _KERNEL
393 hostid = zone_get_hostid(NULL);
394#else /* _KERNEL */
395 /*
396 * We're emulating the system's hostid in userland, so we can't use
397 * zone_get_hostid().
398 */
399 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
400#endif /* _KERNEL */
401 if (hostid != 0) {
402 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
403 hostid) == 0);
404 }
405 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
406 utsname.nodename) == 0);
407
408 if (vd != rvd) {
409 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
410 vd->vdev_top->vdev_guid) == 0);
411 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
412 vd->vdev_guid) == 0);
413 if (vd->vdev_isspare)
414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
415 1ULL) == 0);
416 if (vd->vdev_islog)
417 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
418 1ULL) == 0);
419 vd = vd->vdev_top; /* label contains top config */
420 } else {
421 /*
422 * Only add the (potentially large) split information
423 * in the mos config, and not in the vdev labels
424 */
425 if (spa->spa_config_splitting != NULL)
426 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
427 spa->spa_config_splitting) == 0);
428 }
429
430 /*
431 * Add the top-level config. We even add this on pools which
432 * don't support holes in the namespace.
433 */
434 vdev_top_config_generate(spa, config);
435
436 /*
437 * If we're splitting, record the original pool's guid.
438 */
439 if (spa->spa_config_splitting != NULL &&
440 nvlist_lookup_uint64(spa->spa_config_splitting,
441 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
442 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
443 split_guid) == 0);
444 }
445
446 nvroot = vdev_config_generate(spa, vd, getstats, 0);
447 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
448 nvlist_free(nvroot);
449
450 /*
451 * Store what's necessary for reading the MOS in the label.
452 */
453 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
454 spa->spa_label_features) == 0);
455
456 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
457 ddt_histogram_t *ddh;
458 ddt_stat_t *dds;
459 ddt_object_t *ddo;
460
461 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
462 ddt_get_dedup_histogram(spa, ddh);
463 VERIFY(nvlist_add_uint64_array(config,
464 ZPOOL_CONFIG_DDT_HISTOGRAM,
465 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
466 kmem_free(ddh, sizeof (ddt_histogram_t));
467
468 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
469 ddt_get_dedup_object_stats(spa, ddo);
470 VERIFY(nvlist_add_uint64_array(config,
471 ZPOOL_CONFIG_DDT_OBJ_STATS,
472 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
473 kmem_free(ddo, sizeof (ddt_object_t));
474
475 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
476 ddt_get_dedup_stats(spa, dds);
477 VERIFY(nvlist_add_uint64_array(config,
478 ZPOOL_CONFIG_DDT_STATS,
479 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
480 kmem_free(dds, sizeof (ddt_stat_t));
481 }
482
483 if (locked)
484 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
485
486 return (config);
487}
488
489/*
490 * Update all disk labels, generate a fresh config based on the current
491 * in-core state, and sync the global config cache (do not sync the config
492 * cache if this is a booting rootpool).
493 */
494void
495spa_config_update(spa_t *spa, int what)
496{
497 vdev_t *rvd = spa->spa_root_vdev;
498 uint64_t txg;
499 int c;
500
501 ASSERT(MUTEX_HELD(&spa_namespace_lock));
502
503 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
504 txg = spa_last_synced_txg(spa) + 1;
505 if (what == SPA_CONFIG_UPDATE_POOL) {
506 vdev_config_dirty(rvd);
507 } else {
508 /*
509 * If we have top-level vdevs that were added but have
510 * not yet been prepared for allocation, do that now.
511 * (It's safe now because the config cache is up to date,
512 * so it will be able to translate the new DVAs.)
513 * See comments in spa_vdev_add() for full details.
514 */
515 for (c = 0; c < rvd->vdev_children; c++) {
516 vdev_t *tvd = rvd->vdev_child[c];
517 if (tvd->vdev_ms_array == 0)
518 vdev_metaslab_set_size(tvd);
519 vdev_expand(tvd, txg);
520 }
521 }
522 spa_config_exit(spa, SCL_ALL, FTAG);
523
524 /*
525 * Wait for the mosconfig to be regenerated and synced.
526 */
527 txg_wait_synced(spa->spa_dsl_pool, txg);
528
529 /*
530 * Update the global config cache to reflect the new mosconfig.
531 */
532 if (!spa->spa_is_root)
533 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
534
535 if (what == SPA_CONFIG_UPDATE_POOL)
536 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
537}
|