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) 2011 by Delphix. All rights reserved.
26 */
27
28 #include <sys/spa.h>
29 #include <sys/spa_impl.h>
30 #include <sys/nvpair.h>
31 #include <sys/uio.h>
32 #include <sys/fs/zfs.h>
33 #include <sys/vdev_impl.h>
34 #include <sys/zfs_ioctl.h>
35 #include <sys/utsname.h>
36 #include <sys/systeminfo.h>
37 #include <sys/sunddi.h>
38 #ifdef _KERNEL
39 #include <sys/kobj.h>
40 #include <sys/zone.h>
41 #endif
42
43 /*
44 * Pool configuration repository.
45 *
46 * Pool configuration is stored as a packed nvlist on the filesystem. By
47 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
48 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
49 * property set that allows them to be stored in an alternate location until
50 * the control of external software.
51 *
52 * For each cache file, we have a single nvlist which holds all the
53 * configuration information. When the module loads, we read this information
54 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
55 * maintained independently in spa.c. Whenever the namespace is modified, or
56 * the configuration of a pool is changed, we call spa_config_sync(), which
57 * walks through all the active pools and writes the configuration to disk.
58 */
59
60 static uint64_t spa_config_generation = 1;
61
62 /*
63 * This can be overridden in userland to preserve an alternate namespace for
64 * userland pools when doing testing.
65 */
66 const char *spa_config_path = ZPOOL_CACHE;
67
68 /*
69 * Called when the module is first loaded, this routine loads the configuration
70 * file into the SPA namespace. It does not actually open or load the pools; it
71 * only populates the namespace.
72 */
73 void
74 spa_config_load(void)
75 {
76 void *buf = NULL;
77 nvlist_t *nvlist, *child;
78 nvpair_t *nvpair;
79 char *pathname;
80 struct _buf *file;
81 uint64_t fsize;
82
83 /*
84 * Open the configuration file.
85 */
86 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
87
88 (void) snprintf(pathname, MAXPATHLEN, "%s%s",
89 (rootdir != NULL) ? "./" : "", 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
135 out:
136 if (buf != NULL)
137 kmem_free(buf, fsize);
138
139 kobj_close_file(file);
140 }
141
142 static void
143 spa_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;
150
151 /*
152 * If the nvlist is empty (NULL), then remove the old cachefile.
153 */
154 if (nvl == NULL) {
155 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
156 return;
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
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 }
183 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
184 VN_RELE(vp);
185 }
186
187 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
188
189 kmem_free(buf, buflen);
190 kmem_free(temp, MAXPATHLEN);
191 }
192
193 /*
194 * Synchronize pool configuration to disk. This must be called with the
195 * namespace lock held.
196 */
197 void
198 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
199 {
200 spa_config_dirent_t *dp, *tdp;
201 nvlist_t *nvl;
202
203 ASSERT(MUTEX_HELD(&spa_namespace_lock));
204
205 if (rootdir == NULL || !(spa_mode_global & FWRITE))
206 return;
207
208 /*
209 * Iterate over all cachefiles for the pool, past or present. When the
210 * cachefile is changed, the new one is pushed onto this list, allowing
211 * us to update previous cachefiles that no longer contain this pool.
212 */
213 for (dp = list_head(&target->spa_config_list); dp != NULL;
214 dp = list_next(&target->spa_config_list, dp)) {
215 spa_t *spa = NULL;
216 if (dp->scd_path == NULL)
217 continue;
218
219 /*
220 * Iterate over all pools, adding any matching pools to 'nvl'.
221 */
222 nvl = NULL;
223 while ((spa = spa_next(spa)) != NULL) {
224 if (spa == target && removing)
225 continue;
226
227 mutex_enter(&spa->spa_props_lock);
228 tdp = list_head(&spa->spa_config_list);
229 if (spa->spa_config == NULL ||
230 tdp->scd_path == NULL ||
231 strcmp(tdp->scd_path, dp->scd_path) != 0) {
232 mutex_exit(&spa->spa_props_lock);
233 continue;
234 }
235
236 if (nvl == NULL)
237 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
238 KM_SLEEP) == 0);
239
240 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
241 spa->spa_config) == 0);
242 mutex_exit(&spa->spa_props_lock);
243 }
244
245 spa_config_write(dp, nvl);
246 nvlist_free(nvl);
247 }
248
249 /*
250 * Remove any config entries older than the current one.
251 */
252 dp = list_head(&target->spa_config_list);
253 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
254 list_remove(&target->spa_config_list, tdp);
255 if (tdp->scd_path != NULL)
256 spa_strfree(tdp->scd_path);
257 kmem_free(tdp, sizeof (spa_config_dirent_t));
258 }
259
260 spa_config_generation++;
261
262 if (postsysevent)
263 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
264 }
265
266 /*
267 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
268 * and we don't want to allow the local zone to see all the pools anyway.
269 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
270 * information for all pool visible within the zone.
271 */
272 nvlist_t *
273 spa_all_configs(uint64_t *generation)
274 {
275 nvlist_t *pools;
276 spa_t *spa = NULL;
277
278 if (*generation == spa_config_generation)
279 return (NULL);
280
281 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
282
283 mutex_enter(&spa_namespace_lock);
284 while ((spa = spa_next(spa)) != NULL) {
285 if (INGLOBALZONE(curproc) ||
286 zone_dataset_visible(spa_name(spa), NULL)) {
287 mutex_enter(&spa->spa_props_lock);
288 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
289 spa->spa_config) == 0);
290 mutex_exit(&spa->spa_props_lock);
291 }
292 }
293 *generation = spa_config_generation;
294 mutex_exit(&spa_namespace_lock);
295
296 return (pools);
297 }
298
299 void
300 spa_config_set(spa_t *spa, nvlist_t *config)
301 {
302 mutex_enter(&spa->spa_props_lock);
303 if (spa->spa_config != NULL)
304 nvlist_free(spa->spa_config);
305 spa->spa_config = config;
306 mutex_exit(&spa->spa_props_lock);
307 }
308
309 /*
310 * Generate the pool's configuration based on the current in-core state.
311 * We infer whether to generate a complete config or just one top-level config
312 * based on whether vd is the root vdev.
313 */
314 nvlist_t *
315 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
316 {
317 nvlist_t *config, *nvroot;
318 vdev_t *rvd = spa->spa_root_vdev;
319 unsigned long hostid = 0;
320 boolean_t locked = B_FALSE;
321 uint64_t split_guid;
322
323 if (vd == NULL) {
324 vd = rvd;
325 locked = B_TRUE;
326 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
327 }
328
329 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
330 (SCL_CONFIG | SCL_STATE));
331
332 /*
333 * If txg is -1, report the current value of spa->spa_config_txg.
334 */
335 if (txg == -1ULL)
336 txg = spa->spa_config_txg;
337
338 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
339
340 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
341 spa_version(spa)) == 0);
342 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
343 spa_name(spa)) == 0);
344 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
345 spa_state(spa)) == 0);
346 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
347 txg) == 0);
348 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
349 spa_guid(spa)) == 0);
350 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
351 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
352
353
354 #ifdef _KERNEL
355 hostid = zone_get_hostid(NULL);
356 #else /* _KERNEL */
357 /*
358 * We're emulating the system's hostid in userland, so we can't use
359 * zone_get_hostid().
360 */
361 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
362 #endif /* _KERNEL */
363 if (hostid != 0) {
364 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
365 hostid) == 0);
366 }
367 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
368 utsname.nodename) == 0);
369
370 if (vd != rvd) {
371 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
372 vd->vdev_top->vdev_guid) == 0);
373 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
374 vd->vdev_guid) == 0);
375 if (vd->vdev_isspare)
376 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
377 1ULL) == 0);
378 if (vd->vdev_islog)
379 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
380 1ULL) == 0);
381 vd = vd->vdev_top; /* label contains top config */
382 } else {
383 /*
384 * Only add the (potentially large) split information
385 * in the mos config, and not in the vdev labels
386 */
387 if (spa->spa_config_splitting != NULL)
388 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
389 spa->spa_config_splitting) == 0);
390 }
391
392 /*
393 * Add the top-level config. We even add this on pools which
394 * don't support holes in the namespace.
395 */
396 vdev_top_config_generate(spa, config);
397
398 /*
399 * If we're splitting, record the original pool's guid.
400 */
401 if (spa->spa_config_splitting != NULL &&
402 nvlist_lookup_uint64(spa->spa_config_splitting,
403 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
404 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
405 split_guid) == 0);
406 }
407
408 nvroot = vdev_config_generate(spa, vd, getstats, 0);
409 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
410 nvlist_free(nvroot);
411
412 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
413 ddt_histogram_t *ddh;
414 ddt_stat_t *dds;
415 ddt_object_t *ddo;
416
417 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
418 ddt_get_dedup_histogram(spa, ddh);
419 VERIFY(nvlist_add_uint64_array(config,
420 ZPOOL_CONFIG_DDT_HISTOGRAM,
421 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
422 kmem_free(ddh, sizeof (ddt_histogram_t));
423
424 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
425 ddt_get_dedup_object_stats(spa, ddo);
426 VERIFY(nvlist_add_uint64_array(config,
427 ZPOOL_CONFIG_DDT_OBJ_STATS,
428 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
429 kmem_free(ddo, sizeof (ddt_object_t));
430
431 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
432 ddt_get_dedup_stats(spa, dds);
433 VERIFY(nvlist_add_uint64_array(config,
434 ZPOOL_CONFIG_DDT_STATS,
435 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
436 kmem_free(dds, sizeof (ddt_stat_t));
437 }
438
439 if (locked)
440 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
441
442 return (config);
443 }
444
445 /*
446 * Update all disk labels, generate a fresh config based on the current
447 * in-core state, and sync the global config cache (do not sync the config
448 * cache if this is a booting rootpool).
449 */
450 void
451 spa_config_update(spa_t *spa, int what)
452 {
453 vdev_t *rvd = spa->spa_root_vdev;
454 uint64_t txg;
455 int c;
456
457 ASSERT(MUTEX_HELD(&spa_namespace_lock));
458
459 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
460 txg = spa_last_synced_txg(spa) + 1;
461 if (what == SPA_CONFIG_UPDATE_POOL) {
462 vdev_config_dirty(rvd);
463 } else {
464 /*
465 * If we have top-level vdevs that were added but have
466 * not yet been prepared for allocation, do that now.
467 * (It's safe now because the config cache is up to date,
468 * so it will be able to translate the new DVAs.)
469 * See comments in spa_vdev_add() for full details.
470 */
471 for (c = 0; c < rvd->vdev_children; c++) {
472 vdev_t *tvd = rvd->vdev_child[c];
473 if (tvd->vdev_ms_array == 0)
474 vdev_metaslab_set_size(tvd);
475 vdev_expand(tvd, txg);
476 }
477 }
478 spa_config_exit(spa, SCL_ALL, FTAG);
479
480 /*
481 * Wait for the mosconfig to be regenerated and synced.
482 */
483 txg_wait_synced(spa->spa_dsl_pool, txg);
484
485 /*
486 * Update the global config cache to reflect the new mosconfig.
487 */
488 if (!spa->spa_is_root)
489 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
490
491 if (what == SPA_CONFIG_UPDATE_POOL)
492 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
493 }