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/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 #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
61 static 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 */
67 const 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 */
74 void
75 spa_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%s",
90 (rootdir != NULL) ? "./" : "", 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
136 out:
137 if (buf != NULL)
138 kmem_free(buf, fsize);
139
140 kobj_close_file(file);
141 }
142
143 static void
144 spa_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
152 /*
153 * If the nvlist is empty (NULL), then remove the old cachefile.
154 */
155 if (nvl == NULL) {
156 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
157 return;
158 }
159
160 /*
161 * Pack the configuration into a buffer.
162 */
163 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
164
165 buf = kmem_alloc(buflen, KM_SLEEP);
166 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
167
168 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
169 KM_SLEEP) == 0);
170
171 /*
172 * Write the configuration to disk. We need to do the traditional
173 * 'write to temporary file, sync, move over original' to make sure we
174 * always have a consistent view of the data.
175 */
176 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
177
178 if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
179 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
180 0, RLIM64_INFINITY, kcred, NULL) == 0 &&
181 VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
182 (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
183 }
184 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
185 VN_RELE(vp);
186 }
187
188 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
189
190 kmem_free(buf, buflen);
191 kmem_free(temp, MAXPATHLEN);
192 }
193
194 /*
195 * Synchronize pool configuration to disk. This must be called with the
196 * namespace lock held.
197 */
198 void
199 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
200 {
201 spa_config_dirent_t *dp, *tdp;
202 nvlist_t *nvl;
203
204 ASSERT(MUTEX_HELD(&spa_namespace_lock));
205
206 if (rootdir == NULL || !(spa_mode_global & FWRITE))
207 return;
208
209 /*
210 * Iterate over all cachefiles for the pool, past or present. When the
211 * cachefile is changed, the new one is pushed onto this list, allowing
212 * us to update previous cachefiles that no longer contain this pool.
213 */
214 for (dp = list_head(&target->spa_config_list); dp != NULL;
215 dp = list_next(&target->spa_config_list, dp)) {
216 spa_t *spa = NULL;
217 if (dp->scd_path == NULL)
218 continue;
219
220 /*
221 * Iterate over all pools, adding any matching pools to 'nvl'.
222 */
223 nvl = NULL;
224 while ((spa = spa_next(spa)) != NULL) {
225 /*
226 * Skip over our own pool if we're about to remove
227 * ourselves from the spa namespace or any pool that
228 * is readonly. Since we cannot guarantee that a
229 * readonly pool would successfully import upon reboot,
230 * we don't allow them to be written to the cache file.
231 */
232 if ((spa == target && removing) ||
233 !spa_writeable(spa))
234 continue;
235
236 mutex_enter(&spa->spa_props_lock);
237 tdp = list_head(&spa->spa_config_list);
238 if (spa->spa_config == NULL ||
239 tdp->scd_path == NULL ||
240 strcmp(tdp->scd_path, dp->scd_path) != 0) {
241 mutex_exit(&spa->spa_props_lock);
242 continue;
243 }
244
245 if (nvl == NULL)
246 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
247 KM_SLEEP) == 0);
248
249 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
250 spa->spa_config) == 0);
251 mutex_exit(&spa->spa_props_lock);
252 }
253
254 spa_config_write(dp, nvl);
255 nvlist_free(nvl);
256 }
257
258 /*
259 * Remove any config entries older than the current one.
260 */
261 dp = list_head(&target->spa_config_list);
262 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
263 list_remove(&target->spa_config_list, tdp);
264 if (tdp->scd_path != NULL)
265 spa_strfree(tdp->scd_path);
266 kmem_free(tdp, sizeof (spa_config_dirent_t));
267 }
268
269 spa_config_generation++;
270
271 if (postsysevent)
272 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
273 }
274
275 /*
276 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
277 * and we don't want to allow the local zone to see all the pools anyway.
278 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
279 * information for all pool visible within the zone.
280 */
281 nvlist_t *
282 spa_all_configs(uint64_t *generation)
283 {
284 nvlist_t *pools;
285 spa_t *spa = NULL;
286
287 if (*generation == spa_config_generation)
288 return (NULL);
289
290 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
291
292 mutex_enter(&spa_namespace_lock);
293 while ((spa = spa_next(spa)) != NULL) {
294 if (INGLOBALZONE(curproc) ||
295 zone_dataset_visible(spa_name(spa), NULL)) {
296 mutex_enter(&spa->spa_props_lock);
297 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
298 spa->spa_config) == 0);
299 mutex_exit(&spa->spa_props_lock);
300 }
301 }
302 *generation = spa_config_generation;
303 mutex_exit(&spa_namespace_lock);
304
305 return (pools);
306 }
307
308 void
309 spa_config_set(spa_t *spa, nvlist_t *config)
310 {
311 mutex_enter(&spa->spa_props_lock);
312 if (spa->spa_config != NULL)
313 nvlist_free(spa->spa_config);
314 spa->spa_config = config;
315 mutex_exit(&spa->spa_props_lock);
316 }
317
318 /*
319 * Generate the pool's configuration based on the current in-core state.
320 *
321 * We infer whether to generate a complete config or just one top-level config
322 * based on whether vd is the root vdev.
323 */
324 nvlist_t *
325 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
326 {
327 nvlist_t *config, *nvroot;
328 vdev_t *rvd = spa->spa_root_vdev;
329 unsigned long hostid = 0;
330 boolean_t locked = B_FALSE;
331 uint64_t split_guid;
332
333 if (vd == NULL) {
334 vd = rvd;
335 locked = B_TRUE;
336 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
337 }
338
339 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
340 (SCL_CONFIG | SCL_STATE));
341
342 /*
343 * If txg is -1, report the current value of spa->spa_config_txg.
344 */
345 if (txg == -1ULL)
346 txg = spa->spa_config_txg;
347
348 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
349
350 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
351 spa_version(spa)) == 0);
352 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
353 spa_name(spa)) == 0);
354 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
355 spa_state(spa)) == 0);
356 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
357 txg) == 0);
358 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
359 spa_guid(spa)) == 0);
360 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
361 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
362
363
364 #ifdef _KERNEL
365 hostid = zone_get_hostid(NULL);
366 #else /* _KERNEL */
367 /*
368 * We're emulating the system's hostid in userland, so we can't use
369 * zone_get_hostid().
370 */
371 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
372 #endif /* _KERNEL */
373 if (hostid != 0) {
374 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
375 hostid) == 0);
376 }
377 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
378 utsname.nodename) == 0);
379
380 if (vd != rvd) {
381 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
382 vd->vdev_top->vdev_guid) == 0);
383 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
384 vd->vdev_guid) == 0);
385 if (vd->vdev_isspare)
386 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
387 1ULL) == 0);
388 if (vd->vdev_islog)
389 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
390 1ULL) == 0);
391 vd = vd->vdev_top; /* label contains top config */
392 } else {
393 /*
394 * Only add the (potentially large) split information
395 * in the mos config, and not in the vdev labels
396 */
397 if (spa->spa_config_splitting != NULL)
398 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
399 spa->spa_config_splitting) == 0);
400 }
401
402 /*
403 * Add the top-level config. We even add this on pools which
404 * don't support holes in the namespace.
405 */
406 vdev_top_config_generate(spa, config);
407
408 /*
409 * If we're splitting, record the original pool's guid.
410 */
411 if (spa->spa_config_splitting != NULL &&
412 nvlist_lookup_uint64(spa->spa_config_splitting,
413 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
415 split_guid) == 0);
416 }
417
418 nvroot = vdev_config_generate(spa, vd, getstats, 0);
419 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
420 nvlist_free(nvroot);
421
422 /*
423 * Store what's necessary for reading the MOS in the label.
424 */
425 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
426 spa->spa_label_features) == 0);
427
428 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
429 ddt_histogram_t *ddh;
430 ddt_stat_t *dds;
431 ddt_object_t *ddo;
432
433 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
434 ddt_get_dedup_histogram(spa, ddh);
435 VERIFY(nvlist_add_uint64_array(config,
436 ZPOOL_CONFIG_DDT_HISTOGRAM,
437 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
438 kmem_free(ddh, sizeof (ddt_histogram_t));
439
440 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
441 ddt_get_dedup_object_stats(spa, ddo);
442 VERIFY(nvlist_add_uint64_array(config,
443 ZPOOL_CONFIG_DDT_OBJ_STATS,
444 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
445 kmem_free(ddo, sizeof (ddt_object_t));
446
447 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
448 ddt_get_dedup_stats(spa, dds);
449 VERIFY(nvlist_add_uint64_array(config,
450 ZPOOL_CONFIG_DDT_STATS,
451 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
452 kmem_free(dds, sizeof (ddt_stat_t));
453 }
454
455 if (locked)
456 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
457
458 return (config);
459 }
460
461 /*
462 * Update all disk labels, generate a fresh config based on the current
463 * in-core state, and sync the global config cache (do not sync the config
464 * cache if this is a booting rootpool).
465 */
466 void
467 spa_config_update(spa_t *spa, int what)
468 {
469 vdev_t *rvd = spa->spa_root_vdev;
470 uint64_t txg;
471 int c;
472
473 ASSERT(MUTEX_HELD(&spa_namespace_lock));
474
475 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
476 txg = spa_last_synced_txg(spa) + 1;
477 if (what == SPA_CONFIG_UPDATE_POOL) {
478 vdev_config_dirty(rvd);
479 } else {
480 /*
481 * If we have top-level vdevs that were added but have
482 * not yet been prepared for allocation, do that now.
483 * (It's safe now because the config cache is up to date,
484 * so it will be able to translate the new DVAs.)
485 * See comments in spa_vdev_add() for full details.
486 */
487 for (c = 0; c < rvd->vdev_children; c++) {
488 vdev_t *tvd = rvd->vdev_child[c];
489 if (tvd->vdev_ms_array == 0)
490 vdev_metaslab_set_size(tvd);
491 vdev_expand(tvd, txg);
492 }
493 }
494 spa_config_exit(spa, SCL_ALL, FTAG);
495
496 /*
497 * Wait for the mosconfig to be regenerated and synced.
498 */
499 txg_wait_synced(spa->spa_dsl_pool, txg);
500
501 /*
502 * Update the global config cache to reflect the new mosconfig.
503 */
504 if (!spa->spa_is_root)
505 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
506
507 if (what == SPA_CONFIG_UPDATE_POOL)
508 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
509 }