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/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/systeminfo.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
62 static 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 */
68 const 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 */
75 void
76 spa_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%s",
91 (rootdir != NULL) ? "./" : "", spa_config_path);
92
93 file = kobj_open_file(pathname);
94
95 kmem_free(pathname, MAXPATHLEN);
96
97 if (file == (struct _buf *)-1)
98 return;
99
100 if (kobj_get_filesize(file, &fsize) != 0)
101 goto out;
102
103 buf = kmem_alloc(fsize, KM_SLEEP);
104
105 /*
106 * Read the nvlist from the file.
107 */
108 if (kobj_read_file(file, buf, fsize, 0) < 0)
109 goto out;
110
111 /*
112 * Unpack the nvlist.
113 */
114 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
115 goto out;
116
117 /*
118 * Iterate over all elements in the nvlist, creating a new spa_t for
119 * each one with the specified configuration.
120 */
121 mutex_enter(&spa_namespace_lock);
122 nvpair = NULL;
123 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
124 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
125 continue;
126
127 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
128
129 if (spa_lookup(nvpair_name(nvpair)) != NULL)
130 continue;
131 (void) spa_add(nvpair_name(nvpair), child, NULL);
132 }
133 mutex_exit(&spa_namespace_lock);
134
135 nvlist_free(nvlist);
136
137 out:
138 if (buf != NULL)
139 kmem_free(buf, fsize);
140
141 kobj_close_file(file);
142 }
143
144 static int
145 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
146 {
147 size_t buflen;
148 char *buf;
149 vnode_t *vp;
150 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
151 char *temp;
152 int err;
153
154 /*
155 * If the nvlist is empty (NULL), then remove the old cachefile.
156 */
157 if (nvl == NULL) {
158 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
159 return (err);
160 }
161
162 /*
163 * Pack the configuration into a buffer.
164 */
165 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
166
167 buf = kmem_alloc(buflen, KM_SLEEP);
168 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
169
170 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
171 KM_SLEEP) == 0);
172
173 /*
174 * Write the configuration to disk. We need to do the traditional
175 * 'write to temporary file, sync, move over original' to make sure we
176 * always have a consistent view of the data.
177 */
178 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
179
180 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
181 if (err == 0) {
182 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
183 0, RLIM64_INFINITY, kcred, NULL);
184 if (err == 0)
185 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
186 if (err == 0)
187 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
188 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
189 VN_RELE(vp);
190 }
191
192 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
193
194 kmem_free(buf, buflen);
195 kmem_free(temp, MAXPATHLEN);
196 return (err);
197 }
198
199 /*
200 * Synchronize pool configuration to disk. This must be called with the
201 * namespace lock held.
202 */
203 void
204 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
205 {
206 spa_config_dirent_t *dp, *tdp;
207 nvlist_t *nvl;
208 boolean_t ccw_failure;
209 int error;
210
211 ASSERT(MUTEX_HELD(&spa_namespace_lock));
212
213 if (rootdir == NULL || !(spa_mode_global & FWRITE))
214 return;
215
216 /*
217 * Iterate over all cachefiles for the pool, past or present. When the
218 * cachefile is changed, the new one is pushed onto this list, allowing
219 * us to update previous cachefiles that no longer contain this pool.
220 */
221 ccw_failure = B_FALSE;
222 for (dp = list_head(&target->spa_config_list); dp != NULL;
223 dp = list_next(&target->spa_config_list, dp)) {
224 spa_t *spa = NULL;
225 if (dp->scd_path == NULL)
226 continue;
227
228 /*
229 * Iterate over all pools, adding any matching pools to 'nvl'.
230 */
231 nvl = NULL;
232 while ((spa = spa_next(spa)) != NULL) {
233 /*
234 * Skip over our own pool if we're about to remove
235 * ourselves from the spa namespace or any pool that
236 * is readonly. Since we cannot guarantee that a
237 * readonly pool would successfully import upon reboot,
238 * we don't allow them to be written to the cache file.
239 */
240 if ((spa == target && removing) ||
241 !spa_writeable(spa))
242 continue;
243
244 mutex_enter(&spa->spa_props_lock);
245 tdp = list_head(&spa->spa_config_list);
246 if (spa->spa_config == NULL ||
247 tdp->scd_path == NULL ||
248 strcmp(tdp->scd_path, dp->scd_path) != 0) {
249 mutex_exit(&spa->spa_props_lock);
250 continue;
251 }
252
253 if (nvl == NULL)
254 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
255 KM_SLEEP) == 0);
256
257 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
258 spa->spa_config) == 0);
259 mutex_exit(&spa->spa_props_lock);
260 }
261
262 error = spa_config_write(dp, nvl);
263 if (error != 0)
264 ccw_failure = B_TRUE;
265 nvlist_free(nvl);
266 }
267
268 if (ccw_failure) {
269 /*
270 * Keep trying so that configuration data is
271 * written if/when any temporary filesystem
272 * resource issues are resolved.
273 */
274 if (target->spa_ccw_fail_time == 0) {
275 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
276 target, NULL, NULL, 0, 0);
277 }
278 target->spa_ccw_fail_time = gethrtime();
279 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
280 } else {
281 /*
282 * Do not rate limit future attempts to update
283 * the config cache.
284 */
285 target->spa_ccw_fail_time = 0;
286 }
287
288 /*
289 * Remove any config entries older than the current one.
290 */
291 dp = list_head(&target->spa_config_list);
292 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
293 list_remove(&target->spa_config_list, tdp);
294 if (tdp->scd_path != NULL)
295 spa_strfree(tdp->scd_path);
296 kmem_free(tdp, sizeof (spa_config_dirent_t));
297 }
298
299 spa_config_generation++;
300
301 if (postsysevent)
302 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
303 }
304
305 /*
306 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
307 * and we don't want to allow the local zone to see all the pools anyway.
308 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
309 * information for all pool visible within the zone.
310 */
311 nvlist_t *
312 spa_all_configs(uint64_t *generation)
313 {
314 nvlist_t *pools;
315 spa_t *spa = NULL;
316
317 if (*generation == spa_config_generation)
318 return (NULL);
319
320 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
321
322 mutex_enter(&spa_namespace_lock);
323 while ((spa = spa_next(spa)) != NULL) {
324 if (INGLOBALZONE(curproc) ||
325 zone_dataset_visible(spa_name(spa), NULL)) {
326 mutex_enter(&spa->spa_props_lock);
327 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
328 spa->spa_config) == 0);
329 mutex_exit(&spa->spa_props_lock);
330 }
331 }
332 *generation = spa_config_generation;
333 mutex_exit(&spa_namespace_lock);
334
335 return (pools);
336 }
337
338 void
339 spa_config_set(spa_t *spa, nvlist_t *config)
340 {
341 mutex_enter(&spa->spa_props_lock);
342 if (spa->spa_config != NULL)
343 nvlist_free(spa->spa_config);
344 spa->spa_config = config;
345 mutex_exit(&spa->spa_props_lock);
346 }
347
348 /*
349 * Generate the pool's configuration based on the current in-core state.
350 * We infer whether to generate a complete config or just one top-level config
351 * based on whether vd is the root vdev.
352 */
353 nvlist_t *
354 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
355 {
356 nvlist_t *config, *nvroot;
357 vdev_t *rvd = spa->spa_root_vdev;
358 unsigned long hostid = 0;
359 boolean_t locked = B_FALSE;
360 uint64_t split_guid;
361
362 if (vd == NULL) {
363 vd = rvd;
364 locked = B_TRUE;
365 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
366 }
367
368 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
369 (SCL_CONFIG | SCL_STATE));
370
371 /*
372 * If txg is -1, report the current value of spa->spa_config_txg.
373 */
374 if (txg == -1ULL)
375 txg = spa->spa_config_txg;
376
377 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
378
379 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
380 spa_version(spa)) == 0);
381 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
382 spa_name(spa)) == 0);
383 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
384 spa_state(spa)) == 0);
385 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
386 txg) == 0);
387 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
388 spa_guid(spa)) == 0);
389 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
390 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
391
392
393 #ifdef _KERNEL
394 hostid = zone_get_hostid(NULL);
395 #else /* _KERNEL */
396 /*
397 * We're emulating the system's hostid in userland, so we can't use
398 * zone_get_hostid().
399 */
400 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
401 #endif /* _KERNEL */
402 if (hostid != 0) {
403 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
404 hostid) == 0);
405 }
406 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
407 utsname.nodename) == 0);
408
409 if (vd != rvd) {
410 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
411 vd->vdev_top->vdev_guid) == 0);
412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
413 vd->vdev_guid) == 0);
414 if (vd->vdev_isspare)
415 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
416 1ULL) == 0);
417 if (vd->vdev_islog)
418 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
419 1ULL) == 0);
420 vd = vd->vdev_top; /* label contains top config */
421 } else {
422 /*
423 * Only add the (potentially large) split information
424 * in the mos config, and not in the vdev labels
425 */
426 if (spa->spa_config_splitting != NULL)
427 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
428 spa->spa_config_splitting) == 0);
429 }
430
431 /*
432 * Add the top-level config. We even add this on pools which
433 * don't support holes in the namespace.
434 */
435 vdev_top_config_generate(spa, config);
436
437 /*
438 * If we're splitting, record the original pool's guid.
439 */
440 if (spa->spa_config_splitting != NULL &&
441 nvlist_lookup_uint64(spa->spa_config_splitting,
442 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
443 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
444 split_guid) == 0);
445 }
446
447 nvroot = vdev_config_generate(spa, vd, getstats, 0);
448 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
449 nvlist_free(nvroot);
450
451 /*
452 * Store what's necessary for reading the MOS in the label.
453 */
454 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
455 spa->spa_label_features) == 0);
456
457 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
458 ddt_histogram_t *ddh;
459 ddt_stat_t *dds;
460 ddt_object_t *ddo;
461
462 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
463 ddt_get_dedup_histogram(spa, ddh);
464 VERIFY(nvlist_add_uint64_array(config,
465 ZPOOL_CONFIG_DDT_HISTOGRAM,
466 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
467 kmem_free(ddh, sizeof (ddt_histogram_t));
468
469 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
470 ddt_get_dedup_object_stats(spa, ddo);
471 VERIFY(nvlist_add_uint64_array(config,
472 ZPOOL_CONFIG_DDT_OBJ_STATS,
473 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
474 kmem_free(ddo, sizeof (ddt_object_t));
475
476 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
477 ddt_get_dedup_stats(spa, dds);
478 VERIFY(nvlist_add_uint64_array(config,
479 ZPOOL_CONFIG_DDT_STATS,
480 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
481 kmem_free(dds, sizeof (ddt_stat_t));
482 }
483
484 if (locked)
485 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
486
487 return (config);
488 }
489
490 /*
491 * Update all disk labels, generate a fresh config based on the current
492 * in-core state, and sync the global config cache (do not sync the config
493 * cache if this is a booting rootpool).
494 */
495 void
496 spa_config_update(spa_t *spa, int what)
497 {
498 vdev_t *rvd = spa->spa_root_vdev;
499 uint64_t txg;
500 int c;
501
502 ASSERT(MUTEX_HELD(&spa_namespace_lock));
503
504 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
505 txg = spa_last_synced_txg(spa) + 1;
506 if (what == SPA_CONFIG_UPDATE_POOL) {
507 vdev_config_dirty(rvd);
508 } else {
509 /*
510 * If we have top-level vdevs that were added but have
511 * not yet been prepared for allocation, do that now.
512 * (It's safe now because the config cache is up to date,
513 * so it will be able to translate the new DVAs.)
514 * See comments in spa_vdev_add() for full details.
515 */
516 for (c = 0; c < rvd->vdev_children; c++) {
517 vdev_t *tvd = rvd->vdev_child[c];
518 if (tvd->vdev_ms_array == 0)
519 vdev_metaslab_set_size(tvd);
520 vdev_expand(tvd, txg);
521 }
522 }
523 spa_config_exit(spa, SCL_ALL, FTAG);
524
525 /*
526 * Wait for the mosconfig to be regenerated and synced.
527 */
528 txg_wait_synced(spa->spa_dsl_pool, txg);
529
530 /*
531 * Update the global config cache to reflect the new mosconfig.
532 */
533 if (!spa->spa_is_root)
534 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
535
536 if (what == SPA_CONFIG_UPDATE_POOL)
537 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
538 }