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) 2013 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>
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
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
spa_config_load(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", 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
spa_config_clean(nvlist_t * nvl)144 spa_config_clean(nvlist_t *nvl)
145 {
146 nvlist_t **child;
147 nvlist_t *nvroot = NULL;
148 uint_t c, children;
149
150 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
151 &children) == 0) {
152 for (c = 0; c < children; c++)
153 spa_config_clean(child[c]);
154 }
155
156 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0)
157 spa_config_clean(nvroot);
158
159 nvlist_remove(nvl, ZPOOL_CONFIG_VDEV_STATS, DATA_TYPE_UINT64_ARRAY);
160 nvlist_remove(nvl, ZPOOL_CONFIG_SCAN_STATS, DATA_TYPE_UINT64_ARRAY);
161 }
162
163 static int
spa_config_write(spa_config_dirent_t * dp,nvlist_t * nvl)164 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
165 {
166 size_t buflen;
167 char *buf;
168 vnode_t *vp;
169 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
170 char *temp;
171 int err;
172
173 /*
174 * If the nvlist is empty (NULL), then remove the old cachefile.
175 */
176 if (nvl == NULL) {
177 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
178 return (err);
179 }
180
181 /*
182 * Pack the configuration into a buffer.
183 */
184 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
185
186 buf = kmem_alloc(buflen, KM_SLEEP);
187 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
188
189 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
190 KM_SLEEP) == 0);
191
192 /*
193 * Write the configuration to disk. We need to do the traditional
194 * 'write to temporary file, sync, move over original' to make sure we
195 * always have a consistent view of the data.
196 */
197 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
198
199 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
200 if (err == 0) {
201 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
202 0, RLIM64_INFINITY, kcred, NULL);
203 if (err == 0)
204 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
205 if (err == 0)
206 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
207 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
208 }
209
210 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
211
212 kmem_free(buf, buflen);
213 kmem_free(temp, MAXPATHLEN);
214 return (err);
215 }
216
217 /*
218 * Synchronize pool configuration to disk. This must be called with the
219 * namespace lock held. Synchronizing the pool cache is typically done after
220 * the configuration has been synced to the MOS. This exposes a window where
221 * the MOS config will have been updated but the cache file has not. If
222 * the system were to crash at that instant then the cached config may not
223 * contain the correct information to open the pool and an explicity import
224 * would be required.
225 */
226 void
spa_config_sync(spa_t * target,boolean_t removing,boolean_t postsysevent)227 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
228 {
229 spa_config_dirent_t *dp, *tdp;
230 nvlist_t *nvl;
231 boolean_t ccw_failure;
232 int error;
233
234 ASSERT(MUTEX_HELD(&spa_namespace_lock));
235
236 if (rootdir == NULL || !(spa_mode_global & FWRITE))
237 return;
238
239 /*
240 * Iterate over all cachefiles for the pool, past or present. When the
241 * cachefile is changed, the new one is pushed onto this list, allowing
242 * us to update previous cachefiles that no longer contain this pool.
243 */
244 ccw_failure = B_FALSE;
245 for (dp = list_head(&target->spa_config_list); dp != NULL;
246 dp = list_next(&target->spa_config_list, dp)) {
247 spa_t *spa = NULL;
248 if (dp->scd_path == NULL)
249 continue;
250
251 /*
252 * Iterate over all pools, adding any matching pools to 'nvl'.
253 */
254 nvl = NULL;
255 while ((spa = spa_next(spa)) != NULL) {
256 nvlist_t *nvroot = NULL;
257 /*
258 * Skip over our own pool if we're about to remove
259 * ourselves from the spa namespace or any pool that
260 * is readonly. Since we cannot guarantee that a
261 * readonly pool would successfully import upon reboot,
262 * we don't allow them to be written to the cache file.
263 */
264 if ((spa == target && removing) ||
265 (spa_state(spa) == POOL_STATE_ACTIVE &&
266 !spa_writeable(spa)))
267 continue;
268
269 mutex_enter(&spa->spa_props_lock);
270 tdp = list_head(&spa->spa_config_list);
271 if (spa->spa_config == NULL ||
272 tdp->scd_path == NULL ||
273 strcmp(tdp->scd_path, dp->scd_path) != 0) {
274 mutex_exit(&spa->spa_props_lock);
275 continue;
276 }
277
278 if (nvl == NULL)
279 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
280 KM_SLEEP) == 0);
281
282 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
283 spa->spa_config) == 0);
284 mutex_exit(&spa->spa_props_lock);
285
286 if (nvlist_lookup_nvlist(nvl, spa->spa_name, &nvroot) == 0)
287 spa_config_clean(nvroot);
288 }
289
290 error = spa_config_write(dp, nvl);
291 if (error != 0)
292 ccw_failure = B_TRUE;
293 nvlist_free(nvl);
294 }
295
296 if (ccw_failure) {
297 /*
298 * Keep trying so that configuration data is
299 * written if/when any temporary filesystem
300 * resource issues are resolved.
301 */
302 if (target->spa_ccw_fail_time == 0) {
303 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
304 target, NULL, NULL, 0, 0);
305 }
306 target->spa_ccw_fail_time = gethrtime();
307 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
308 } else {
309 /*
310 * Do not rate limit future attempts to update
311 * the config cache.
312 */
313 target->spa_ccw_fail_time = 0;
314 }
315
316 /*
317 * Remove any config entries older than the current one.
318 */
319 dp = list_head(&target->spa_config_list);
320 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
321 list_remove(&target->spa_config_list, tdp);
322 if (tdp->scd_path != NULL)
323 spa_strfree(tdp->scd_path);
324 kmem_free(tdp, sizeof (spa_config_dirent_t));
325 }
326
327 spa_config_generation++;
328
329 if (postsysevent)
330 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
331 }
332
333 /*
334 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
335 * and we don't want to allow the local zone to see all the pools anyway.
336 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
337 * information for all pool visible within the zone.
338 */
339 nvlist_t *
spa_all_configs(uint64_t * generation)340 spa_all_configs(uint64_t *generation)
341 {
342 nvlist_t *pools;
343 spa_t *spa = NULL;
344
345 if (*generation == spa_config_generation)
346 return (NULL);
347
348 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
349
350 mutex_enter(&spa_namespace_lock);
351 while ((spa = spa_next(spa)) != NULL) {
352 if (INGLOBALZONE(curthread) ||
353 zone_dataset_visible(spa_name(spa), NULL)) {
354 mutex_enter(&spa->spa_props_lock);
355 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
356 spa->spa_config) == 0);
357 mutex_exit(&spa->spa_props_lock);
358 }
359 }
360 *generation = spa_config_generation;
361 mutex_exit(&spa_namespace_lock);
362
363 return (pools);
364 }
365
366 void
spa_config_set(spa_t * spa,nvlist_t * config)367 spa_config_set(spa_t *spa, nvlist_t *config)
368 {
369 mutex_enter(&spa->spa_props_lock);
370 if (spa->spa_config != NULL)
371 nvlist_free(spa->spa_config);
372 spa->spa_config = config;
373 mutex_exit(&spa->spa_props_lock);
374 }
375
376 /*
377 * Generate the pool's configuration based on the current in-core state.
378 *
379 * We infer whether to generate a complete config or just one top-level config
380 * based on whether vd is the root vdev.
381 */
382 nvlist_t *
spa_config_generate(spa_t * spa,vdev_t * vd,uint64_t txg,int getstats)383 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
384 {
385 nvlist_t *config, *nvroot;
386 vdev_t *rvd = spa->spa_root_vdev;
387 unsigned long hostid = 0;
388 boolean_t locked = B_FALSE;
389 uint64_t split_guid;
390
391 if (vd == NULL) {
392 vd = rvd;
393 locked = B_TRUE;
394 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
395 }
396
397 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
398 (SCL_CONFIG | SCL_STATE));
399
400 /*
401 * If txg is -1, report the current value of spa->spa_config_txg.
402 */
403 if (txg == -1ULL)
404 txg = spa->spa_config_txg;
405
406 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
407
408 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
409 spa_version(spa)) == 0);
410 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
411 spa_name(spa)) == 0);
412 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
413 spa_state(spa)) == 0);
414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
415 txg) == 0);
416 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
417 spa_guid(spa)) == 0);
418 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
419 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
420
421
422 #ifdef _KERNEL
423 hostid = zone_get_hostid(NULL);
424 #else /* _KERNEL */
425 /*
426 * We're emulating the system's hostid in userland, so we can't use
427 * zone_get_hostid().
428 */
429 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
430 #endif /* _KERNEL */
431 if (hostid != 0) {
432 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
433 hostid) == 0);
434 }
435 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
436 utsname.nodename) == 0);
437
438 if (vd != rvd) {
439 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
440 vd->vdev_top->vdev_guid) == 0);
441 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
442 vd->vdev_guid) == 0);
443 if (vd->vdev_isspare)
444 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
445 1ULL) == 0);
446 if (vd->vdev_islog)
447 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
448 1ULL) == 0);
449 vd = vd->vdev_top; /* label contains top config */
450 } else {
451 /*
452 * Only add the (potentially large) split information
453 * in the mos config, and not in the vdev labels
454 */
455 if (spa->spa_config_splitting != NULL)
456 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
457 spa->spa_config_splitting) == 0);
458 }
459
460 /*
461 * Add the top-level config. We even add this on pools which
462 * don't support holes in the namespace.
463 */
464 vdev_top_config_generate(spa, config);
465
466 /*
467 * If we're splitting, record the original pool's guid.
468 */
469 if (spa->spa_config_splitting != NULL &&
470 nvlist_lookup_uint64(spa->spa_config_splitting,
471 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
472 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
473 split_guid) == 0);
474 }
475
476 nvroot = vdev_config_generate(spa, vd, getstats, 0);
477 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
478 nvlist_free(nvroot);
479
480 /*
481 * Store what's necessary for reading the MOS in the label.
482 */
483 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
484 spa->spa_label_features) == 0);
485
486 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
487 ddt_histogram_t *ddh;
488 ddt_stat_t *dds;
489 ddt_object_t *ddo;
490
491 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
492 ddt_get_dedup_histogram(spa, ddh);
493 VERIFY(nvlist_add_uint64_array(config,
494 ZPOOL_CONFIG_DDT_HISTOGRAM,
495 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
496 kmem_free(ddh, sizeof (ddt_histogram_t));
497
498 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
499 ddt_get_dedup_object_stats(spa, ddo);
500 VERIFY(nvlist_add_uint64_array(config,
501 ZPOOL_CONFIG_DDT_OBJ_STATS,
502 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
503 kmem_free(ddo, sizeof (ddt_object_t));
504
505 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
506 ddt_get_dedup_stats(spa, dds);
507 VERIFY(nvlist_add_uint64_array(config,
508 ZPOOL_CONFIG_DDT_STATS,
509 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
510 kmem_free(dds, sizeof (ddt_stat_t));
511 }
512
513 if (locked)
514 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
515
516 return (config);
517 }
518
519 /*
520 * Update all disk labels, generate a fresh config based on the current
521 * in-core state, and sync the global config cache (do not sync the config
522 * cache if this is a booting rootpool).
523 */
524 void
spa_config_update(spa_t * spa,int what)525 spa_config_update(spa_t *spa, int what)
526 {
527 vdev_t *rvd = spa->spa_root_vdev;
528 uint64_t txg;
529 int c;
530
531 ASSERT(MUTEX_HELD(&spa_namespace_lock));
532
533 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
534 txg = spa_last_synced_txg(spa) + 1;
535 if (what == SPA_CONFIG_UPDATE_POOL) {
536 vdev_config_dirty(rvd);
537 } else {
538 /*
539 * If we have top-level vdevs that were added but have
540 * not yet been prepared for allocation, do that now.
541 * (It's safe now because the config cache is up to date,
542 * so it will be able to translate the new DVAs.)
543 * See comments in spa_vdev_add() for full details.
544 */
545 for (c = 0; c < rvd->vdev_children; c++) {
546 vdev_t *tvd = rvd->vdev_child[c];
547 if (tvd->vdev_ms_array == 0) {
548 vdev_ashift_optimize(tvd);
549 vdev_metaslab_set_size(tvd);
550 }
551 vdev_expand(tvd, txg);
552 }
553 }
554 spa_config_exit(spa, SCL_ALL, FTAG);
555
556 /*
557 * Wait for the mosconfig to be regenerated and synced.
558 */
559 txg_wait_synced(spa->spa_dsl_pool, txg);
560
561 /*
562 * Update the global config cache to reflect the new mosconfig.
563 */
564 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
565
566 if (what == SPA_CONFIG_UPDATE_POOL)
567 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
568 }
569