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, 2018 by Delphix. All rights reserved.
26 * Copyright 2017 Joyent, Inc.
27 */
28
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/systeminfo.h>
39 #include <sys/sunddi.h>
40 #include <sys/zfeature.h>
41 #ifdef _KERNEL
42 #include <sys/kobj.h>
43 #include <sys/zone.h>
44 #endif
45
46 /*
47 * Pool configuration repository.
48 *
49 * Pool configuration is stored as a packed nvlist on the filesystem. By
50 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
51 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
52 * property set that allows them to be stored in an alternate location until
53 * the control of external software.
54 *
55 * For each cache file, we have a single nvlist which holds all the
56 * configuration information. When the module loads, we read this information
57 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
58 * maintained independently in spa.c. Whenever the namespace is modified, or
59 * the configuration of a pool is changed, we call spa_write_cachefile(), which
60 * walks through all the active pools and writes the configuration to disk.
61 */
62
63 static uint64_t spa_config_generation = 1;
64
65 /*
66 * This can be overridden in userland to preserve an alternate namespace for
67 * userland pools when doing testing.
68 */
69 const char *spa_config_path = ZPOOL_CACHE;
70
71 /*
72 * Called when the module is first loaded, this routine loads the configuration
73 * file into the SPA namespace. It does not actually open or load the pools; it
74 * only populates the namespace.
75 */
76 void
spa_config_load(void)77 spa_config_load(void)
78 {
79 void *buf = NULL;
80 nvlist_t *nvlist, *child;
81 nvpair_t *nvpair;
82 char *pathname;
83 struct _buf *file;
84 uint64_t fsize;
85
86 /*
87 * Open the configuration file.
88 */
89 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
90
91 (void) snprintf(pathname, MAXPATHLEN, "%s%s",
92 (rootdir != NULL) ? "./" : "", spa_config_path);
93
94 file = kobj_open_file(pathname);
95
96 kmem_free(pathname, MAXPATHLEN);
97
98 if (file == (struct _buf *)-1)
99 return;
100
101 if (kobj_get_filesize(file, &fsize) != 0)
102 goto out;
103
104 buf = kmem_alloc(fsize, KM_SLEEP);
105
106 /*
107 * Read the nvlist from the file.
108 */
109 if (kobj_read_file(file, buf, fsize, 0) < 0)
110 goto out;
111
112 /*
113 * Unpack the nvlist.
114 */
115 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
116 goto out;
117
118 /*
119 * Iterate over all elements in the nvlist, creating a new spa_t for
120 * each one with the specified configuration.
121 */
122 mutex_enter(&spa_namespace_lock);
123 nvpair = NULL;
124 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
125 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
126 continue;
127
128 child = fnvpair_value_nvlist(nvpair);
129
130 if (spa_lookup(nvpair_name(nvpair)) != NULL)
131 continue;
132 (void) spa_add(nvpair_name(nvpair), child, NULL);
133 }
134 mutex_exit(&spa_namespace_lock);
135
136 nvlist_free(nvlist);
137
138 out:
139 if (buf != NULL)
140 kmem_free(buf, fsize);
141
142 kobj_close_file(file);
143 }
144
145 static int
spa_config_write(spa_config_dirent_t * dp,nvlist_t * nvl)146 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
147 {
148 size_t buflen;
149 char *buf;
150 vnode_t *vp;
151 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
152 char *temp;
153 int err;
154
155 /*
156 * If the nvlist is empty (NULL), then remove the old cachefile.
157 */
158 if (nvl == NULL) {
159 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
160 return (err);
161 }
162
163 /*
164 * Pack the configuration into a buffer.
165 */
166 buf = fnvlist_pack(nvl, &buflen);
167 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
168
169 /*
170 * Write the configuration to disk. We need to do the traditional
171 * 'write to temporary file, sync, move over original' to make sure we
172 * always have a consistent view of the data.
173 */
174 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
175
176 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
177 if (err == 0) {
178 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
179 0, RLIM64_INFINITY, kcred, NULL);
180 if (err == 0)
181 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
182 if (err == 0)
183 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
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 fnvlist_pack_free(buf, buflen);
191 kmem_free(temp, MAXPATHLEN);
192 return (err);
193 }
194
195 /*
196 * Synchronize pool configuration to disk. This must be called with the
197 * namespace lock held. Synchronizing the pool cache is typically done after
198 * the configuration has been synced to the MOS. This exposes a window where
199 * the MOS config will have been updated but the cache file has not. If
200 * the system were to crash at that instant then the cached config may not
201 * contain the correct information to open the pool and an explicit import
202 * would be required.
203 */
204 void
spa_write_cachefile(spa_t * target,boolean_t removing,boolean_t postsysevent)205 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent)
206 {
207 spa_config_dirent_t *dp, *tdp;
208 nvlist_t *nvl;
209 boolean_t ccw_failure;
210 int error;
211 char *pool_name;
212
213 ASSERT(MUTEX_HELD(&spa_namespace_lock));
214
215 if (rootdir == NULL || !(spa_mode_global & FWRITE))
216 return;
217
218 /*
219 * Iterate over all cachefiles for the pool, past or present. When the
220 * cachefile is changed, the new one is pushed onto this list, allowing
221 * us to update previous cachefiles that no longer contain this pool.
222 */
223 ccw_failure = B_FALSE;
224 for (dp = list_head(&target->spa_config_list); dp != NULL;
225 dp = list_next(&target->spa_config_list, dp)) {
226 spa_t *spa = NULL;
227 if (dp->scd_path == NULL)
228 continue;
229
230 /*
231 * Iterate over all pools, adding any matching pools to 'nvl'.
232 */
233 nvl = NULL;
234 while ((spa = spa_next(spa)) != NULL) {
235 /*
236 * Skip over our own pool if we're about to remove
237 * ourselves from the spa namespace or any pool that
238 * is readonly. Since we cannot guarantee that a
239 * readonly pool would successfully import upon reboot,
240 * we don't allow them to be written to the cache file.
241 */
242 if ((spa == target && removing) ||
243 !spa_writeable(spa))
244 continue;
245
246 mutex_enter(&spa->spa_props_lock);
247 tdp = list_head(&spa->spa_config_list);
248 if (spa->spa_config == NULL ||
249 tdp->scd_path == NULL ||
250 strcmp(tdp->scd_path, dp->scd_path) != 0) {
251 mutex_exit(&spa->spa_props_lock);
252 continue;
253 }
254
255 if (nvl == NULL)
256 nvl = fnvlist_alloc();
257
258 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
259 pool_name = fnvlist_lookup_string(
260 spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
261 } else {
262 pool_name = spa_name(spa);
263 }
264
265 fnvlist_add_nvlist(nvl, pool_name,
266 spa->spa_config);
267 mutex_exit(&spa->spa_props_lock);
268 }
269
270 error = spa_config_write(dp, nvl);
271 if (error != 0)
272 ccw_failure = B_TRUE;
273 nvlist_free(nvl);
274 }
275
276 if (ccw_failure) {
277 /*
278 * Keep trying so that configuration data is
279 * written if/when any temporary filesystem
280 * resource issues are resolved.
281 */
282 if (target->spa_ccw_fail_time == 0) {
283 (void) zfs_ereport_post(
284 FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
285 target, NULL, NULL, NULL, 0, 0);
286 }
287 target->spa_ccw_fail_time = gethrtime();
288 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
289 } else {
290 /*
291 * Do not rate limit future attempts to update
292 * the config cache.
293 */
294 target->spa_ccw_fail_time = 0;
295 }
296
297 /*
298 * Remove any config entries older than the current one.
299 */
300 dp = list_head(&target->spa_config_list);
301 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
302 list_remove(&target->spa_config_list, tdp);
303 if (tdp->scd_path != NULL)
304 spa_strfree(tdp->scd_path);
305 kmem_free(tdp, sizeof (spa_config_dirent_t));
306 }
307
308 spa_config_generation++;
309
310 if (postsysevent)
311 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
312 }
313
314 /*
315 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
316 * and we don't want to allow the local zone to see all the pools anyway.
317 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
318 * information for all pool visible within the zone.
319 */
320 nvlist_t *
spa_all_configs(uint64_t * generation)321 spa_all_configs(uint64_t *generation)
322 {
323 nvlist_t *pools;
324 spa_t *spa = NULL;
325
326 if (*generation == spa_config_generation)
327 return (NULL);
328
329 pools = fnvlist_alloc();
330
331 mutex_enter(&spa_namespace_lock);
332 while ((spa = spa_next(spa)) != NULL) {
333 if (INGLOBALZONE(curproc) ||
334 zone_dataset_visible(spa_name(spa), NULL)) {
335 mutex_enter(&spa->spa_props_lock);
336 fnvlist_add_nvlist(pools, spa_name(spa),
337 spa->spa_config);
338 mutex_exit(&spa->spa_props_lock);
339 }
340 }
341 *generation = spa_config_generation;
342 mutex_exit(&spa_namespace_lock);
343
344 return (pools);
345 }
346
347 void
spa_config_set(spa_t * spa,nvlist_t * config)348 spa_config_set(spa_t *spa, nvlist_t *config)
349 {
350 mutex_enter(&spa->spa_props_lock);
351 if (spa->spa_config != NULL && spa->spa_config != config)
352 nvlist_free(spa->spa_config);
353 spa->spa_config = config;
354 mutex_exit(&spa->spa_props_lock);
355 }
356
357 /*
358 * Generate the pool's configuration based on the current in-core state.
359 *
360 * We infer whether to generate a complete config or just one top-level config
361 * based on whether vd is the root vdev.
362 */
363 nvlist_t *
spa_config_generate(spa_t * spa,vdev_t * vd,uint64_t txg,int getstats)364 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
365 {
366 nvlist_t *config, *nvroot;
367 vdev_t *rvd = spa->spa_root_vdev;
368 unsigned long hostid = 0;
369 boolean_t locked = B_FALSE;
370 uint64_t split_guid;
371 char *pool_name;
372
373 if (vd == NULL) {
374 vd = rvd;
375 locked = B_TRUE;
376 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
377 }
378
379 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
380 (SCL_CONFIG | SCL_STATE));
381
382 /*
383 * If txg is -1, report the current value of spa->spa_config_txg.
384 */
385 if (txg == -1ULL)
386 txg = spa->spa_config_txg;
387
388 /*
389 * Originally, users had to handle spa namespace collisions by either
390 * exporting the already imported pool or by specifying a new name for
391 * the pool with a conflicting name. In the case of root pools from
392 * virtual guests, neither approach to collision resolution is
393 * reasonable. This is addressed by extending the new name syntax with
394 * an option to specify that the new name is temporary. When specified,
395 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
396 * to use the previous name, which we do below.
397 */
398 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
399 pool_name = fnvlist_lookup_string(spa->spa_config,
400 ZPOOL_CONFIG_POOL_NAME);
401 } else {
402 pool_name = spa_name(spa);
403 }
404
405 config = fnvlist_alloc();
406
407 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
408 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
409 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
410 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
411 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
412 fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
413 if (spa->spa_comment != NULL) {
414 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
415 spa->spa_comment);
416 }
417
418 hostid = spa_get_hostid();
419 if (hostid != 0) {
420 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
421 }
422 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname.nodename);
423
424 int config_gen_flags = 0;
425 if (vd != rvd) {
426 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
427 vd->vdev_top->vdev_guid);
428 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
429 vd->vdev_guid);
430 if (vd->vdev_isspare) {
431 fnvlist_add_uint64(config,
432 ZPOOL_CONFIG_IS_SPARE, 1ULL);
433 }
434 if (vd->vdev_islog) {
435 fnvlist_add_uint64(config,
436 ZPOOL_CONFIG_IS_LOG, 1ULL);
437 }
438 vd = vd->vdev_top; /* label contains top config */
439 } else {
440 /*
441 * Only add the (potentially large) split information
442 * in the mos config, and not in the vdev labels
443 */
444 if (spa->spa_config_splitting != NULL)
445 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
446 spa->spa_config_splitting);
447 fnvlist_add_boolean(config,
448 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
449
450 config_gen_flags |= VDEV_CONFIG_MOS;
451 }
452
453 /*
454 * Add the top-level config. We even add this on pools which
455 * don't support holes in the namespace.
456 */
457 vdev_top_config_generate(spa, config);
458
459 /*
460 * If we're splitting, record the original pool's guid.
461 */
462 if (spa->spa_config_splitting != NULL &&
463 nvlist_lookup_uint64(spa->spa_config_splitting,
464 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
465 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
466 split_guid);
467 }
468
469 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
470 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
471 nvlist_free(nvroot);
472
473 /*
474 * Store what's necessary for reading the MOS in the label.
475 */
476 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
477 spa->spa_label_features);
478
479 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
480 ddt_histogram_t *ddh;
481 ddt_stat_t *dds;
482 ddt_object_t *ddo;
483
484 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
485 ddt_get_dedup_histogram(spa, ddh);
486 fnvlist_add_uint64_array(config,
487 ZPOOL_CONFIG_DDT_HISTOGRAM,
488 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
489 kmem_free(ddh, sizeof (ddt_histogram_t));
490
491 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
492 ddt_get_dedup_object_stats(spa, ddo);
493 fnvlist_add_uint64_array(config,
494 ZPOOL_CONFIG_DDT_OBJ_STATS,
495 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
496 kmem_free(ddo, sizeof (ddt_object_t));
497
498 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
499 ddt_get_dedup_stats(spa, dds);
500 fnvlist_add_uint64_array(config,
501 ZPOOL_CONFIG_DDT_STATS,
502 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
503 kmem_free(dds, sizeof (ddt_stat_t));
504 }
505
506 if (locked)
507 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
508
509 return (config);
510 }
511
512 /*
513 * Update all disk labels, generate a fresh config based on the current
514 * in-core state, and sync the global config cache (do not sync the config
515 * cache if this is a booting rootpool).
516 */
517 void
spa_config_update(spa_t * spa,int what)518 spa_config_update(spa_t *spa, int what)
519 {
520 vdev_t *rvd = spa->spa_root_vdev;
521 uint64_t txg;
522 int c;
523
524 ASSERT(MUTEX_HELD(&spa_namespace_lock));
525
526 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
527 txg = spa_last_synced_txg(spa) + 1;
528 if (what == SPA_CONFIG_UPDATE_POOL) {
529 vdev_config_dirty(rvd);
530 } else {
531 /*
532 * If we have top-level vdevs that were added but have
533 * not yet been prepared for allocation, do that now.
534 * (It's safe now because the config cache is up to date,
535 * so it will be able to translate the new DVAs.)
536 * See comments in spa_vdev_add() for full details.
537 */
538 for (c = 0; c < rvd->vdev_children; c++) {
539 vdev_t *tvd = rvd->vdev_child[c];
540
541 /*
542 * Explicitly skip vdevs that are indirect or
543 * log vdevs that are being removed. The reason
544 * is that both of those can have vdev_ms_array
545 * set to 0 and we wouldn't want to change their
546 * metaslab size nor call vdev_expand() on them.
547 */
548 if (!vdev_is_concrete(tvd) ||
549 (tvd->vdev_islog && tvd->vdev_removing))
550 continue;
551
552 if (tvd->vdev_ms_array == 0)
553 vdev_metaslab_set_size(tvd);
554 vdev_expand(tvd, txg);
555 }
556 }
557 spa_config_exit(spa, SCL_ALL, FTAG);
558
559 /*
560 * Wait for the mosconfig to be regenerated and synced.
561 */
562 txg_wait_synced(spa->spa_dsl_pool, txg);
563
564 /*
565 * Update the global config cache to reflect the new mosconfig.
566 */
567 if (!spa->spa_is_root) {
568 spa_write_cachefile(spa, B_FALSE,
569 what != SPA_CONFIG_UPDATE_POOL);
570 }
571
572 if (what == SPA_CONFIG_UPDATE_POOL)
573 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
574 }
575