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/zfs_context.h>
30 #include <sys/spa.h>
31 #include <sys/fm/fs/zfs.h>
32 #include <sys/spa_impl.h>
33 #include <sys/nvpair.h>
34 #include <sys/uio.h>
35 #include <sys/fs/zfs.h>
36 #include <sys/vdev_impl.h>
37 #include <sys/zfs_ioctl.h>
38 #include <sys/utsname.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", 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 child = fnvpair_value_nvlist(nvpair);
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 void
spa_config_clean(nvlist_t * nvl)145 spa_config_clean(nvlist_t *nvl)
146 {
147 nvlist_t **child;
148 nvlist_t *nvroot = NULL;
149 uint_t c, children;
150
151 if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
152 &children) == 0) {
153 for (c = 0; c < children; c++)
154 spa_config_clean(child[c]);
155 }
156
157 if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0)
158 spa_config_clean(nvroot);
159
160 nvlist_remove(nvl, ZPOOL_CONFIG_VDEV_STATS, DATA_TYPE_UINT64_ARRAY);
161 nvlist_remove(nvl, ZPOOL_CONFIG_SCAN_STATS, DATA_TYPE_UINT64_ARRAY);
162 }
163
164 static int
spa_config_write(spa_config_dirent_t * dp,nvlist_t * nvl)165 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
166 {
167 size_t buflen;
168 char *buf;
169 vnode_t *vp;
170 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
171 char *temp;
172 int err;
173
174 /*
175 * If the nvlist is empty (NULL), then remove the old cachefile.
176 */
177 if (nvl == NULL) {
178 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
179 return (err);
180 }
181
182 /*
183 * Pack the configuration into a buffer.
184 */
185 buf = fnvlist_pack(nvl, &buflen);
186 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
187
188 /*
189 * Write the configuration to disk. We need to do the traditional
190 * 'write to temporary file, sync, move over original' to make sure we
191 * always have a consistent view of the data.
192 */
193 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
194
195 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
196 if (err == 0) {
197 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
198 0, RLIM64_INFINITY, kcred, NULL);
199 if (err == 0)
200 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
201 if (err == 0)
202 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
203 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
204 }
205
206 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
207
208 fnvlist_pack_free(buf, buflen);
209 kmem_free(temp, MAXPATHLEN);
210 return (err);
211 }
212
213 /*
214 * Synchronize pool configuration to disk. This must be called with the
215 * namespace lock held. Synchronizing the pool cache is typically done after
216 * the configuration has been synced to the MOS. This exposes a window where
217 * the MOS config will have been updated but the cache file has not. If
218 * the system were to crash at that instant then the cached config may not
219 * contain the correct information to open the pool and an explicit import
220 * would be required.
221 */
222 void
spa_write_cachefile(spa_t * target,boolean_t removing,boolean_t postsysevent)223 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent)
224 {
225 spa_config_dirent_t *dp, *tdp;
226 nvlist_t *nvl;
227 boolean_t ccw_failure;
228 int error;
229 char *pool_name;
230
231 ASSERT(MUTEX_HELD(&spa_namespace_lock));
232
233 if (rootdir == NULL || !(spa_mode_global & FWRITE))
234 return;
235
236 /*
237 * Iterate over all cachefiles for the pool, past or present. When the
238 * cachefile is changed, the new one is pushed onto this list, allowing
239 * us to update previous cachefiles that no longer contain this pool.
240 */
241 ccw_failure = B_FALSE;
242 for (dp = list_head(&target->spa_config_list); dp != NULL;
243 dp = list_next(&target->spa_config_list, dp)) {
244 spa_t *spa = NULL;
245 if (dp->scd_path == NULL)
246 continue;
247
248 /*
249 * Iterate over all pools, adding any matching pools to 'nvl'.
250 */
251 nvl = NULL;
252 while ((spa = spa_next(spa)) != NULL) {
253 nvlist_t *nvroot = NULL;
254 /*
255 * Skip over our own pool if we're about to remove
256 * ourselves from the spa namespace or any pool that
257 * is readonly. Since we cannot guarantee that a
258 * readonly pool would successfully import upon reboot,
259 * we don't allow them to be written to the cache file.
260 */
261 if ((spa == target && removing) ||
262 (spa_state(spa) == POOL_STATE_ACTIVE &&
263 !spa_writeable(spa)))
264 continue;
265
266 mutex_enter(&spa->spa_props_lock);
267 tdp = list_head(&spa->spa_config_list);
268 if (spa->spa_config == NULL ||
269 tdp->scd_path == NULL ||
270 strcmp(tdp->scd_path, dp->scd_path) != 0) {
271 mutex_exit(&spa->spa_props_lock);
272 continue;
273 }
274
275 if (nvl == NULL)
276 nvl = fnvlist_alloc();
277
278 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
279 pool_name = fnvlist_lookup_string(
280 spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
281 } else {
282 pool_name = spa_name(spa);
283 }
284
285 fnvlist_add_nvlist(nvl, pool_name,
286 spa->spa_config);
287 mutex_exit(&spa->spa_props_lock);
288
289 if (nvlist_lookup_nvlist(nvl, pool_name, &nvroot) == 0)
290 spa_config_clean(nvroot);
291 }
292
293 error = spa_config_write(dp, nvl);
294 if (error != 0)
295 ccw_failure = B_TRUE;
296 nvlist_free(nvl);
297 }
298
299 if (ccw_failure) {
300 /*
301 * Keep trying so that configuration data is
302 * written if/when any temporary filesystem
303 * resource issues are resolved.
304 */
305 if (target->spa_ccw_fail_time == 0) {
306 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
307 target, NULL, NULL, 0, 0);
308 }
309 target->spa_ccw_fail_time = gethrtime();
310 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
311 } else {
312 /*
313 * Do not rate limit future attempts to update
314 * the config cache.
315 */
316 target->spa_ccw_fail_time = 0;
317 }
318
319 /*
320 * Remove any config entries older than the current one.
321 */
322 dp = list_head(&target->spa_config_list);
323 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
324 list_remove(&target->spa_config_list, tdp);
325 if (tdp->scd_path != NULL)
326 spa_strfree(tdp->scd_path);
327 kmem_free(tdp, sizeof (spa_config_dirent_t));
328 }
329
330 spa_config_generation++;
331
332 if (postsysevent)
333 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
334 }
335
336 /*
337 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
338 * and we don't want to allow the local zone to see all the pools anyway.
339 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
340 * information for all pool visible within the zone.
341 */
342 nvlist_t *
spa_all_configs(uint64_t * generation)343 spa_all_configs(uint64_t *generation)
344 {
345 nvlist_t *pools;
346 spa_t *spa = NULL;
347
348 if (*generation == spa_config_generation)
349 return (NULL);
350
351 pools = fnvlist_alloc();
352
353 mutex_enter(&spa_namespace_lock);
354 while ((spa = spa_next(spa)) != NULL) {
355 if (INGLOBALZONE(curthread) ||
356 zone_dataset_visible(spa_name(spa), NULL)) {
357 mutex_enter(&spa->spa_props_lock);
358 fnvlist_add_nvlist(pools, spa_name(spa),
359 spa->spa_config);
360 mutex_exit(&spa->spa_props_lock);
361 }
362 }
363 *generation = spa_config_generation;
364 mutex_exit(&spa_namespace_lock);
365
366 return (pools);
367 }
368
369 void
spa_config_set(spa_t * spa,nvlist_t * config)370 spa_config_set(spa_t *spa, nvlist_t *config)
371 {
372 mutex_enter(&spa->spa_props_lock);
373 if (spa->spa_config != NULL && spa->spa_config != config)
374 nvlist_free(spa->spa_config);
375 spa->spa_config = config;
376 mutex_exit(&spa->spa_props_lock);
377 }
378
379 /*
380 * Generate the pool's configuration based on the current in-core state.
381 *
382 * We infer whether to generate a complete config or just one top-level config
383 * based on whether vd is the root vdev.
384 */
385 nvlist_t *
spa_config_generate(spa_t * spa,vdev_t * vd,uint64_t txg,int getstats)386 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
387 {
388 nvlist_t *config, *nvroot;
389 vdev_t *rvd = spa->spa_root_vdev;
390 unsigned long hostid = 0;
391 boolean_t locked = B_FALSE;
392 uint64_t split_guid;
393 char *pool_name;
394
395 if (vd == NULL) {
396 vd = rvd;
397 locked = B_TRUE;
398 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
399 }
400
401 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
402 (SCL_CONFIG | SCL_STATE));
403
404 /*
405 * If txg is -1, report the current value of spa->spa_config_txg.
406 */
407 if (txg == -1ULL)
408 txg = spa->spa_config_txg;
409
410 /*
411 * Originally, users had to handle spa namespace collisions by either
412 * exporting the already imported pool or by specifying a new name for
413 * the pool with a conflicting name. In the case of root pools from
414 * virtual guests, neither approach to collision resolution is
415 * reasonable. This is addressed by extending the new name syntax with
416 * an option to specify that the new name is temporary. When specified,
417 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
418 * to use the previous name, which we do below.
419 */
420 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
421 pool_name = fnvlist_lookup_string(spa->spa_config,
422 ZPOOL_CONFIG_POOL_NAME);
423 } else {
424 pool_name = spa_name(spa);
425 }
426
427 config = fnvlist_alloc();
428
429 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
430 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
431 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
432 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
433 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
434 if (spa->spa_comment != NULL) {
435 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
436 spa->spa_comment);
437 }
438
439 hostid = zone_get_hostid(NULL);
440
441 if (hostid != 0) {
442 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
443 }
444 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname.nodename);
445
446 int config_gen_flags = 0;
447 if (vd != rvd) {
448 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
449 vd->vdev_top->vdev_guid);
450 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
451 vd->vdev_guid);
452 if (vd->vdev_isspare) {
453 fnvlist_add_uint64(config,
454 ZPOOL_CONFIG_IS_SPARE, 1ULL);
455 }
456 if (vd->vdev_islog) {
457 fnvlist_add_uint64(config,
458 ZPOOL_CONFIG_IS_LOG, 1ULL);
459 }
460 vd = vd->vdev_top; /* label contains top config */
461 } else {
462 /*
463 * Only add the (potentially large) split information
464 * in the mos config, and not in the vdev labels
465 */
466 if (spa->spa_config_splitting != NULL)
467 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
468 spa->spa_config_splitting);
469 fnvlist_add_boolean(config,
470 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
471
472 config_gen_flags |= VDEV_CONFIG_MOS;
473 }
474
475 /*
476 * Add the top-level config. We even add this on pools which
477 * don't support holes in the namespace.
478 */
479 vdev_top_config_generate(spa, config);
480
481 /*
482 * If we're splitting, record the original pool's guid.
483 */
484 if (spa->spa_config_splitting != NULL &&
485 nvlist_lookup_uint64(spa->spa_config_splitting,
486 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
487 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
488 split_guid);
489 }
490
491 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
492 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
493 nvlist_free(nvroot);
494
495 /*
496 * Store what's necessary for reading the MOS in the label.
497 */
498 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
499 spa->spa_label_features);
500
501 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
502 ddt_histogram_t *ddh;
503 ddt_stat_t *dds;
504 ddt_object_t *ddo;
505
506 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
507 ddt_get_dedup_histogram(spa, ddh);
508 fnvlist_add_uint64_array(config,
509 ZPOOL_CONFIG_DDT_HISTOGRAM,
510 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
511 kmem_free(ddh, sizeof (ddt_histogram_t));
512
513 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
514 ddt_get_dedup_object_stats(spa, ddo);
515 fnvlist_add_uint64_array(config,
516 ZPOOL_CONFIG_DDT_OBJ_STATS,
517 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
518 kmem_free(ddo, sizeof (ddt_object_t));
519
520 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
521 ddt_get_dedup_stats(spa, dds);
522 fnvlist_add_uint64_array(config,
523 ZPOOL_CONFIG_DDT_STATS,
524 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
525 kmem_free(dds, sizeof (ddt_stat_t));
526 }
527
528 if (locked)
529 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
530
531 return (config);
532 }
533
534 /*
535 * Update all disk labels, generate a fresh config based on the current
536 * in-core state, and sync the global config cache (do not sync the config
537 * cache if this is a booting rootpool).
538 */
539 void
spa_config_update(spa_t * spa,int what)540 spa_config_update(spa_t *spa, int what)
541 {
542 vdev_t *rvd = spa->spa_root_vdev;
543 uint64_t txg;
544 int c;
545
546 ASSERT(MUTEX_HELD(&spa_namespace_lock));
547
548 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
549 txg = spa_last_synced_txg(spa) + 1;
550 if (what == SPA_CONFIG_UPDATE_POOL) {
551 vdev_config_dirty(rvd);
552 } else {
553 /*
554 * If we have top-level vdevs that were added but have
555 * not yet been prepared for allocation, do that now.
556 * (It's safe now because the config cache is up to date,
557 * so it will be able to translate the new DVAs.)
558 * See comments in spa_vdev_add() for full details.
559 */
560 for (c = 0; c < rvd->vdev_children; c++) {
561 vdev_t *tvd = rvd->vdev_child[c];
562
563 /*
564 * Explicitly skip vdevs that are indirect or
565 * log vdevs that are being removed. The reason
566 * is that both of those can have vdev_ms_array
567 * set to 0 and we wouldn't want to change their
568 * metaslab size nor call vdev_expand() on them.
569 */
570 if (!vdev_is_concrete(tvd) ||
571 (tvd->vdev_islog && tvd->vdev_removing))
572 continue;
573
574 if (tvd->vdev_ms_array == 0) {
575 vdev_ashift_optimize(tvd);
576 vdev_metaslab_set_size(tvd);
577 }
578 vdev_expand(tvd, txg);
579 }
580 }
581 spa_config_exit(spa, SCL_ALL, FTAG);
582
583 /*
584 * Wait for the mosconfig to be regenerated and synced.
585 */
586 txg_wait_synced(spa->spa_dsl_pool, txg);
587
588 /*
589 * Update the global config cache to reflect the new mosconfig.
590 */
591 spa_write_cachefile(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
592
593 if (what == SPA_CONFIG_UPDATE_POOL)
594 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
595 }
596