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 2013 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2011, 2015 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_config_sync(), 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
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 uint64_t guid = 0;
126
127 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
128 continue;
129
130 child = fnvpair_value_nvlist(nvpair);
131 /* a zero guid means we simply will ignore the check later */
132 (void) nvlist_lookup_uint64(child, ZPOOL_CONFIG_POOL_GUID,
133 &guid);
134
135 if (spa_lookup(nvpair_name(nvpair)) != NULL ||
136 spa_config_guid_exists(guid)) {
137 continue;
138 }
139 (void) spa_add(nvpair_name(nvpair), child, NULL);
140 }
141 mutex_exit(&spa_namespace_lock);
142
143 nvlist_free(nvlist);
144
145 out:
146 if (buf != NULL)
147 kmem_free(buf, fsize);
148
149 kobj_close_file(file);
150 }
151
152 static int
153 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
154 {
155 size_t buflen;
156 char *buf;
157 vnode_t *vp;
158 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
159 char *temp;
160 int err;
161
162 /*
163 * If the nvlist is empty (NULL), then remove the old cachefile.
164 */
165 if (nvl == NULL) {
166 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
167 return (err);
168 }
169
170 /*
171 * Pack the configuration into a buffer.
172 */
173 buf = fnvlist_pack(nvl, &buflen);
174 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
175
176 /*
177 * Write the configuration to disk. We need to do the traditional
178 * 'write to temporary file, sync, move over original' to make sure we
179 * always have a consistent view of the data.
180 */
181 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
182
183 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
184 if (err == 0) {
185 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
186 0, RLIM64_INFINITY, kcred, NULL);
187 if (err == 0)
188 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
189 if (err == 0)
190 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
191 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
192 VN_RELE(vp);
193 }
194
195 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
196
197 fnvlist_pack_free(buf, buflen);
198 kmem_free(temp, MAXPATHLEN);
199 return (err);
200 }
201
202 /*
203 * Synchronize pool configuration to disk. This must be called with the
204 * namespace lock held. Synchronizing the pool cache is typically done after
205 * the configuration has been synced to the MOS. This exposes a window where
206 * the MOS config will have been updated but the cache file has not. If
207 * the system were to crash at that instant then the cached config may not
208 * contain the correct information to open the pool and an explicity import
209 * would be required.
210 */
211 void
212 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
213 {
214 spa_config_dirent_t *dp, *tdp;
215 nvlist_t *nvl;
216 boolean_t ccw_failure;
217 int error;
218
219 ASSERT(MUTEX_HELD(&spa_namespace_lock));
220
221 if (rootdir == NULL || !(spa_mode_global & FWRITE))
222 return;
223
224 /*
225 * Iterate over all cachefiles for the pool, past or present. When the
226 * cachefile is changed, the new one is pushed onto this list, allowing
227 * us to update previous cachefiles that no longer contain this pool.
228 */
229 ccw_failure = B_FALSE;
230 for (dp = list_head(&target->spa_config_list); dp != NULL;
231 dp = list_next(&target->spa_config_list, dp)) {
232 spa_t *spa = NULL;
233 if (dp->scd_path == NULL)
234 continue;
235
236 /*
237 * Iterate over all pools, adding any matching pools to 'nvl'.
238 */
239 nvl = NULL;
240 while ((spa = spa_next(spa)) != NULL) {
241 /*
242 * Skip over our own pool if we're about to remove
243 * ourselves from the spa namespace or any pool that
244 * is readonly. Since we cannot guarantee that a
245 * readonly pool would successfully import upon reboot,
246 * we don't allow them to be written to the cache file.
247 */
248 if ((spa == target && removing) ||
249 !spa_writeable(spa))
250 continue;
251
252 mutex_enter(&spa->spa_props_lock);
253 tdp = list_head(&spa->spa_config_list);
254 if (spa->spa_config == NULL ||
255 tdp->scd_path == NULL ||
256 strcmp(tdp->scd_path, dp->scd_path) != 0) {
257 mutex_exit(&spa->spa_props_lock);
258 continue;
259 }
260
261 if (nvl == NULL)
262 nvl = fnvlist_alloc();
263
264 fnvlist_add_nvlist(nvl, spa->spa_name,
265 spa->spa_config);
266 mutex_exit(&spa->spa_props_lock);
267 }
268
269 error = spa_config_write(dp, nvl);
270 if (error != 0)
271 ccw_failure = B_TRUE;
272 nvlist_free(nvl);
273 }
274
275 if (ccw_failure) {
276 /*
277 * Keep trying so that configuration data is
278 * written if/when any temporary filesystem
279 * resource issues are resolved.
280 */
281 if (target->spa_ccw_fail_time == 0) {
282 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
283 target, NULL, NULL, 0, 0);
284 }
285 target->spa_ccw_fail_time = gethrtime();
286 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
287 } else {
288 /*
289 * Do not rate limit future attempts to update
290 * the config cache.
291 */
292 target->spa_ccw_fail_time = 0;
293 }
294
295 /*
296 * Remove any config entries older than the current one.
297 */
298 dp = list_head(&target->spa_config_list);
299 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
300 list_remove(&target->spa_config_list, tdp);
301 if (tdp->scd_path != NULL)
302 spa_strfree(tdp->scd_path);
303 kmem_free(tdp, sizeof (spa_config_dirent_t));
304 }
305
306 spa_config_generation++;
307
308 if (postsysevent)
309 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
310 }
311
312 /*
313 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
314 * and we don't want to allow the local zone to see all the pools anyway.
315 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
316 * information for all pool visible within the zone.
317 */
318 nvlist_t *
319 spa_all_configs(uint64_t *generation)
320 {
321 nvlist_t *pools;
322 spa_t *spa = NULL;
323
324 if (*generation == spa_config_generation)
325 return (NULL);
326
327 pools = fnvlist_alloc();
328
329 mutex_enter(&spa_namespace_lock);
330 while ((spa = spa_next(spa)) != NULL) {
331 if (INGLOBALZONE(curproc) ||
332 zone_dataset_visible(spa_name(spa), NULL)) {
333 mutex_enter(&spa->spa_props_lock);
334 fnvlist_add_nvlist(pools, spa_name(spa),
335 spa->spa_config);
336 mutex_exit(&spa->spa_props_lock);
337 }
338 }
339 *generation = spa_config_generation;
340 mutex_exit(&spa_namespace_lock);
341
342 return (pools);
343 }
344
345 void
346 spa_config_set(spa_t *spa, nvlist_t *config)
347 {
348 mutex_enter(&spa->spa_props_lock);
349 nvlist_free(spa->spa_config);
350 spa->spa_config = config;
351 mutex_exit(&spa->spa_props_lock);
352 }
353
354 /*
355 * Generate the pool's configuration based on the current in-core state.
356 *
357 * We infer whether to generate a complete config or just one top-level config
358 * based on whether vd is the root vdev.
359 */
360 nvlist_t *
361 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
362 {
363 nvlist_t *config, *nvroot;
364 vdev_t *rvd = spa->spa_root_vdev;
365 unsigned long hostid = 0;
366 boolean_t locked = B_FALSE;
367 uint64_t split_guid;
368
369 if (vd == NULL) {
370 vd = rvd;
371 locked = B_TRUE;
372 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
373 }
374
375 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
376 (SCL_CONFIG | SCL_STATE));
377
378 /*
379 * If txg is -1, report the current value of spa->spa_config_txg.
380 */
381 if (txg == -1ULL)
382 txg = spa->spa_config_txg;
383
384 config = fnvlist_alloc();
385
386 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
387 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, spa_name(spa));
388 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
389 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
390 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
391 if (spa->spa_comment != NULL) {
392 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
393 spa->spa_comment);
394 }
395
396 #ifdef _KERNEL
397 hostid = zone_get_hostid(NULL);
398 #else /* _KERNEL */
399 /*
400 * We're emulating the system's hostid in userland, so we can't use
401 * zone_get_hostid().
402 */
403 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
404 #endif /* _KERNEL */
405 if (hostid != 0) {
406 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
407 }
408 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname.nodename);
409
410 int config_gen_flags = 0;
411 if (vd != rvd) {
412 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
413 vd->vdev_top->vdev_guid);
414 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
415 vd->vdev_guid);
416 if (vd->vdev_isspare) {
417 fnvlist_add_uint64(config,
418 ZPOOL_CONFIG_IS_SPARE, 1ULL);
419 }
420 if (vd->vdev_isspecial)
421 fnvlist_add_uint64(config,
422 ZPOOL_CONFIG_IS_SPECIAL, 1ULL);
423 if (vd->vdev_islog) {
424 fnvlist_add_uint64(config,
425 ZPOOL_CONFIG_IS_LOG, 1ULL);
426 }
427 vd = vd->vdev_top; /* label contains top config */
428 } else {
429 /*
430 * Only add the (potentially large) split information
431 * in the mos config, and not in the vdev labels
432 */
433 if (spa->spa_config_splitting != NULL)
434 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
435 spa->spa_config_splitting);
436 fnvlist_add_boolean(config,
437 ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
438
439 config_gen_flags |= VDEV_CONFIG_MOS;
440 }
441
442 /*
443 * Add the top-level config. We even add this on pools which
444 * don't support holes in the namespace.
445 */
446 vdev_top_config_generate(spa, config);
447
448 /*
449 * If we're splitting, record the original pool's guid.
450 */
451 if (spa->spa_config_splitting != NULL &&
452 nvlist_lookup_uint64(spa->spa_config_splitting,
453 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
454 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
455 split_guid);
456 }
457
458 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
459 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
460 nvlist_free(nvroot);
461
462 /* If we're getting stats, calculate trim progress from leaf vdevs. */
463 if (getstats) {
464 uint64_t prog, rate, start_time, stop_time;
465
466 spa_get_trim_prog(spa, &prog, &rate, &start_time, &stop_time);
467 fnvlist_add_uint64(config, ZPOOL_CONFIG_TRIM_PROG, prog);
468 fnvlist_add_uint64(config, ZPOOL_CONFIG_TRIM_RATE, rate);
469 fnvlist_add_uint64(config, ZPOOL_CONFIG_TRIM_START_TIME,
470 start_time);
471 fnvlist_add_uint64(config, ZPOOL_CONFIG_TRIM_STOP_TIME,
472 stop_time);
473 }
474
475 /*
476 * Store what's necessary for reading the MOS in the label.
477 */
478 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
479 spa->spa_label_features);
480
481 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
482 ddt_histogram_t *ddh;
483 ddt_stat_t *dds;
484 ddt_object_t *ddo;
485
486 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
487 ddt_get_dedup_histogram(spa, ddh);
488 fnvlist_add_uint64_array(config,
489 ZPOOL_CONFIG_DDT_HISTOGRAM,
490 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
491 kmem_free(ddh, sizeof (ddt_histogram_t));
492
493 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
494 ddt_get_dedup_object_stats(spa, ddo);
495 fnvlist_add_uint64_array(config,
496 ZPOOL_CONFIG_DDT_OBJ_STATS,
497 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
498 kmem_free(ddo, sizeof (ddt_object_t));
499
500 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
501 ddt_get_dedup_stats(spa, dds);
502 fnvlist_add_uint64_array(config,
503 ZPOOL_CONFIG_DDT_STATS,
504 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
505 kmem_free(dds, sizeof (ddt_stat_t));
506 }
507
508 if (locked)
509 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
510
511 return (config);
512 }
513
514 /*
515 * Update all disk labels, generate a fresh config based on the current
516 * in-core state, and sync the global config cache (do not sync the config
517 * cache if this is a booting rootpool).
518 */
519 void
520 spa_config_update(spa_t *spa, int what)
521 {
522 vdev_t *rvd = spa->spa_root_vdev;
523 uint64_t txg;
524 int c;
525
526 ASSERT(MUTEX_HELD(&spa_namespace_lock));
527
528 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
529 txg = spa_last_synced_txg(spa) + 1;
530 if (what == SPA_CONFIG_UPDATE_POOL) {
531 vdev_config_dirty(rvd);
532 } else {
533 /*
534 * If we have top-level vdevs that were added but have
535 * not yet been prepared for allocation, do that now.
536 * (It's safe now because the config cache is up to date,
537 * so it will be able to translate the new DVAs.)
538 * See comments in spa_vdev_add() for full details.
539 */
540 for (c = 0; c < rvd->vdev_children; c++) {
541 vdev_t *tvd = rvd->vdev_child[c];
542 if (tvd->vdev_ms_array == 0)
543 vdev_metaslab_set_size(tvd);
544 vdev_expand(tvd, txg);
545 }
546 }
547 spa_config_exit(spa, SCL_ALL, FTAG);
548
549 /*
550 * Wait for the mosconfig to be regenerated and synced.
551 */
552 txg_wait_synced(spa->spa_dsl_pool, txg);
553
554 /*
555 * Update the global config cache to reflect the new mosconfig.
556 */
557 if (!spa->spa_is_root)
558 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
559
560 if (what == SPA_CONFIG_UPDATE_POOL)
561 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
562 }