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, 2015 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 	if (fsize == 0)
103 		goto out;
104 
105 	buf = kmem_alloc(fsize, KM_SLEEP);
106 
107 	/*
108 	 * Read the nvlist from the file.
109 	 */
110 	if (kobj_read_file(file, buf, fsize, 0) < 0)
111 		goto out;
112 
113 	/*
114 	 * Unpack the nvlist.
115 	 */
116 	if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
117 		goto out;
118 
119 	/*
120 	 * Iterate over all elements in the nvlist, creating a new spa_t for
121 	 * each one with the specified configuration.
122 	 */
123 	mutex_enter(&spa_namespace_lock);
124 	nvpair = NULL;
125 	while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
126 		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
127 			continue;
128 
129 		child = fnvpair_value_nvlist(nvpair);
130 
131 		if (spa_lookup(nvpair_name(nvpair)) != NULL)
132 			continue;
133 		(void) spa_add(nvpair_name(nvpair), child, NULL);
134 	}
135 	mutex_exit(&spa_namespace_lock);
136 
137 	nvlist_free(nvlist);
138 
139 out:
140 	if (buf != NULL)
141 		kmem_free(buf, fsize);
142 
143 	kobj_close_file(file);
144 }
145 
146 static void
spa_config_clean(nvlist_t * nvl)147 spa_config_clean(nvlist_t *nvl)
148 {
149 	nvlist_t **child;
150 	nvlist_t *nvroot = NULL;
151 	uint_t c, children;
152 
153 	if (nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
154 	    &children) == 0) {
155 		for (c = 0; c < children; c++)
156 			spa_config_clean(child[c]);
157 	}
158 
159 	if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0)
160 		spa_config_clean(nvroot);
161 
162 	nvlist_remove(nvl, ZPOOL_CONFIG_VDEV_STATS, DATA_TYPE_UINT64_ARRAY);
163 	nvlist_remove(nvl, ZPOOL_CONFIG_SCAN_STATS, DATA_TYPE_UINT64_ARRAY);
164 }
165 
166 static int
spa_config_write(spa_config_dirent_t * dp,nvlist_t * nvl)167 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
168 {
169 	size_t buflen;
170 	char *buf;
171 	vnode_t *vp;
172 	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
173 	char *temp;
174 	int err;
175 
176 	/*
177 	 * If the nvlist is empty (NULL), then remove the old cachefile.
178 	 */
179 	if (nvl == NULL) {
180 		err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
181 		return (err);
182 	}
183 
184 	/*
185 	 * Pack the configuration into a buffer.
186 	 */
187 	buf = fnvlist_pack(nvl, &buflen);
188 	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
189 
190 	/*
191 	 * Write the configuration to disk.  We need to do the traditional
192 	 * 'write to temporary file, sync, move over original' to make sure we
193 	 * always have a consistent view of the data.
194 	 */
195 	(void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
196 
197 	err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
198 	if (err == 0) {
199 		err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
200 		    0, RLIM64_INFINITY, kcred, NULL);
201 		if (err == 0)
202 			err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
203 		if (err == 0)
204 			err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
205 		(void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
206 	}
207 
208 	(void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
209 
210 	fnvlist_pack_free(buf, buflen);
211 	kmem_free(temp, MAXPATHLEN);
212 	return (err);
213 }
214 
215 /*
216  * Synchronize pool configuration to disk.  This must be called with the
217  * namespace lock held. Synchronizing the pool cache is typically done after
218  * the configuration has been synced to the MOS. This exposes a window where
219  * the MOS config will have been updated but the cache file has not. If
220  * the system were to crash at that instant then the cached config may not
221  * contain the correct information to open the pool and an explicity import
222  * would be required.
223  */
224 void
spa_config_sync(spa_t * target,boolean_t removing,boolean_t postsysevent)225 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
226 {
227 	spa_config_dirent_t *dp, *tdp;
228 	nvlist_t *nvl;
229 	boolean_t ccw_failure;
230 	int error;
231 
232 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
233 
234 	if (rootdir == NULL || !(spa_mode_global & FWRITE))
235 		return;
236 
237 	/*
238 	 * Iterate over all cachefiles for the pool, past or present.  When the
239 	 * cachefile is changed, the new one is pushed onto this list, allowing
240 	 * us to update previous cachefiles that no longer contain this pool.
241 	 */
242 	ccw_failure = B_FALSE;
243 	for (dp = list_head(&target->spa_config_list); dp != NULL;
244 	    dp = list_next(&target->spa_config_list, dp)) {
245 		spa_t *spa = NULL;
246 		if (dp->scd_path == NULL)
247 			continue;
248 
249 		/*
250 		 * Iterate over all pools, adding any matching pools to 'nvl'.
251 		 */
252 		nvl = NULL;
253 		while ((spa = spa_next(spa)) != NULL) {
254 			nvlist_t *nvroot = NULL;
255 			/*
256 			 * Skip over our own pool if we're about to remove
257 			 * ourselves from the spa namespace or any pool that
258 			 * is readonly. Since we cannot guarantee that a
259 			 * readonly pool would successfully import upon reboot,
260 			 * we don't allow them to be written to the cache file.
261 			 */
262 			if ((spa == target && removing) ||
263 			    (spa_state(spa) == POOL_STATE_ACTIVE &&
264 			    !spa_writeable(spa)))
265 				continue;
266 
267 			mutex_enter(&spa->spa_props_lock);
268 			tdp = list_head(&spa->spa_config_list);
269 			if (spa->spa_config == NULL ||
270 			    tdp->scd_path == NULL ||
271 			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
272 				mutex_exit(&spa->spa_props_lock);
273 				continue;
274 			}
275 
276 			if (nvl == NULL)
277 				nvl = fnvlist_alloc();
278 
279 			fnvlist_add_nvlist(nvl, spa->spa_name,
280 			    spa->spa_config);
281 			mutex_exit(&spa->spa_props_lock);
282 
283 			if (nvlist_lookup_nvlist(nvl, spa->spa_name, &nvroot) == 0)
284 				spa_config_clean(nvroot);
285 		}
286 
287 		error = spa_config_write(dp, nvl);
288 		if (error != 0)
289 			ccw_failure = B_TRUE;
290 		nvlist_free(nvl);
291 	}
292 
293 	if (ccw_failure) {
294 		/*
295 		 * Keep trying so that configuration data is
296 		 * written if/when any temporary filesystem
297 		 * resource issues are resolved.
298 		 */
299 		if (target->spa_ccw_fail_time == 0) {
300 			zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
301 			    target, NULL, NULL, 0, 0);
302 		}
303 		target->spa_ccw_fail_time = gethrtime();
304 		spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
305 	} else {
306 		/*
307 		 * Do not rate limit future attempts to update
308 		 * the config cache.
309 		 */
310 		target->spa_ccw_fail_time = 0;
311 	}
312 
313 	/*
314 	 * Remove any config entries older than the current one.
315 	 */
316 	dp = list_head(&target->spa_config_list);
317 	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
318 		list_remove(&target->spa_config_list, tdp);
319 		if (tdp->scd_path != NULL)
320 			spa_strfree(tdp->scd_path);
321 		kmem_free(tdp, sizeof (spa_config_dirent_t));
322 	}
323 
324 	spa_config_generation++;
325 
326 	if (postsysevent)
327 		spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
328 }
329 
330 /*
331  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
332  * and we don't want to allow the local zone to see all the pools anyway.
333  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
334  * information for all pool visible within the zone.
335  */
336 nvlist_t *
spa_all_configs(uint64_t * generation)337 spa_all_configs(uint64_t *generation)
338 {
339 	nvlist_t *pools;
340 	spa_t *spa = NULL;
341 
342 	if (*generation == spa_config_generation)
343 		return (NULL);
344 
345 	pools = fnvlist_alloc();
346 
347 	mutex_enter(&spa_namespace_lock);
348 	while ((spa = spa_next(spa)) != NULL) {
349 		if (INGLOBALZONE(curthread) ||
350 		    zone_dataset_visible(spa_name(spa), NULL)) {
351 			mutex_enter(&spa->spa_props_lock);
352 			fnvlist_add_nvlist(pools, spa_name(spa),
353 			    spa->spa_config);
354 			mutex_exit(&spa->spa_props_lock);
355 		}
356 	}
357 	*generation = spa_config_generation;
358 	mutex_exit(&spa_namespace_lock);
359 
360 	return (pools);
361 }
362 
363 void
spa_config_set(spa_t * spa,nvlist_t * config)364 spa_config_set(spa_t *spa, nvlist_t *config)
365 {
366 	mutex_enter(&spa->spa_props_lock);
367 	nvlist_free(spa->spa_config);
368 	spa->spa_config = config;
369 	mutex_exit(&spa->spa_props_lock);
370 }
371 
372 /*
373  * Generate the pool's configuration based on the current in-core state.
374  *
375  * We infer whether to generate a complete config or just one top-level config
376  * based on whether vd is the root vdev.
377  */
378 nvlist_t *
spa_config_generate(spa_t * spa,vdev_t * vd,uint64_t txg,int getstats)379 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
380 {
381 	nvlist_t *config, *nvroot;
382 	vdev_t *rvd = spa->spa_root_vdev;
383 	unsigned long hostid = 0;
384 	boolean_t locked = B_FALSE;
385 	uint64_t split_guid;
386 
387 	if (vd == NULL) {
388 		vd = rvd;
389 		locked = B_TRUE;
390 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
391 	}
392 
393 	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
394 	    (SCL_CONFIG | SCL_STATE));
395 
396 	/*
397 	 * If txg is -1, report the current value of spa->spa_config_txg.
398 	 */
399 	if (txg == -1ULL)
400 		txg = spa->spa_config_txg;
401 
402 	config = fnvlist_alloc();
403 
404 	fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
405 	fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, spa_name(spa));
406 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
407 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
408 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
409 	if (spa->spa_comment != NULL) {
410 		fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
411 		    spa->spa_comment);
412 	}
413 
414 #ifdef	_KERNEL
415 	hostid = zone_get_hostid(NULL);
416 #else	/* _KERNEL */
417 	/*
418 	 * We're emulating the system's hostid in userland, so we can't use
419 	 * zone_get_hostid().
420 	 */
421 	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
422 #endif	/* _KERNEL */
423 	if (hostid != 0) {
424 		fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
425 	}
426 	fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname.nodename);
427 
428 	int config_gen_flags = 0;
429 	if (vd != rvd) {
430 		fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
431 		    vd->vdev_top->vdev_guid);
432 		fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
433 		    vd->vdev_guid);
434 		if (vd->vdev_isspare) {
435 			fnvlist_add_uint64(config,
436 			    ZPOOL_CONFIG_IS_SPARE, 1ULL);
437 		}
438 		if (vd->vdev_islog) {
439 			fnvlist_add_uint64(config,
440 			    ZPOOL_CONFIG_IS_LOG, 1ULL);
441 		}
442 		vd = vd->vdev_top;		/* label contains top config */
443 	} else {
444 		/*
445 		 * Only add the (potentially large) split information
446 		 * in the mos config, and not in the vdev labels
447 		 */
448 		if (spa->spa_config_splitting != NULL)
449 			fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
450 			    spa->spa_config_splitting);
451 		fnvlist_add_boolean(config,
452 		    ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
453 
454 		config_gen_flags |= VDEV_CONFIG_MOS;
455 	}
456 
457 	/*
458 	 * Add the top-level config.  We even add this on pools which
459 	 * don't support holes in the namespace.
460 	 */
461 	vdev_top_config_generate(spa, config);
462 
463 	/*
464 	 * If we're splitting, record the original pool's guid.
465 	 */
466 	if (spa->spa_config_splitting != NULL &&
467 	    nvlist_lookup_uint64(spa->spa_config_splitting,
468 	    ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
469 		fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
470 		    split_guid);
471 	}
472 
473 	nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
474 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
475 	nvlist_free(nvroot);
476 
477 	/*
478 	 * Store what's necessary for reading the MOS in the label.
479 	 */
480 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
481 	    spa->spa_label_features);
482 
483 	if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
484 		ddt_histogram_t *ddh;
485 		ddt_stat_t *dds;
486 		ddt_object_t *ddo;
487 
488 		ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
489 		ddt_get_dedup_histogram(spa, ddh);
490 		fnvlist_add_uint64_array(config,
491 		    ZPOOL_CONFIG_DDT_HISTOGRAM,
492 		    (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
493 		kmem_free(ddh, sizeof (ddt_histogram_t));
494 
495 		ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
496 		ddt_get_dedup_object_stats(spa, ddo);
497 		fnvlist_add_uint64_array(config,
498 		    ZPOOL_CONFIG_DDT_OBJ_STATS,
499 		    (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
500 		kmem_free(ddo, sizeof (ddt_object_t));
501 
502 		dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
503 		ddt_get_dedup_stats(spa, dds);
504 		fnvlist_add_uint64_array(config,
505 		    ZPOOL_CONFIG_DDT_STATS,
506 		    (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
507 		kmem_free(dds, sizeof (ddt_stat_t));
508 	}
509 
510 	if (locked)
511 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
512 
513 	return (config);
514 }
515 
516 /*
517  * Update all disk labels, generate a fresh config based on the current
518  * in-core state, and sync the global config cache (do not sync the config
519  * cache if this is a booting rootpool).
520  */
521 void
spa_config_update(spa_t * spa,int what)522 spa_config_update(spa_t *spa, int what)
523 {
524 	vdev_t *rvd = spa->spa_root_vdev;
525 	uint64_t txg;
526 	int c;
527 
528 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
529 
530 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
531 	txg = spa_last_synced_txg(spa) + 1;
532 	if (what == SPA_CONFIG_UPDATE_POOL) {
533 		vdev_config_dirty(rvd);
534 	} else {
535 		/*
536 		 * If we have top-level vdevs that were added but have
537 		 * not yet been prepared for allocation, do that now.
538 		 * (It's safe now because the config cache is up to date,
539 		 * so it will be able to translate the new DVAs.)
540 		 * See comments in spa_vdev_add() for full details.
541 		 */
542 		for (c = 0; c < rvd->vdev_children; c++) {
543 			vdev_t *tvd = rvd->vdev_child[c];
544 			if (tvd->vdev_ms_array == 0) {
545 				vdev_ashift_optimize(tvd);
546 				vdev_metaslab_set_size(tvd);
547 			}
548 			vdev_expand(tvd, txg);
549 		}
550 	}
551 	spa_config_exit(spa, SCL_ALL, FTAG);
552 
553 	/*
554 	 * Wait for the mosconfig to be regenerated and synced.
555 	 */
556 	txg_wait_synced(spa->spa_dsl_pool, txg);
557 
558 	/*
559 	 * Update the global config cache to reflect the new mosconfig.
560 	 */
561 	spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
562 
563 	if (what == SPA_CONFIG_UPDATE_POOL)
564 		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
565 }
566