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