xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa_config.c (revision 8b80e8cb6855118d46f605e91b5ed4ce83417395)
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 2008 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include <sys/spa.h>
28 #include <sys/spa_impl.h>
29 #include <sys/nvpair.h>
30 #include <sys/uio.h>
31 #include <sys/fs/zfs.h>
32 #include <sys/vdev_impl.h>
33 #include <sys/zfs_ioctl.h>
34 #include <sys/utsname.h>
35 #include <sys/systeminfo.h>
36 #include <sys/sunddi.h>
37 #ifdef _KERNEL
38 #include <sys/kobj.h>
39 #endif
40 
41 /*
42  * Pool configuration repository.
43  *
44  * Pool configuration is stored as a packed nvlist on the filesystem.  By
45  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
46  * (when the ZFS module is loaded).  Pools can also have the 'cachefile'
47  * property set that allows them to be stored in an alternate location until
48  * the control of external software.
49  *
50  * For each cache file, we have a single nvlist which holds all the
51  * configuration information.  When the module loads, we read this information
52  * from /etc/zfs/zpool.cache and populate the SPA namespace.  This namespace is
53  * maintained independently in spa.c.  Whenever the namespace is modified, or
54  * the configuration of a pool is changed, we call spa_config_sync(), which
55  * walks through all the active pools and writes the configuration to disk.
56  */
57 
58 static uint64_t spa_config_generation = 1;
59 
60 /*
61  * This can be overridden in userland to preserve an alternate namespace for
62  * userland pools when doing testing.
63  */
64 const char *spa_config_path = ZPOOL_CACHE;
65 
66 /*
67  * Called when the module is first loaded, this routine loads the configuration
68  * file into the SPA namespace.  It does not actually open or load the pools; it
69  * only populates the namespace.
70  */
71 void
72 spa_config_load(void)
73 {
74 	void *buf = NULL;
75 	nvlist_t *nvlist, *child;
76 	nvpair_t *nvpair;
77 	spa_t *spa;
78 	char *pathname;
79 	struct _buf *file;
80 	uint64_t fsize;
81 
82 	/*
83 	 * Open the configuration file.
84 	 */
85 	pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
86 
87 	(void) snprintf(pathname, MAXPATHLEN, "%s%s",
88 	    (rootdir != NULL) ? "./" : "", spa_config_path);
89 
90 	file = kobj_open_file(pathname);
91 
92 	kmem_free(pathname, MAXPATHLEN);
93 
94 	if (file == (struct _buf *)-1)
95 		return;
96 
97 	if (kobj_get_filesize(file, &fsize) != 0)
98 		goto out;
99 
100 	buf = kmem_alloc(fsize, KM_SLEEP);
101 
102 	/*
103 	 * Read the nvlist from the file.
104 	 */
105 	if (kobj_read_file(file, buf, fsize, 0) < 0)
106 		goto out;
107 
108 	/*
109 	 * Unpack the nvlist.
110 	 */
111 	if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
112 		goto out;
113 
114 	/*
115 	 * Iterate over all elements in the nvlist, creating a new spa_t for
116 	 * each one with the specified configuration.
117 	 */
118 	mutex_enter(&spa_namespace_lock);
119 	nvpair = NULL;
120 	while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
121 
122 		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
123 			continue;
124 
125 		VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
126 
127 		if (spa_lookup(nvpair_name(nvpair)) != NULL)
128 			continue;
129 		spa = spa_add(nvpair_name(nvpair), NULL);
130 
131 		/*
132 		 * We blindly duplicate the configuration here.  If it's
133 		 * invalid, we will catch it when the pool is first opened.
134 		 */
135 		VERIFY(nvlist_dup(child, &spa->spa_config, 0) == 0);
136 	}
137 	mutex_exit(&spa_namespace_lock);
138 
139 	nvlist_free(nvlist);
140 
141 out:
142 	if (buf != NULL)
143 		kmem_free(buf, fsize);
144 
145 	kobj_close_file(file);
146 }
147 
148 static void
149 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
150 {
151 	size_t buflen;
152 	char *buf;
153 	vnode_t *vp;
154 	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
155 	char *temp;
156 
157 	/*
158 	 * If the nvlist is empty (NULL), then remove the old cachefile.
159 	 */
160 	if (nvl == NULL) {
161 		(void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
162 		return;
163 	}
164 
165 	/*
166 	 * Pack the configuration into a buffer.
167 	 */
168 	VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
169 
170 	buf = kmem_alloc(buflen, KM_SLEEP);
171 	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
172 
173 	VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
174 	    KM_SLEEP) == 0);
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 	if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
184 		if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
185 		    0, RLIM64_INFINITY, kcred, NULL) == 0 &&
186 		    VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
187 			(void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
188 		}
189 		(void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
190 		VN_RELE(vp);
191 	}
192 
193 	(void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
194 
195 	kmem_free(buf, buflen);
196 	kmem_free(temp, MAXPATHLEN);
197 }
198 
199 /*
200  * Synchronize pool configuration to disk.  This must be called with the
201  * namespace lock held.
202  */
203 void
204 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
205 {
206 	spa_config_dirent_t *dp, *tdp;
207 	nvlist_t *nvl;
208 
209 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
210 
211 	/*
212 	 * Iterate over all cachefiles for the pool, past or present.  When the
213 	 * cachefile is changed, the new one is pushed onto this list, allowing
214 	 * us to update previous cachefiles that no longer contain this pool.
215 	 */
216 	for (dp = list_head(&target->spa_config_list); dp != NULL;
217 	    dp = list_next(&target->spa_config_list, dp)) {
218 		spa_t *spa = NULL;
219 		if (dp->scd_path == NULL)
220 			continue;
221 
222 		/*
223 		 * Iterate over all pools, adding any matching pools to 'nvl'.
224 		 */
225 		nvl = NULL;
226 		while ((spa = spa_next(spa)) != NULL) {
227 			if (spa == target && removing)
228 				continue;
229 
230 			mutex_enter(&spa->spa_props_lock);
231 			tdp = list_head(&spa->spa_config_list);
232 			if (spa->spa_config == NULL ||
233 			    tdp->scd_path == NULL ||
234 			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
235 				mutex_exit(&spa->spa_props_lock);
236 				continue;
237 			}
238 
239 			if (nvl == NULL)
240 				VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
241 				    KM_SLEEP) == 0);
242 
243 			VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
244 			    spa->spa_config) == 0);
245 			mutex_exit(&spa->spa_props_lock);
246 		}
247 
248 		spa_config_write(dp, nvl);
249 		nvlist_free(nvl);
250 	}
251 
252 	/*
253 	 * Remove any config entries older than the current one.
254 	 */
255 	dp = list_head(&target->spa_config_list);
256 	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
257 		list_remove(&target->spa_config_list, tdp);
258 		if (tdp->scd_path != NULL)
259 			spa_strfree(tdp->scd_path);
260 		kmem_free(tdp, sizeof (spa_config_dirent_t));
261 	}
262 
263 	spa_config_generation++;
264 
265 	if (postsysevent)
266 		spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
267 }
268 
269 /*
270  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
271  * and we don't want to allow the local zone to see all the pools anyway.
272  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
273  * information for all pool visible within the zone.
274  */
275 nvlist_t *
276 spa_all_configs(uint64_t *generation)
277 {
278 	nvlist_t *pools;
279 	spa_t *spa = NULL;
280 
281 	if (*generation == spa_config_generation)
282 		return (NULL);
283 
284 	VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
285 
286 	mutex_enter(&spa_namespace_lock);
287 	while ((spa = spa_next(spa)) != NULL) {
288 		if (INGLOBALZONE(curproc) ||
289 		    zone_dataset_visible(spa_name(spa), NULL)) {
290 			mutex_enter(&spa->spa_props_lock);
291 			VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
292 			    spa->spa_config) == 0);
293 			mutex_exit(&spa->spa_props_lock);
294 		}
295 	}
296 	*generation = spa_config_generation;
297 	mutex_exit(&spa_namespace_lock);
298 
299 	return (pools);
300 }
301 
302 void
303 spa_config_set(spa_t *spa, nvlist_t *config)
304 {
305 	mutex_enter(&spa->spa_props_lock);
306 	if (spa->spa_config != NULL)
307 		nvlist_free(spa->spa_config);
308 	spa->spa_config = config;
309 	mutex_exit(&spa->spa_props_lock);
310 }
311 
312 /*
313  * Generate the pool's configuration based on the current in-core state.
314  * We infer whether to generate a complete config or just one top-level config
315  * based on whether vd is the root vdev.
316  */
317 nvlist_t *
318 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
319 {
320 	nvlist_t *config, *nvroot;
321 	vdev_t *rvd = spa->spa_root_vdev;
322 	unsigned long hostid = 0;
323 	boolean_t locked = B_FALSE;
324 
325 	if (vd == NULL) {
326 		vd = rvd;
327 		locked = B_TRUE;
328 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
329 	}
330 
331 	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
332 	    (SCL_CONFIG | SCL_STATE));
333 
334 	/*
335 	 * If txg is -1, report the current value of spa->spa_config_txg.
336 	 */
337 	if (txg == -1ULL)
338 		txg = spa->spa_config_txg;
339 
340 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
341 
342 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
343 	    spa_version(spa)) == 0);
344 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
345 	    spa_name(spa)) == 0);
346 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
347 	    spa_state(spa)) == 0);
348 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
349 	    txg) == 0);
350 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
351 	    spa_guid(spa)) == 0);
352 	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
353 	if (hostid != 0) {
354 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
355 		    hostid) == 0);
356 	}
357 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
358 	    utsname.nodename) == 0);
359 
360 	if (vd != rvd) {
361 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
362 		    vd->vdev_top->vdev_guid) == 0);
363 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
364 		    vd->vdev_guid) == 0);
365 		if (vd->vdev_isspare)
366 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
367 			    1ULL) == 0);
368 		if (vd->vdev_islog)
369 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
370 			    1ULL) == 0);
371 		vd = vd->vdev_top;		/* label contains top config */
372 	}
373 
374 	nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE);
375 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
376 	nvlist_free(nvroot);
377 
378 	if (locked)
379 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
380 
381 	return (config);
382 }
383 
384 /*
385  * For a pool that's not currently a booting rootpool, update all disk labels,
386  * generate a fresh config based on the current in-core state, and sync the
387  * global config cache.
388  */
389 void
390 spa_config_update(spa_t *spa, int what)
391 {
392 	spa_config_update_common(spa, what, FALSE);
393 }
394 
395 /*
396  * Update all disk labels, generate a fresh config based on the current
397  * in-core state, and sync the global config cache (do not sync the config
398  * cache if this is a booting rootpool).
399  */
400 void
401 spa_config_update_common(spa_t *spa, int what, boolean_t isroot)
402 {
403 	vdev_t *rvd = spa->spa_root_vdev;
404 	uint64_t txg;
405 	int c;
406 
407 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
408 
409 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
410 	txg = spa_last_synced_txg(spa) + 1;
411 	if (what == SPA_CONFIG_UPDATE_POOL) {
412 		vdev_config_dirty(rvd);
413 	} else {
414 		/*
415 		 * If we have top-level vdevs that were added but have
416 		 * not yet been prepared for allocation, do that now.
417 		 * (It's safe now because the config cache is up to date,
418 		 * so it will be able to translate the new DVAs.)
419 		 * See comments in spa_vdev_add() for full details.
420 		 */
421 		for (c = 0; c < rvd->vdev_children; c++) {
422 			vdev_t *tvd = rvd->vdev_child[c];
423 			if (tvd->vdev_ms_array == 0) {
424 				vdev_init(tvd, txg);
425 				vdev_config_dirty(tvd);
426 			}
427 		}
428 	}
429 	spa_config_exit(spa, SCL_ALL, FTAG);
430 
431 	/*
432 	 * Wait for the mosconfig to be regenerated and synced.
433 	 */
434 	txg_wait_synced(spa->spa_dsl_pool, txg);
435 
436 	/*
437 	 * Update the global config cache to reflect the new mosconfig.
438 	 */
439 	if (!isroot)
440 		spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
441 
442 	if (what == SPA_CONFIG_UPDATE_POOL)
443 		spa_config_update_common(spa, SPA_CONFIG_UPDATE_VDEVS, isroot);
444 }
445