xref: /illumos-gate/usr/src/uts/common/fs/zfs/spa_config.c (revision c5904d138f3bdf0762dbf452a43d5a5c387ea6a8)
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 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/spa.h>
30 #include <sys/spa_impl.h>
31 #include <sys/nvpair.h>
32 #include <sys/uio.h>
33 #include <sys/fs/zfs.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/zfs_ioctl.h>
36 #include <sys/utsname.h>
37 #include <sys/systeminfo.h>
38 #include <sys/sunddi.h>
39 #ifdef _KERNEL
40 #include <sys/kobj.h>
41 #endif
42 
43 /*
44  * Pool configuration repository.
45  *
46  * Pool configuration is stored as a packed nvlist on the filesystem.  By
47  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
48  * (when the ZFS module is loaded).  Pools can also have the 'cachefile'
49  * property set that allows them to be stored in an alternate location until
50  * the control of external software.
51  *
52  * For each cache file, we have a single nvlist which holds all the
53  * configuration information.  When the module loads, we read this information
54  * from /etc/zfs/zpool.cache and populate the SPA namespace.  This namespace is
55  * maintained independently in spa.c.  Whenever the namespace is modified, or
56  * the configuration of a pool is changed, we call spa_config_sync(), which
57  * walks through all the active pools and writes the configuration to disk.
58  */
59 
60 static uint64_t spa_config_generation = 1;
61 
62 /*
63  * This can be overridden in userland to preserve an alternate namespace for
64  * userland pools when doing testing.
65  */
66 const char *spa_config_path = ZPOOL_CACHE;
67 
68 /*
69  * Called when the module is first loaded, this routine loads the configuration
70  * file into the SPA namespace.  It does not actually open or load the pools; it
71  * only populates the namespace.
72  */
73 void
74 spa_config_load(void)
75 {
76 	void *buf = NULL;
77 	nvlist_t *nvlist, *child;
78 	nvpair_t *nvpair;
79 	spa_t *spa;
80 	char pathname[128];
81 	struct _buf *file;
82 	uint64_t fsize;
83 
84 	/*
85 	 * Open the configuration file.
86 	 */
87 	(void) snprintf(pathname, sizeof (pathname), "%s%s",
88 	    (rootdir != NULL) ? "./" : "", spa_config_path);
89 
90 	file = kobj_open_file(pathname);
91 	if (file == (struct _buf *)-1)
92 		return;
93 
94 	if (kobj_get_filesize(file, &fsize) != 0)
95 		goto out;
96 
97 	buf = kmem_alloc(fsize, KM_SLEEP);
98 
99 	/*
100 	 * Read the nvlist from the file.
101 	 */
102 	if (kobj_read_file(file, buf, fsize, 0) < 0)
103 		goto out;
104 
105 	/*
106 	 * Unpack the nvlist.
107 	 */
108 	if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
109 		goto out;
110 
111 	/*
112 	 * Iterate over all elements in the nvlist, creating a new spa_t for
113 	 * each one with the specified configuration.
114 	 */
115 	mutex_enter(&spa_namespace_lock);
116 	nvpair = NULL;
117 	while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
118 
119 		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
120 			continue;
121 
122 		VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
123 
124 		if (spa_lookup(nvpair_name(nvpair)) != NULL)
125 			continue;
126 		spa = spa_add(nvpair_name(nvpair), NULL);
127 
128 		/*
129 		 * We blindly duplicate the configuration here.  If it's
130 		 * invalid, we will catch it when the pool is first opened.
131 		 */
132 		VERIFY(nvlist_dup(child, &spa->spa_config, 0) == 0);
133 	}
134 	mutex_exit(&spa_namespace_lock);
135 
136 	nvlist_free(nvlist);
137 
138 out:
139 	if (buf != NULL)
140 		kmem_free(buf, fsize);
141 
142 	kobj_close_file(file);
143 }
144 
145 static void
146 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
147 {
148 	size_t buflen;
149 	char *buf;
150 	vnode_t *vp;
151 	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
152 	char tempname[128];
153 
154 	/*
155 	 * If the nvlist is empty (NULL), then remove the old cachefile.
156 	 */
157 	if (nvl == NULL) {
158 		(void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
159 		return;
160 	}
161 
162 	/*
163 	 * Pack the configuration into a buffer.
164 	 */
165 	VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
166 
167 	buf = kmem_alloc(buflen, 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(tempname, sizeof (tempname), "%s.tmp", dp->scd_path);
178 
179 	if (vn_open(tempname, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) != 0)
180 		goto out;
181 
182 	if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
183 	    0, RLIM64_INFINITY, kcred, NULL) == 0 &&
184 	    VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
185 		(void) vn_rename(tempname, dp->scd_path, UIO_SYSSPACE);
186 	}
187 
188 	(void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
189 	VN_RELE(vp);
190 
191 out:
192 	(void) vn_remove(tempname, UIO_SYSSPACE, RMFILE);
193 	kmem_free(buf, buflen);
194 }
195 
196 /*
197  * Synchronize pool configuration to disk.  This must be called with the
198  * namespace lock held.
199  */
200 void
201 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
202 {
203 	spa_t *spa = NULL;
204 	spa_config_dirent_t *dp, *tdp;
205 	nvlist_t *nvl;
206 
207 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
208 
209 	/*
210 	 * Iterate over all cachefiles for the pool, past or present.  When the
211 	 * cachefile is changed, the new one is pushed onto this list, allowing
212 	 * us to update previous cachefiles that no longer contain this pool.
213 	 */
214 	for (dp = list_head(&target->spa_config_list); dp != NULL;
215 	    dp = list_next(&target->spa_config_list, dp)) {
216 		spa = NULL;
217 		if (dp->scd_path == NULL)
218 			continue;
219 
220 		/*
221 		 * Iterate over all pools, adding any matching pools to 'nvl'.
222 		 */
223 		nvl = NULL;
224 		while ((spa = spa_next(spa)) != NULL) {
225 			if (spa->spa_config == NULL || spa->spa_name == NULL)
226 				continue;
227 
228 			if (spa == target && removing)
229 				continue;
230 
231 			tdp = list_head(&spa->spa_config_list);
232 			ASSERT(tdp != NULL);
233 			if (tdp->scd_path == NULL ||
234 			    strcmp(tdp->scd_path, dp->scd_path) != 0)
235 				continue;
236 
237 			if (nvl == NULL)
238 				VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
239 				    KM_SLEEP) == 0);
240 
241 			VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
242 			    spa->spa_config) == 0);
243 		}
244 
245 		spa_config_write(dp, nvl);
246 		nvlist_free(nvl);
247 	}
248 
249 	/*
250 	 * Remove any config entries older than the current one.
251 	 */
252 	dp = list_head(&target->spa_config_list);
253 	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
254 		list_remove(&target->spa_config_list, tdp);
255 		if (tdp->scd_path != NULL)
256 			spa_strfree(tdp->scd_path);
257 		kmem_free(tdp, sizeof (spa_config_dirent_t));
258 	}
259 
260 	spa_config_generation++;
261 
262 	if (postsysevent)
263 		spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
264 }
265 
266 /*
267  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
268  * and we don't want to allow the local zone to see all the pools anyway.
269  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
270  * information for all pool visible within the zone.
271  */
272 nvlist_t *
273 spa_all_configs(uint64_t *generation)
274 {
275 	nvlist_t *pools;
276 	spa_t *spa;
277 
278 	if (*generation == spa_config_generation)
279 		return (NULL);
280 
281 	VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
282 
283 	spa = NULL;
284 	mutex_enter(&spa_namespace_lock);
285 	while ((spa = spa_next(spa)) != NULL) {
286 		if (INGLOBALZONE(curproc) ||
287 		    zone_dataset_visible(spa_name(spa), NULL)) {
288 			mutex_enter(&spa->spa_config_cache_lock);
289 			VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
290 			    spa->spa_config) == 0);
291 			mutex_exit(&spa->spa_config_cache_lock);
292 		}
293 	}
294 	mutex_exit(&spa_namespace_lock);
295 
296 	*generation = spa_config_generation;
297 
298 	return (pools);
299 }
300 
301 void
302 spa_config_set(spa_t *spa, nvlist_t *config)
303 {
304 	mutex_enter(&spa->spa_config_cache_lock);
305 	if (spa->spa_config != NULL)
306 		nvlist_free(spa->spa_config);
307 	spa->spa_config = config;
308 	mutex_exit(&spa->spa_config_cache_lock);
309 }
310 
311 /*
312  * Generate the pool's configuration based on the current in-core state.
313  * We infer whether to generate a complete config or just one top-level config
314  * based on whether vd is the root vdev.
315  */
316 nvlist_t *
317 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
318 {
319 	nvlist_t *config, *nvroot;
320 	vdev_t *rvd = spa->spa_root_vdev;
321 	unsigned long hostid = 0;
322 
323 	ASSERT(spa_config_held(spa, RW_READER) ||
324 	    spa_config_held(spa, RW_WRITER));
325 
326 	if (vd == NULL)
327 		vd = rvd;
328 
329 	/*
330 	 * If txg is -1, report the current value of spa->spa_config_txg.
331 	 */
332 	if (txg == -1ULL)
333 		txg = spa->spa_config_txg;
334 
335 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
336 
337 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
338 	    spa_version(spa)) == 0);
339 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
340 	    spa_name(spa)) == 0);
341 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
342 	    spa_state(spa)) == 0);
343 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
344 	    txg) == 0);
345 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
346 	    spa_guid(spa)) == 0);
347 	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
348 	if (hostid != 0) {
349 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
350 		    hostid) == 0);
351 	}
352 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
353 	    utsname.nodename) == 0);
354 
355 	if (vd != rvd) {
356 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
357 		    vd->vdev_top->vdev_guid) == 0);
358 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
359 		    vd->vdev_guid) == 0);
360 		if (vd->vdev_isspare)
361 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
362 			    1ULL) == 0);
363 		if (vd->vdev_islog)
364 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
365 			    1ULL) == 0);
366 		vd = vd->vdev_top;		/* label contains top config */
367 	}
368 
369 	nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE);
370 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
371 	nvlist_free(nvroot);
372 
373 	return (config);
374 }
375 
376 /*
377  * For a pool that's not currently a booting rootpool, update all disk labels,
378  * generate a fresh config based on the current in-core state, and sync the
379  * global config cache.
380  */
381 void
382 spa_config_update(spa_t *spa, int what)
383 {
384 	spa_config_update_common(spa, what, FALSE);
385 }
386 
387 /*
388  * Update all disk labels, generate a fresh config based on the current
389  * in-core state, and sync the global config cache (do not sync the config
390  * cache if this is a booting rootpool).
391  */
392 void
393 spa_config_update_common(spa_t *spa, int what, boolean_t isroot)
394 {
395 	vdev_t *rvd = spa->spa_root_vdev;
396 	uint64_t txg;
397 	int c;
398 
399 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
400 
401 	spa_config_enter(spa, RW_WRITER, FTAG);
402 	txg = spa_last_synced_txg(spa) + 1;
403 	if (what == SPA_CONFIG_UPDATE_POOL) {
404 		vdev_config_dirty(rvd);
405 	} else {
406 		/*
407 		 * If we have top-level vdevs that were added but have
408 		 * not yet been prepared for allocation, do that now.
409 		 * (It's safe now because the config cache is up to date,
410 		 * so it will be able to translate the new DVAs.)
411 		 * See comments in spa_vdev_add() for full details.
412 		 */
413 		for (c = 0; c < rvd->vdev_children; c++) {
414 			vdev_t *tvd = rvd->vdev_child[c];
415 			if (tvd->vdev_ms_array == 0) {
416 				vdev_init(tvd, txg);
417 				vdev_config_dirty(tvd);
418 			}
419 		}
420 	}
421 	spa_config_exit(spa, FTAG);
422 
423 	/*
424 	 * Wait for the mosconfig to be regenerated and synced.
425 	 */
426 	txg_wait_synced(spa->spa_dsl_pool, txg);
427 
428 	/*
429 	 * Update the global config cache to reflect the new mosconfig.
430 	 */
431 	if (!isroot)
432 		spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
433 
434 	if (what == SPA_CONFIG_UPDATE_POOL)
435 		spa_config_update_common(spa, SPA_CONFIG_UPDATE_VDEVS, isroot);
436 }
437