xref: /titanic_50/usr/src/uts/common/fs/zfs/spa_config.c (revision 75e1bcde77704c349b44a76fabb51e083ea9a755)
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 2009 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 #include <sys/zone.h>
40 #endif
41 
42 /*
43  * Pool configuration repository.
44  *
45  * Pool configuration is stored as a packed nvlist on the filesystem.  By
46  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
47  * (when the ZFS module is loaded).  Pools can also have the 'cachefile'
48  * property set that allows them to be stored in an alternate location until
49  * the control of external software.
50  *
51  * For each cache file, we have a single nvlist which holds all the
52  * configuration information.  When the module loads, we read this information
53  * from /etc/zfs/zpool.cache and populate the SPA namespace.  This namespace is
54  * maintained independently in spa.c.  Whenever the namespace is modified, or
55  * the configuration of a pool is changed, we call spa_config_sync(), which
56  * walks through all the active pools and writes the configuration to disk.
57  */
58 
59 static uint64_t spa_config_generation = 1;
60 
61 /*
62  * This can be overridden in userland to preserve an alternate namespace for
63  * userland pools when doing testing.
64  */
65 const char *spa_config_path = ZPOOL_CACHE;
66 
67 /*
68  * Called when the module is first loaded, this routine loads the configuration
69  * file into the SPA namespace.  It does not actually open or load the pools; it
70  * only populates the namespace.
71  */
72 void
73 spa_config_load(void)
74 {
75 	void *buf = NULL;
76 	nvlist_t *nvlist, *child;
77 	nvpair_t *nvpair;
78 	spa_t *spa;
79 	char *pathname;
80 	struct _buf *file;
81 	uint64_t fsize;
82 
83 	/*
84 	 * Open the configuration file.
85 	 */
86 	pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
87 
88 	(void) snprintf(pathname, MAXPATHLEN, "%s%s",
89 	    (rootdir != NULL) ? "./" : "", spa_config_path);
90 
91 	file = kobj_open_file(pathname);
92 
93 	kmem_free(pathname, MAXPATHLEN);
94 
95 	if (file == (struct _buf *)-1)
96 		return;
97 
98 	if (kobj_get_filesize(file, &fsize) != 0)
99 		goto out;
100 
101 	buf = kmem_alloc(fsize, KM_SLEEP);
102 
103 	/*
104 	 * Read the nvlist from the file.
105 	 */
106 	if (kobj_read_file(file, buf, fsize, 0) < 0)
107 		goto out;
108 
109 	/*
110 	 * Unpack the nvlist.
111 	 */
112 	if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
113 		goto out;
114 
115 	/*
116 	 * Iterate over all elements in the nvlist, creating a new spa_t for
117 	 * each one with the specified configuration.
118 	 */
119 	mutex_enter(&spa_namespace_lock);
120 	nvpair = NULL;
121 	while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
122 
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 		spa = spa_add(nvpair_name(nvpair), NULL);
131 
132 		/*
133 		 * We blindly duplicate the configuration here.  If it's
134 		 * invalid, we will catch it when the pool is first opened.
135 		 */
136 		VERIFY(nvlist_dup(child, &spa->spa_config, 0) == 0);
137 	}
138 	mutex_exit(&spa_namespace_lock);
139 
140 	nvlist_free(nvlist);
141 
142 out:
143 	if (buf != NULL)
144 		kmem_free(buf, fsize);
145 
146 	kobj_close_file(file);
147 }
148 
149 static void
150 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
151 {
152 	size_t buflen;
153 	char *buf;
154 	vnode_t *vp;
155 	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
156 	char *temp;
157 
158 	/*
159 	 * If the nvlist is empty (NULL), then remove the old cachefile.
160 	 */
161 	if (nvl == NULL) {
162 		(void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
163 		return;
164 	}
165 
166 	/*
167 	 * Pack the configuration into a buffer.
168 	 */
169 	VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
170 
171 	buf = kmem_alloc(buflen, KM_SLEEP);
172 	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
173 
174 	VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
175 	    KM_SLEEP) == 0);
176 
177 	/*
178 	 * Write the configuration to disk.  We need to do the traditional
179 	 * 'write to temporary file, sync, move over original' to make sure we
180 	 * always have a consistent view of the data.
181 	 */
182 	(void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
183 
184 	if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
185 		if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
186 		    0, RLIM64_INFINITY, kcred, NULL) == 0 &&
187 		    VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
188 			(void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
189 		}
190 		(void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
191 		VN_RELE(vp);
192 	}
193 
194 	(void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
195 
196 	kmem_free(buf, buflen);
197 	kmem_free(temp, MAXPATHLEN);
198 }
199 
200 /*
201  * Synchronize pool configuration to disk.  This must be called with the
202  * namespace lock held.
203  */
204 void
205 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
206 {
207 	spa_config_dirent_t *dp, *tdp;
208 	nvlist_t *nvl;
209 
210 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
211 
212 	if (rootdir == NULL || !(spa_mode_global & FWRITE))
213 		return;
214 
215 	/*
216 	 * Iterate over all cachefiles for the pool, past or present.  When the
217 	 * cachefile is changed, the new one is pushed onto this list, allowing
218 	 * us to update previous cachefiles that no longer contain this pool.
219 	 */
220 	for (dp = list_head(&target->spa_config_list); dp != NULL;
221 	    dp = list_next(&target->spa_config_list, dp)) {
222 		spa_t *spa = NULL;
223 		if (dp->scd_path == NULL)
224 			continue;
225 
226 		/*
227 		 * Iterate over all pools, adding any matching pools to 'nvl'.
228 		 */
229 		nvl = NULL;
230 		while ((spa = spa_next(spa)) != NULL) {
231 			if (spa == target && removing)
232 				continue;
233 
234 			mutex_enter(&spa->spa_props_lock);
235 			tdp = list_head(&spa->spa_config_list);
236 			if (spa->spa_config == NULL ||
237 			    tdp->scd_path == NULL ||
238 			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
239 				mutex_exit(&spa->spa_props_lock);
240 				continue;
241 			}
242 
243 			if (nvl == NULL)
244 				VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
245 				    KM_SLEEP) == 0);
246 
247 			VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
248 			    spa->spa_config) == 0);
249 			mutex_exit(&spa->spa_props_lock);
250 		}
251 
252 		spa_config_write(dp, nvl);
253 		nvlist_free(nvl);
254 	}
255 
256 	/*
257 	 * Remove any config entries older than the current one.
258 	 */
259 	dp = list_head(&target->spa_config_list);
260 	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
261 		list_remove(&target->spa_config_list, tdp);
262 		if (tdp->scd_path != NULL)
263 			spa_strfree(tdp->scd_path);
264 		kmem_free(tdp, sizeof (spa_config_dirent_t));
265 	}
266 
267 	spa_config_generation++;
268 
269 	if (postsysevent)
270 		spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
271 }
272 
273 /*
274  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
275  * and we don't want to allow the local zone to see all the pools anyway.
276  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
277  * information for all pool visible within the zone.
278  */
279 nvlist_t *
280 spa_all_configs(uint64_t *generation)
281 {
282 	nvlist_t *pools;
283 	spa_t *spa = NULL;
284 
285 	if (*generation == spa_config_generation)
286 		return (NULL);
287 
288 	VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
289 
290 	mutex_enter(&spa_namespace_lock);
291 	while ((spa = spa_next(spa)) != NULL) {
292 		if (INGLOBALZONE(curproc) ||
293 		    zone_dataset_visible(spa_name(spa), NULL)) {
294 			mutex_enter(&spa->spa_props_lock);
295 			VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
296 			    spa->spa_config) == 0);
297 			mutex_exit(&spa->spa_props_lock);
298 		}
299 	}
300 	*generation = spa_config_generation;
301 	mutex_exit(&spa_namespace_lock);
302 
303 	return (pools);
304 }
305 
306 void
307 spa_config_set(spa_t *spa, nvlist_t *config)
308 {
309 	mutex_enter(&spa->spa_props_lock);
310 	if (spa->spa_config != NULL)
311 		nvlist_free(spa->spa_config);
312 	spa->spa_config = config;
313 	mutex_exit(&spa->spa_props_lock);
314 }
315 
316 /*
317  * Generate the pool's configuration based on the current in-core state.
318  * We infer whether to generate a complete config or just one top-level config
319  * based on whether vd is the root vdev.
320  */
321 nvlist_t *
322 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
323 {
324 	nvlist_t *config, *nvroot;
325 	vdev_t *rvd = spa->spa_root_vdev;
326 	unsigned long hostid = 0;
327 	boolean_t locked = B_FALSE;
328 
329 	if (vd == NULL) {
330 		vd = rvd;
331 		locked = B_TRUE;
332 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
333 	}
334 
335 	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
336 	    (SCL_CONFIG | SCL_STATE));
337 
338 	/*
339 	 * If txg is -1, report the current value of spa->spa_config_txg.
340 	 */
341 	if (txg == -1ULL)
342 		txg = spa->spa_config_txg;
343 
344 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
345 
346 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
347 	    spa_version(spa)) == 0);
348 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
349 	    spa_name(spa)) == 0);
350 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
351 	    spa_state(spa)) == 0);
352 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
353 	    txg) == 0);
354 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
355 	    spa_guid(spa)) == 0);
356 #ifdef	_KERNEL
357 	hostid = zone_get_hostid(NULL);
358 #else	/* _KERNEL */
359 	/*
360 	 * We're emulating the system's hostid in userland, so we can't use
361 	 * zone_get_hostid().
362 	 */
363 	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
364 #endif	/* _KERNEL */
365 	if (hostid != 0) {
366 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
367 		    hostid) == 0);
368 	}
369 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
370 	    utsname.nodename) == 0);
371 
372 	if (vd != rvd) {
373 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
374 		    vd->vdev_top->vdev_guid) == 0);
375 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
376 		    vd->vdev_guid) == 0);
377 		if (vd->vdev_isspare)
378 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
379 			    1ULL) == 0);
380 		if (vd->vdev_islog)
381 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
382 			    1ULL) == 0);
383 		vd = vd->vdev_top;		/* label contains top config */
384 	}
385 
386 	nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE);
387 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
388 	nvlist_free(nvroot);
389 
390 	if (locked)
391 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
392 
393 	return (config);
394 }
395 
396 /*
397  * For a pool that's not currently a booting rootpool, update all disk labels,
398  * generate a fresh config based on the current in-core state, and sync the
399  * global config cache.
400  */
401 void
402 spa_config_update(spa_t *spa, int what)
403 {
404 	spa_config_update_common(spa, what, FALSE);
405 }
406 
407 /*
408  * Update all disk labels, generate a fresh config based on the current
409  * in-core state, and sync the global config cache (do not sync the config
410  * cache if this is a booting rootpool).
411  */
412 void
413 spa_config_update_common(spa_t *spa, int what, boolean_t isroot)
414 {
415 	vdev_t *rvd = spa->spa_root_vdev;
416 	uint64_t txg;
417 	int c;
418 
419 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
420 
421 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
422 	txg = spa_last_synced_txg(spa) + 1;
423 	if (what == SPA_CONFIG_UPDATE_POOL) {
424 		vdev_config_dirty(rvd);
425 	} else {
426 		/*
427 		 * If we have top-level vdevs that were added but have
428 		 * not yet been prepared for allocation, do that now.
429 		 * (It's safe now because the config cache is up to date,
430 		 * so it will be able to translate the new DVAs.)
431 		 * See comments in spa_vdev_add() for full details.
432 		 */
433 		for (c = 0; c < rvd->vdev_children; c++) {
434 			vdev_t *tvd = rvd->vdev_child[c];
435 			if (tvd->vdev_ms_array == 0)
436 				vdev_metaslab_set_size(tvd);
437 			vdev_expand(tvd, txg);
438 		}
439 	}
440 	spa_config_exit(spa, SCL_ALL, FTAG);
441 
442 	/*
443 	 * Wait for the mosconfig to be regenerated and synced.
444 	 */
445 	txg_wait_synced(spa->spa_dsl_pool, txg);
446 
447 	/*
448 	 * Update the global config cache to reflect the new mosconfig.
449 	 */
450 	if (!isroot)
451 		spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
452 
453 	if (what == SPA_CONFIG_UPDATE_POOL)
454 		spa_config_update_common(spa, SPA_CONFIG_UPDATE_VDEVS, isroot);
455 }
456