xref: /titanic_51/usr/src/uts/common/fs/zfs/spa_config.c (revision c8e880c1386b032ac975c61826ba3bc0d8dce5ac)
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 2007 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  * The configuration for all pools, in addition to being stored on disk, is
47  * stored in /etc/zfs/zpool.cache as a packed nvlist.  The kernel maintains
48  * this list as pools are created, destroyed, or modified.
49  *
50  * We have a single nvlist which holds all the configuration information.  When
51  * the module loads, we read this information from the cache and populate the
52  * SPA namespace.  This namespace is maintained independently in spa.c.
53  * Whenever the namespace is modified, or the configuration of a pool is
54  * changed, we call spa_config_sync(), which walks through all the active pools
55  * 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_dir = ZPOOL_CACHE_DIR;
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[128];
79 	struct _buf *file;
80 	uint64_t fsize;
81 
82 	/*
83 	 * Open the configuration file.
84 	 */
85 	(void) snprintf(pathname, sizeof (pathname), "%s%s/%s",
86 	    (rootdir != NULL) ? "./" : "", spa_config_dir, ZPOOL_CACHE_FILE);
87 
88 	file = kobj_open_file(pathname);
89 	if (file == (struct _buf *)-1)
90 		return;
91 
92 	if (kobj_get_filesize(file, &fsize) != 0)
93 		goto out;
94 
95 	buf = kmem_alloc(fsize, KM_SLEEP);
96 
97 	/*
98 	 * Read the nvlist from the file.
99 	 */
100 	if (kobj_read_file(file, buf, fsize, 0) < 0)
101 		goto out;
102 
103 	/*
104 	 * Unpack the nvlist.
105 	 */
106 	if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
107 		goto out;
108 
109 	/*
110 	 * Iterate over all elements in the nvlist, creating a new spa_t for
111 	 * each one with the specified configuration.
112 	 */
113 	mutex_enter(&spa_namespace_lock);
114 	nvpair = NULL;
115 	while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
116 
117 		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
118 			continue;
119 
120 		VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
121 
122 		if (spa_lookup(nvpair_name(nvpair)) != NULL)
123 			continue;
124 		spa = spa_add(nvpair_name(nvpair), NULL);
125 
126 		/*
127 		 * We blindly duplicate the configuration here.  If it's
128 		 * invalid, we will catch it when the pool is first opened.
129 		 */
130 		VERIFY(nvlist_dup(child, &spa->spa_config, 0) == 0);
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 /*
144  * Synchronize all pools to disk.  This must be called with the namespace lock
145  * held.
146  */
147 void
148 spa_config_sync(void)
149 {
150 	spa_t *spa = NULL;
151 	nvlist_t *config;
152 	size_t buflen;
153 	char *buf;
154 	vnode_t *vp;
155 	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
156 	char pathname[128];
157 	char pathname2[128];
158 
159 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
160 
161 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
162 
163 	/*
164 	 * Add all known pools to the configuration list, ignoring those with
165 	 * alternate root paths.
166 	 */
167 	spa = NULL;
168 	while ((spa = spa_next(spa)) != NULL) {
169 		mutex_enter(&spa->spa_config_cache_lock);
170 		if (spa->spa_config && spa->spa_name && spa->spa_root == NULL)
171 			VERIFY(nvlist_add_nvlist(config, spa->spa_name,
172 			    spa->spa_config) == 0);
173 		mutex_exit(&spa->spa_config_cache_lock);
174 	}
175 
176 	/*
177 	 * Pack the configuration into a buffer.
178 	 */
179 	VERIFY(nvlist_size(config, &buflen, NV_ENCODE_XDR) == 0);
180 
181 	buf = kmem_alloc(buflen, KM_SLEEP);
182 
183 	VERIFY(nvlist_pack(config, &buf, &buflen, NV_ENCODE_XDR,
184 	    KM_SLEEP) == 0);
185 
186 	/*
187 	 * Write the configuration to disk.  We need to do the traditional
188 	 * 'write to temporary file, sync, move over original' to make sure we
189 	 * always have a consistent view of the data.
190 	 */
191 	(void) snprintf(pathname, sizeof (pathname), "%s/%s", spa_config_dir,
192 	    ZPOOL_CACHE_TMP);
193 
194 	if (vn_open(pathname, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) != 0)
195 		goto out;
196 
197 	if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
198 	    0, RLIM64_INFINITY, kcred, NULL) == 0 &&
199 	    VOP_FSYNC(vp, FSYNC, kcred) == 0) {
200 		(void) snprintf(pathname2, sizeof (pathname2), "%s/%s",
201 		    spa_config_dir, ZPOOL_CACHE_FILE);
202 		(void) vn_rename(pathname, pathname2, UIO_SYSSPACE);
203 	}
204 
205 	(void) VOP_CLOSE(vp, oflags, 1, 0, kcred);
206 	VN_RELE(vp);
207 
208 out:
209 	(void) vn_remove(pathname, UIO_SYSSPACE, RMFILE);
210 	spa_config_generation++;
211 
212 	kmem_free(buf, buflen);
213 	nvlist_free(config);
214 }
215 
216 /*
217  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
218  * and we don't want to allow the local zone to see all the pools anyway.
219  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
220  * information for all pool visible within the zone.
221  */
222 nvlist_t *
223 spa_all_configs(uint64_t *generation)
224 {
225 	nvlist_t *pools;
226 	spa_t *spa;
227 
228 	if (*generation == spa_config_generation)
229 		return (NULL);
230 
231 	VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
232 
233 	spa = NULL;
234 	mutex_enter(&spa_namespace_lock);
235 	while ((spa = spa_next(spa)) != NULL) {
236 		if (INGLOBALZONE(curproc) ||
237 		    zone_dataset_visible(spa_name(spa), NULL)) {
238 			mutex_enter(&spa->spa_config_cache_lock);
239 			VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
240 			    spa->spa_config) == 0);
241 			mutex_exit(&spa->spa_config_cache_lock);
242 		}
243 	}
244 	mutex_exit(&spa_namespace_lock);
245 
246 	*generation = spa_config_generation;
247 
248 	return (pools);
249 }
250 
251 void
252 spa_config_set(spa_t *spa, nvlist_t *config)
253 {
254 	mutex_enter(&spa->spa_config_cache_lock);
255 	if (spa->spa_config != NULL)
256 		nvlist_free(spa->spa_config);
257 	spa->spa_config = config;
258 	mutex_exit(&spa->spa_config_cache_lock);
259 }
260 
261 /*
262  * Generate the pool's configuration based on the current in-core state.
263  * We infer whether to generate a complete config or just one top-level config
264  * based on whether vd is the root vdev.
265  */
266 nvlist_t *
267 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
268 {
269 	nvlist_t *config, *nvroot;
270 	vdev_t *rvd = spa->spa_root_vdev;
271 	unsigned long hostid = 0;
272 
273 	ASSERT(spa_config_held(spa, RW_READER) ||
274 	    spa_config_held(spa, RW_WRITER));
275 
276 	if (vd == NULL)
277 		vd = rvd;
278 
279 	/*
280 	 * If txg is -1, report the current value of spa->spa_config_txg.
281 	 */
282 	if (txg == -1ULL)
283 		txg = spa->spa_config_txg;
284 
285 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
286 
287 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
288 	    spa_version(spa)) == 0);
289 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
290 	    spa_name(spa)) == 0);
291 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
292 	    spa_state(spa)) == 0);
293 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
294 	    txg) == 0);
295 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
296 	    spa_guid(spa)) == 0);
297 	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
298 	if (hostid != 0) {
299 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
300 		    hostid) == 0);
301 	}
302 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
303 	    utsname.nodename) == 0);
304 
305 	if (vd != rvd) {
306 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
307 		    vd->vdev_top->vdev_guid) == 0);
308 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
309 		    vd->vdev_guid) == 0);
310 		if (vd->vdev_isspare)
311 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
312 			    1ULL) == 0);
313 		if (vd->vdev_islog)
314 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
315 			    1ULL) == 0);
316 		vd = vd->vdev_top;		/* label contains top config */
317 	}
318 
319 	nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE);
320 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
321 	nvlist_free(nvroot);
322 
323 	return (config);
324 }
325 
326 /*
327  * Update all disk labels, generate a fresh config based on the current
328  * in-core state, and sync the global config cache.
329  */
330 void
331 spa_config_update(spa_t *spa, int what)
332 {
333 	vdev_t *rvd = spa->spa_root_vdev;
334 	uint64_t txg;
335 	int c;
336 
337 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
338 
339 	spa_config_enter(spa, RW_WRITER, FTAG);
340 	txg = spa_last_synced_txg(spa) + 1;
341 	if (what == SPA_CONFIG_UPDATE_POOL) {
342 		vdev_config_dirty(rvd);
343 	} else {
344 		/*
345 		 * If we have top-level vdevs that were added but have
346 		 * not yet been prepared for allocation, do that now.
347 		 * (It's safe now because the config cache is up to date,
348 		 * so it will be able to translate the new DVAs.)
349 		 * See comments in spa_vdev_add() for full details.
350 		 */
351 		for (c = 0; c < rvd->vdev_children; c++) {
352 			vdev_t *tvd = rvd->vdev_child[c];
353 			if (tvd->vdev_ms_array == 0) {
354 				vdev_init(tvd, txg);
355 				vdev_config_dirty(tvd);
356 			}
357 		}
358 	}
359 	spa_config_exit(spa, FTAG);
360 
361 	/*
362 	 * Wait for the mosconfig to be regenerated and synced.
363 	 */
364 	txg_wait_synced(spa->spa_dsl_pool, txg);
365 
366 	/*
367 	 * Update the global config cache to reflect the new mosconfig.
368 	 */
369 	spa_config_sync();
370 
371 	if (what == SPA_CONFIG_UPDATE_POOL)
372 		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
373 }
374