xref: /titanic_41/usr/src/uts/common/fs/zfs/spa_config.c (revision dbed73cbda2229fd1aa6dc5743993cae7f0a7ee9)
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 	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 		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
122 			continue;
123 
124 		VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
125 
126 		if (spa_lookup(nvpair_name(nvpair)) != NULL)
127 			continue;
128 		(void) spa_add(nvpair_name(nvpair), child, NULL);
129 	}
130 	mutex_exit(&spa_namespace_lock);
131 
132 	nvlist_free(nvlist);
133 
134 out:
135 	if (buf != NULL)
136 		kmem_free(buf, fsize);
137 
138 	kobj_close_file(file);
139 }
140 
141 static void
142 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
143 {
144 	size_t buflen;
145 	char *buf;
146 	vnode_t *vp;
147 	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
148 	char *temp;
149 
150 	/*
151 	 * If the nvlist is empty (NULL), then remove the old cachefile.
152 	 */
153 	if (nvl == NULL) {
154 		(void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
155 		return;
156 	}
157 
158 	/*
159 	 * Pack the configuration into a buffer.
160 	 */
161 	VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
162 
163 	buf = kmem_alloc(buflen, KM_SLEEP);
164 	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
165 
166 	VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
167 	    KM_SLEEP) == 0);
168 
169 	/*
170 	 * Write the configuration to disk.  We need to do the traditional
171 	 * 'write to temporary file, sync, move over original' to make sure we
172 	 * always have a consistent view of the data.
173 	 */
174 	(void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
175 
176 	if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
177 		if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
178 		    0, RLIM64_INFINITY, kcred, NULL) == 0 &&
179 		    VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
180 			(void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
181 		}
182 		(void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
183 		VN_RELE(vp);
184 	}
185 
186 	(void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
187 
188 	kmem_free(buf, buflen);
189 	kmem_free(temp, MAXPATHLEN);
190 }
191 
192 /*
193  * Synchronize pool configuration to disk.  This must be called with the
194  * namespace lock held.
195  */
196 void
197 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
198 {
199 	spa_config_dirent_t *dp, *tdp;
200 	nvlist_t *nvl;
201 
202 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
203 
204 	if (rootdir == NULL || !(spa_mode_global & FWRITE))
205 		return;
206 
207 	/*
208 	 * Iterate over all cachefiles for the pool, past or present.  When the
209 	 * cachefile is changed, the new one is pushed onto this list, allowing
210 	 * us to update previous cachefiles that no longer contain this pool.
211 	 */
212 	for (dp = list_head(&target->spa_config_list); dp != NULL;
213 	    dp = list_next(&target->spa_config_list, dp)) {
214 		spa_t *spa = NULL;
215 		if (dp->scd_path == NULL)
216 			continue;
217 
218 		/*
219 		 * Iterate over all pools, adding any matching pools to 'nvl'.
220 		 */
221 		nvl = NULL;
222 		while ((spa = spa_next(spa)) != NULL) {
223 			if (spa == target && removing)
224 				continue;
225 
226 			mutex_enter(&spa->spa_props_lock);
227 			tdp = list_head(&spa->spa_config_list);
228 			if (spa->spa_config == NULL ||
229 			    tdp->scd_path == NULL ||
230 			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
231 				mutex_exit(&spa->spa_props_lock);
232 				continue;
233 			}
234 
235 			if (nvl == NULL)
236 				VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
237 				    KM_SLEEP) == 0);
238 
239 			VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
240 			    spa->spa_config) == 0);
241 			mutex_exit(&spa->spa_props_lock);
242 		}
243 
244 		spa_config_write(dp, nvl);
245 		nvlist_free(nvl);
246 	}
247 
248 	/*
249 	 * Remove any config entries older than the current one.
250 	 */
251 	dp = list_head(&target->spa_config_list);
252 	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
253 		list_remove(&target->spa_config_list, tdp);
254 		if (tdp->scd_path != NULL)
255 			spa_strfree(tdp->scd_path);
256 		kmem_free(tdp, sizeof (spa_config_dirent_t));
257 	}
258 
259 	spa_config_generation++;
260 
261 	if (postsysevent)
262 		spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
263 }
264 
265 /*
266  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
267  * and we don't want to allow the local zone to see all the pools anyway.
268  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
269  * information for all pool visible within the zone.
270  */
271 nvlist_t *
272 spa_all_configs(uint64_t *generation)
273 {
274 	nvlist_t *pools;
275 	spa_t *spa = NULL;
276 
277 	if (*generation == spa_config_generation)
278 		return (NULL);
279 
280 	VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
281 
282 	mutex_enter(&spa_namespace_lock);
283 	while ((spa = spa_next(spa)) != NULL) {
284 		if (INGLOBALZONE(curproc) ||
285 		    zone_dataset_visible(spa_name(spa), NULL)) {
286 			mutex_enter(&spa->spa_props_lock);
287 			VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
288 			    spa->spa_config) == 0);
289 			mutex_exit(&spa->spa_props_lock);
290 		}
291 	}
292 	*generation = spa_config_generation;
293 	mutex_exit(&spa_namespace_lock);
294 
295 	return (pools);
296 }
297 
298 void
299 spa_config_set(spa_t *spa, nvlist_t *config)
300 {
301 	mutex_enter(&spa->spa_props_lock);
302 	if (spa->spa_config != NULL)
303 		nvlist_free(spa->spa_config);
304 	spa->spa_config = config;
305 	mutex_exit(&spa->spa_props_lock);
306 }
307 
308 /* Add discovered rewind info, if any to the provided nvlist */
309 void
310 spa_rewind_data_to_nvlist(spa_t *spa, nvlist_t *tonvl)
311 {
312 	int64_t loss = 0;
313 
314 	if (tonvl == NULL || spa->spa_load_txg == 0)
315 		return;
316 
317 	VERIFY(nvlist_add_uint64(tonvl, ZPOOL_CONFIG_LOAD_TIME,
318 	    spa->spa_load_txg_ts) == 0);
319 	if (spa->spa_last_ubsync_txg)
320 		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
321 	VERIFY(nvlist_add_int64(tonvl, ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
322 	VERIFY(nvlist_add_uint64(tonvl, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
323 	    spa->spa_load_data_errors) == 0);
324 }
325 
326 /*
327  * Generate the pool's configuration based on the current in-core state.
328  * We infer whether to generate a complete config or just one top-level config
329  * based on whether vd is the root vdev.
330  */
331 nvlist_t *
332 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
333 {
334 	nvlist_t *config, *nvroot;
335 	vdev_t *rvd = spa->spa_root_vdev;
336 	unsigned long hostid = 0;
337 	boolean_t locked = B_FALSE;
338 
339 	if (vd == NULL) {
340 		vd = rvd;
341 		locked = B_TRUE;
342 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
343 	}
344 
345 	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
346 	    (SCL_CONFIG | SCL_STATE));
347 
348 	/*
349 	 * If txg is -1, report the current value of spa->spa_config_txg.
350 	 */
351 	if (txg == -1ULL)
352 		txg = spa->spa_config_txg;
353 
354 	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
355 
356 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
357 	    spa_version(spa)) == 0);
358 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
359 	    spa_name(spa)) == 0);
360 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
361 	    spa_state(spa)) == 0);
362 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
363 	    txg) == 0);
364 	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
365 	    spa_guid(spa)) == 0);
366 #ifdef	_KERNEL
367 	hostid = zone_get_hostid(NULL);
368 #else	/* _KERNEL */
369 	/*
370 	 * We're emulating the system's hostid in userland, so we can't use
371 	 * zone_get_hostid().
372 	 */
373 	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
374 #endif	/* _KERNEL */
375 	if (hostid != 0) {
376 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
377 		    hostid) == 0);
378 	}
379 	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
380 	    utsname.nodename) == 0);
381 
382 	if (vd != rvd) {
383 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
384 		    vd->vdev_top->vdev_guid) == 0);
385 		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
386 		    vd->vdev_guid) == 0);
387 		if (vd->vdev_isspare)
388 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
389 			    1ULL) == 0);
390 		if (vd->vdev_islog)
391 			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
392 			    1ULL) == 0);
393 		vd = vd->vdev_top;		/* label contains top config */
394 	}
395 
396 	/*
397 	 * Add the top-level config.  We even add this on pools which
398 	 * don't support holes in the namespace as older pools will
399 	 * just ignore it.
400 	 */
401 	vdev_top_config_generate(spa, config);
402 
403 	nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE);
404 	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
405 	nvlist_free(nvroot);
406 
407 	spa_rewind_data_to_nvlist(spa, config);
408 
409 	if (locked)
410 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
411 
412 	return (config);
413 }
414 
415 /*
416  * Update all disk labels, generate a fresh config based on the current
417  * in-core state, and sync the global config cache (do not sync the config
418  * cache if this is a booting rootpool).
419  */
420 void
421 spa_config_update(spa_t *spa, int what)
422 {
423 	vdev_t *rvd = spa->spa_root_vdev;
424 	uint64_t txg;
425 	int c;
426 
427 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
428 
429 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
430 	txg = spa_last_synced_txg(spa) + 1;
431 	if (what == SPA_CONFIG_UPDATE_POOL) {
432 		vdev_config_dirty(rvd);
433 	} else {
434 		/*
435 		 * If we have top-level vdevs that were added but have
436 		 * not yet been prepared for allocation, do that now.
437 		 * (It's safe now because the config cache is up to date,
438 		 * so it will be able to translate the new DVAs.)
439 		 * See comments in spa_vdev_add() for full details.
440 		 */
441 		for (c = 0; c < rvd->vdev_children; c++) {
442 			vdev_t *tvd = rvd->vdev_child[c];
443 			if (tvd->vdev_ms_array == 0)
444 				vdev_metaslab_set_size(tvd);
445 			vdev_expand(tvd, txg);
446 		}
447 	}
448 	spa_config_exit(spa, SCL_ALL, FTAG);
449 
450 	/*
451 	 * Wait for the mosconfig to be regenerated and synced.
452 	 */
453 	txg_wait_synced(spa->spa_dsl_pool, txg);
454 
455 	/*
456 	 * Update the global config cache to reflect the new mosconfig.
457 	 */
458 	if (!spa->spa_is_root)
459 		spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
460 
461 	if (what == SPA_CONFIG_UPDATE_POOL)
462 		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
463 }
464