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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 */
25
26#include <sys/spa.h>
27#include <sys/spa_impl.h>
28#include <sys/nvpair.h>
29#include <sys/uio.h>
30#include <sys/fs/zfs.h>
31#include <sys/vdev_impl.h>
32#include <sys/zfs_ioctl.h>
33#include <sys/utsname.h>
34#include <sys/systeminfo.h>
35#include <sys/sunddi.h>
36#ifdef _KERNEL
37#include <sys/kobj.h>
38#include <sys/zone.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
58static 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 */
64const 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 */
71void
72spa_config_load(void)
73{
74	void *buf = NULL;
75	nvlist_t *nvlist, *child;
76	nvpair_t *nvpair;
77	char *pathname;
78	struct _buf *file;
79	uint64_t fsize;
80
81	/*
82	 * Open the configuration file.
83	 */
84	pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
85
86	(void) snprintf(pathname, MAXPATHLEN, "%s%s",
87	    (rootdir != NULL) ? "./" : "", spa_config_path);
88
89	file = kobj_open_file(pathname);
90
91	kmem_free(pathname, MAXPATHLEN);
92
93	if (file == (struct _buf *)-1)
94		return;
95
96	if (kobj_get_filesize(file, &fsize) != 0)
97		goto out;
98
99	buf = kmem_alloc(fsize, KM_SLEEP);
100
101	/*
102	 * Read the nvlist from the file.
103	 */
104	if (kobj_read_file(file, buf, fsize, 0) < 0)
105		goto out;
106
107	/*
108	 * Unpack the nvlist.
109	 */
110	if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
111		goto out;
112
113	/*
114	 * Iterate over all elements in the nvlist, creating a new spa_t for
115	 * each one with the specified configuration.
116	 */
117	mutex_enter(&spa_namespace_lock);
118	nvpair = NULL;
119	while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
120		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
121			continue;
122
123		VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
124
125		if (spa_lookup(nvpair_name(nvpair)) != NULL)
126			continue;
127		(void) spa_add(nvpair_name(nvpair), child, NULL);
128	}
129	mutex_exit(&spa_namespace_lock);
130
131	nvlist_free(nvlist);
132
133out:
134	if (buf != NULL)
135		kmem_free(buf, fsize);
136
137	kobj_close_file(file);
138}
139
140static void
141spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
142{
143	size_t buflen;
144	char *buf;
145	vnode_t *vp;
146	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
147	char *temp;
148
149	/*
150	 * If the nvlist is empty (NULL), then remove the old cachefile.
151	 */
152	if (nvl == NULL) {
153		(void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
154		return;
155	}
156
157	/*
158	 * Pack the configuration into a buffer.
159	 */
160	VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
161
162	buf = kmem_alloc(buflen, KM_SLEEP);
163	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
164
165	VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
166	    KM_SLEEP) == 0);
167
168	/*
169	 * Write the configuration to disk.  We need to do the traditional
170	 * 'write to temporary file, sync, move over original' to make sure we
171	 * always have a consistent view of the data.
172	 */
173	(void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
174
175	if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
176		if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
177		    0, RLIM64_INFINITY, kcred, NULL) == 0 &&
178		    VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
179			(void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
180		}
181		(void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
182		VN_RELE(vp);
183	}
184
185	(void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
186
187	kmem_free(buf, buflen);
188	kmem_free(temp, MAXPATHLEN);
189}
190
191/*
192 * Synchronize pool configuration to disk.  This must be called with the
193 * namespace lock held.
194 */
195void
196spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
197{
198	spa_config_dirent_t *dp, *tdp;
199	nvlist_t *nvl;
200
201	ASSERT(MUTEX_HELD(&spa_namespace_lock));
202
203	if (rootdir == NULL || !(spa_mode_global & FWRITE))
204		return;
205
206	/*
207	 * Iterate over all cachefiles for the pool, past or present.  When the
208	 * cachefile is changed, the new one is pushed onto this list, allowing
209	 * us to update previous cachefiles that no longer contain this pool.
210	 */
211	for (dp = list_head(&target->spa_config_list); dp != NULL;
212	    dp = list_next(&target->spa_config_list, dp)) {
213		spa_t *spa = NULL;
214		if (dp->scd_path == NULL)
215			continue;
216
217		/*
218		 * Iterate over all pools, adding any matching pools to 'nvl'.
219		 */
220		nvl = NULL;
221		while ((spa = spa_next(spa)) != NULL) {
222			if (spa == target && removing)
223				continue;
224
225			mutex_enter(&spa->spa_props_lock);
226			tdp = list_head(&spa->spa_config_list);
227			if (spa->spa_config == NULL ||
228			    tdp->scd_path == NULL ||
229			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
230				mutex_exit(&spa->spa_props_lock);
231				continue;
232			}
233
234			if (nvl == NULL)
235				VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
236				    KM_SLEEP) == 0);
237
238			VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
239			    spa->spa_config) == 0);
240			mutex_exit(&spa->spa_props_lock);
241		}
242
243		spa_config_write(dp, nvl);
244		nvlist_free(nvl);
245	}
246
247	/*
248	 * Remove any config entries older than the current one.
249	 */
250	dp = list_head(&target->spa_config_list);
251	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
252		list_remove(&target->spa_config_list, tdp);
253		if (tdp->scd_path != NULL)
254			spa_strfree(tdp->scd_path);
255		kmem_free(tdp, sizeof (spa_config_dirent_t));
256	}
257
258	spa_config_generation++;
259
260	if (postsysevent)
261		spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
262}
263
264/*
265 * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
266 * and we don't want to allow the local zone to see all the pools anyway.
267 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
268 * information for all pool visible within the zone.
269 */
270nvlist_t *
271spa_all_configs(uint64_t *generation)
272{
273	nvlist_t *pools;
274	spa_t *spa = NULL;
275
276	if (*generation == spa_config_generation)
277		return (NULL);
278
279	VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
280
281	mutex_enter(&spa_namespace_lock);
282	while ((spa = spa_next(spa)) != NULL) {
283		if (INGLOBALZONE(curproc) ||
284		    zone_dataset_visible(spa_name(spa), NULL)) {
285			mutex_enter(&spa->spa_props_lock);
286			VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
287			    spa->spa_config) == 0);
288			mutex_exit(&spa->spa_props_lock);
289		}
290	}
291	*generation = spa_config_generation;
292	mutex_exit(&spa_namespace_lock);
293
294	return (pools);
295}
296
297void
298spa_config_set(spa_t *spa, nvlist_t *config)
299{
300	mutex_enter(&spa->spa_props_lock);
301	if (spa->spa_config != NULL)
302		nvlist_free(spa->spa_config);
303	spa->spa_config = config;
304	mutex_exit(&spa->spa_props_lock);
305}
306
307/*
308 * Generate the pool's configuration based on the current in-core state.
309 * We infer whether to generate a complete config or just one top-level config
310 * based on whether vd is the root vdev.
311 */
312nvlist_t *
313spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
314{
315	nvlist_t *config, *nvroot;
316	vdev_t *rvd = spa->spa_root_vdev;
317	unsigned long hostid = 0;
318	boolean_t locked = B_FALSE;
319	uint64_t split_guid;
320
321	if (vd == NULL) {
322		vd = rvd;
323		locked = B_TRUE;
324		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
325	}
326
327	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
328	    (SCL_CONFIG | SCL_STATE));
329
330	/*
331	 * If txg is -1, report the current value of spa->spa_config_txg.
332	 */
333	if (txg == -1ULL)
334		txg = spa->spa_config_txg;
335
336	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
337
338	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
339	    spa_version(spa)) == 0);
340	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
341	    spa_name(spa)) == 0);
342	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
343	    spa_state(spa)) == 0);
344	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
345	    txg) == 0);
346	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
347	    spa_guid(spa)) == 0);
348#ifdef	_KERNEL
349	hostid = zone_get_hostid(NULL);
350#else	/* _KERNEL */
351	/*
352	 * We're emulating the system's hostid in userland, so we can't use
353	 * zone_get_hostid().
354	 */
355	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
356#endif	/* _KERNEL */
357	if (hostid != 0) {
358		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
359		    hostid) == 0);
360	}
361	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
362	    utsname.nodename) == 0);
363
364	if (vd != rvd) {
365		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
366		    vd->vdev_top->vdev_guid) == 0);
367		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
368		    vd->vdev_guid) == 0);
369		if (vd->vdev_isspare)
370			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
371			    1ULL) == 0);
372		if (vd->vdev_islog)
373			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
374			    1ULL) == 0);
375		vd = vd->vdev_top;		/* label contains top config */
376	} else {
377		/*
378		 * Only add the (potentially large) split information
379		 * in the mos config, and not in the vdev labels
380		 */
381		if (spa->spa_config_splitting != NULL)
382			VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
383			    spa->spa_config_splitting) == 0);
384	}
385
386	/*
387	 * Add the top-level config.  We even add this on pools which
388	 * don't support holes in the namespace.
389	 */
390	vdev_top_config_generate(spa, config);
391
392	/*
393	 * If we're splitting, record the original pool's guid.
394	 */
395	if (spa->spa_config_splitting != NULL &&
396	    nvlist_lookup_uint64(spa->spa_config_splitting,
397	    ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
398		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
399		    split_guid) == 0);
400	}
401
402	nvroot = vdev_config_generate(spa, vd, getstats, 0);
403	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
404	nvlist_free(nvroot);
405
406	if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
407		ddt_histogram_t *ddh;
408		ddt_stat_t *dds;
409		ddt_object_t *ddo;
410
411		ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
412		ddt_get_dedup_histogram(spa, ddh);
413		VERIFY(nvlist_add_uint64_array(config,
414		    ZPOOL_CONFIG_DDT_HISTOGRAM,
415		    (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
416		kmem_free(ddh, sizeof (ddt_histogram_t));
417
418		ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
419		ddt_get_dedup_object_stats(spa, ddo);
420		VERIFY(nvlist_add_uint64_array(config,
421		    ZPOOL_CONFIG_DDT_OBJ_STATS,
422		    (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
423		kmem_free(ddo, sizeof (ddt_object_t));
424
425		dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
426		ddt_get_dedup_stats(spa, dds);
427		VERIFY(nvlist_add_uint64_array(config,
428		    ZPOOL_CONFIG_DDT_STATS,
429		    (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
430		kmem_free(dds, sizeof (ddt_stat_t));
431	}
432
433	if (locked)
434		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
435
436	return (config);
437}
438
439/*
440 * Update all disk labels, generate a fresh config based on the current
441 * in-core state, and sync the global config cache (do not sync the config
442 * cache if this is a booting rootpool).
443 */
444void
445spa_config_update(spa_t *spa, int what)
446{
447	vdev_t *rvd = spa->spa_root_vdev;
448	uint64_t txg;
449	int c;
450
451	ASSERT(MUTEX_HELD(&spa_namespace_lock));
452
453	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
454	txg = spa_last_synced_txg(spa) + 1;
455	if (what == SPA_CONFIG_UPDATE_POOL) {
456		vdev_config_dirty(rvd);
457	} else {
458		/*
459		 * If we have top-level vdevs that were added but have
460		 * not yet been prepared for allocation, do that now.
461		 * (It's safe now because the config cache is up to date,
462		 * so it will be able to translate the new DVAs.)
463		 * See comments in spa_vdev_add() for full details.
464		 */
465		for (c = 0; c < rvd->vdev_children; c++) {
466			vdev_t *tvd = rvd->vdev_child[c];
467			if (tvd->vdev_ms_array == 0)
468				vdev_metaslab_set_size(tvd);
469			vdev_expand(tvd, txg);
470		}
471	}
472	spa_config_exit(spa, SCL_ALL, FTAG);
473
474	/*
475	 * Wait for the mosconfig to be regenerated and synced.
476	 */
477	txg_wait_synced(spa->spa_dsl_pool, txg);
478
479	/*
480	 * Update the global config cache to reflect the new mosconfig.
481	 */
482	if (!spa->spa_is_root)
483		spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
484
485	if (what == SPA_CONFIG_UPDATE_POOL)
486		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
487}
488