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 2010 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
59static 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 */
65const 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 */
72void
73spa_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	(void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path);
87
88	file = kobj_open_file(pathname);
89
90	kmem_free(pathname, MAXPATHLEN);
91
92	if (file == (struct _buf *)-1)
93		return;
94
95	if (kobj_get_filesize(file, &fsize) != 0)
96		goto out;
97
98	if (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		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
123			continue;
124
125		VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
126
127		if (spa_lookup(nvpair_name(nvpair)) != NULL)
128			continue;
129		(void) spa_add(nvpair_name(nvpair), child, NULL);
130	}
131	mutex_exit(&spa_namespace_lock);
132
133	nvlist_free(nvlist);
134
135out:
136	if (buf != NULL)
137		kmem_free(buf, fsize);
138
139	kobj_close_file(file);
140}
141
142static void
143spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
144{
145	size_t buflen;
146	char *buf;
147	vnode_t *vp;
148	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
149	char *temp;
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	}
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			mutex_enter(&spa->spa_props_lock);
225			tdp = list_head(&spa->spa_config_list);
226			if (spa->spa_config == NULL ||
227			    tdp->scd_path == NULL ||
228			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
229				mutex_exit(&spa->spa_props_lock);
230				continue;
231			}
232
233			if (nvl == NULL)
234				VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
235				    KM_SLEEP) == 0);
236
237			VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
238			    spa->spa_config) == 0);
239			mutex_exit(&spa->spa_props_lock);
240		}
241
242		spa_config_write(dp, nvl);
243		nvlist_free(nvl);
244	}
245
246	/*
247	 * Remove any config entries older than the current one.
248	 */
249	dp = list_head(&target->spa_config_list);
250	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
251		list_remove(&target->spa_config_list, tdp);
252		if (tdp->scd_path != NULL)
253			spa_strfree(tdp->scd_path);
254		kmem_free(tdp, sizeof (spa_config_dirent_t));
255	}
256
257	spa_config_generation++;
258
259	if (postsysevent)
260		spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
261}
262
263/*
264 * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
265 * and we don't want to allow the local zone to see all the pools anyway.
266 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
267 * information for all pool visible within the zone.
268 */
269nvlist_t *
270spa_all_configs(uint64_t *generation)
271{
272	nvlist_t *pools;
273	spa_t *spa = NULL;
274
275	if (*generation == spa_config_generation)
276		return (NULL);
277
278	VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
279
280	mutex_enter(&spa_namespace_lock);
281	while ((spa = spa_next(spa)) != NULL) {
282		if (INGLOBALZONE(curproc) ||
283		    zone_dataset_visible(spa_name(spa), NULL)) {
284			mutex_enter(&spa->spa_props_lock);
285			VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
286			    spa->spa_config) == 0);
287			mutex_exit(&spa->spa_props_lock);
288		}
289	}
290	*generation = spa_config_generation;
291	mutex_exit(&spa_namespace_lock);
292
293	return (pools);
294}
295
296void
297spa_config_set(spa_t *spa, nvlist_t *config)
298{
299	mutex_enter(&spa->spa_props_lock);
300	if (spa->spa_config != NULL)
301		nvlist_free(spa->spa_config);
302	spa->spa_config = config;
303	mutex_exit(&spa->spa_props_lock);
304}
305
306/* Add discovered rewind info, if any to the provided nvlist */
307void
308spa_rewind_data_to_nvlist(spa_t *spa, nvlist_t *tonvl)
309{
310	int64_t loss = 0;
311
312	if (tonvl == NULL || spa->spa_load_txg == 0)
313		return;
314
315	VERIFY(nvlist_add_uint64(tonvl, ZPOOL_CONFIG_LOAD_TIME,
316	    spa->spa_load_txg_ts) == 0);
317	if (spa->spa_last_ubsync_txg)
318		loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
319	VERIFY(nvlist_add_int64(tonvl, ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
320	VERIFY(nvlist_add_uint64(tonvl, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
321	    spa->spa_load_data_errors) == 0);
322}
323
324/*
325 * Generate the pool's configuration based on the current in-core state.
326 * We infer whether to generate a complete config or just one top-level config
327 * based on whether vd is the root vdev.
328 */
329nvlist_t *
330spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
331{
332	nvlist_t *config, *nvroot;
333	vdev_t *rvd = spa->spa_root_vdev;
334	unsigned long hostid = 0;
335	boolean_t locked = B_FALSE;
336	uint64_t split_guid;
337
338	if (vd == NULL) {
339		vd = rvd;
340		locked = B_TRUE;
341		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
342	}
343
344	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
345	    (SCL_CONFIG | SCL_STATE));
346
347	/*
348	 * If txg is -1, report the current value of spa->spa_config_txg.
349	 */
350	if (txg == -1ULL)
351		txg = spa->spa_config_txg;
352
353	VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
354
355	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
356	    spa_version(spa)) == 0);
357	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
358	    spa_name(spa)) == 0);
359	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
360	    spa_state(spa)) == 0);
361	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
362	    txg) == 0);
363	VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
364	    spa_guid(spa)) == 0);
365#ifdef	_KERNEL
366	hostid = zone_get_hostid(NULL);
367#else	/* _KERNEL */
368	/*
369	 * We're emulating the system's hostid in userland, so we can't use
370	 * zone_get_hostid().
371	 */
372	(void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
373#endif	/* _KERNEL */
374	if (hostid != 0) {
375		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
376		    hostid) == 0);
377	}
378	VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
379	    utsname.nodename) == 0);
380
381	if (vd != rvd) {
382		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
383		    vd->vdev_top->vdev_guid) == 0);
384		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
385		    vd->vdev_guid) == 0);
386		if (vd->vdev_isspare)
387			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
388			    1ULL) == 0);
389		if (vd->vdev_islog)
390			VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
391			    1ULL) == 0);
392		vd = vd->vdev_top;		/* label contains top config */
393	} else {
394		/*
395		 * Only add the (potentially large) split information
396		 * in the mos config, and not in the vdev labels
397		 */
398		if (spa->spa_config_splitting != NULL)
399			VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
400			    spa->spa_config_splitting) == 0);
401	}
402
403	/*
404	 * Add the top-level config.  We even add this on pools which
405	 * don't support holes in the namespace as older pools will
406	 * just ignore it.
407	 */
408	vdev_top_config_generate(spa, config);
409
410	/*
411	 * If we're splitting, record the original pool's guid.
412	 */
413	if (spa->spa_config_splitting != NULL &&
414	    nvlist_lookup_uint64(spa->spa_config_splitting,
415	    ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
416		VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
417		    split_guid) == 0);
418	}
419
420	nvroot = vdev_config_generate(spa, vd, getstats, B_FALSE, B_FALSE);
421	VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
422	nvlist_free(nvroot);
423
424	if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
425		ddt_histogram_t *ddh;
426		ddt_stat_t *dds;
427		ddt_object_t *ddo;
428
429		ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
430		ddt_get_dedup_histogram(spa, ddh);
431		VERIFY(nvlist_add_uint64_array(config,
432		    ZPOOL_CONFIG_DDT_HISTOGRAM,
433		    (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
434		kmem_free(ddh, sizeof (ddt_histogram_t));
435
436		ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
437		ddt_get_dedup_object_stats(spa, ddo);
438		VERIFY(nvlist_add_uint64_array(config,
439		    ZPOOL_CONFIG_DDT_OBJ_STATS,
440		    (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
441		kmem_free(ddo, sizeof (ddt_object_t));
442
443		dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
444		ddt_get_dedup_stats(spa, dds);
445		VERIFY(nvlist_add_uint64_array(config,
446		    ZPOOL_CONFIG_DDT_STATS,
447		    (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
448		kmem_free(dds, sizeof (ddt_stat_t));
449	}
450
451	spa_rewind_data_to_nvlist(spa, config);
452
453	if (locked)
454		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
455
456	return (config);
457}
458
459/*
460 * Update all disk labels, generate a fresh config based on the current
461 * in-core state, and sync the global config cache (do not sync the config
462 * cache if this is a booting rootpool).
463 */
464void
465spa_config_update(spa_t *spa, int what)
466{
467	vdev_t *rvd = spa->spa_root_vdev;
468	uint64_t txg;
469	int c;
470
471	ASSERT(MUTEX_HELD(&spa_namespace_lock));
472
473	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
474	txg = spa_last_synced_txg(spa) + 1;
475	if (what == SPA_CONFIG_UPDATE_POOL) {
476		vdev_config_dirty(rvd);
477	} else {
478		/*
479		 * If we have top-level vdevs that were added but have
480		 * not yet been prepared for allocation, do that now.
481		 * (It's safe now because the config cache is up to date,
482		 * so it will be able to translate the new DVAs.)
483		 * See comments in spa_vdev_add() for full details.
484		 */
485		for (c = 0; c < rvd->vdev_children; c++) {
486			vdev_t *tvd = rvd->vdev_child[c];
487			if (tvd->vdev_ms_array == 0)
488				vdev_metaslab_set_size(tvd);
489			vdev_expand(tvd, txg);
490		}
491	}
492	spa_config_exit(spa, SCL_ALL, FTAG);
493
494	/*
495	 * Wait for the mosconfig to be regenerated and synced.
496	 */
497	txg_wait_synced(spa->spa_dsl_pool, txg);
498
499	/*
500	 * Update the global config cache to reflect the new mosconfig.
501	 */
502	if (!spa->spa_is_root)
503		spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
504
505	if (what == SPA_CONFIG_UPDATE_POOL)
506		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
507}
508