]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/spa_config.c
Merge branch 'kmem-rework'
[mirror_zfs.git] / module / zfs / spa_config.c
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 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
26 */
27
28 #include <sys/spa.h>
29 #include <sys/spa_impl.h>
30 #include <sys/nvpair.h>
31 #include <sys/uio.h>
32 #include <sys/fs/zfs.h>
33 #include <sys/vdev_impl.h>
34 #include <sys/zfs_ioctl.h>
35 #include <sys/systeminfo.h>
36 #include <sys/sunddi.h>
37 #include <sys/zfeature.h>
38 #ifdef _KERNEL
39 #include <sys/kobj.h>
40 #include <sys/zone.h>
41 #endif
42
43 /*
44 * Pool configuration repository.
45 *
46 * Pool configuration is stored as a packed nvlist on the filesystem. By
47 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
48 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
49 * property set that allows them to be stored in an alternate location until
50 * the control of external software.
51 *
52 * For each cache file, we have a single nvlist which holds all the
53 * configuration information. When the module loads, we read this information
54 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
55 * maintained independently in spa.c. Whenever the namespace is modified, or
56 * the configuration of a pool is changed, we call spa_config_sync(), which
57 * walks through all the active pools and writes the configuration to disk.
58 */
59
60 static uint64_t spa_config_generation = 1;
61
62 /*
63 * This can be overridden in userland to preserve an alternate namespace for
64 * userland pools when doing testing.
65 */
66 char *spa_config_path = ZPOOL_CACHE;
67 int zfs_autoimport_disable = 0;
68
69 /*
70 * Called when the module is first loaded, this routine loads the configuration
71 * file into the SPA namespace. It does not actually open or load the pools; it
72 * only populates the namespace.
73 */
74 void
75 spa_config_load(void)
76 {
77 void *buf = NULL;
78 nvlist_t *nvlist, *child;
79 nvpair_t *nvpair;
80 char *pathname;
81 struct _buf *file;
82 uint64_t fsize;
83
84 if (zfs_autoimport_disable)
85 return;
86
87 /*
88 * Open the configuration file.
89 */
90 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
91
92 (void) snprintf(pathname, MAXPATHLEN, "%s%s",
93 (rootdir != NULL) ? "./" : "", spa_config_path);
94
95 file = kobj_open_file(pathname);
96
97 kmem_free(pathname, MAXPATHLEN);
98
99 if (file == (struct _buf *)-1)
100 return;
101
102 if (kobj_get_filesize(file, &fsize) != 0)
103 goto out;
104
105 buf = kmem_alloc(fsize, KM_SLEEP);
106
107 /*
108 * Read the nvlist from the file.
109 */
110 if (kobj_read_file(file, buf, fsize, 0) < 0)
111 goto out;
112
113 /*
114 * Unpack the nvlist.
115 */
116 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
117 goto out;
118
119 /*
120 * Iterate over all elements in the nvlist, creating a new spa_t for
121 * each one with the specified configuration.
122 */
123 mutex_enter(&spa_namespace_lock);
124 nvpair = NULL;
125 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
126 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
127 continue;
128
129 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
130
131 if (spa_lookup(nvpair_name(nvpair)) != NULL)
132 continue;
133 (void) spa_add(nvpair_name(nvpair), child, NULL);
134 }
135 mutex_exit(&spa_namespace_lock);
136
137 nvlist_free(nvlist);
138
139 out:
140 if (buf != NULL)
141 kmem_free(buf, fsize);
142
143 kobj_close_file(file);
144 }
145
146 static void
147 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
148 {
149 size_t buflen;
150 char *buf;
151 vnode_t *vp;
152 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
153 char *temp;
154
155 /*
156 * If the nvlist is empty (NULL), then remove the old cachefile.
157 */
158 if (nvl == NULL) {
159 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
160 return;
161 }
162
163 /*
164 * Pack the configuration into a buffer.
165 */
166 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
167
168 buf = kmem_alloc(buflen, KM_SLEEP);
169 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
170
171 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
172 KM_SLEEP) == 0);
173
174 /*
175 * Write the configuration to disk. We need to do the traditional
176 * 'write to temporary file, sync, move over original' to make sure we
177 * always have a consistent view of the data.
178 */
179 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
180
181 if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
182 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
183 0, RLIM64_INFINITY, kcred, NULL) == 0 &&
184 VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
185 (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
186 }
187 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
188 }
189
190 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
191
192 kmem_free(buf, buflen);
193 kmem_free(temp, MAXPATHLEN);
194 }
195
196 /*
197 * Synchronize pool configuration to disk. This must be called with the
198 * namespace lock held. Synchronizing the pool cache is typically done after
199 * the configuration has been synced to the MOS. This exposes a window where
200 * the MOS config will have been updated but the cache file has not. If
201 * the system were to crash at that instant then the cached config may not
202 * contain the correct information to open the pool and an explicity import
203 * would be required.
204 */
205 void
206 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
207 {
208 spa_config_dirent_t *dp, *tdp;
209 nvlist_t *nvl;
210 char *pool_name;
211
212 ASSERT(MUTEX_HELD(&spa_namespace_lock));
213
214 if (rootdir == NULL || !(spa_mode_global & FWRITE))
215 return;
216
217 /*
218 * Iterate over all cachefiles for the pool, past or present. When the
219 * cachefile is changed, the new one is pushed onto this list, allowing
220 * us to update previous cachefiles that no longer contain this pool.
221 */
222 for (dp = list_head(&target->spa_config_list); dp != NULL;
223 dp = list_next(&target->spa_config_list, dp)) {
224 spa_t *spa = NULL;
225 if (dp->scd_path == NULL)
226 continue;
227
228 /*
229 * Iterate over all pools, adding any matching pools to 'nvl'.
230 */
231 nvl = NULL;
232 while ((spa = spa_next(spa)) != NULL) {
233 /*
234 * Skip over our own pool if we're about to remove
235 * ourselves from the spa namespace or any pool that
236 * is readonly. Since we cannot guarantee that a
237 * readonly pool would successfully import upon reboot,
238 * we don't allow them to be written to the cache file.
239 */
240 if ((spa == target && removing) ||
241 !spa_writeable(spa))
242 continue;
243
244 mutex_enter(&spa->spa_props_lock);
245 tdp = list_head(&spa->spa_config_list);
246 if (spa->spa_config == NULL ||
247 tdp->scd_path == NULL ||
248 strcmp(tdp->scd_path, dp->scd_path) != 0) {
249 mutex_exit(&spa->spa_props_lock);
250 continue;
251 }
252
253 if (nvl == NULL)
254 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
255 KM_SLEEP) == 0);
256
257 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
258 VERIFY0(nvlist_lookup_string(spa->spa_config,
259 ZPOOL_CONFIG_POOL_NAME, &pool_name));
260 } else
261 pool_name = spa_name(spa);
262
263 VERIFY(nvlist_add_nvlist(nvl, pool_name,
264 spa->spa_config) == 0);
265 mutex_exit(&spa->spa_props_lock);
266 }
267
268 spa_config_write(dp, nvl);
269 nvlist_free(nvl);
270 }
271
272 /*
273 * Remove any config entries older than the current one.
274 */
275 dp = list_head(&target->spa_config_list);
276 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
277 list_remove(&target->spa_config_list, tdp);
278 if (tdp->scd_path != NULL)
279 spa_strfree(tdp->scd_path);
280 kmem_free(tdp, sizeof (spa_config_dirent_t));
281 }
282
283 spa_config_generation++;
284
285 if (postsysevent)
286 spa_event_notify(target, NULL, FM_EREPORT_ZFS_CONFIG_SYNC);
287 }
288
289 /*
290 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
291 * and we don't want to allow the local zone to see all the pools anyway.
292 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
293 * information for all pool visible within the zone.
294 */
295 nvlist_t *
296 spa_all_configs(uint64_t *generation)
297 {
298 nvlist_t *pools;
299 spa_t *spa = NULL;
300
301 if (*generation == spa_config_generation)
302 return (NULL);
303
304 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
305
306 mutex_enter(&spa_namespace_lock);
307 while ((spa = spa_next(spa)) != NULL) {
308 if (INGLOBALZONE(curproc) ||
309 zone_dataset_visible(spa_name(spa), NULL)) {
310 mutex_enter(&spa->spa_props_lock);
311 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
312 spa->spa_config) == 0);
313 mutex_exit(&spa->spa_props_lock);
314 }
315 }
316 *generation = spa_config_generation;
317 mutex_exit(&spa_namespace_lock);
318
319 return (pools);
320 }
321
322 void
323 spa_config_set(spa_t *spa, nvlist_t *config)
324 {
325 mutex_enter(&spa->spa_props_lock);
326 if (spa->spa_config != NULL)
327 nvlist_free(spa->spa_config);
328 spa->spa_config = config;
329 mutex_exit(&spa->spa_props_lock);
330 }
331
332 /*
333 * Generate the pool's configuration based on the current in-core state.
334 *
335 * We infer whether to generate a complete config or just one top-level config
336 * based on whether vd is the root vdev.
337 */
338 nvlist_t *
339 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
340 {
341 nvlist_t *config, *nvroot;
342 vdev_t *rvd = spa->spa_root_vdev;
343 unsigned long hostid = 0;
344 boolean_t locked = B_FALSE;
345 uint64_t split_guid;
346 char *pool_name;
347
348 if (vd == NULL) {
349 vd = rvd;
350 locked = B_TRUE;
351 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
352 }
353
354 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
355 (SCL_CONFIG | SCL_STATE));
356
357 /*
358 * If txg is -1, report the current value of spa->spa_config_txg.
359 */
360 if (txg == -1ULL)
361 txg = spa->spa_config_txg;
362
363 /*
364 * Originally, users had to handle spa namespace collisions by either
365 * exporting the already imported pool or by specifying a new name for
366 * the pool with a conflicting name. In the case of root pools from
367 * virtual guests, neither approach to collision resolution is
368 * reasonable. This is addressed by extending the new name syntax with
369 * an option to specify that the new name is temporary. When specified,
370 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
371 * to use the previous name, which we do below.
372 */
373 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
374 VERIFY0(nvlist_lookup_string(spa->spa_config,
375 ZPOOL_CONFIG_POOL_NAME, &pool_name));
376 } else
377 pool_name = spa_name(spa);
378
379 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
380
381 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
382 spa_version(spa)) == 0);
383 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
384 pool_name) == 0);
385 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
386 spa_state(spa)) == 0);
387 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
388 txg) == 0);
389 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
390 spa_guid(spa)) == 0);
391 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
392 spa->spa_errata) == 0);
393 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
394 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
395
396
397 #ifdef _KERNEL
398 hostid = zone_get_hostid(NULL);
399 #else /* _KERNEL */
400 /*
401 * We're emulating the system's hostid in userland, so we can't use
402 * zone_get_hostid().
403 */
404 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
405 #endif /* _KERNEL */
406 if (hostid != 0) {
407 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
408 hostid) == 0);
409 }
410 VERIFY0(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
411 utsname()->nodename));
412
413 if (vd != rvd) {
414 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
415 vd->vdev_top->vdev_guid) == 0);
416 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
417 vd->vdev_guid) == 0);
418 if (vd->vdev_isspare)
419 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
420 1ULL) == 0);
421 if (vd->vdev_islog)
422 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
423 1ULL) == 0);
424 vd = vd->vdev_top; /* label contains top config */
425 } else {
426 /*
427 * Only add the (potentially large) split information
428 * in the mos config, and not in the vdev labels
429 */
430 if (spa->spa_config_splitting != NULL)
431 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
432 spa->spa_config_splitting) == 0);
433 }
434
435 /*
436 * Add the top-level config. We even add this on pools which
437 * don't support holes in the namespace.
438 */
439 vdev_top_config_generate(spa, config);
440
441 /*
442 * If we're splitting, record the original pool's guid.
443 */
444 if (spa->spa_config_splitting != NULL &&
445 nvlist_lookup_uint64(spa->spa_config_splitting,
446 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
447 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
448 split_guid) == 0);
449 }
450
451 nvroot = vdev_config_generate(spa, vd, getstats, 0);
452 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
453 nvlist_free(nvroot);
454
455 /*
456 * Store what's necessary for reading the MOS in the label.
457 */
458 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
459 spa->spa_label_features) == 0);
460
461 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
462 ddt_histogram_t *ddh;
463 ddt_stat_t *dds;
464 ddt_object_t *ddo;
465
466 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
467 ddt_get_dedup_histogram(spa, ddh);
468 VERIFY(nvlist_add_uint64_array(config,
469 ZPOOL_CONFIG_DDT_HISTOGRAM,
470 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
471 kmem_free(ddh, sizeof (ddt_histogram_t));
472
473 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
474 ddt_get_dedup_object_stats(spa, ddo);
475 VERIFY(nvlist_add_uint64_array(config,
476 ZPOOL_CONFIG_DDT_OBJ_STATS,
477 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
478 kmem_free(ddo, sizeof (ddt_object_t));
479
480 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
481 ddt_get_dedup_stats(spa, dds);
482 VERIFY(nvlist_add_uint64_array(config,
483 ZPOOL_CONFIG_DDT_STATS,
484 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
485 kmem_free(dds, sizeof (ddt_stat_t));
486 }
487
488 if (locked)
489 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
490
491 return (config);
492 }
493
494 /*
495 * Update all disk labels, generate a fresh config based on the current
496 * in-core state, and sync the global config cache (do not sync the config
497 * cache if this is a booting rootpool).
498 */
499 void
500 spa_config_update(spa_t *spa, int what)
501 {
502 vdev_t *rvd = spa->spa_root_vdev;
503 uint64_t txg;
504 int c;
505
506 ASSERT(MUTEX_HELD(&spa_namespace_lock));
507
508 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
509 txg = spa_last_synced_txg(spa) + 1;
510 if (what == SPA_CONFIG_UPDATE_POOL) {
511 vdev_config_dirty(rvd);
512 } else {
513 /*
514 * If we have top-level vdevs that were added but have
515 * not yet been prepared for allocation, do that now.
516 * (It's safe now because the config cache is up to date,
517 * so it will be able to translate the new DVAs.)
518 * See comments in spa_vdev_add() for full details.
519 */
520 for (c = 0; c < rvd->vdev_children; c++) {
521 vdev_t *tvd = rvd->vdev_child[c];
522 if (tvd->vdev_ms_array == 0)
523 vdev_metaslab_set_size(tvd);
524 vdev_expand(tvd, txg);
525 }
526 }
527 spa_config_exit(spa, SCL_ALL, FTAG);
528
529 /*
530 * Wait for the mosconfig to be regenerated and synced.
531 */
532 txg_wait_synced(spa->spa_dsl_pool, txg);
533
534 /*
535 * Update the global config cache to reflect the new mosconfig.
536 */
537 if (!spa->spa_is_root)
538 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
539
540 if (what == SPA_CONFIG_UPDATE_POOL)
541 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
542 }
543
544 #if defined(_KERNEL) && defined(HAVE_SPL)
545 EXPORT_SYMBOL(spa_config_sync);
546 EXPORT_SYMBOL(spa_config_load);
547 EXPORT_SYMBOL(spa_all_configs);
548 EXPORT_SYMBOL(spa_config_set);
549 EXPORT_SYMBOL(spa_config_generate);
550 EXPORT_SYMBOL(spa_config_update);
551
552 module_param(spa_config_path, charp, 0444);
553 MODULE_PARM_DESC(spa_config_path, "SPA config file (/etc/zfs/zpool.cache)");
554
555 module_param(zfs_autoimport_disable, int, 0644);
556 MODULE_PARM_DESC(zfs_autoimport_disable, "Disable pool import at module load");
557
558 #endif