]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/spa_config.c
Illumos 3749 - zfs event processing should work on R/O root filesystems
[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/fm/fs/zfs.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/systeminfo.h>
37 #include <sys/sunddi.h>
38 #include <sys/zfeature.h>
39 #ifdef _KERNEL
40 #include <sys/kobj.h>
41 #include <sys/zone.h>
42 #endif
43
44 /*
45 * Pool configuration repository.
46 *
47 * Pool configuration is stored as a packed nvlist on the filesystem. By
48 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
49 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
50 * property set that allows them to be stored in an alternate location until
51 * the control of external software.
52 *
53 * For each cache file, we have a single nvlist which holds all the
54 * configuration information. When the module loads, we read this information
55 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
56 * maintained independently in spa.c. Whenever the namespace is modified, or
57 * the configuration of a pool is changed, we call spa_config_sync(), which
58 * walks through all the active pools and writes the configuration to disk.
59 */
60
61 static uint64_t spa_config_generation = 1;
62
63 /*
64 * This can be overridden in userland to preserve an alternate namespace for
65 * userland pools when doing testing.
66 */
67 char *spa_config_path = ZPOOL_CACHE;
68 int zfs_autoimport_disable = 1;
69
70 /*
71 * Called when the module is first loaded, this routine loads the configuration
72 * file into the SPA namespace. It does not actually open or load the pools; it
73 * only populates the namespace.
74 */
75 void
76 spa_config_load(void)
77 {
78 void *buf = NULL;
79 nvlist_t *nvlist, *child;
80 nvpair_t *nvpair;
81 char *pathname;
82 struct _buf *file;
83 uint64_t fsize;
84
85 #ifdef _KERNEL
86 if (zfs_autoimport_disable)
87 return;
88 #endif
89
90 /*
91 * Open the configuration file.
92 */
93 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
94
95 (void) snprintf(pathname, MAXPATHLEN, "%s%s",
96 (rootdir != NULL) ? "./" : "", spa_config_path);
97
98 file = kobj_open_file(pathname);
99
100 kmem_free(pathname, MAXPATHLEN);
101
102 if (file == (struct _buf *)-1)
103 return;
104
105 if (kobj_get_filesize(file, &fsize) != 0)
106 goto out;
107
108 buf = kmem_alloc(fsize, KM_SLEEP);
109
110 /*
111 * Read the nvlist from the file.
112 */
113 if (kobj_read_file(file, buf, fsize, 0) < 0)
114 goto out;
115
116 /*
117 * Unpack the nvlist.
118 */
119 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
120 goto out;
121
122 /*
123 * Iterate over all elements in the nvlist, creating a new spa_t for
124 * each one with the specified configuration.
125 */
126 mutex_enter(&spa_namespace_lock);
127 nvpair = NULL;
128 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
129 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
130 continue;
131
132 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
133
134 if (spa_lookup(nvpair_name(nvpair)) != NULL)
135 continue;
136 (void) spa_add(nvpair_name(nvpair), child, NULL);
137 }
138 mutex_exit(&spa_namespace_lock);
139
140 nvlist_free(nvlist);
141
142 out:
143 if (buf != NULL)
144 kmem_free(buf, fsize);
145
146 kobj_close_file(file);
147 }
148
149 static int
150 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
151 {
152 size_t buflen;
153 char *buf;
154 vnode_t *vp;
155 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
156 char *temp;
157 int err;
158
159 /*
160 * If the nvlist is empty (NULL), then remove the old cachefile.
161 */
162 if (nvl == NULL) {
163 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
164 return (err);
165 }
166
167 /*
168 * Pack the configuration into a buffer.
169 */
170 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
171
172 buf = vmem_alloc(buflen, KM_SLEEP);
173 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
174
175 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
176 KM_SLEEP) == 0);
177
178 #if defined(__linux__) && defined(_KERNEL)
179 /*
180 * Write the configuration to disk. Due to the complexity involved
181 * in performing a rename from within the kernel the file is truncated
182 * and overwritten in place. In the event of an error the file is
183 * unlinked to make sure we always have a consistent view of the data.
184 */
185 err = vn_open(dp->scd_path, UIO_SYSSPACE, oflags, 0644, &vp, 0, 0);
186 if (err == 0) {
187 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0,
188 UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, NULL);
189 if (err == 0)
190 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
191
192 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
193
194 if (err)
195 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
196 }
197 #else
198 /*
199 * Write the configuration to disk. We need to do the traditional
200 * 'write to temporary file, sync, move over original' to make sure we
201 * always have a consistent view of the data.
202 */
203 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
204
205 err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
206 if (err == 0) {
207 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
208 0, RLIM64_INFINITY, kcred, NULL);
209 if (err == 0)
210 err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
211 if (err == 0)
212 err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
213 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
214 VN_RELE(vp);
215 }
216
217 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
218 #endif
219
220 vmem_free(buf, buflen);
221 kmem_free(temp, MAXPATHLEN);
222 return (err);
223 }
224
225 /*
226 * Synchronize pool configuration to disk. This must be called with the
227 * namespace lock held. Synchronizing the pool cache is typically done after
228 * the configuration has been synced to the MOS. This exposes a window where
229 * the MOS config will have been updated but the cache file has not. If
230 * the system were to crash at that instant then the cached config may not
231 * contain the correct information to open the pool and an explicity import
232 * would be required.
233 */
234 void
235 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
236 {
237 spa_config_dirent_t *dp, *tdp;
238 nvlist_t *nvl;
239 char *pool_name;
240 boolean_t ccw_failure;
241 int error = 0;
242
243 ASSERT(MUTEX_HELD(&spa_namespace_lock));
244
245 if (rootdir == NULL || !(spa_mode_global & FWRITE))
246 return;
247
248 /*
249 * Iterate over all cachefiles for the pool, past or present. When the
250 * cachefile is changed, the new one is pushed onto this list, allowing
251 * us to update previous cachefiles that no longer contain this pool.
252 */
253 ccw_failure = B_FALSE;
254 for (dp = list_head(&target->spa_config_list); dp != NULL;
255 dp = list_next(&target->spa_config_list, dp)) {
256 spa_t *spa = NULL;
257 if (dp->scd_path == NULL)
258 continue;
259
260 /*
261 * Iterate over all pools, adding any matching pools to 'nvl'.
262 */
263 nvl = NULL;
264 while ((spa = spa_next(spa)) != NULL) {
265 /*
266 * Skip over our own pool if we're about to remove
267 * ourselves from the spa namespace or any pool that
268 * is readonly. Since we cannot guarantee that a
269 * readonly pool would successfully import upon reboot,
270 * we don't allow them to be written to the cache file.
271 */
272 if ((spa == target && removing) ||
273 !spa_writeable(spa))
274 continue;
275
276 mutex_enter(&spa->spa_props_lock);
277 tdp = list_head(&spa->spa_config_list);
278 if (spa->spa_config == NULL ||
279 tdp->scd_path == NULL ||
280 strcmp(tdp->scd_path, dp->scd_path) != 0) {
281 mutex_exit(&spa->spa_props_lock);
282 continue;
283 }
284
285 if (nvl == NULL)
286 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
287 KM_SLEEP) == 0);
288
289 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
290 VERIFY0(nvlist_lookup_string(spa->spa_config,
291 ZPOOL_CONFIG_POOL_NAME, &pool_name));
292 } else
293 pool_name = spa_name(spa);
294
295 VERIFY(nvlist_add_nvlist(nvl, pool_name,
296 spa->spa_config) == 0);
297 mutex_exit(&spa->spa_props_lock);
298 }
299
300 error = spa_config_write(dp, nvl);
301 if (error != 0)
302 ccw_failure = B_TRUE;
303 nvlist_free(nvl);
304 }
305
306 if (ccw_failure) {
307 /*
308 * Keep trying so that configuration data is
309 * written if/when any temporary filesystem
310 * resource issues are resolved.
311 */
312 if (target->spa_ccw_fail_time == 0) {
313 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
314 target, NULL, NULL, 0, 0);
315 }
316 target->spa_ccw_fail_time = gethrtime();
317 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
318 } else {
319 /*
320 * Do not rate limit future attempts to update
321 * the config cache.
322 */
323 target->spa_ccw_fail_time = 0;
324 }
325
326 /*
327 * Remove any config entries older than the current one.
328 */
329 dp = list_head(&target->spa_config_list);
330 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
331 list_remove(&target->spa_config_list, tdp);
332 if (tdp->scd_path != NULL)
333 spa_strfree(tdp->scd_path);
334 kmem_free(tdp, sizeof (spa_config_dirent_t));
335 }
336
337 spa_config_generation++;
338
339 if (postsysevent)
340 spa_event_notify(target, NULL, FM_EREPORT_ZFS_CONFIG_SYNC);
341 }
342
343 /*
344 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
345 * and we don't want to allow the local zone to see all the pools anyway.
346 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
347 * information for all pool visible within the zone.
348 */
349 nvlist_t *
350 spa_all_configs(uint64_t *generation)
351 {
352 nvlist_t *pools;
353 spa_t *spa = NULL;
354
355 if (*generation == spa_config_generation)
356 return (NULL);
357
358 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
359
360 mutex_enter(&spa_namespace_lock);
361 while ((spa = spa_next(spa)) != NULL) {
362 if (INGLOBALZONE(curproc) ||
363 zone_dataset_visible(spa_name(spa), NULL)) {
364 mutex_enter(&spa->spa_props_lock);
365 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
366 spa->spa_config) == 0);
367 mutex_exit(&spa->spa_props_lock);
368 }
369 }
370 *generation = spa_config_generation;
371 mutex_exit(&spa_namespace_lock);
372
373 return (pools);
374 }
375
376 void
377 spa_config_set(spa_t *spa, nvlist_t *config)
378 {
379 mutex_enter(&spa->spa_props_lock);
380 if (spa->spa_config != NULL)
381 nvlist_free(spa->spa_config);
382 spa->spa_config = config;
383 mutex_exit(&spa->spa_props_lock);
384 }
385
386 /*
387 * Generate the pool's configuration based on the current in-core state.
388 *
389 * We infer whether to generate a complete config or just one top-level config
390 * based on whether vd is the root vdev.
391 */
392 nvlist_t *
393 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
394 {
395 nvlist_t *config, *nvroot;
396 vdev_t *rvd = spa->spa_root_vdev;
397 unsigned long hostid = 0;
398 boolean_t locked = B_FALSE;
399 uint64_t split_guid;
400 char *pool_name;
401
402 if (vd == NULL) {
403 vd = rvd;
404 locked = B_TRUE;
405 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
406 }
407
408 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
409 (SCL_CONFIG | SCL_STATE));
410
411 /*
412 * If txg is -1, report the current value of spa->spa_config_txg.
413 */
414 if (txg == -1ULL)
415 txg = spa->spa_config_txg;
416
417 /*
418 * Originally, users had to handle spa namespace collisions by either
419 * exporting the already imported pool or by specifying a new name for
420 * the pool with a conflicting name. In the case of root pools from
421 * virtual guests, neither approach to collision resolution is
422 * reasonable. This is addressed by extending the new name syntax with
423 * an option to specify that the new name is temporary. When specified,
424 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
425 * to use the previous name, which we do below.
426 */
427 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
428 VERIFY0(nvlist_lookup_string(spa->spa_config,
429 ZPOOL_CONFIG_POOL_NAME, &pool_name));
430 } else
431 pool_name = spa_name(spa);
432
433 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
434
435 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
436 spa_version(spa)) == 0);
437 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
438 pool_name) == 0);
439 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
440 spa_state(spa)) == 0);
441 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
442 txg) == 0);
443 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
444 spa_guid(spa)) == 0);
445 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA,
446 spa->spa_errata) == 0);
447 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
448 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
449
450
451 #ifdef _KERNEL
452 hostid = zone_get_hostid(NULL);
453 #else /* _KERNEL */
454 /*
455 * We're emulating the system's hostid in userland, so we can't use
456 * zone_get_hostid().
457 */
458 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
459 #endif /* _KERNEL */
460 if (hostid != 0) {
461 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
462 hostid) == 0);
463 }
464 VERIFY0(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
465 utsname()->nodename));
466
467 if (vd != rvd) {
468 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
469 vd->vdev_top->vdev_guid) == 0);
470 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
471 vd->vdev_guid) == 0);
472 if (vd->vdev_isspare)
473 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
474 1ULL) == 0);
475 if (vd->vdev_islog)
476 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
477 1ULL) == 0);
478 vd = vd->vdev_top; /* label contains top config */
479 } else {
480 /*
481 * Only add the (potentially large) split information
482 * in the mos config, and not in the vdev labels
483 */
484 if (spa->spa_config_splitting != NULL)
485 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
486 spa->spa_config_splitting) == 0);
487 }
488
489 /*
490 * Add the top-level config. We even add this on pools which
491 * don't support holes in the namespace.
492 */
493 vdev_top_config_generate(spa, config);
494
495 /*
496 * If we're splitting, record the original pool's guid.
497 */
498 if (spa->spa_config_splitting != NULL &&
499 nvlist_lookup_uint64(spa->spa_config_splitting,
500 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
501 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
502 split_guid) == 0);
503 }
504
505 nvroot = vdev_config_generate(spa, vd, getstats, 0);
506 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
507 nvlist_free(nvroot);
508
509 /*
510 * Store what's necessary for reading the MOS in the label.
511 */
512 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
513 spa->spa_label_features) == 0);
514
515 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
516 ddt_histogram_t *ddh;
517 ddt_stat_t *dds;
518 ddt_object_t *ddo;
519
520 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
521 ddt_get_dedup_histogram(spa, ddh);
522 VERIFY(nvlist_add_uint64_array(config,
523 ZPOOL_CONFIG_DDT_HISTOGRAM,
524 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
525 kmem_free(ddh, sizeof (ddt_histogram_t));
526
527 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
528 ddt_get_dedup_object_stats(spa, ddo);
529 VERIFY(nvlist_add_uint64_array(config,
530 ZPOOL_CONFIG_DDT_OBJ_STATS,
531 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
532 kmem_free(ddo, sizeof (ddt_object_t));
533
534 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
535 ddt_get_dedup_stats(spa, dds);
536 VERIFY(nvlist_add_uint64_array(config,
537 ZPOOL_CONFIG_DDT_STATS,
538 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
539 kmem_free(dds, sizeof (ddt_stat_t));
540 }
541
542 if (locked)
543 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
544
545 return (config);
546 }
547
548 /*
549 * Update all disk labels, generate a fresh config based on the current
550 * in-core state, and sync the global config cache (do not sync the config
551 * cache if this is a booting rootpool).
552 */
553 void
554 spa_config_update(spa_t *spa, int what)
555 {
556 vdev_t *rvd = spa->spa_root_vdev;
557 uint64_t txg;
558 int c;
559
560 ASSERT(MUTEX_HELD(&spa_namespace_lock));
561
562 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
563 txg = spa_last_synced_txg(spa) + 1;
564 if (what == SPA_CONFIG_UPDATE_POOL) {
565 vdev_config_dirty(rvd);
566 } else {
567 /*
568 * If we have top-level vdevs that were added but have
569 * not yet been prepared for allocation, do that now.
570 * (It's safe now because the config cache is up to date,
571 * so it will be able to translate the new DVAs.)
572 * See comments in spa_vdev_add() for full details.
573 */
574 for (c = 0; c < rvd->vdev_children; c++) {
575 vdev_t *tvd = rvd->vdev_child[c];
576 if (tvd->vdev_ms_array == 0)
577 vdev_metaslab_set_size(tvd);
578 vdev_expand(tvd, txg);
579 }
580 }
581 spa_config_exit(spa, SCL_ALL, FTAG);
582
583 /*
584 * Wait for the mosconfig to be regenerated and synced.
585 */
586 txg_wait_synced(spa->spa_dsl_pool, txg);
587
588 /*
589 * Update the global config cache to reflect the new mosconfig.
590 */
591 if (!spa->spa_is_root)
592 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
593
594 if (what == SPA_CONFIG_UPDATE_POOL)
595 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
596 }
597
598 #if defined(_KERNEL) && defined(HAVE_SPL)
599 EXPORT_SYMBOL(spa_config_sync);
600 EXPORT_SYMBOL(spa_config_load);
601 EXPORT_SYMBOL(spa_all_configs);
602 EXPORT_SYMBOL(spa_config_set);
603 EXPORT_SYMBOL(spa_config_generate);
604 EXPORT_SYMBOL(spa_config_update);
605
606 module_param(spa_config_path, charp, 0444);
607 MODULE_PARM_DESC(spa_config_path, "SPA config file (/etc/zfs/zpool.cache)");
608
609 module_param(zfs_autoimport_disable, int, 0644);
610 MODULE_PARM_DESC(zfs_autoimport_disable, "Disable pool import at module load");
611
612 #endif