Print this page
2619 asynchronous destruction of ZFS file systems
2747 SPA versioning with zfs feature flags
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <gwilson@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com>
Approved by: Dan McDonald <danmcd@nexenta.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/spa_config.c
+++ new/usr/src/uts/common/fs/zfs/spa_config.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
|
↓ open down ↓ |
14 lines elided |
↑ open up ↑ |
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 - * Copyright (c) 2011 by Delphix. All rights reserved.
25 + * Copyright (c) 2012 by Delphix. All rights reserved.
26 26 */
27 27
28 28 #include <sys/spa.h>
29 29 #include <sys/spa_impl.h>
30 30 #include <sys/nvpair.h>
31 31 #include <sys/uio.h>
32 32 #include <sys/fs/zfs.h>
33 33 #include <sys/vdev_impl.h>
34 34 #include <sys/zfs_ioctl.h>
35 35 #include <sys/utsname.h>
36 36 #include <sys/systeminfo.h>
37 37 #include <sys/sunddi.h>
38 +#include <sys/zfeature.h>
38 39 #ifdef _KERNEL
39 40 #include <sys/kobj.h>
40 41 #include <sys/zone.h>
41 42 #endif
42 43
43 44 /*
44 45 * Pool configuration repository.
45 46 *
46 47 * Pool configuration is stored as a packed nvlist on the filesystem. By
47 48 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
48 49 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
49 50 * property set that allows them to be stored in an alternate location until
50 51 * the control of external software.
51 52 *
52 53 * For each cache file, we have a single nvlist which holds all the
53 54 * configuration information. When the module loads, we read this information
54 55 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
55 56 * maintained independently in spa.c. Whenever the namespace is modified, or
56 57 * the configuration of a pool is changed, we call spa_config_sync(), which
57 58 * walks through all the active pools and writes the configuration to disk.
58 59 */
59 60
60 61 static uint64_t spa_config_generation = 1;
61 62
62 63 /*
63 64 * This can be overridden in userland to preserve an alternate namespace for
64 65 * userland pools when doing testing.
65 66 */
66 67 const char *spa_config_path = ZPOOL_CACHE;
67 68
68 69 /*
69 70 * Called when the module is first loaded, this routine loads the configuration
70 71 * file into the SPA namespace. It does not actually open or load the pools; it
71 72 * only populates the namespace.
72 73 */
73 74 void
74 75 spa_config_load(void)
75 76 {
76 77 void *buf = NULL;
77 78 nvlist_t *nvlist, *child;
78 79 nvpair_t *nvpair;
79 80 char *pathname;
80 81 struct _buf *file;
81 82 uint64_t fsize;
82 83
83 84 /*
84 85 * Open the configuration file.
85 86 */
86 87 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
87 88
88 89 (void) snprintf(pathname, MAXPATHLEN, "%s%s",
89 90 (rootdir != NULL) ? "./" : "", spa_config_path);
90 91
91 92 file = kobj_open_file(pathname);
92 93
93 94 kmem_free(pathname, MAXPATHLEN);
94 95
95 96 if (file == (struct _buf *)-1)
96 97 return;
97 98
98 99 if (kobj_get_filesize(file, &fsize) != 0)
99 100 goto out;
100 101
101 102 buf = kmem_alloc(fsize, KM_SLEEP);
102 103
103 104 /*
104 105 * Read the nvlist from the file.
105 106 */
106 107 if (kobj_read_file(file, buf, fsize, 0) < 0)
107 108 goto out;
108 109
109 110 /*
110 111 * Unpack the nvlist.
111 112 */
112 113 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
113 114 goto out;
114 115
115 116 /*
116 117 * Iterate over all elements in the nvlist, creating a new spa_t for
117 118 * each one with the specified configuration.
118 119 */
119 120 mutex_enter(&spa_namespace_lock);
120 121 nvpair = NULL;
121 122 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
122 123 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
123 124 continue;
124 125
125 126 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
126 127
127 128 if (spa_lookup(nvpair_name(nvpair)) != NULL)
128 129 continue;
129 130 (void) spa_add(nvpair_name(nvpair), child, NULL);
130 131 }
131 132 mutex_exit(&spa_namespace_lock);
132 133
133 134 nvlist_free(nvlist);
134 135
135 136 out:
136 137 if (buf != NULL)
137 138 kmem_free(buf, fsize);
138 139
139 140 kobj_close_file(file);
140 141 }
141 142
142 143 static void
143 144 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
144 145 {
145 146 size_t buflen;
146 147 char *buf;
147 148 vnode_t *vp;
148 149 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
149 150 char *temp;
150 151
151 152 /*
152 153 * If the nvlist is empty (NULL), then remove the old cachefile.
153 154 */
154 155 if (nvl == NULL) {
155 156 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
156 157 return;
157 158 }
158 159
159 160 /*
160 161 * Pack the configuration into a buffer.
161 162 */
162 163 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
163 164
164 165 buf = kmem_alloc(buflen, KM_SLEEP);
165 166 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
166 167
167 168 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
168 169 KM_SLEEP) == 0);
169 170
170 171 /*
171 172 * Write the configuration to disk. We need to do the traditional
172 173 * 'write to temporary file, sync, move over original' to make sure we
173 174 * always have a consistent view of the data.
174 175 */
175 176 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
176 177
177 178 if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
178 179 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
179 180 0, RLIM64_INFINITY, kcred, NULL) == 0 &&
180 181 VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
181 182 (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
182 183 }
183 184 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
184 185 VN_RELE(vp);
185 186 }
186 187
187 188 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
188 189
189 190 kmem_free(buf, buflen);
190 191 kmem_free(temp, MAXPATHLEN);
191 192 }
192 193
193 194 /*
194 195 * Synchronize pool configuration to disk. This must be called with the
195 196 * namespace lock held.
196 197 */
197 198 void
198 199 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
199 200 {
200 201 spa_config_dirent_t *dp, *tdp;
201 202 nvlist_t *nvl;
202 203
203 204 ASSERT(MUTEX_HELD(&spa_namespace_lock));
204 205
205 206 if (rootdir == NULL || !(spa_mode_global & FWRITE))
206 207 return;
207 208
208 209 /*
209 210 * Iterate over all cachefiles for the pool, past or present. When the
210 211 * cachefile is changed, the new one is pushed onto this list, allowing
211 212 * us to update previous cachefiles that no longer contain this pool.
212 213 */
213 214 for (dp = list_head(&target->spa_config_list); dp != NULL;
214 215 dp = list_next(&target->spa_config_list, dp)) {
215 216 spa_t *spa = NULL;
216 217 if (dp->scd_path == NULL)
217 218 continue;
218 219
219 220 /*
220 221 * Iterate over all pools, adding any matching pools to 'nvl'.
221 222 */
222 223 nvl = NULL;
223 224 while ((spa = spa_next(spa)) != NULL) {
224 225 if (spa == target && removing)
225 226 continue;
226 227
227 228 mutex_enter(&spa->spa_props_lock);
228 229 tdp = list_head(&spa->spa_config_list);
229 230 if (spa->spa_config == NULL ||
230 231 tdp->scd_path == NULL ||
231 232 strcmp(tdp->scd_path, dp->scd_path) != 0) {
232 233 mutex_exit(&spa->spa_props_lock);
233 234 continue;
234 235 }
235 236
236 237 if (nvl == NULL)
237 238 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
238 239 KM_SLEEP) == 0);
239 240
240 241 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
241 242 spa->spa_config) == 0);
242 243 mutex_exit(&spa->spa_props_lock);
243 244 }
244 245
245 246 spa_config_write(dp, nvl);
246 247 nvlist_free(nvl);
247 248 }
248 249
249 250 /*
250 251 * Remove any config entries older than the current one.
251 252 */
252 253 dp = list_head(&target->spa_config_list);
253 254 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
254 255 list_remove(&target->spa_config_list, tdp);
255 256 if (tdp->scd_path != NULL)
256 257 spa_strfree(tdp->scd_path);
257 258 kmem_free(tdp, sizeof (spa_config_dirent_t));
258 259 }
259 260
260 261 spa_config_generation++;
261 262
262 263 if (postsysevent)
263 264 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
264 265 }
265 266
266 267 /*
267 268 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
268 269 * and we don't want to allow the local zone to see all the pools anyway.
269 270 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
270 271 * information for all pool visible within the zone.
271 272 */
272 273 nvlist_t *
273 274 spa_all_configs(uint64_t *generation)
274 275 {
275 276 nvlist_t *pools;
276 277 spa_t *spa = NULL;
277 278
278 279 if (*generation == spa_config_generation)
279 280 return (NULL);
280 281
281 282 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
282 283
283 284 mutex_enter(&spa_namespace_lock);
284 285 while ((spa = spa_next(spa)) != NULL) {
285 286 if (INGLOBALZONE(curproc) ||
286 287 zone_dataset_visible(spa_name(spa), NULL)) {
287 288 mutex_enter(&spa->spa_props_lock);
288 289 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
289 290 spa->spa_config) == 0);
290 291 mutex_exit(&spa->spa_props_lock);
291 292 }
292 293 }
293 294 *generation = spa_config_generation;
294 295 mutex_exit(&spa_namespace_lock);
295 296
296 297 return (pools);
297 298 }
298 299
299 300 void
300 301 spa_config_set(spa_t *spa, nvlist_t *config)
301 302 {
302 303 mutex_enter(&spa->spa_props_lock);
303 304 if (spa->spa_config != NULL)
304 305 nvlist_free(spa->spa_config);
305 306 spa->spa_config = config;
306 307 mutex_exit(&spa->spa_props_lock);
307 308 }
308 309
309 310 /*
310 311 * Generate the pool's configuration based on the current in-core state.
311 312 * We infer whether to generate a complete config or just one top-level config
312 313 * based on whether vd is the root vdev.
313 314 */
314 315 nvlist_t *
315 316 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
316 317 {
317 318 nvlist_t *config, *nvroot;
318 319 vdev_t *rvd = spa->spa_root_vdev;
319 320 unsigned long hostid = 0;
320 321 boolean_t locked = B_FALSE;
321 322 uint64_t split_guid;
322 323
323 324 if (vd == NULL) {
324 325 vd = rvd;
325 326 locked = B_TRUE;
326 327 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
327 328 }
328 329
329 330 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
330 331 (SCL_CONFIG | SCL_STATE));
331 332
332 333 /*
333 334 * If txg is -1, report the current value of spa->spa_config_txg.
334 335 */
335 336 if (txg == -1ULL)
336 337 txg = spa->spa_config_txg;
337 338
338 339 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
339 340
340 341 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
341 342 spa_version(spa)) == 0);
342 343 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
343 344 spa_name(spa)) == 0);
344 345 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
345 346 spa_state(spa)) == 0);
346 347 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
347 348 txg) == 0);
348 349 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
349 350 spa_guid(spa)) == 0);
350 351 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
351 352 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
352 353
353 354
354 355 #ifdef _KERNEL
355 356 hostid = zone_get_hostid(NULL);
356 357 #else /* _KERNEL */
357 358 /*
358 359 * We're emulating the system's hostid in userland, so we can't use
359 360 * zone_get_hostid().
360 361 */
361 362 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
362 363 #endif /* _KERNEL */
363 364 if (hostid != 0) {
364 365 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
365 366 hostid) == 0);
366 367 }
367 368 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
368 369 utsname.nodename) == 0);
369 370
370 371 if (vd != rvd) {
371 372 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
372 373 vd->vdev_top->vdev_guid) == 0);
373 374 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
374 375 vd->vdev_guid) == 0);
375 376 if (vd->vdev_isspare)
376 377 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
377 378 1ULL) == 0);
378 379 if (vd->vdev_islog)
379 380 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
380 381 1ULL) == 0);
381 382 vd = vd->vdev_top; /* label contains top config */
382 383 } else {
383 384 /*
384 385 * Only add the (potentially large) split information
385 386 * in the mos config, and not in the vdev labels
386 387 */
387 388 if (spa->spa_config_splitting != NULL)
388 389 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
389 390 spa->spa_config_splitting) == 0);
390 391 }
391 392
392 393 /*
393 394 * Add the top-level config. We even add this on pools which
394 395 * don't support holes in the namespace.
395 396 */
396 397 vdev_top_config_generate(spa, config);
397 398
398 399 /*
399 400 * If we're splitting, record the original pool's guid.
400 401 */
401 402 if (spa->spa_config_splitting != NULL &&
|
↓ open down ↓ |
354 lines elided |
↑ open up ↑ |
402 403 nvlist_lookup_uint64(spa->spa_config_splitting,
403 404 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
404 405 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
405 406 split_guid) == 0);
406 407 }
407 408
408 409 nvroot = vdev_config_generate(spa, vd, getstats, 0);
409 410 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
410 411 nvlist_free(nvroot);
411 412
413 + /*
414 + * Store what's necessary for reading the MOS in the label.
415 + */
416 + VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
417 + spa->spa_label_features) == 0);
418 +
412 419 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
413 420 ddt_histogram_t *ddh;
414 421 ddt_stat_t *dds;
415 422 ddt_object_t *ddo;
416 423
417 424 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
418 425 ddt_get_dedup_histogram(spa, ddh);
419 426 VERIFY(nvlist_add_uint64_array(config,
420 427 ZPOOL_CONFIG_DDT_HISTOGRAM,
421 428 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
422 429 kmem_free(ddh, sizeof (ddt_histogram_t));
423 430
424 431 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
425 432 ddt_get_dedup_object_stats(spa, ddo);
426 433 VERIFY(nvlist_add_uint64_array(config,
427 434 ZPOOL_CONFIG_DDT_OBJ_STATS,
428 435 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
429 436 kmem_free(ddo, sizeof (ddt_object_t));
430 437
431 438 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
432 439 ddt_get_dedup_stats(spa, dds);
433 440 VERIFY(nvlist_add_uint64_array(config,
434 441 ZPOOL_CONFIG_DDT_STATS,
435 442 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
436 443 kmem_free(dds, sizeof (ddt_stat_t));
437 444 }
438 445
439 446 if (locked)
440 447 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
441 448
442 449 return (config);
443 450 }
444 451
445 452 /*
446 453 * Update all disk labels, generate a fresh config based on the current
447 454 * in-core state, and sync the global config cache (do not sync the config
448 455 * cache if this is a booting rootpool).
449 456 */
450 457 void
451 458 spa_config_update(spa_t *spa, int what)
452 459 {
453 460 vdev_t *rvd = spa->spa_root_vdev;
454 461 uint64_t txg;
455 462 int c;
456 463
457 464 ASSERT(MUTEX_HELD(&spa_namespace_lock));
458 465
459 466 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
460 467 txg = spa_last_synced_txg(spa) + 1;
461 468 if (what == SPA_CONFIG_UPDATE_POOL) {
462 469 vdev_config_dirty(rvd);
463 470 } else {
464 471 /*
465 472 * If we have top-level vdevs that were added but have
466 473 * not yet been prepared for allocation, do that now.
467 474 * (It's safe now because the config cache is up to date,
468 475 * so it will be able to translate the new DVAs.)
469 476 * See comments in spa_vdev_add() for full details.
470 477 */
471 478 for (c = 0; c < rvd->vdev_children; c++) {
472 479 vdev_t *tvd = rvd->vdev_child[c];
473 480 if (tvd->vdev_ms_array == 0)
474 481 vdev_metaslab_set_size(tvd);
475 482 vdev_expand(tvd, txg);
476 483 }
477 484 }
478 485 spa_config_exit(spa, SCL_ALL, FTAG);
479 486
480 487 /*
481 488 * Wait for the mosconfig to be regenerated and synced.
482 489 */
483 490 txg_wait_synced(spa->spa_dsl_pool, txg);
484 491
485 492 /*
486 493 * Update the global config cache to reflect the new mosconfig.
487 494 */
488 495 if (!spa->spa_is_root)
489 496 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
490 497
491 498 if (what == SPA_CONFIG_UPDATE_POOL)
492 499 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
493 500 }
|
↓ open down ↓ |
72 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX