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