Print this page
3749 zfs event processing should work on R/O root filesystems
Submitted by: Justin Gibbs <justing@spectralogic.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.
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 *
|
↓ open down ↓ |
18 lines elided |
↑ open up ↑ |
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) 2012 by Delphix. All rights reserved.
26 26 */
27 27
28 28 #include <sys/spa.h>
29 +#include <sys/fm/fs/zfs.h>
29 30 #include <sys/spa_impl.h>
30 31 #include <sys/nvpair.h>
31 32 #include <sys/uio.h>
32 33 #include <sys/fs/zfs.h>
33 34 #include <sys/vdev_impl.h>
34 35 #include <sys/zfs_ioctl.h>
35 36 #include <sys/utsname.h>
36 37 #include <sys/systeminfo.h>
37 38 #include <sys/sunddi.h>
38 39 #include <sys/zfeature.h>
39 40 #ifdef _KERNEL
40 41 #include <sys/kobj.h>
41 42 #include <sys/zone.h>
42 43 #endif
43 44
44 45 /*
45 46 * Pool configuration repository.
46 47 *
47 48 * Pool configuration is stored as a packed nvlist on the filesystem. By
48 49 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
49 50 * (when the ZFS module is loaded). Pools can also have the 'cachefile'
50 51 * property set that allows them to be stored in an alternate location until
51 52 * the control of external software.
52 53 *
53 54 * For each cache file, we have a single nvlist which holds all the
54 55 * configuration information. When the module loads, we read this information
55 56 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is
56 57 * maintained independently in spa.c. Whenever the namespace is modified, or
57 58 * the configuration of a pool is changed, we call spa_config_sync(), which
58 59 * walks through all the active pools and writes the configuration to disk.
59 60 */
60 61
61 62 static uint64_t spa_config_generation = 1;
62 63
63 64 /*
64 65 * This can be overridden in userland to preserve an alternate namespace for
65 66 * userland pools when doing testing.
66 67 */
67 68 const char *spa_config_path = ZPOOL_CACHE;
68 69
69 70 /*
70 71 * Called when the module is first loaded, this routine loads the configuration
71 72 * file into the SPA namespace. It does not actually open or load the pools; it
72 73 * only populates the namespace.
73 74 */
74 75 void
75 76 spa_config_load(void)
76 77 {
77 78 void *buf = NULL;
78 79 nvlist_t *nvlist, *child;
79 80 nvpair_t *nvpair;
80 81 char *pathname;
81 82 struct _buf *file;
82 83 uint64_t fsize;
83 84
84 85 /*
85 86 * Open the configuration file.
86 87 */
87 88 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
88 89
89 90 (void) snprintf(pathname, MAXPATHLEN, "%s%s",
90 91 (rootdir != NULL) ? "./" : "", spa_config_path);
91 92
92 93 file = kobj_open_file(pathname);
93 94
94 95 kmem_free(pathname, MAXPATHLEN);
95 96
96 97 if (file == (struct _buf *)-1)
97 98 return;
98 99
99 100 if (kobj_get_filesize(file, &fsize) != 0)
100 101 goto out;
101 102
102 103 buf = kmem_alloc(fsize, KM_SLEEP);
103 104
104 105 /*
105 106 * Read the nvlist from the file.
106 107 */
107 108 if (kobj_read_file(file, buf, fsize, 0) < 0)
108 109 goto out;
109 110
110 111 /*
111 112 * Unpack the nvlist.
112 113 */
113 114 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
114 115 goto out;
115 116
116 117 /*
117 118 * Iterate over all elements in the nvlist, creating a new spa_t for
118 119 * each one with the specified configuration.
119 120 */
120 121 mutex_enter(&spa_namespace_lock);
121 122 nvpair = NULL;
122 123 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
123 124 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
124 125 continue;
125 126
126 127 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
127 128
128 129 if (spa_lookup(nvpair_name(nvpair)) != NULL)
129 130 continue;
130 131 (void) spa_add(nvpair_name(nvpair), child, NULL);
131 132 }
132 133 mutex_exit(&spa_namespace_lock);
|
↓ open down ↓ |
94 lines elided |
↑ open up ↑ |
133 134
134 135 nvlist_free(nvlist);
135 136
136 137 out:
137 138 if (buf != NULL)
138 139 kmem_free(buf, fsize);
139 140
140 141 kobj_close_file(file);
141 142 }
142 143
143 -static void
144 +static int
144 145 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
145 146 {
146 147 size_t buflen;
147 148 char *buf;
148 149 vnode_t *vp;
149 150 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
150 151 char *temp;
152 + int err;
151 153
152 154 /*
153 155 * If the nvlist is empty (NULL), then remove the old cachefile.
154 156 */
155 157 if (nvl == NULL) {
156 - (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
157 - return;
158 + err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
159 + return (err);
158 160 }
159 161
160 162 /*
161 163 * Pack the configuration into a buffer.
162 164 */
163 165 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
164 166
165 167 buf = kmem_alloc(buflen, KM_SLEEP);
166 168 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
167 169
168 170 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
169 171 KM_SLEEP) == 0);
170 172
171 173 /*
172 174 * Write the configuration to disk. We need to do the traditional
173 175 * 'write to temporary file, sync, move over original' to make sure we
174 176 * always have a consistent view of the data.
175 177 */
176 178 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
177 179
178 - if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) {
179 - if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
180 - 0, RLIM64_INFINITY, kcred, NULL) == 0 &&
181 - VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) {
182 - (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
183 - }
180 + err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
181 + if (err == 0) {
182 + err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
183 + 0, RLIM64_INFINITY, kcred, NULL);
184 + if (err == 0)
185 + err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
186 + if (err == 0)
187 + err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
184 188 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
185 189 VN_RELE(vp);
186 190 }
187 191
188 192 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
189 193
190 194 kmem_free(buf, buflen);
191 195 kmem_free(temp, MAXPATHLEN);
196 + return (err);
192 197 }
193 198
194 199 /*
195 200 * Synchronize pool configuration to disk. This must be called with the
196 201 * namespace lock held.
197 202 */
198 203 void
199 204 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
200 205 {
201 206 spa_config_dirent_t *dp, *tdp;
202 207 nvlist_t *nvl;
208 + boolean_t ccw_failure;
209 + int error;
203 210
204 211 ASSERT(MUTEX_HELD(&spa_namespace_lock));
205 212
206 213 if (rootdir == NULL || !(spa_mode_global & FWRITE))
207 214 return;
208 215
209 216 /*
210 217 * Iterate over all cachefiles for the pool, past or present. When the
211 218 * cachefile is changed, the new one is pushed onto this list, allowing
212 219 * us to update previous cachefiles that no longer contain this pool.
213 220 */
221 + ccw_failure = B_FALSE;
214 222 for (dp = list_head(&target->spa_config_list); dp != NULL;
215 223 dp = list_next(&target->spa_config_list, dp)) {
216 224 spa_t *spa = NULL;
217 225 if (dp->scd_path == NULL)
218 226 continue;
219 227
220 228 /*
221 229 * Iterate over all pools, adding any matching pools to 'nvl'.
222 230 */
223 231 nvl = NULL;
224 232 while ((spa = spa_next(spa)) != NULL) {
225 233 /*
226 234 * Skip over our own pool if we're about to remove
227 235 * ourselves from the spa namespace or any pool that
228 236 * is readonly. Since we cannot guarantee that a
229 237 * readonly pool would successfully import upon reboot,
230 238 * we don't allow them to be written to the cache file.
231 239 */
232 240 if ((spa == target && removing) ||
233 241 !spa_writeable(spa))
234 242 continue;
235 243
236 244 mutex_enter(&spa->spa_props_lock);
237 245 tdp = list_head(&spa->spa_config_list);
238 246 if (spa->spa_config == NULL ||
239 247 tdp->scd_path == NULL ||
240 248 strcmp(tdp->scd_path, dp->scd_path) != 0) {
241 249 mutex_exit(&spa->spa_props_lock);
242 250 continue;
243 251 }
|
↓ open down ↓ |
20 lines elided |
↑ open up ↑ |
244 252
245 253 if (nvl == NULL)
246 254 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
247 255 KM_SLEEP) == 0);
248 256
249 257 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
250 258 spa->spa_config) == 0);
251 259 mutex_exit(&spa->spa_props_lock);
252 260 }
253 261
254 - spa_config_write(dp, nvl);
262 + error = spa_config_write(dp, nvl);
263 + if (error != 0)
264 + ccw_failure = B_TRUE;
255 265 nvlist_free(nvl);
256 266 }
257 267
268 + if (ccw_failure) {
269 + /*
270 + * Keep trying so that configuration data is
271 + * written if/when any temporary filesystem
272 + * resource issues are resolved.
273 + */
274 + target->spa_ccw_fail_time = ddi_get_lbolt64();
275 + spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
276 + zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
277 + target, NULL, NULL, 0, 0);
278 + } else {
279 + /*
280 + * Do not rate limit future attempts to update
281 + * the config cache.
282 + */
283 + target->spa_ccw_fail_time = 0;
284 + }
285 +
258 286 /*
259 287 * Remove any config entries older than the current one.
260 288 */
261 289 dp = list_head(&target->spa_config_list);
262 290 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
263 291 list_remove(&target->spa_config_list, tdp);
264 292 if (tdp->scd_path != NULL)
265 293 spa_strfree(tdp->scd_path);
266 294 kmem_free(tdp, sizeof (spa_config_dirent_t));
267 295 }
268 296
269 297 spa_config_generation++;
270 298
271 299 if (postsysevent)
272 300 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
273 301 }
274 302
275 303 /*
276 304 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
277 305 * and we don't want to allow the local zone to see all the pools anyway.
278 306 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
279 307 * information for all pool visible within the zone.
280 308 */
281 309 nvlist_t *
282 310 spa_all_configs(uint64_t *generation)
283 311 {
284 312 nvlist_t *pools;
285 313 spa_t *spa = NULL;
286 314
287 315 if (*generation == spa_config_generation)
288 316 return (NULL);
289 317
290 318 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
291 319
292 320 mutex_enter(&spa_namespace_lock);
293 321 while ((spa = spa_next(spa)) != NULL) {
294 322 if (INGLOBALZONE(curproc) ||
295 323 zone_dataset_visible(spa_name(spa), NULL)) {
296 324 mutex_enter(&spa->spa_props_lock);
297 325 VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
298 326 spa->spa_config) == 0);
299 327 mutex_exit(&spa->spa_props_lock);
300 328 }
301 329 }
302 330 *generation = spa_config_generation;
303 331 mutex_exit(&spa_namespace_lock);
304 332
305 333 return (pools);
306 334 }
307 335
308 336 void
309 337 spa_config_set(spa_t *spa, nvlist_t *config)
310 338 {
311 339 mutex_enter(&spa->spa_props_lock);
312 340 if (spa->spa_config != NULL)
313 341 nvlist_free(spa->spa_config);
314 342 spa->spa_config = config;
315 343 mutex_exit(&spa->spa_props_lock);
316 344 }
317 345
318 346 /*
319 347 * Generate the pool's configuration based on the current in-core state.
320 348 * We infer whether to generate a complete config or just one top-level config
321 349 * based on whether vd is the root vdev.
322 350 */
323 351 nvlist_t *
324 352 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
325 353 {
326 354 nvlist_t *config, *nvroot;
327 355 vdev_t *rvd = spa->spa_root_vdev;
328 356 unsigned long hostid = 0;
329 357 boolean_t locked = B_FALSE;
330 358 uint64_t split_guid;
331 359
332 360 if (vd == NULL) {
333 361 vd = rvd;
334 362 locked = B_TRUE;
335 363 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
336 364 }
337 365
338 366 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
339 367 (SCL_CONFIG | SCL_STATE));
340 368
341 369 /*
342 370 * If txg is -1, report the current value of spa->spa_config_txg.
343 371 */
344 372 if (txg == -1ULL)
345 373 txg = spa->spa_config_txg;
346 374
347 375 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
348 376
349 377 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
350 378 spa_version(spa)) == 0);
351 379 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
352 380 spa_name(spa)) == 0);
353 381 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
354 382 spa_state(spa)) == 0);
355 383 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
356 384 txg) == 0);
357 385 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
358 386 spa_guid(spa)) == 0);
359 387 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
360 388 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
361 389
362 390
363 391 #ifdef _KERNEL
364 392 hostid = zone_get_hostid(NULL);
365 393 #else /* _KERNEL */
366 394 /*
367 395 * We're emulating the system's hostid in userland, so we can't use
368 396 * zone_get_hostid().
369 397 */
370 398 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
371 399 #endif /* _KERNEL */
372 400 if (hostid != 0) {
373 401 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
374 402 hostid) == 0);
375 403 }
376 404 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
377 405 utsname.nodename) == 0);
378 406
379 407 if (vd != rvd) {
380 408 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
381 409 vd->vdev_top->vdev_guid) == 0);
382 410 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
383 411 vd->vdev_guid) == 0);
384 412 if (vd->vdev_isspare)
385 413 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
386 414 1ULL) == 0);
387 415 if (vd->vdev_islog)
388 416 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
389 417 1ULL) == 0);
390 418 vd = vd->vdev_top; /* label contains top config */
391 419 } else {
392 420 /*
393 421 * Only add the (potentially large) split information
394 422 * in the mos config, and not in the vdev labels
395 423 */
396 424 if (spa->spa_config_splitting != NULL)
397 425 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
398 426 spa->spa_config_splitting) == 0);
399 427 }
400 428
401 429 /*
402 430 * Add the top-level config. We even add this on pools which
403 431 * don't support holes in the namespace.
404 432 */
405 433 vdev_top_config_generate(spa, config);
406 434
407 435 /*
408 436 * If we're splitting, record the original pool's guid.
409 437 */
410 438 if (spa->spa_config_splitting != NULL &&
411 439 nvlist_lookup_uint64(spa->spa_config_splitting,
412 440 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
413 441 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
414 442 split_guid) == 0);
415 443 }
416 444
417 445 nvroot = vdev_config_generate(spa, vd, getstats, 0);
418 446 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
419 447 nvlist_free(nvroot);
420 448
421 449 /*
422 450 * Store what's necessary for reading the MOS in the label.
423 451 */
424 452 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
425 453 spa->spa_label_features) == 0);
426 454
427 455 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
428 456 ddt_histogram_t *ddh;
429 457 ddt_stat_t *dds;
430 458 ddt_object_t *ddo;
431 459
432 460 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
433 461 ddt_get_dedup_histogram(spa, ddh);
434 462 VERIFY(nvlist_add_uint64_array(config,
435 463 ZPOOL_CONFIG_DDT_HISTOGRAM,
436 464 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
437 465 kmem_free(ddh, sizeof (ddt_histogram_t));
438 466
439 467 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
440 468 ddt_get_dedup_object_stats(spa, ddo);
441 469 VERIFY(nvlist_add_uint64_array(config,
442 470 ZPOOL_CONFIG_DDT_OBJ_STATS,
443 471 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
444 472 kmem_free(ddo, sizeof (ddt_object_t));
445 473
446 474 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
447 475 ddt_get_dedup_stats(spa, dds);
448 476 VERIFY(nvlist_add_uint64_array(config,
449 477 ZPOOL_CONFIG_DDT_STATS,
450 478 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
451 479 kmem_free(dds, sizeof (ddt_stat_t));
452 480 }
453 481
454 482 if (locked)
455 483 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
456 484
457 485 return (config);
458 486 }
459 487
460 488 /*
461 489 * Update all disk labels, generate a fresh config based on the current
462 490 * in-core state, and sync the global config cache (do not sync the config
463 491 * cache if this is a booting rootpool).
464 492 */
465 493 void
466 494 spa_config_update(spa_t *spa, int what)
467 495 {
468 496 vdev_t *rvd = spa->spa_root_vdev;
469 497 uint64_t txg;
470 498 int c;
471 499
472 500 ASSERT(MUTEX_HELD(&spa_namespace_lock));
473 501
474 502 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
475 503 txg = spa_last_synced_txg(spa) + 1;
476 504 if (what == SPA_CONFIG_UPDATE_POOL) {
477 505 vdev_config_dirty(rvd);
478 506 } else {
479 507 /*
480 508 * If we have top-level vdevs that were added but have
481 509 * not yet been prepared for allocation, do that now.
482 510 * (It's safe now because the config cache is up to date,
483 511 * so it will be able to translate the new DVAs.)
484 512 * See comments in spa_vdev_add() for full details.
485 513 */
486 514 for (c = 0; c < rvd->vdev_children; c++) {
487 515 vdev_t *tvd = rvd->vdev_child[c];
488 516 if (tvd->vdev_ms_array == 0)
489 517 vdev_metaslab_set_size(tvd);
490 518 vdev_expand(tvd, txg);
491 519 }
492 520 }
493 521 spa_config_exit(spa, SCL_ALL, FTAG);
494 522
495 523 /*
496 524 * Wait for the mosconfig to be regenerated and synced.
497 525 */
498 526 txg_wait_synced(spa->spa_dsl_pool, txg);
499 527
500 528 /*
501 529 * Update the global config cache to reflect the new mosconfig.
502 530 */
503 531 if (!spa->spa_is_root)
504 532 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
505 533
506 534 if (what == SPA_CONFIG_UPDATE_POOL)
507 535 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
508 536 }
|
↓ open down ↓ |
241 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX