]>
Commit | Line | Data |
---|---|---|
428870ff BB |
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 | * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved. | |
a6255b7f | 23 | * Copyright (c) 2011, 2016 by Delphix. All rights reserved. |
7c9abfa7 | 24 | * Copyright 2016 Gary Mills |
0ea05c64 | 25 | * Copyright (c) 2017 Datto Inc. |
12fa0466 | 26 | * Copyright 2017 Joyent, Inc. |
428870ff BB |
27 | */ |
28 | ||
29 | #include <sys/dsl_scan.h> | |
30 | #include <sys/dsl_pool.h> | |
31 | #include <sys/dsl_dataset.h> | |
32 | #include <sys/dsl_prop.h> | |
33 | #include <sys/dsl_dir.h> | |
34 | #include <sys/dsl_synctask.h> | |
35 | #include <sys/dnode.h> | |
36 | #include <sys/dmu_tx.h> | |
37 | #include <sys/dmu_objset.h> | |
38 | #include <sys/arc.h> | |
39 | #include <sys/zap.h> | |
40 | #include <sys/zio.h> | |
41 | #include <sys/zfs_context.h> | |
42 | #include <sys/fs/zfs.h> | |
43 | #include <sys/zfs_znode.h> | |
44 | #include <sys/spa_impl.h> | |
45 | #include <sys/vdev_impl.h> | |
46 | #include <sys/zil_impl.h> | |
47 | #include <sys/zio_checksum.h> | |
48 | #include <sys/ddt.h> | |
49 | #include <sys/sa.h> | |
50 | #include <sys/sa_impl.h> | |
9ae529ec | 51 | #include <sys/zfeature.h> |
a6255b7f | 52 | #include <sys/abd.h> |
428870ff BB |
53 | #ifdef _KERNEL |
54 | #include <sys/zfs_vfsops.h> | |
55 | #endif | |
56 | ||
5dbd68a3 MA |
57 | typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, |
58 | const zbookmark_phys_t *); | |
428870ff | 59 | |
428870ff | 60 | static scan_cb_t dsl_scan_scrub_cb; |
13fe0198 | 61 | static void dsl_scan_cancel_sync(void *, dmu_tx_t *); |
784d15c1 NR |
62 | static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *); |
63 | static boolean_t dsl_scan_restarting(dsl_scan_t *, dmu_tx_t *); | |
428870ff | 64 | |
572e2857 BB |
65 | int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */ |
66 | int zfs_resilver_delay = 2; /* number of ticks to delay resilver */ | |
67 | int zfs_scrub_delay = 4; /* number of ticks to delay scrub */ | |
68 | int zfs_scan_idle = 50; /* idle window in clock ticks */ | |
69 | ||
428870ff BB |
70 | int zfs_scan_min_time_ms = 1000; /* min millisecs to scrub per txg */ |
71 | int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */ | |
72 | int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */ | |
c409e464 | 73 | int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */ |
fbeddd60 | 74 | int zfs_no_scrub_prefetch = B_FALSE; /* set to disable scrub prefetch */ |
428870ff BB |
75 | enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE; |
76 | int dsl_scan_delay_completion = B_FALSE; /* set to delay scan completion */ | |
36283ca2 | 77 | /* max number of blocks to free in a single TXG */ |
02730c33 | 78 | unsigned long zfs_free_max_blocks = 100000; |
428870ff BB |
79 | |
80 | #define DSL_SCAN_IS_SCRUB_RESILVER(scn) \ | |
81 | ((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \ | |
82 | (scn)->scn_phys.scn_func == POOL_SCAN_RESILVER) | |
83 | ||
ba5ad9a4 GW |
84 | /* |
85 | * Enable/disable the processing of the free_bpobj object. | |
86 | */ | |
87 | int zfs_free_bpobj_enabled = 1; | |
88 | ||
428870ff BB |
89 | /* the order has to match pool_scan_type */ |
90 | static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = { | |
91 | NULL, | |
92 | dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */ | |
93 | dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */ | |
94 | }; | |
95 | ||
96 | int | |
97 | dsl_scan_init(dsl_pool_t *dp, uint64_t txg) | |
98 | { | |
99 | int err; | |
100 | dsl_scan_t *scn; | |
101 | spa_t *spa = dp->dp_spa; | |
102 | uint64_t f; | |
103 | ||
104 | scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP); | |
105 | scn->scn_dp = dp; | |
106 | ||
2696dfaf GW |
107 | /* |
108 | * It's possible that we're resuming a scan after a reboot so | |
109 | * make sure that the scan_async_destroying flag is initialized | |
110 | * appropriately. | |
111 | */ | |
112 | ASSERT(!scn->scn_async_destroying); | |
113 | scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa, | |
fa86b5db | 114 | SPA_FEATURE_ASYNC_DESTROY); |
2696dfaf | 115 | |
428870ff BB |
116 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, |
117 | "scrub_func", sizeof (uint64_t), 1, &f); | |
118 | if (err == 0) { | |
119 | /* | |
120 | * There was an old-style scrub in progress. Restart a | |
121 | * new-style scrub from the beginning. | |
122 | */ | |
123 | scn->scn_restart_txg = txg; | |
124 | zfs_dbgmsg("old-style scrub was in progress; " | |
125 | "restarting new-style scrub in txg %llu", | |
126 | scn->scn_restart_txg); | |
127 | ||
128 | /* | |
129 | * Load the queue obj from the old location so that it | |
130 | * can be freed by dsl_scan_done(). | |
131 | */ | |
132 | (void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
133 | "scrub_queue", sizeof (uint64_t), 1, | |
134 | &scn->scn_phys.scn_queue_obj); | |
135 | } else { | |
136 | err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT, | |
137 | DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, | |
138 | &scn->scn_phys); | |
4f2dcb3e RY |
139 | /* |
140 | * Detect if the pool contains the signature of #2094. If it | |
141 | * does properly update the scn->scn_phys structure and notify | |
142 | * the administrator by setting an errata for the pool. | |
143 | */ | |
144 | if (err == EOVERFLOW) { | |
145 | uint64_t zaptmp[SCAN_PHYS_NUMINTS + 1]; | |
146 | VERIFY3S(SCAN_PHYS_NUMINTS, ==, 24); | |
147 | VERIFY3S(offsetof(dsl_scan_phys_t, scn_flags), ==, | |
148 | (23 * sizeof (uint64_t))); | |
149 | ||
150 | err = zap_lookup(dp->dp_meta_objset, | |
151 | DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SCAN, | |
152 | sizeof (uint64_t), SCAN_PHYS_NUMINTS + 1, &zaptmp); | |
153 | if (err == 0) { | |
154 | uint64_t overflow = zaptmp[SCAN_PHYS_NUMINTS]; | |
155 | ||
156 | if (overflow & ~DSL_SCAN_FLAGS_MASK || | |
157 | scn->scn_async_destroying) { | |
158 | spa->spa_errata = | |
159 | ZPOOL_ERRATA_ZOL_2094_ASYNC_DESTROY; | |
ecb2b7dc | 160 | return (SET_ERROR(EOVERFLOW)); |
4f2dcb3e RY |
161 | } |
162 | ||
163 | bcopy(zaptmp, &scn->scn_phys, | |
164 | SCAN_PHYS_NUMINTS * sizeof (uint64_t)); | |
165 | scn->scn_phys.scn_flags = overflow; | |
166 | ||
167 | /* Required scrub already in progress. */ | |
168 | if (scn->scn_phys.scn_state == DSS_FINISHED || | |
169 | scn->scn_phys.scn_state == DSS_CANCELED) | |
170 | spa->spa_errata = | |
171 | ZPOOL_ERRATA_ZOL_2094_SCRUB; | |
172 | } | |
173 | } | |
174 | ||
428870ff BB |
175 | if (err == ENOENT) |
176 | return (0); | |
177 | else if (err) | |
178 | return (err); | |
179 | ||
180 | if (scn->scn_phys.scn_state == DSS_SCANNING && | |
181 | spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) { | |
182 | /* | |
183 | * A new-type scrub was in progress on an old | |
184 | * pool, and the pool was accessed by old | |
185 | * software. Restart from the beginning, since | |
186 | * the old software may have changed the pool in | |
187 | * the meantime. | |
188 | */ | |
189 | scn->scn_restart_txg = txg; | |
190 | zfs_dbgmsg("new-style scrub was modified " | |
191 | "by old software; restarting in txg %llu", | |
192 | scn->scn_restart_txg); | |
193 | } | |
194 | } | |
195 | ||
196 | spa_scan_stat_init(spa); | |
197 | return (0); | |
198 | } | |
199 | ||
200 | void | |
201 | dsl_scan_fini(dsl_pool_t *dp) | |
202 | { | |
203 | if (dp->dp_scan) { | |
204 | kmem_free(dp->dp_scan, sizeof (dsl_scan_t)); | |
205 | dp->dp_scan = NULL; | |
206 | } | |
207 | } | |
208 | ||
209 | /* ARGSUSED */ | |
210 | static int | |
13fe0198 | 211 | dsl_scan_setup_check(void *arg, dmu_tx_t *tx) |
428870ff | 212 | { |
13fe0198 | 213 | dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; |
428870ff BB |
214 | |
215 | if (scn->scn_phys.scn_state == DSS_SCANNING) | |
2e528b49 | 216 | return (SET_ERROR(EBUSY)); |
428870ff BB |
217 | |
218 | return (0); | |
219 | } | |
220 | ||
428870ff | 221 | static void |
13fe0198 | 222 | dsl_scan_setup_sync(void *arg, dmu_tx_t *tx) |
428870ff | 223 | { |
13fe0198 MA |
224 | dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; |
225 | pool_scan_func_t *funcp = arg; | |
428870ff BB |
226 | dmu_object_type_t ot = 0; |
227 | dsl_pool_t *dp = scn->scn_dp; | |
228 | spa_t *spa = dp->dp_spa; | |
229 | ||
230 | ASSERT(scn->scn_phys.scn_state != DSS_SCANNING); | |
231 | ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS); | |
232 | bzero(&scn->scn_phys, sizeof (scn->scn_phys)); | |
233 | scn->scn_phys.scn_func = *funcp; | |
234 | scn->scn_phys.scn_state = DSS_SCANNING; | |
235 | scn->scn_phys.scn_min_txg = 0; | |
236 | scn->scn_phys.scn_max_txg = tx->tx_txg; | |
237 | scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */ | |
238 | scn->scn_phys.scn_start_time = gethrestime_sec(); | |
239 | scn->scn_phys.scn_errors = 0; | |
240 | scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc; | |
241 | scn->scn_restart_txg = 0; | |
5d1f7fb6 | 242 | scn->scn_done_txg = 0; |
428870ff BB |
243 | spa_scan_stat_init(spa); |
244 | ||
245 | if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { | |
246 | scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max; | |
247 | ||
248 | /* rewrite all disk labels */ | |
249 | vdev_config_dirty(spa->spa_root_vdev); | |
250 | ||
251 | if (vdev_resilver_needed(spa->spa_root_vdev, | |
252 | &scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) { | |
12fa0466 DE |
253 | spa_event_notify(spa, NULL, NULL, |
254 | ESC_ZFS_RESILVER_START); | |
428870ff | 255 | } else { |
12fa0466 | 256 | spa_event_notify(spa, NULL, NULL, ESC_ZFS_SCRUB_START); |
428870ff BB |
257 | } |
258 | ||
259 | spa->spa_scrub_started = B_TRUE; | |
260 | /* | |
261 | * If this is an incremental scrub, limit the DDT scrub phase | |
262 | * to just the auto-ditto class (for correctness); the rest | |
263 | * of the scrub should go faster using top-down pruning. | |
264 | */ | |
265 | if (scn->scn_phys.scn_min_txg > TXG_INITIAL) | |
266 | scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO; | |
267 | ||
268 | } | |
269 | ||
270 | /* back to the generic stuff */ | |
271 | ||
272 | if (dp->dp_blkstats == NULL) { | |
79c76d5b BB |
273 | dp->dp_blkstats = |
274 | vmem_alloc(sizeof (zfs_all_blkstats_t), KM_SLEEP); | |
428870ff BB |
275 | } |
276 | bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t)); | |
277 | ||
278 | if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) | |
279 | ot = DMU_OT_ZAP_OTHER; | |
280 | ||
281 | scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset, | |
282 | ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx); | |
283 | ||
284 | dsl_scan_sync_state(scn, tx); | |
285 | ||
6f1ffb06 | 286 | spa_history_log_internal(spa, "scan setup", tx, |
428870ff BB |
287 | "func=%u mintxg=%llu maxtxg=%llu", |
288 | *funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg); | |
289 | } | |
290 | ||
291 | /* ARGSUSED */ | |
292 | static void | |
293 | dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx) | |
294 | { | |
295 | static const char *old_names[] = { | |
296 | "scrub_bookmark", | |
297 | "scrub_ddt_bookmark", | |
298 | "scrub_ddt_class_max", | |
299 | "scrub_queue", | |
300 | "scrub_min_txg", | |
301 | "scrub_max_txg", | |
302 | "scrub_func", | |
303 | "scrub_errors", | |
304 | NULL | |
305 | }; | |
306 | ||
307 | dsl_pool_t *dp = scn->scn_dp; | |
308 | spa_t *spa = dp->dp_spa; | |
309 | int i; | |
310 | ||
311 | /* Remove any remnants of an old-style scrub. */ | |
312 | for (i = 0; old_names[i]; i++) { | |
313 | (void) zap_remove(dp->dp_meta_objset, | |
314 | DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx); | |
315 | } | |
316 | ||
317 | if (scn->scn_phys.scn_queue_obj != 0) { | |
318 | VERIFY(0 == dmu_object_free(dp->dp_meta_objset, | |
319 | scn->scn_phys.scn_queue_obj, tx)); | |
320 | scn->scn_phys.scn_queue_obj = 0; | |
321 | } | |
322 | ||
0ea05c64 AP |
323 | scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED; |
324 | ||
428870ff BB |
325 | /* |
326 | * If we were "restarted" from a stopped state, don't bother | |
327 | * with anything else. | |
328 | */ | |
329 | if (scn->scn_phys.scn_state != DSS_SCANNING) | |
330 | return; | |
331 | ||
332 | if (complete) | |
333 | scn->scn_phys.scn_state = DSS_FINISHED; | |
334 | else | |
335 | scn->scn_phys.scn_state = DSS_CANCELED; | |
336 | ||
784d15c1 NR |
337 | if (dsl_scan_restarting(scn, tx)) |
338 | spa_history_log_internal(spa, "scan aborted, restarting", tx, | |
339 | "errors=%llu", spa_get_errlog_size(spa)); | |
340 | else if (!complete) | |
341 | spa_history_log_internal(spa, "scan cancelled", tx, | |
342 | "errors=%llu", spa_get_errlog_size(spa)); | |
343 | else | |
344 | spa_history_log_internal(spa, "scan done", tx, | |
345 | "errors=%llu", spa_get_errlog_size(spa)); | |
428870ff BB |
346 | |
347 | if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { | |
348 | mutex_enter(&spa->spa_scrub_lock); | |
349 | while (spa->spa_scrub_inflight > 0) { | |
350 | cv_wait(&spa->spa_scrub_io_cv, | |
351 | &spa->spa_scrub_lock); | |
352 | } | |
353 | mutex_exit(&spa->spa_scrub_lock); | |
354 | spa->spa_scrub_started = B_FALSE; | |
355 | spa->spa_scrub_active = B_FALSE; | |
356 | ||
357 | /* | |
358 | * If the scrub/resilver completed, update all DTLs to | |
359 | * reflect this. Whether it succeeded or not, vacate | |
360 | * all temporary scrub DTLs. | |
361 | */ | |
362 | vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg, | |
363 | complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE); | |
364 | if (complete) { | |
12fa0466 DE |
365 | spa_event_notify(spa, NULL, NULL, |
366 | scn->scn_phys.scn_min_txg ? | |
fb390aaf | 367 | ESC_ZFS_RESILVER_FINISH : ESC_ZFS_SCRUB_FINISH); |
428870ff BB |
368 | } |
369 | spa_errlog_rotate(spa); | |
370 | ||
371 | /* | |
372 | * We may have finished replacing a device. | |
373 | * Let the async thread assess this and handle the detach. | |
374 | */ | |
375 | spa_async_request(spa, SPA_ASYNC_RESILVER_DONE); | |
376 | } | |
377 | ||
378 | scn->scn_phys.scn_end_time = gethrestime_sec(); | |
4f2dcb3e RY |
379 | |
380 | if (spa->spa_errata == ZPOOL_ERRATA_ZOL_2094_SCRUB) | |
381 | spa->spa_errata = 0; | |
428870ff BB |
382 | } |
383 | ||
384 | /* ARGSUSED */ | |
385 | static int | |
13fe0198 | 386 | dsl_scan_cancel_check(void *arg, dmu_tx_t *tx) |
428870ff | 387 | { |
13fe0198 | 388 | dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; |
428870ff BB |
389 | |
390 | if (scn->scn_phys.scn_state != DSS_SCANNING) | |
2e528b49 | 391 | return (SET_ERROR(ENOENT)); |
428870ff BB |
392 | return (0); |
393 | } | |
394 | ||
395 | /* ARGSUSED */ | |
396 | static void | |
13fe0198 | 397 | dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx) |
428870ff | 398 | { |
13fe0198 | 399 | dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan; |
428870ff BB |
400 | |
401 | dsl_scan_done(scn, B_FALSE, tx); | |
402 | dsl_scan_sync_state(scn, tx); | |
403 | } | |
404 | ||
405 | int | |
406 | dsl_scan_cancel(dsl_pool_t *dp) | |
407 | { | |
13fe0198 | 408 | return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check, |
3d45fdd6 | 409 | dsl_scan_cancel_sync, NULL, 3, ZFS_SPACE_CHECK_RESERVED)); |
428870ff BB |
410 | } |
411 | ||
0ea05c64 AP |
412 | boolean_t |
413 | dsl_scan_is_paused_scrub(const dsl_scan_t *scn) | |
414 | { | |
415 | if (dsl_scan_scrubbing(scn->scn_dp) && | |
416 | scn->scn_phys.scn_flags & DSF_SCRUB_PAUSED) | |
417 | return (B_TRUE); | |
418 | ||
419 | return (B_FALSE); | |
420 | } | |
421 | ||
422 | static int | |
423 | dsl_scrub_pause_resume_check(void *arg, dmu_tx_t *tx) | |
424 | { | |
425 | pool_scrub_cmd_t *cmd = arg; | |
426 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
427 | dsl_scan_t *scn = dp->dp_scan; | |
428 | ||
429 | if (*cmd == POOL_SCRUB_PAUSE) { | |
430 | /* can't pause a scrub when there is no in-progress scrub */ | |
431 | if (!dsl_scan_scrubbing(dp)) | |
432 | return (SET_ERROR(ENOENT)); | |
433 | ||
434 | /* can't pause a paused scrub */ | |
435 | if (dsl_scan_is_paused_scrub(scn)) | |
436 | return (SET_ERROR(EBUSY)); | |
437 | } else if (*cmd != POOL_SCRUB_NORMAL) { | |
438 | return (SET_ERROR(ENOTSUP)); | |
439 | } | |
440 | ||
441 | return (0); | |
442 | } | |
443 | ||
444 | static void | |
445 | dsl_scrub_pause_resume_sync(void *arg, dmu_tx_t *tx) | |
446 | { | |
447 | pool_scrub_cmd_t *cmd = arg; | |
448 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
449 | spa_t *spa = dp->dp_spa; | |
450 | dsl_scan_t *scn = dp->dp_scan; | |
451 | ||
0ea05c64 AP |
452 | if (*cmd == POOL_SCRUB_PAUSE) { |
453 | /* can't pause a scrub when there is no in-progress scrub */ | |
454 | spa->spa_scan_pass_scrub_pause = gethrestime_sec(); | |
455 | scn->scn_phys.scn_flags |= DSF_SCRUB_PAUSED; | |
456 | dsl_scan_sync_state(scn, tx); | |
457 | } else { | |
458 | ASSERT3U(*cmd, ==, POOL_SCRUB_NORMAL); | |
459 | if (dsl_scan_is_paused_scrub(scn)) { | |
460 | /* | |
461 | * We need to keep track of how much time we spend | |
462 | * paused per pass so that we can adjust the scrub rate | |
463 | * shown in the output of 'zpool status' | |
464 | */ | |
465 | spa->spa_scan_pass_scrub_spent_paused += | |
466 | gethrestime_sec() - spa->spa_scan_pass_scrub_pause; | |
467 | spa->spa_scan_pass_scrub_pause = 0; | |
468 | scn->scn_phys.scn_flags &= ~DSF_SCRUB_PAUSED; | |
469 | dsl_scan_sync_state(scn, tx); | |
470 | } | |
471 | } | |
472 | } | |
473 | ||
474 | /* | |
475 | * Set scrub pause/resume state if it makes sense to do so | |
476 | */ | |
477 | int | |
478 | dsl_scrub_set_pause_resume(const dsl_pool_t *dp, pool_scrub_cmd_t cmd) | |
479 | { | |
480 | return (dsl_sync_task(spa_name(dp->dp_spa), | |
481 | dsl_scrub_pause_resume_check, dsl_scrub_pause_resume_sync, &cmd, 3, | |
482 | ZFS_SPACE_CHECK_RESERVED)); | |
483 | } | |
484 | ||
485 | boolean_t | |
486 | dsl_scan_scrubbing(const dsl_pool_t *dp) | |
487 | { | |
488 | dsl_scan_t *scn = dp->dp_scan; | |
489 | ||
490 | if (scn->scn_phys.scn_state == DSS_SCANNING && | |
491 | scn->scn_phys.scn_func == POOL_SCAN_SCRUB) | |
492 | return (B_TRUE); | |
493 | ||
494 | return (B_FALSE); | |
495 | } | |
496 | ||
ebcf4936 MA |
497 | static void dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb, |
498 | dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn, | |
499 | dmu_objset_type_t ostype, dmu_tx_t *tx); | |
10be533e BB |
500 | inline __attribute__((always_inline)) static void dsl_scan_visitdnode( |
501 | dsl_scan_t *, dsl_dataset_t *ds, dmu_objset_type_t ostype, | |
ebcf4936 | 502 | dnode_phys_t *dnp, uint64_t object, dmu_tx_t *tx); |
428870ff BB |
503 | |
504 | void | |
505 | dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp) | |
506 | { | |
507 | zio_free(dp->dp_spa, txg, bp); | |
508 | } | |
509 | ||
510 | void | |
511 | dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp) | |
512 | { | |
513 | ASSERT(dsl_pool_sync_context(dp)); | |
514 | zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags)); | |
515 | } | |
516 | ||
428870ff BB |
517 | static uint64_t |
518 | dsl_scan_ds_maxtxg(dsl_dataset_t *ds) | |
519 | { | |
520 | uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg; | |
0c66c32d | 521 | if (ds->ds_is_snapshot) |
d683ddbb | 522 | return (MIN(smt, dsl_dataset_phys(ds)->ds_creation_txg)); |
428870ff BB |
523 | return (smt); |
524 | } | |
525 | ||
526 | static void | |
527 | dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx) | |
528 | { | |
13fe0198 | 529 | VERIFY0(zap_update(scn->scn_dp->dp_meta_objset, |
428870ff BB |
530 | DMU_POOL_DIRECTORY_OBJECT, |
531 | DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS, | |
532 | &scn->scn_phys, tx)); | |
533 | } | |
534 | ||
10400bfe MA |
535 | extern int zfs_vdev_async_write_active_min_dirty_percent; |
536 | ||
428870ff | 537 | static boolean_t |
0ea05c64 | 538 | dsl_scan_check_suspend(dsl_scan_t *scn, const zbookmark_phys_t *zb) |
428870ff BB |
539 | { |
540 | uint64_t elapsed_nanosecs; | |
541 | int mintime; | |
10400bfe | 542 | int dirty_pct; |
428870ff BB |
543 | |
544 | /* we never skip user/group accounting objects */ | |
545 | if (zb && (int64_t)zb->zb_object < 0) | |
546 | return (B_FALSE); | |
547 | ||
0ea05c64 AP |
548 | if (scn->scn_suspending) |
549 | return (B_TRUE); /* we're already suspending */ | |
428870ff | 550 | |
9ae529ec | 551 | if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark)) |
428870ff BB |
552 | return (B_FALSE); /* we're resuming */ |
553 | ||
554 | /* We only know how to resume from level-0 blocks. */ | |
555 | if (zb && zb->zb_level != 0) | |
556 | return (B_FALSE); | |
557 | ||
10400bfe | 558 | /* |
0ea05c64 | 559 | * We suspend if: |
10400bfe MA |
560 | * - we have scanned for the maximum time: an entire txg |
561 | * timeout (default 5 sec) | |
562 | * or | |
563 | * - we have scanned for at least the minimum time (default 1 sec | |
564 | * for scrub, 3 sec for resilver), and either we have sufficient | |
565 | * dirty data that we are starting to write more quickly | |
566 | * (default 30%), or someone is explicitly waiting for this txg | |
567 | * to complete. | |
568 | * or | |
569 | * - the spa is shutting down because this pool is being exported | |
570 | * or the machine is rebooting. | |
571 | */ | |
428870ff BB |
572 | mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ? |
573 | zfs_resilver_min_time_ms : zfs_scan_min_time_ms; | |
574 | elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; | |
10400bfe MA |
575 | dirty_pct = scn->scn_dp->dp_dirty_total * 100 / zfs_dirty_data_max; |
576 | if (elapsed_nanosecs / NANOSEC >= zfs_txg_timeout || | |
63fd3c6c | 577 | (NSEC2MSEC(elapsed_nanosecs) > mintime && |
10400bfe MA |
578 | (txg_sync_waiting(scn->scn_dp) || |
579 | dirty_pct >= zfs_vdev_async_write_active_min_dirty_percent)) || | |
428870ff BB |
580 | spa_shutting_down(scn->scn_dp->dp_spa)) { |
581 | if (zb) { | |
0ea05c64 | 582 | dprintf("suspending at bookmark %llx/%llx/%llx/%llx\n", |
428870ff BB |
583 | (longlong_t)zb->zb_objset, |
584 | (longlong_t)zb->zb_object, | |
585 | (longlong_t)zb->zb_level, | |
586 | (longlong_t)zb->zb_blkid); | |
587 | scn->scn_phys.scn_bookmark = *zb; | |
588 | } | |
0ea05c64 | 589 | dprintf("suspending at DDT bookmark %llx/%llx/%llx/%llx\n", |
428870ff BB |
590 | (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class, |
591 | (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type, | |
592 | (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum, | |
593 | (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor); | |
0ea05c64 | 594 | scn->scn_suspending = B_TRUE; |
428870ff BB |
595 | return (B_TRUE); |
596 | } | |
597 | return (B_FALSE); | |
598 | } | |
599 | ||
600 | typedef struct zil_scan_arg { | |
601 | dsl_pool_t *zsa_dp; | |
602 | zil_header_t *zsa_zh; | |
603 | } zil_scan_arg_t; | |
604 | ||
605 | /* ARGSUSED */ | |
606 | static int | |
607 | dsl_scan_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg) | |
608 | { | |
609 | zil_scan_arg_t *zsa = arg; | |
610 | dsl_pool_t *dp = zsa->zsa_dp; | |
611 | dsl_scan_t *scn = dp->dp_scan; | |
612 | zil_header_t *zh = zsa->zsa_zh; | |
5dbd68a3 | 613 | zbookmark_phys_t zb; |
428870ff | 614 | |
b0bc7a84 | 615 | if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) |
428870ff BB |
616 | return (0); |
617 | ||
618 | /* | |
619 | * One block ("stubby") can be allocated a long time ago; we | |
620 | * want to visit that one because it has been allocated | |
621 | * (on-disk) even if it hasn't been claimed (even though for | |
622 | * scrub there's nothing to do to it). | |
623 | */ | |
624 | if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(dp->dp_spa)) | |
625 | return (0); | |
626 | ||
627 | SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], | |
628 | ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]); | |
629 | ||
630 | VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); | |
631 | return (0); | |
632 | } | |
633 | ||
634 | /* ARGSUSED */ | |
635 | static int | |
636 | dsl_scan_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg) | |
637 | { | |
638 | if (lrc->lrc_txtype == TX_WRITE) { | |
639 | zil_scan_arg_t *zsa = arg; | |
640 | dsl_pool_t *dp = zsa->zsa_dp; | |
641 | dsl_scan_t *scn = dp->dp_scan; | |
642 | zil_header_t *zh = zsa->zsa_zh; | |
643 | lr_write_t *lr = (lr_write_t *)lrc; | |
644 | blkptr_t *bp = &lr->lr_blkptr; | |
5dbd68a3 | 645 | zbookmark_phys_t zb; |
428870ff | 646 | |
b0bc7a84 MG |
647 | if (BP_IS_HOLE(bp) || |
648 | bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) | |
428870ff BB |
649 | return (0); |
650 | ||
651 | /* | |
652 | * birth can be < claim_txg if this record's txg is | |
653 | * already txg sync'ed (but this log block contains | |
654 | * other records that are not synced) | |
655 | */ | |
656 | if (claim_txg == 0 || bp->blk_birth < claim_txg) | |
657 | return (0); | |
658 | ||
659 | SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET], | |
660 | lr->lr_foid, ZB_ZIL_LEVEL, | |
661 | lr->lr_offset / BP_GET_LSIZE(bp)); | |
662 | ||
663 | VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb)); | |
664 | } | |
665 | return (0); | |
666 | } | |
667 | ||
668 | static void | |
669 | dsl_scan_zil(dsl_pool_t *dp, zil_header_t *zh) | |
670 | { | |
671 | uint64_t claim_txg = zh->zh_claim_txg; | |
672 | zil_scan_arg_t zsa = { dp, zh }; | |
673 | zilog_t *zilog; | |
674 | ||
675 | /* | |
676 | * We only want to visit blocks that have been claimed but not yet | |
677 | * replayed (or, in read-only mode, blocks that *would* be claimed). | |
678 | */ | |
679 | if (claim_txg == 0 && spa_writeable(dp->dp_spa)) | |
680 | return; | |
681 | ||
682 | zilog = zil_alloc(dp->dp_meta_objset, zh); | |
683 | ||
684 | (void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa, | |
b5256303 | 685 | claim_txg, B_FALSE); |
428870ff BB |
686 | |
687 | zil_free(zilog); | |
688 | } | |
689 | ||
690 | /* ARGSUSED */ | |
691 | static void | |
692 | dsl_scan_prefetch(dsl_scan_t *scn, arc_buf_t *buf, blkptr_t *bp, | |
693 | uint64_t objset, uint64_t object, uint64_t blkid) | |
694 | { | |
5dbd68a3 | 695 | zbookmark_phys_t czb; |
2a432414 | 696 | arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH; |
b5256303 | 697 | int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD; |
428870ff BB |
698 | |
699 | if (zfs_no_scrub_prefetch) | |
700 | return; | |
701 | ||
702 | if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_min_txg || | |
703 | (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)) | |
704 | return; | |
705 | ||
b5256303 TC |
706 | if (BP_IS_PROTECTED(bp)) { |
707 | ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_DNODE); | |
708 | ASSERT3U(BP_GET_LEVEL(bp), ==, 0); | |
709 | zio_flags |= ZIO_FLAG_RAW; | |
710 | } | |
711 | ||
428870ff BB |
712 | SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid); |
713 | ||
428870ff | 714 | (void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp, |
b5256303 | 715 | NULL, NULL, ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, &czb); |
428870ff BB |
716 | } |
717 | ||
718 | static boolean_t | |
719 | dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp, | |
5dbd68a3 | 720 | const zbookmark_phys_t *zb) |
428870ff BB |
721 | { |
722 | /* | |
723 | * We never skip over user/group accounting objects (obj<0) | |
724 | */ | |
9ae529ec | 725 | if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) && |
428870ff BB |
726 | (int64_t)zb->zb_object >= 0) { |
727 | /* | |
728 | * If we already visited this bp & everything below (in | |
729 | * a prior txg sync), don't bother doing it again. | |
730 | */ | |
fcff0f35 PD |
731 | if (zbookmark_subtree_completed(dnp, zb, |
732 | &scn->scn_phys.scn_bookmark)) | |
428870ff BB |
733 | return (B_TRUE); |
734 | ||
735 | /* | |
736 | * If we found the block we're trying to resume from, or | |
737 | * we went past it to a different object, zero it out to | |
0ea05c64 | 738 | * indicate that it's OK to start checking for suspending |
428870ff BB |
739 | * again. |
740 | */ | |
741 | if (bcmp(zb, &scn->scn_phys.scn_bookmark, sizeof (*zb)) == 0 || | |
742 | zb->zb_object > scn->scn_phys.scn_bookmark.zb_object) { | |
743 | dprintf("resuming at %llx/%llx/%llx/%llx\n", | |
744 | (longlong_t)zb->zb_objset, | |
745 | (longlong_t)zb->zb_object, | |
746 | (longlong_t)zb->zb_level, | |
747 | (longlong_t)zb->zb_blkid); | |
748 | bzero(&scn->scn_phys.scn_bookmark, sizeof (*zb)); | |
749 | } | |
750 | } | |
751 | return (B_FALSE); | |
752 | } | |
753 | ||
754 | /* | |
755 | * Return nonzero on i/o error. | |
756 | * Return new buf to write out in *bufp. | |
757 | */ | |
10be533e | 758 | inline __attribute__((always_inline)) static int |
428870ff BB |
759 | dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype, |
760 | dnode_phys_t *dnp, const blkptr_t *bp, | |
ebcf4936 | 761 | const zbookmark_phys_t *zb, dmu_tx_t *tx) |
428870ff BB |
762 | { |
763 | dsl_pool_t *dp = scn->scn_dp; | |
572e2857 | 764 | int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD; |
428870ff BB |
765 | int err; |
766 | ||
767 | if (BP_GET_LEVEL(bp) > 0) { | |
2a432414 | 768 | arc_flags_t flags = ARC_FLAG_WAIT; |
428870ff BB |
769 | int i; |
770 | blkptr_t *cbp; | |
771 | int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT; | |
ebcf4936 | 772 | arc_buf_t *buf; |
428870ff | 773 | |
ebcf4936 | 774 | err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, |
572e2857 | 775 | ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); |
428870ff BB |
776 | if (err) { |
777 | scn->scn_phys.scn_errors++; | |
778 | return (err); | |
779 | } | |
ebcf4936 MA |
780 | for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) { |
781 | dsl_scan_prefetch(scn, buf, cbp, zb->zb_objset, | |
428870ff BB |
782 | zb->zb_object, zb->zb_blkid * epb + i); |
783 | } | |
ebcf4936 | 784 | for (i = 0, cbp = buf->b_data; i < epb; i++, cbp++) { |
5dbd68a3 | 785 | zbookmark_phys_t czb; |
428870ff BB |
786 | |
787 | SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object, | |
788 | zb->zb_level - 1, | |
789 | zb->zb_blkid * epb + i); | |
790 | dsl_scan_visitbp(cbp, &czb, dnp, | |
ebcf4936 | 791 | ds, scn, ostype, tx); |
428870ff | 792 | } |
d3c2ae1c | 793 | arc_buf_destroy(buf, &buf); |
428870ff | 794 | } else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) { |
2a432414 | 795 | arc_flags_t flags = ARC_FLAG_WAIT; |
428870ff BB |
796 | dnode_phys_t *cdnp; |
797 | int i, j; | |
798 | int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT; | |
ebcf4936 | 799 | arc_buf_t *buf; |
428870ff | 800 | |
b5256303 TC |
801 | if (BP_IS_PROTECTED(bp)) { |
802 | ASSERT3U(BP_GET_COMPRESS(bp), ==, ZIO_COMPRESS_OFF); | |
803 | zio_flags |= ZIO_FLAG_RAW; | |
804 | } | |
805 | ||
ebcf4936 | 806 | err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, |
572e2857 | 807 | ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); |
428870ff BB |
808 | if (err) { |
809 | scn->scn_phys.scn_errors++; | |
810 | return (err); | |
811 | } | |
50c957f7 NB |
812 | for (i = 0, cdnp = buf->b_data; i < epb; |
813 | i += cdnp->dn_extra_slots + 1, | |
814 | cdnp += cdnp->dn_extra_slots + 1) { | |
428870ff BB |
815 | for (j = 0; j < cdnp->dn_nblkptr; j++) { |
816 | blkptr_t *cbp = &cdnp->dn_blkptr[j]; | |
ebcf4936 | 817 | dsl_scan_prefetch(scn, buf, cbp, |
428870ff BB |
818 | zb->zb_objset, zb->zb_blkid * epb + i, j); |
819 | } | |
820 | } | |
50c957f7 NB |
821 | for (i = 0, cdnp = buf->b_data; i < epb; |
822 | i += cdnp->dn_extra_slots + 1, | |
823 | cdnp += cdnp->dn_extra_slots + 1) { | |
428870ff | 824 | dsl_scan_visitdnode(scn, ds, ostype, |
ebcf4936 | 825 | cdnp, zb->zb_blkid * epb + i, tx); |
428870ff BB |
826 | } |
827 | ||
d3c2ae1c | 828 | arc_buf_destroy(buf, &buf); |
428870ff | 829 | } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) { |
2a432414 | 830 | arc_flags_t flags = ARC_FLAG_WAIT; |
428870ff | 831 | objset_phys_t *osp; |
ebcf4936 | 832 | arc_buf_t *buf; |
428870ff | 833 | |
ebcf4936 | 834 | err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, &buf, |
572e2857 | 835 | ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb); |
428870ff BB |
836 | if (err) { |
837 | scn->scn_phys.scn_errors++; | |
838 | return (err); | |
839 | } | |
840 | ||
ebcf4936 | 841 | osp = buf->b_data; |
428870ff | 842 | |
428870ff | 843 | dsl_scan_visitdnode(scn, ds, osp->os_type, |
ebcf4936 | 844 | &osp->os_meta_dnode, DMU_META_DNODE_OBJECT, tx); |
428870ff | 845 | |
ebcf4936 | 846 | if (OBJSET_BUF_HAS_USERUSED(buf)) { |
428870ff BB |
847 | /* |
848 | * We also always visit user/group accounting | |
849 | * objects, and never skip them, even if we are | |
0ea05c64 | 850 | * suspending. This is necessary so that the space |
428870ff BB |
851 | * deltas from this txg get integrated. |
852 | */ | |
853 | dsl_scan_visitdnode(scn, ds, osp->os_type, | |
ebcf4936 | 854 | &osp->os_groupused_dnode, |
428870ff BB |
855 | DMU_GROUPUSED_OBJECT, tx); |
856 | dsl_scan_visitdnode(scn, ds, osp->os_type, | |
ebcf4936 | 857 | &osp->os_userused_dnode, |
428870ff BB |
858 | DMU_USERUSED_OBJECT, tx); |
859 | } | |
d3c2ae1c | 860 | arc_buf_destroy(buf, &buf); |
428870ff BB |
861 | } |
862 | ||
863 | return (0); | |
864 | } | |
865 | ||
10be533e | 866 | inline __attribute__((always_inline)) static void |
428870ff | 867 | dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds, |
ebcf4936 | 868 | dmu_objset_type_t ostype, dnode_phys_t *dnp, |
428870ff BB |
869 | uint64_t object, dmu_tx_t *tx) |
870 | { | |
871 | int j; | |
872 | ||
873 | for (j = 0; j < dnp->dn_nblkptr; j++) { | |
5dbd68a3 | 874 | zbookmark_phys_t czb; |
428870ff BB |
875 | |
876 | SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, | |
877 | dnp->dn_nlevels - 1, j); | |
878 | dsl_scan_visitbp(&dnp->dn_blkptr[j], | |
ebcf4936 | 879 | &czb, dnp, ds, scn, ostype, tx); |
428870ff BB |
880 | } |
881 | ||
882 | if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) { | |
5dbd68a3 | 883 | zbookmark_phys_t czb; |
428870ff BB |
884 | SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object, |
885 | 0, DMU_SPILL_BLKID); | |
50c957f7 | 886 | dsl_scan_visitbp(DN_SPILL_BLKPTR(dnp), |
ebcf4936 | 887 | &czb, dnp, ds, scn, ostype, tx); |
428870ff BB |
888 | } |
889 | } | |
890 | ||
891 | /* | |
892 | * The arguments are in this order because mdb can only print the | |
893 | * first 5; we want them to be useful. | |
894 | */ | |
895 | static void | |
5dbd68a3 | 896 | dsl_scan_visitbp(blkptr_t *bp, const zbookmark_phys_t *zb, |
ebcf4936 MA |
897 | dnode_phys_t *dnp, dsl_dataset_t *ds, dsl_scan_t *scn, |
898 | dmu_objset_type_t ostype, dmu_tx_t *tx) | |
428870ff BB |
899 | { |
900 | dsl_pool_t *dp = scn->scn_dp; | |
161ce7ce BB |
901 | blkptr_t *bp_toread; |
902 | ||
79c76d5b | 903 | bp_toread = kmem_alloc(sizeof (blkptr_t), KM_SLEEP); |
161ce7ce | 904 | *bp_toread = *bp; |
428870ff BB |
905 | |
906 | /* ASSERT(pbuf == NULL || arc_released(pbuf)); */ | |
907 | ||
0ea05c64 | 908 | if (dsl_scan_check_suspend(scn, zb)) |
161ce7ce | 909 | goto out; |
428870ff BB |
910 | |
911 | if (dsl_scan_check_resume(scn, dnp, zb)) | |
161ce7ce | 912 | goto out; |
428870ff | 913 | |
b0bc7a84 | 914 | if (BP_IS_HOLE(bp)) |
161ce7ce | 915 | goto out; |
428870ff BB |
916 | |
917 | scn->scn_visited_this_txg++; | |
918 | ||
b81c4ac9 BB |
919 | /* |
920 | * This debugging is commented out to conserve stack space. This | |
921 | * function is called recursively and the debugging addes several | |
922 | * bytes to the stack for each call. It can be commented back in | |
923 | * if required to debug an issue in dsl_scan_visitbp(). | |
924 | * | |
925 | * dprintf_bp(bp, | |
ebcf4936 | 926 | * "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx bp=%p", |
b81c4ac9 BB |
927 | * ds, ds ? ds->ds_object : 0, |
928 | * zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid, | |
ebcf4936 | 929 | * bp); |
b81c4ac9 | 930 | */ |
428870ff BB |
931 | |
932 | if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg) | |
161ce7ce | 933 | goto out; |
428870ff | 934 | |
ebcf4936 | 935 | if (dsl_scan_recurse(scn, ds, ostype, dnp, bp_toread, zb, tx) != 0) |
161ce7ce | 936 | goto out; |
428870ff BB |
937 | |
938 | /* | |
4e33ba4c | 939 | * If dsl_scan_ddt() has already visited this block, it will have |
428870ff BB |
940 | * already done any translations or scrubbing, so don't call the |
941 | * callback again. | |
942 | */ | |
943 | if (ddt_class_contains(dp->dp_spa, | |
944 | scn->scn_phys.scn_ddt_class_max, bp)) { | |
161ce7ce | 945 | goto out; |
428870ff BB |
946 | } |
947 | ||
948 | /* | |
949 | * If this block is from the future (after cur_max_txg), then we | |
950 | * are doing this on behalf of a deleted snapshot, and we will | |
951 | * revisit the future block on the next pass of this dataset. | |
952 | * Don't scan it now unless we need to because something | |
953 | * under it was modified. | |
954 | */ | |
5d1f7fb6 | 955 | if (BP_PHYSICAL_BIRTH(bp) <= scn->scn_phys.scn_cur_max_txg) { |
428870ff BB |
956 | scan_funcs[scn->scn_phys.scn_func](dp, bp, zb); |
957 | } | |
161ce7ce | 958 | out: |
d1d7e268 | 959 | kmem_free(bp_toread, sizeof (blkptr_t)); |
428870ff BB |
960 | } |
961 | ||
962 | static void | |
963 | dsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp, | |
964 | dmu_tx_t *tx) | |
965 | { | |
5dbd68a3 | 966 | zbookmark_phys_t zb; |
428870ff BB |
967 | |
968 | SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET, | |
969 | ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID); | |
ebcf4936 | 970 | dsl_scan_visitbp(bp, &zb, NULL, |
428870ff BB |
971 | ds, scn, DMU_OST_NONE, tx); |
972 | ||
973 | dprintf_ds(ds, "finished scan%s", ""); | |
974 | } | |
975 | ||
976 | void | |
977 | dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx) | |
978 | { | |
979 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
980 | dsl_scan_t *scn = dp->dp_scan; | |
981 | uint64_t mintxg; | |
982 | ||
983 | if (scn->scn_phys.scn_state != DSS_SCANNING) | |
984 | return; | |
985 | ||
986 | if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) { | |
0c66c32d | 987 | if (ds->ds_is_snapshot) { |
b77222c8 MA |
988 | /* |
989 | * Note: | |
990 | * - scn_cur_{min,max}_txg stays the same. | |
991 | * - Setting the flag is not really necessary if | |
992 | * scn_cur_max_txg == scn_max_txg, because there | |
993 | * is nothing after this snapshot that we care | |
994 | * about. However, we set it anyway and then | |
995 | * ignore it when we retraverse it in | |
996 | * dsl_scan_visitds(). | |
997 | */ | |
428870ff | 998 | scn->scn_phys.scn_bookmark.zb_objset = |
d683ddbb | 999 | dsl_dataset_phys(ds)->ds_next_snap_obj; |
428870ff BB |
1000 | zfs_dbgmsg("destroying ds %llu; currently traversing; " |
1001 | "reset zb_objset to %llu", | |
1002 | (u_longlong_t)ds->ds_object, | |
d683ddbb JG |
1003 | (u_longlong_t)dsl_dataset_phys(ds)-> |
1004 | ds_next_snap_obj); | |
428870ff BB |
1005 | scn->scn_phys.scn_flags |= DSF_VISIT_DS_AGAIN; |
1006 | } else { | |
1007 | SET_BOOKMARK(&scn->scn_phys.scn_bookmark, | |
1008 | ZB_DESTROYED_OBJSET, 0, 0, 0); | |
1009 | zfs_dbgmsg("destroying ds %llu; currently traversing; " | |
1010 | "reset bookmark to -1,0,0,0", | |
1011 | (u_longlong_t)ds->ds_object); | |
1012 | } | |
1013 | } else if (zap_lookup_int_key(dp->dp_meta_objset, | |
1014 | scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { | |
d683ddbb | 1015 | ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1); |
428870ff BB |
1016 | VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, |
1017 | scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); | |
0c66c32d | 1018 | if (ds->ds_is_snapshot) { |
428870ff BB |
1019 | /* |
1020 | * We keep the same mintxg; it could be > | |
1021 | * ds_creation_txg if the previous snapshot was | |
1022 | * deleted too. | |
1023 | */ | |
1024 | VERIFY(zap_add_int_key(dp->dp_meta_objset, | |
1025 | scn->scn_phys.scn_queue_obj, | |
d683ddbb JG |
1026 | dsl_dataset_phys(ds)->ds_next_snap_obj, |
1027 | mintxg, tx) == 0); | |
428870ff BB |
1028 | zfs_dbgmsg("destroying ds %llu; in queue; " |
1029 | "replacing with %llu", | |
1030 | (u_longlong_t)ds->ds_object, | |
d683ddbb JG |
1031 | (u_longlong_t)dsl_dataset_phys(ds)-> |
1032 | ds_next_snap_obj); | |
428870ff BB |
1033 | } else { |
1034 | zfs_dbgmsg("destroying ds %llu; in queue; removing", | |
1035 | (u_longlong_t)ds->ds_object); | |
1036 | } | |
428870ff BB |
1037 | } |
1038 | ||
1039 | /* | |
1040 | * dsl_scan_sync() should be called after this, and should sync | |
1041 | * out our changed state, but just to be safe, do it here. | |
1042 | */ | |
1043 | dsl_scan_sync_state(scn, tx); | |
1044 | } | |
1045 | ||
1046 | void | |
1047 | dsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx) | |
1048 | { | |
1049 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
1050 | dsl_scan_t *scn = dp->dp_scan; | |
1051 | uint64_t mintxg; | |
1052 | ||
1053 | if (scn->scn_phys.scn_state != DSS_SCANNING) | |
1054 | return; | |
1055 | ||
d683ddbb | 1056 | ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0); |
428870ff BB |
1057 | |
1058 | if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) { | |
1059 | scn->scn_phys.scn_bookmark.zb_objset = | |
d683ddbb | 1060 | dsl_dataset_phys(ds)->ds_prev_snap_obj; |
428870ff BB |
1061 | zfs_dbgmsg("snapshotting ds %llu; currently traversing; " |
1062 | "reset zb_objset to %llu", | |
1063 | (u_longlong_t)ds->ds_object, | |
d683ddbb | 1064 | (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj); |
428870ff BB |
1065 | } else if (zap_lookup_int_key(dp->dp_meta_objset, |
1066 | scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) { | |
1067 | VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, | |
1068 | scn->scn_phys.scn_queue_obj, ds->ds_object, tx)); | |
1069 | VERIFY(zap_add_int_key(dp->dp_meta_objset, | |
1070 | scn->scn_phys.scn_queue_obj, | |
d683ddbb | 1071 | dsl_dataset_phys(ds)->ds_prev_snap_obj, mintxg, tx) == 0); |
428870ff BB |
1072 | zfs_dbgmsg("snapshotting ds %llu; in queue; " |
1073 | "replacing with %llu", | |
1074 | (u_longlong_t)ds->ds_object, | |
d683ddbb | 1075 | (u_longlong_t)dsl_dataset_phys(ds)->ds_prev_snap_obj); |
428870ff BB |
1076 | } |
1077 | dsl_scan_sync_state(scn, tx); | |
1078 | } | |
1079 | ||
1080 | void | |
1081 | dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx) | |
1082 | { | |
1083 | dsl_pool_t *dp = ds1->ds_dir->dd_pool; | |
1084 | dsl_scan_t *scn = dp->dp_scan; | |
1085 | uint64_t mintxg; | |
1086 | ||
1087 | if (scn->scn_phys.scn_state != DSS_SCANNING) | |
1088 | return; | |
1089 | ||
1090 | if (scn->scn_phys.scn_bookmark.zb_objset == ds1->ds_object) { | |
1091 | scn->scn_phys.scn_bookmark.zb_objset = ds2->ds_object; | |
1092 | zfs_dbgmsg("clone_swap ds %llu; currently traversing; " | |
1093 | "reset zb_objset to %llu", | |
1094 | (u_longlong_t)ds1->ds_object, | |
1095 | (u_longlong_t)ds2->ds_object); | |
1096 | } else if (scn->scn_phys.scn_bookmark.zb_objset == ds2->ds_object) { | |
1097 | scn->scn_phys.scn_bookmark.zb_objset = ds1->ds_object; | |
1098 | zfs_dbgmsg("clone_swap ds %llu; currently traversing; " | |
1099 | "reset zb_objset to %llu", | |
1100 | (u_longlong_t)ds2->ds_object, | |
1101 | (u_longlong_t)ds1->ds_object); | |
1102 | } | |
1103 | ||
1104 | if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, | |
1105 | ds1->ds_object, &mintxg) == 0) { | |
1106 | int err; | |
1107 | ||
d683ddbb JG |
1108 | ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); |
1109 | ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); | |
428870ff BB |
1110 | VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, |
1111 | scn->scn_phys.scn_queue_obj, ds1->ds_object, tx)); | |
1112 | err = zap_add_int_key(dp->dp_meta_objset, | |
1113 | scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg, tx); | |
1114 | VERIFY(err == 0 || err == EEXIST); | |
1115 | if (err == EEXIST) { | |
1116 | /* Both were there to begin with */ | |
1117 | VERIFY(0 == zap_add_int_key(dp->dp_meta_objset, | |
1118 | scn->scn_phys.scn_queue_obj, | |
1119 | ds1->ds_object, mintxg, tx)); | |
1120 | } | |
1121 | zfs_dbgmsg("clone_swap ds %llu; in queue; " | |
1122 | "replacing with %llu", | |
1123 | (u_longlong_t)ds1->ds_object, | |
1124 | (u_longlong_t)ds2->ds_object); | |
1125 | } else if (zap_lookup_int_key(dp->dp_meta_objset, | |
1126 | scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg) == 0) { | |
d683ddbb JG |
1127 | ASSERT3U(mintxg, ==, dsl_dataset_phys(ds1)->ds_prev_snap_txg); |
1128 | ASSERT3U(mintxg, ==, dsl_dataset_phys(ds2)->ds_prev_snap_txg); | |
428870ff BB |
1129 | VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, |
1130 | scn->scn_phys.scn_queue_obj, ds2->ds_object, tx)); | |
1131 | VERIFY(0 == zap_add_int_key(dp->dp_meta_objset, | |
1132 | scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx)); | |
1133 | zfs_dbgmsg("clone_swap ds %llu; in queue; " | |
1134 | "replacing with %llu", | |
1135 | (u_longlong_t)ds2->ds_object, | |
1136 | (u_longlong_t)ds1->ds_object); | |
1137 | } | |
1138 | ||
1139 | dsl_scan_sync_state(scn, tx); | |
1140 | } | |
1141 | ||
1142 | struct enqueue_clones_arg { | |
1143 | dmu_tx_t *tx; | |
1144 | uint64_t originobj; | |
1145 | }; | |
1146 | ||
1147 | /* ARGSUSED */ | |
1148 | static int | |
13fe0198 | 1149 | enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) |
428870ff BB |
1150 | { |
1151 | struct enqueue_clones_arg *eca = arg; | |
1152 | dsl_dataset_t *ds; | |
1153 | int err; | |
428870ff BB |
1154 | dsl_scan_t *scn = dp->dp_scan; |
1155 | ||
d683ddbb | 1156 | if (dsl_dir_phys(hds->ds_dir)->dd_origin_obj != eca->originobj) |
13fe0198 MA |
1157 | return (0); |
1158 | ||
1159 | err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); | |
428870ff BB |
1160 | if (err) |
1161 | return (err); | |
1162 | ||
d683ddbb | 1163 | while (dsl_dataset_phys(ds)->ds_prev_snap_obj != eca->originobj) { |
13fe0198 MA |
1164 | dsl_dataset_t *prev; |
1165 | err = dsl_dataset_hold_obj(dp, | |
d683ddbb | 1166 | dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); |
428870ff | 1167 | |
13fe0198 MA |
1168 | dsl_dataset_rele(ds, FTAG); |
1169 | if (err) | |
1170 | return (err); | |
1171 | ds = prev; | |
428870ff | 1172 | } |
13fe0198 MA |
1173 | VERIFY(zap_add_int_key(dp->dp_meta_objset, |
1174 | scn->scn_phys.scn_queue_obj, ds->ds_object, | |
d683ddbb | 1175 | dsl_dataset_phys(ds)->ds_prev_snap_txg, eca->tx) == 0); |
428870ff BB |
1176 | dsl_dataset_rele(ds, FTAG); |
1177 | return (0); | |
1178 | } | |
1179 | ||
1180 | static void | |
1181 | dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx) | |
1182 | { | |
1183 | dsl_pool_t *dp = scn->scn_dp; | |
1184 | dsl_dataset_t *ds; | |
572e2857 | 1185 | objset_t *os; |
d6320ddb | 1186 | char *dsname; |
428870ff BB |
1187 | |
1188 | VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); | |
1189 | ||
b77222c8 MA |
1190 | if (scn->scn_phys.scn_cur_min_txg >= |
1191 | scn->scn_phys.scn_max_txg) { | |
1192 | /* | |
1193 | * This can happen if this snapshot was created after the | |
1194 | * scan started, and we already completed a previous snapshot | |
1195 | * that was created after the scan started. This snapshot | |
1196 | * only references blocks with: | |
1197 | * | |
1198 | * birth < our ds_creation_txg | |
1199 | * cur_min_txg is no less than ds_creation_txg. | |
1200 | * We have already visited these blocks. | |
1201 | * or | |
1202 | * birth > scn_max_txg | |
1203 | * The scan requested not to visit these blocks. | |
1204 | * | |
1205 | * Subsequent snapshots (and clones) can reference our | |
1206 | * blocks, or blocks with even higher birth times. | |
1207 | * Therefore we do not need to visit them either, | |
1208 | * so we do not add them to the work queue. | |
1209 | * | |
1210 | * Note that checking for cur_min_txg >= cur_max_txg | |
1211 | * is not sufficient, because in that case we may need to | |
1212 | * visit subsequent snapshots. This happens when min_txg > 0, | |
1213 | * which raises cur_min_txg. In this case we will visit | |
1214 | * this dataset but skip all of its blocks, because the | |
1215 | * rootbp's birth time is < cur_min_txg. Then we will | |
1216 | * add the next snapshots/clones to the work queue. | |
1217 | */ | |
eca7b760 | 1218 | char *dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
b77222c8 MA |
1219 | dsl_dataset_name(ds, dsname); |
1220 | zfs_dbgmsg("scanning dataset %llu (%s) is unnecessary because " | |
1221 | "cur_min_txg (%llu) >= max_txg (%llu)", | |
1222 | dsobj, dsname, | |
1223 | scn->scn_phys.scn_cur_min_txg, | |
1224 | scn->scn_phys.scn_max_txg); | |
1225 | kmem_free(dsname, MAXNAMELEN); | |
1226 | ||
1227 | goto out; | |
1228 | } | |
1229 | ||
572e2857 BB |
1230 | if (dmu_objset_from_ds(ds, &os)) |
1231 | goto out; | |
1232 | ||
1233 | /* | |
1234 | * Only the ZIL in the head (non-snapshot) is valid. Even though | |
1235 | * snapshots can have ZIL block pointers (which may be the same | |
1236 | * BP as in the head), they must be ignored. So we traverse the | |
1237 | * ZIL here, rather than in scan_recurse(), because the regular | |
1238 | * snapshot block-sharing rules don't apply to it. | |
1239 | */ | |
0c66c32d | 1240 | if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !ds->ds_is_snapshot) |
572e2857 BB |
1241 | dsl_scan_zil(dp, &os->os_zil_header); |
1242 | ||
428870ff BB |
1243 | /* |
1244 | * Iterate over the bps in this ds. | |
1245 | */ | |
1246 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
cc9bb3e5 | 1247 | rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG); |
d683ddbb | 1248 | dsl_scan_visit_rootbp(scn, ds, &dsl_dataset_phys(ds)->ds_bp, tx); |
cc9bb3e5 | 1249 | rrw_exit(&ds->ds_bp_rwlock, FTAG); |
428870ff | 1250 | |
eca7b760 | 1251 | dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP); |
428870ff BB |
1252 | dsl_dataset_name(ds, dsname); |
1253 | zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; " | |
0ea05c64 | 1254 | "suspending=%u", |
428870ff BB |
1255 | (longlong_t)dsobj, dsname, |
1256 | (longlong_t)scn->scn_phys.scn_cur_min_txg, | |
1257 | (longlong_t)scn->scn_phys.scn_cur_max_txg, | |
0ea05c64 | 1258 | (int)scn->scn_suspending); |
eca7b760 | 1259 | kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN); |
428870ff | 1260 | |
0ea05c64 | 1261 | if (scn->scn_suspending) |
428870ff BB |
1262 | goto out; |
1263 | ||
1264 | /* | |
1265 | * We've finished this pass over this dataset. | |
1266 | */ | |
1267 | ||
1268 | /* | |
1269 | * If we did not completely visit this dataset, do another pass. | |
1270 | */ | |
1271 | if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) { | |
1272 | zfs_dbgmsg("incomplete pass; visiting again"); | |
1273 | scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN; | |
1274 | VERIFY(zap_add_int_key(dp->dp_meta_objset, | |
1275 | scn->scn_phys.scn_queue_obj, ds->ds_object, | |
1276 | scn->scn_phys.scn_cur_max_txg, tx) == 0); | |
1277 | goto out; | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * Add descendent datasets to work queue. | |
1282 | */ | |
d683ddbb | 1283 | if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) { |
428870ff | 1284 | VERIFY(zap_add_int_key(dp->dp_meta_objset, |
d683ddbb JG |
1285 | scn->scn_phys.scn_queue_obj, |
1286 | dsl_dataset_phys(ds)->ds_next_snap_obj, | |
1287 | dsl_dataset_phys(ds)->ds_creation_txg, tx) == 0); | |
428870ff | 1288 | } |
d683ddbb | 1289 | if (dsl_dataset_phys(ds)->ds_num_children > 1) { |
428870ff | 1290 | boolean_t usenext = B_FALSE; |
d683ddbb | 1291 | if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) { |
428870ff BB |
1292 | uint64_t count; |
1293 | /* | |
1294 | * A bug in a previous version of the code could | |
1295 | * cause upgrade_clones_cb() to not set | |
1296 | * ds_next_snap_obj when it should, leading to a | |
1297 | * missing entry. Therefore we can only use the | |
1298 | * next_clones_obj when its count is correct. | |
1299 | */ | |
1300 | int err = zap_count(dp->dp_meta_objset, | |
d683ddbb | 1301 | dsl_dataset_phys(ds)->ds_next_clones_obj, &count); |
428870ff | 1302 | if (err == 0 && |
d683ddbb | 1303 | count == dsl_dataset_phys(ds)->ds_num_children - 1) |
428870ff BB |
1304 | usenext = B_TRUE; |
1305 | } | |
1306 | ||
1307 | if (usenext) { | |
13fe0198 | 1308 | VERIFY0(zap_join_key(dp->dp_meta_objset, |
d683ddbb | 1309 | dsl_dataset_phys(ds)->ds_next_clones_obj, |
428870ff | 1310 | scn->scn_phys.scn_queue_obj, |
d683ddbb | 1311 | dsl_dataset_phys(ds)->ds_creation_txg, tx)); |
428870ff BB |
1312 | } else { |
1313 | struct enqueue_clones_arg eca; | |
1314 | eca.tx = tx; | |
1315 | eca.originobj = ds->ds_object; | |
1316 | ||
13fe0198 MA |
1317 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
1318 | enqueue_clones_cb, &eca, DS_FIND_CHILDREN)); | |
428870ff BB |
1319 | } |
1320 | } | |
1321 | ||
1322 | out: | |
1323 | dsl_dataset_rele(ds, FTAG); | |
1324 | } | |
1325 | ||
1326 | /* ARGSUSED */ | |
1327 | static int | |
13fe0198 | 1328 | enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg) |
428870ff BB |
1329 | { |
1330 | dmu_tx_t *tx = arg; | |
1331 | dsl_dataset_t *ds; | |
1332 | int err; | |
428870ff BB |
1333 | dsl_scan_t *scn = dp->dp_scan; |
1334 | ||
13fe0198 | 1335 | err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds); |
428870ff BB |
1336 | if (err) |
1337 | return (err); | |
1338 | ||
d683ddbb | 1339 | while (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) { |
428870ff | 1340 | dsl_dataset_t *prev; |
d683ddbb JG |
1341 | err = dsl_dataset_hold_obj(dp, |
1342 | dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev); | |
428870ff BB |
1343 | if (err) { |
1344 | dsl_dataset_rele(ds, FTAG); | |
1345 | return (err); | |
1346 | } | |
1347 | ||
1348 | /* | |
1349 | * If this is a clone, we don't need to worry about it for now. | |
1350 | */ | |
d683ddbb | 1351 | if (dsl_dataset_phys(prev)->ds_next_snap_obj != ds->ds_object) { |
428870ff BB |
1352 | dsl_dataset_rele(ds, FTAG); |
1353 | dsl_dataset_rele(prev, FTAG); | |
1354 | return (0); | |
1355 | } | |
1356 | dsl_dataset_rele(ds, FTAG); | |
1357 | ds = prev; | |
1358 | } | |
1359 | ||
1360 | VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj, | |
d683ddbb | 1361 | ds->ds_object, dsl_dataset_phys(ds)->ds_prev_snap_txg, tx) == 0); |
428870ff BB |
1362 | dsl_dataset_rele(ds, FTAG); |
1363 | return (0); | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Scrub/dedup interaction. | |
1368 | * | |
1369 | * If there are N references to a deduped block, we don't want to scrub it | |
1370 | * N times -- ideally, we should scrub it exactly once. | |
1371 | * | |
1372 | * We leverage the fact that the dde's replication class (enum ddt_class) | |
1373 | * is ordered from highest replication class (DDT_CLASS_DITTO) to lowest | |
1374 | * (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order. | |
1375 | * | |
1376 | * To prevent excess scrubbing, the scrub begins by walking the DDT | |
1377 | * to find all blocks with refcnt > 1, and scrubs each of these once. | |
1378 | * Since there are two replication classes which contain blocks with | |
1379 | * refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first. | |
1380 | * Finally the top-down scrub begins, only visiting blocks with refcnt == 1. | |
1381 | * | |
1382 | * There would be nothing more to say if a block's refcnt couldn't change | |
1383 | * during a scrub, but of course it can so we must account for changes | |
1384 | * in a block's replication class. | |
1385 | * | |
1386 | * Here's an example of what can occur: | |
1387 | * | |
1388 | * If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1 | |
1389 | * when visited during the top-down scrub phase, it will be scrubbed twice. | |
1390 | * This negates our scrub optimization, but is otherwise harmless. | |
1391 | * | |
1392 | * If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1 | |
1393 | * on each visit during the top-down scrub phase, it will never be scrubbed. | |
1394 | * To catch this, ddt_sync_entry() notifies the scrub code whenever a block's | |
1395 | * reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to | |
1396 | * DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1 | |
1397 | * while a scrub is in progress, it scrubs the block right then. | |
1398 | */ | |
1399 | static void | |
1400 | dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx) | |
1401 | { | |
1402 | ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark; | |
2598c001 | 1403 | ddt_entry_t dde; |
428870ff BB |
1404 | int error; |
1405 | uint64_t n = 0; | |
1406 | ||
2598c001 BB |
1407 | bzero(&dde, sizeof (ddt_entry_t)); |
1408 | ||
428870ff BB |
1409 | while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) { |
1410 | ddt_t *ddt; | |
1411 | ||
1412 | if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max) | |
1413 | break; | |
1414 | dprintf("visiting ddb=%llu/%llu/%llu/%llx\n", | |
1415 | (longlong_t)ddb->ddb_class, | |
1416 | (longlong_t)ddb->ddb_type, | |
1417 | (longlong_t)ddb->ddb_checksum, | |
1418 | (longlong_t)ddb->ddb_cursor); | |
1419 | ||
1420 | /* There should be no pending changes to the dedup table */ | |
1421 | ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum]; | |
1422 | ASSERT(avl_first(&ddt->ddt_tree) == NULL); | |
1423 | ||
1424 | dsl_scan_ddt_entry(scn, ddb->ddb_checksum, &dde, tx); | |
1425 | n++; | |
1426 | ||
0ea05c64 | 1427 | if (dsl_scan_check_suspend(scn, NULL)) |
428870ff BB |
1428 | break; |
1429 | } | |
1430 | ||
0ea05c64 AP |
1431 | zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; " |
1432 | "suspending=%u", (longlong_t)n, | |
1433 | (int)scn->scn_phys.scn_ddt_class_max, (int)scn->scn_suspending); | |
428870ff BB |
1434 | |
1435 | ASSERT(error == 0 || error == ENOENT); | |
1436 | ASSERT(error != ENOENT || | |
1437 | ddb->ddb_class > scn->scn_phys.scn_ddt_class_max); | |
1438 | } | |
1439 | ||
1440 | /* ARGSUSED */ | |
1441 | void | |
1442 | dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum, | |
1443 | ddt_entry_t *dde, dmu_tx_t *tx) | |
1444 | { | |
1445 | const ddt_key_t *ddk = &dde->dde_key; | |
1446 | ddt_phys_t *ddp = dde->dde_phys; | |
1447 | blkptr_t bp; | |
5dbd68a3 | 1448 | zbookmark_phys_t zb = { 0 }; |
d6320ddb | 1449 | int p; |
428870ff BB |
1450 | |
1451 | if (scn->scn_phys.scn_state != DSS_SCANNING) | |
1452 | return; | |
1453 | ||
d6320ddb | 1454 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff | 1455 | if (ddp->ddp_phys_birth == 0 || |
5d1f7fb6 | 1456 | ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg) |
428870ff BB |
1457 | continue; |
1458 | ddt_bp_create(checksum, ddk, ddp, &bp); | |
1459 | ||
1460 | scn->scn_visited_this_txg++; | |
1461 | scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb); | |
1462 | } | |
1463 | } | |
1464 | ||
1465 | static void | |
1466 | dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx) | |
1467 | { | |
1468 | dsl_pool_t *dp = scn->scn_dp; | |
40a39e11 BB |
1469 | zap_cursor_t *zc; |
1470 | zap_attribute_t *za; | |
428870ff BB |
1471 | |
1472 | if (scn->scn_phys.scn_ddt_bookmark.ddb_class <= | |
1473 | scn->scn_phys.scn_ddt_class_max) { | |
1474 | scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; | |
1475 | scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; | |
1476 | dsl_scan_ddt(scn, tx); | |
0ea05c64 | 1477 | if (scn->scn_suspending) |
428870ff BB |
1478 | return; |
1479 | } | |
1480 | ||
1481 | if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) { | |
1482 | /* First do the MOS & ORIGIN */ | |
1483 | ||
1484 | scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg; | |
1485 | scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg; | |
1486 | dsl_scan_visit_rootbp(scn, NULL, | |
1487 | &dp->dp_meta_rootbp, tx); | |
1488 | spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp); | |
0ea05c64 | 1489 | if (scn->scn_suspending) |
428870ff BB |
1490 | return; |
1491 | ||
1492 | if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) { | |
13fe0198 MA |
1493 | VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj, |
1494 | enqueue_cb, tx, DS_FIND_CHILDREN)); | |
428870ff BB |
1495 | } else { |
1496 | dsl_scan_visitds(scn, | |
1497 | dp->dp_origin_snap->ds_object, tx); | |
1498 | } | |
0ea05c64 | 1499 | ASSERT(!scn->scn_suspending); |
428870ff BB |
1500 | } else if (scn->scn_phys.scn_bookmark.zb_objset != |
1501 | ZB_DESTROYED_OBJSET) { | |
1502 | /* | |
0ea05c64 AP |
1503 | * If we were suspended, continue from here. Note if the |
1504 | * ds we were suspended on was deleted, the zb_objset may | |
428870ff BB |
1505 | * be -1, so we will skip this and find a new objset |
1506 | * below. | |
1507 | */ | |
1508 | dsl_scan_visitds(scn, scn->scn_phys.scn_bookmark.zb_objset, tx); | |
0ea05c64 | 1509 | if (scn->scn_suspending) |
428870ff BB |
1510 | return; |
1511 | } | |
1512 | ||
1513 | /* | |
0ea05c64 | 1514 | * In case we were suspended right at the end of the ds, zero the |
428870ff BB |
1515 | * bookmark so we don't think that we're still trying to resume. |
1516 | */ | |
5dbd68a3 | 1517 | bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_phys_t)); |
79c76d5b BB |
1518 | zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP); |
1519 | za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); | |
428870ff BB |
1520 | |
1521 | /* keep pulling things out of the zap-object-as-queue */ | |
40a39e11 | 1522 | while (zap_cursor_init(zc, dp->dp_meta_objset, |
428870ff | 1523 | scn->scn_phys.scn_queue_obj), |
40a39e11 | 1524 | zap_cursor_retrieve(zc, za) == 0) { |
428870ff BB |
1525 | dsl_dataset_t *ds; |
1526 | uint64_t dsobj; | |
1527 | ||
e19572e4 | 1528 | dsobj = zfs_strtonum(za->za_name, NULL); |
428870ff BB |
1529 | VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, |
1530 | scn->scn_phys.scn_queue_obj, dsobj, tx)); | |
1531 | ||
1532 | /* Set up min/max txg */ | |
1533 | VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); | |
40a39e11 | 1534 | if (za->za_first_integer != 0) { |
428870ff BB |
1535 | scn->scn_phys.scn_cur_min_txg = |
1536 | MAX(scn->scn_phys.scn_min_txg, | |
40a39e11 | 1537 | za->za_first_integer); |
428870ff BB |
1538 | } else { |
1539 | scn->scn_phys.scn_cur_min_txg = | |
1540 | MAX(scn->scn_phys.scn_min_txg, | |
d683ddbb | 1541 | dsl_dataset_phys(ds)->ds_prev_snap_txg); |
428870ff BB |
1542 | } |
1543 | scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds); | |
1544 | dsl_dataset_rele(ds, FTAG); | |
1545 | ||
1546 | dsl_scan_visitds(scn, dsobj, tx); | |
40a39e11 | 1547 | zap_cursor_fini(zc); |
0ea05c64 | 1548 | if (scn->scn_suspending) |
40a39e11 | 1549 | goto out; |
428870ff | 1550 | } |
40a39e11 BB |
1551 | zap_cursor_fini(zc); |
1552 | out: | |
d1d7e268 MK |
1553 | kmem_free(za, sizeof (zap_attribute_t)); |
1554 | kmem_free(zc, sizeof (zap_cursor_t)); | |
428870ff BB |
1555 | } |
1556 | ||
9ae529ec | 1557 | static boolean_t |
0ea05c64 | 1558 | dsl_scan_free_should_suspend(dsl_scan_t *scn) |
428870ff | 1559 | { |
428870ff BB |
1560 | uint64_t elapsed_nanosecs; |
1561 | ||
78e2739d MA |
1562 | if (zfs_recover) |
1563 | return (B_FALSE); | |
1564 | ||
36283ca2 MG |
1565 | if (scn->scn_visited_this_txg >= zfs_free_max_blocks) |
1566 | return (B_TRUE); | |
1567 | ||
428870ff | 1568 | elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time; |
9ae529ec | 1569 | return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout || |
63fd3c6c | 1570 | (NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms && |
428870ff | 1571 | txg_sync_waiting(scn->scn_dp)) || |
9ae529ec CS |
1572 | spa_shutting_down(scn->scn_dp->dp_spa)); |
1573 | } | |
1574 | ||
1575 | static int | |
1576 | dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
1577 | { | |
1578 | dsl_scan_t *scn = arg; | |
1579 | ||
1580 | if (!scn->scn_is_bptree || | |
1581 | (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) { | |
0ea05c64 | 1582 | if (dsl_scan_free_should_suspend(scn)) |
2e528b49 | 1583 | return (SET_ERROR(ERESTART)); |
9ae529ec | 1584 | } |
428870ff BB |
1585 | |
1586 | zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa, | |
1587 | dmu_tx_get_txg(tx), bp, 0)); | |
1588 | dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD, | |
1589 | -bp_get_dsize_sync(scn->scn_dp->dp_spa, bp), | |
1590 | -BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx); | |
1591 | scn->scn_visited_this_txg++; | |
1592 | return (0); | |
1593 | } | |
1594 | ||
1595 | boolean_t | |
1596 | dsl_scan_active(dsl_scan_t *scn) | |
1597 | { | |
1598 | spa_t *spa = scn->scn_dp->dp_spa; | |
1599 | uint64_t used = 0, comp, uncomp; | |
1600 | ||
1601 | if (spa->spa_load_state != SPA_LOAD_NONE) | |
1602 | return (B_FALSE); | |
1603 | if (spa_shutting_down(spa)) | |
1604 | return (B_FALSE); | |
0ea05c64 AP |
1605 | if ((scn->scn_phys.scn_state == DSS_SCANNING && |
1606 | !dsl_scan_is_paused_scrub(scn)) || | |
fbeddd60 | 1607 | (scn->scn_async_destroying && !scn->scn_async_stalled)) |
428870ff BB |
1608 | return (B_TRUE); |
1609 | ||
1610 | if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) { | |
1611 | (void) bpobj_space(&scn->scn_dp->dp_free_bpobj, | |
1612 | &used, &comp, &uncomp); | |
1613 | } | |
1614 | return (used != 0); | |
1615 | } | |
1616 | ||
00710365 | 1617 | /* Called whenever a txg syncs. */ |
428870ff BB |
1618 | void |
1619 | dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx) | |
1620 | { | |
1621 | dsl_scan_t *scn = dp->dp_scan; | |
1622 | spa_t *spa = dp->dp_spa; | |
fbeddd60 | 1623 | int err = 0; |
428870ff BB |
1624 | |
1625 | /* | |
1626 | * Check for scn_restart_txg before checking spa_load_state, so | |
1627 | * that we can restart an old-style scan while the pool is being | |
1628 | * imported (see dsl_scan_init). | |
1629 | */ | |
784d15c1 | 1630 | if (dsl_scan_restarting(scn, tx)) { |
428870ff BB |
1631 | pool_scan_func_t func = POOL_SCAN_SCRUB; |
1632 | dsl_scan_done(scn, B_FALSE, tx); | |
1633 | if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) | |
1634 | func = POOL_SCAN_RESILVER; | |
1635 | zfs_dbgmsg("restarting scan func=%u txg=%llu", | |
1636 | func, tx->tx_txg); | |
13fe0198 | 1637 | dsl_scan_setup_sync(&func, tx); |
428870ff BB |
1638 | } |
1639 | ||
2e8efe1b GW |
1640 | /* |
1641 | * Only process scans in sync pass 1. | |
1642 | */ | |
1643 | if (spa_sync_pass(dp->dp_spa) > 1) | |
1644 | return; | |
1645 | ||
1646 | /* | |
1647 | * If the spa is shutting down, then stop scanning. This will | |
1648 | * ensure that the scan does not dirty any new data during the | |
1649 | * shutdown phase. | |
1650 | */ | |
1651 | if (spa_shutting_down(spa)) | |
1652 | return; | |
1653 | ||
fbeddd60 MA |
1654 | /* |
1655 | * If the scan is inactive due to a stalled async destroy, try again. | |
1656 | */ | |
2e8efe1b | 1657 | if (!scn->scn_async_stalled && !dsl_scan_active(scn)) |
428870ff BB |
1658 | return; |
1659 | ||
1660 | scn->scn_visited_this_txg = 0; | |
0ea05c64 | 1661 | scn->scn_suspending = B_FALSE; |
428870ff BB |
1662 | scn->scn_sync_start_time = gethrtime(); |
1663 | spa->spa_scrub_active = B_TRUE; | |
1664 | ||
1665 | /* | |
0ea05c64 | 1666 | * First process the async destroys. If we suspend, don't do |
fbeddd60 MA |
1667 | * any scrubbing or resilvering. This ensures that there are no |
1668 | * async destroys while we are scanning, so the scan code doesn't | |
1669 | * have to worry about traversing it. It is also faster to free the | |
1670 | * blocks than to scrub them. | |
428870ff | 1671 | */ |
ba5ad9a4 GW |
1672 | if (zfs_free_bpobj_enabled && |
1673 | spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) { | |
9ae529ec | 1674 | scn->scn_is_bptree = B_FALSE; |
428870ff BB |
1675 | scn->scn_zio_root = zio_root(dp->dp_spa, NULL, |
1676 | NULL, ZIO_FLAG_MUSTSUCCEED); | |
1677 | err = bpobj_iterate(&dp->dp_free_bpobj, | |
9ae529ec | 1678 | dsl_scan_free_block_cb, scn, tx); |
428870ff | 1679 | VERIFY3U(0, ==, zio_wait(scn->scn_zio_root)); |
9ae529ec | 1680 | |
fbeddd60 MA |
1681 | if (err != 0 && err != ERESTART) |
1682 | zfs_panic_recover("error %u from bpobj_iterate()", err); | |
1683 | } | |
13fe0198 | 1684 | |
fbeddd60 MA |
1685 | if (err == 0 && spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) { |
1686 | ASSERT(scn->scn_async_destroying); | |
1687 | scn->scn_is_bptree = B_TRUE; | |
1688 | scn->scn_zio_root = zio_root(dp->dp_spa, NULL, | |
1689 | NULL, ZIO_FLAG_MUSTSUCCEED); | |
1690 | err = bptree_iterate(dp->dp_meta_objset, | |
1691 | dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb, scn, tx); | |
1692 | VERIFY0(zio_wait(scn->scn_zio_root)); | |
1693 | ||
1694 | if (err == EIO || err == ECKSUM) { | |
1695 | err = 0; | |
1696 | } else if (err != 0 && err != ERESTART) { | |
1697 | zfs_panic_recover("error %u from " | |
1698 | "traverse_dataset_destroyed()", err); | |
9ae529ec | 1699 | } |
fbeddd60 | 1700 | |
fbeddd60 MA |
1701 | if (bptree_is_empty(dp->dp_meta_objset, dp->dp_bptree_obj)) { |
1702 | /* finished; deactivate async destroy feature */ | |
1703 | spa_feature_decr(spa, SPA_FEATURE_ASYNC_DESTROY, tx); | |
1704 | ASSERT(!spa_feature_is_active(spa, | |
1705 | SPA_FEATURE_ASYNC_DESTROY)); | |
1706 | VERIFY0(zap_remove(dp->dp_meta_objset, | |
1707 | DMU_POOL_DIRECTORY_OBJECT, | |
1708 | DMU_POOL_BPTREE_OBJ, tx)); | |
1709 | VERIFY0(bptree_free(dp->dp_meta_objset, | |
1710 | dp->dp_bptree_obj, tx)); | |
1711 | dp->dp_bptree_obj = 0; | |
1712 | scn->scn_async_destroying = B_FALSE; | |
905edb40 | 1713 | scn->scn_async_stalled = B_FALSE; |
89b1cd65 | 1714 | } else { |
1715 | /* | |
905edb40 MA |
1716 | * If we didn't make progress, mark the async |
1717 | * destroy as stalled, so that we will not initiate | |
1718 | * a spa_sync() on its behalf. Note that we only | |
1719 | * check this if we are not finished, because if the | |
1720 | * bptree had no blocks for us to visit, we can | |
1721 | * finish without "making progress". | |
89b1cd65 | 1722 | */ |
1723 | scn->scn_async_stalled = | |
1724 | (scn->scn_visited_this_txg == 0); | |
428870ff | 1725 | } |
fbeddd60 MA |
1726 | } |
1727 | if (scn->scn_visited_this_txg) { | |
1728 | zfs_dbgmsg("freed %llu blocks in %llums from " | |
1729 | "free_bpobj/bptree txg %llu; err=%u", | |
1730 | (longlong_t)scn->scn_visited_this_txg, | |
1731 | (longlong_t) | |
1732 | NSEC2MSEC(gethrtime() - scn->scn_sync_start_time), | |
1733 | (longlong_t)tx->tx_txg, err); | |
1734 | scn->scn_visited_this_txg = 0; | |
1735 | ||
1736 | /* | |
1737 | * Write out changes to the DDT that may be required as a | |
1738 | * result of the blocks freed. This ensures that the DDT | |
1739 | * is clean when a scrub/resilver runs. | |
1740 | */ | |
1741 | ddt_sync(spa, tx->tx_txg); | |
1742 | } | |
1743 | if (err != 0) | |
1744 | return; | |
7c9abfa7 GM |
1745 | if (dp->dp_free_dir != NULL && !scn->scn_async_destroying && |
1746 | zfs_free_leak_on_eio && | |
d683ddbb JG |
1747 | (dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes != 0 || |
1748 | dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes != 0 || | |
1749 | dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes != 0)) { | |
fbeddd60 MA |
1750 | /* |
1751 | * We have finished background destroying, but there is still | |
1752 | * some space left in the dp_free_dir. Transfer this leaked | |
1753 | * space to the dp_leak_dir. | |
1754 | */ | |
1755 | if (dp->dp_leak_dir == NULL) { | |
1756 | rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); | |
1757 | (void) dsl_dir_create_sync(dp, dp->dp_root_dir, | |
1758 | LEAK_DIR_NAME, tx); | |
1759 | VERIFY0(dsl_pool_open_special_dir(dp, | |
1760 | LEAK_DIR_NAME, &dp->dp_leak_dir)); | |
1761 | rrw_exit(&dp->dp_config_rwlock, FTAG); | |
1762 | } | |
1763 | dsl_dir_diduse_space(dp->dp_leak_dir, DD_USED_HEAD, | |
d683ddbb JG |
1764 | dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes, |
1765 | dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes, | |
1766 | dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx); | |
fbeddd60 | 1767 | dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD, |
d683ddbb JG |
1768 | -dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes, |
1769 | -dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes, | |
1770 | -dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes, tx); | |
fbeddd60 | 1771 | } |
7c9abfa7 | 1772 | if (dp->dp_free_dir != NULL && !scn->scn_async_destroying) { |
9b67f605 | 1773 | /* finished; verify that space accounting went to zero */ |
d683ddbb JG |
1774 | ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_used_bytes); |
1775 | ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_compressed_bytes); | |
1776 | ASSERT0(dsl_dir_phys(dp->dp_free_dir)->dd_uncompressed_bytes); | |
428870ff BB |
1777 | } |
1778 | ||
1779 | if (scn->scn_phys.scn_state != DSS_SCANNING) | |
1780 | return; | |
1781 | ||
5d1f7fb6 | 1782 | if (scn->scn_done_txg == tx->tx_txg) { |
0ea05c64 | 1783 | ASSERT(!scn->scn_suspending); |
5d1f7fb6 GW |
1784 | /* finished with scan. */ |
1785 | zfs_dbgmsg("txg %llu scan complete", tx->tx_txg); | |
1786 | dsl_scan_done(scn, B_TRUE, tx); | |
1787 | ASSERT3U(spa->spa_scrub_inflight, ==, 0); | |
1788 | dsl_scan_sync_state(scn, tx); | |
1789 | return; | |
1790 | } | |
1791 | ||
0ea05c64 AP |
1792 | if (dsl_scan_is_paused_scrub(scn)) |
1793 | return; | |
1794 | ||
428870ff BB |
1795 | if (scn->scn_phys.scn_ddt_bookmark.ddb_class <= |
1796 | scn->scn_phys.scn_ddt_class_max) { | |
1797 | zfs_dbgmsg("doing scan sync txg %llu; " | |
1798 | "ddt bm=%llu/%llu/%llu/%llx", | |
1799 | (longlong_t)tx->tx_txg, | |
1800 | (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class, | |
1801 | (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type, | |
1802 | (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum, | |
1803 | (longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor); | |
1804 | ASSERT(scn->scn_phys.scn_bookmark.zb_objset == 0); | |
1805 | ASSERT(scn->scn_phys.scn_bookmark.zb_object == 0); | |
1806 | ASSERT(scn->scn_phys.scn_bookmark.zb_level == 0); | |
1807 | ASSERT(scn->scn_phys.scn_bookmark.zb_blkid == 0); | |
1808 | } else { | |
1809 | zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu", | |
1810 | (longlong_t)tx->tx_txg, | |
1811 | (longlong_t)scn->scn_phys.scn_bookmark.zb_objset, | |
1812 | (longlong_t)scn->scn_phys.scn_bookmark.zb_object, | |
1813 | (longlong_t)scn->scn_phys.scn_bookmark.zb_level, | |
1814 | (longlong_t)scn->scn_phys.scn_bookmark.zb_blkid); | |
1815 | } | |
1816 | ||
1817 | scn->scn_zio_root = zio_root(dp->dp_spa, NULL, | |
1818 | NULL, ZIO_FLAG_CANFAIL); | |
13fe0198 | 1819 | dsl_pool_config_enter(dp, FTAG); |
428870ff | 1820 | dsl_scan_visit(scn, tx); |
13fe0198 | 1821 | dsl_pool_config_exit(dp, FTAG); |
428870ff BB |
1822 | (void) zio_wait(scn->scn_zio_root); |
1823 | scn->scn_zio_root = NULL; | |
1824 | ||
1825 | zfs_dbgmsg("visited %llu blocks in %llums", | |
1826 | (longlong_t)scn->scn_visited_this_txg, | |
63fd3c6c | 1827 | (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time)); |
428870ff | 1828 | |
0ea05c64 | 1829 | if (!scn->scn_suspending) { |
5d1f7fb6 GW |
1830 | scn->scn_done_txg = tx->tx_txg + 1; |
1831 | zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu", | |
1832 | tx->tx_txg, scn->scn_done_txg); | |
428870ff BB |
1833 | } |
1834 | ||
1835 | if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) { | |
1836 | mutex_enter(&spa->spa_scrub_lock); | |
1837 | while (spa->spa_scrub_inflight > 0) { | |
1838 | cv_wait(&spa->spa_scrub_io_cv, | |
1839 | &spa->spa_scrub_lock); | |
1840 | } | |
1841 | mutex_exit(&spa->spa_scrub_lock); | |
1842 | } | |
1843 | ||
1844 | dsl_scan_sync_state(scn, tx); | |
1845 | } | |
1846 | ||
1847 | /* | |
1848 | * This will start a new scan, or restart an existing one. | |
1849 | */ | |
1850 | void | |
1851 | dsl_resilver_restart(dsl_pool_t *dp, uint64_t txg) | |
1852 | { | |
1853 | if (txg == 0) { | |
1854 | dmu_tx_t *tx; | |
1855 | tx = dmu_tx_create_dd(dp->dp_mos_dir); | |
1856 | VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT)); | |
1857 | ||
1858 | txg = dmu_tx_get_txg(tx); | |
1859 | dp->dp_scan->scn_restart_txg = txg; | |
1860 | dmu_tx_commit(tx); | |
1861 | } else { | |
1862 | dp->dp_scan->scn_restart_txg = txg; | |
1863 | } | |
1864 | zfs_dbgmsg("restarting resilver txg=%llu", txg); | |
1865 | } | |
1866 | ||
1867 | boolean_t | |
1868 | dsl_scan_resilvering(dsl_pool_t *dp) | |
1869 | { | |
1870 | return (dp->dp_scan->scn_phys.scn_state == DSS_SCANNING && | |
1871 | dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER); | |
1872 | } | |
1873 | ||
1874 | /* | |
1875 | * scrub consumers | |
1876 | */ | |
1877 | ||
1878 | static void | |
1879 | count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp) | |
1880 | { | |
1881 | int i; | |
1882 | ||
1883 | /* | |
1884 | * If we resume after a reboot, zab will be NULL; don't record | |
1885 | * incomplete stats in that case. | |
1886 | */ | |
1887 | if (zab == NULL) | |
1888 | return; | |
1889 | ||
1890 | for (i = 0; i < 4; i++) { | |
1891 | int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS; | |
1892 | int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL; | |
428870ff | 1893 | int equal; |
9ae529ec CS |
1894 | zfs_blkstat_t *zb; |
1895 | ||
1896 | if (t & DMU_OT_NEWTYPE) | |
1897 | t = DMU_OT_OTHER; | |
428870ff | 1898 | |
9ae529ec | 1899 | zb = &zab->zab_type[l][t]; |
428870ff BB |
1900 | zb->zb_count++; |
1901 | zb->zb_asize += BP_GET_ASIZE(bp); | |
1902 | zb->zb_lsize += BP_GET_LSIZE(bp); | |
1903 | zb->zb_psize += BP_GET_PSIZE(bp); | |
1904 | zb->zb_gangs += BP_COUNT_GANG(bp); | |
1905 | ||
1906 | switch (BP_GET_NDVAS(bp)) { | |
1907 | case 2: | |
1908 | if (DVA_GET_VDEV(&bp->blk_dva[0]) == | |
1909 | DVA_GET_VDEV(&bp->blk_dva[1])) | |
1910 | zb->zb_ditto_2_of_2_samevdev++; | |
1911 | break; | |
1912 | case 3: | |
1913 | equal = (DVA_GET_VDEV(&bp->blk_dva[0]) == | |
1914 | DVA_GET_VDEV(&bp->blk_dva[1])) + | |
1915 | (DVA_GET_VDEV(&bp->blk_dva[0]) == | |
1916 | DVA_GET_VDEV(&bp->blk_dva[2])) + | |
1917 | (DVA_GET_VDEV(&bp->blk_dva[1]) == | |
1918 | DVA_GET_VDEV(&bp->blk_dva[2])); | |
1919 | if (equal == 1) | |
1920 | zb->zb_ditto_2_of_3_samevdev++; | |
1921 | else if (equal == 3) | |
1922 | zb->zb_ditto_3_of_3_samevdev++; | |
1923 | break; | |
1924 | } | |
1925 | } | |
1926 | } | |
1927 | ||
1928 | static void | |
1929 | dsl_scan_scrub_done(zio_t *zio) | |
1930 | { | |
1931 | spa_t *spa = zio->io_spa; | |
1932 | ||
a6255b7f | 1933 | abd_free(zio->io_abd); |
428870ff BB |
1934 | |
1935 | mutex_enter(&spa->spa_scrub_lock); | |
1936 | spa->spa_scrub_inflight--; | |
1937 | cv_broadcast(&spa->spa_scrub_io_cv); | |
1938 | ||
1939 | if (zio->io_error && (zio->io_error != ECKSUM || | |
1940 | !(zio->io_flags & ZIO_FLAG_SPECULATIVE))) { | |
1941 | spa->spa_dsl_pool->dp_scan->scn_phys.scn_errors++; | |
1942 | } | |
1943 | mutex_exit(&spa->spa_scrub_lock); | |
1944 | } | |
1945 | ||
3d6da72d IH |
1946 | static boolean_t |
1947 | dsl_scan_need_resilver(spa_t *spa, const dva_t *dva, size_t psize, | |
1948 | uint64_t phys_birth) | |
1949 | { | |
1950 | vdev_t *vd; | |
1951 | ||
1952 | if (DVA_GET_GANG(dva)) { | |
1953 | /* | |
1954 | * Gang members may be spread across multiple | |
1955 | * vdevs, so the best estimate we have is the | |
1956 | * scrub range, which has already been checked. | |
1957 | * XXX -- it would be better to change our | |
1958 | * allocation policy to ensure that all | |
1959 | * gang members reside on the same vdev. | |
1960 | */ | |
1961 | return (B_TRUE); | |
1962 | } | |
1963 | ||
1964 | vd = vdev_lookup_top(spa, DVA_GET_VDEV(dva)); | |
1965 | ||
1966 | /* | |
1967 | * Check if the txg falls within the range which must be | |
1968 | * resilvered. DVAs outside this range can always be skipped. | |
1969 | */ | |
1970 | if (!vdev_dtl_contains(vd, DTL_PARTIAL, phys_birth, 1)) | |
1971 | return (B_FALSE); | |
1972 | ||
1973 | /* | |
1974 | * Check if the top-level vdev must resilver this offset. | |
1975 | * When the offset does not intersect with a dirty leaf DTL | |
1976 | * then it may be possible to skip the resilver IO. The psize | |
1977 | * is provided instead of asize to simplify the check for RAIDZ. | |
1978 | */ | |
1979 | if (!vdev_dtl_need_resilver(vd, DVA_GET_OFFSET(dva), psize)) | |
1980 | return (B_FALSE); | |
1981 | ||
1982 | return (B_TRUE); | |
1983 | } | |
1984 | ||
428870ff BB |
1985 | static int |
1986 | dsl_scan_scrub_cb(dsl_pool_t *dp, | |
5dbd68a3 | 1987 | const blkptr_t *bp, const zbookmark_phys_t *zb) |
428870ff BB |
1988 | { |
1989 | dsl_scan_t *scn = dp->dp_scan; | |
3d6da72d | 1990 | size_t psize = BP_GET_PSIZE(bp); |
428870ff BB |
1991 | spa_t *spa = dp->dp_spa; |
1992 | uint64_t phys_birth = BP_PHYSICAL_BIRTH(bp); | |
d6320ddb | 1993 | boolean_t needs_io = B_FALSE; |
572e2857 | 1994 | int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL; |
572e2857 | 1995 | int scan_delay = 0; |
d6320ddb | 1996 | int d; |
428870ff BB |
1997 | |
1998 | if (phys_birth <= scn->scn_phys.scn_min_txg || | |
1999 | phys_birth >= scn->scn_phys.scn_max_txg) | |
2000 | return (0); | |
2001 | ||
2002 | count_block(dp->dp_blkstats, bp); | |
2003 | ||
9b67f605 MA |
2004 | if (BP_IS_EMBEDDED(bp)) |
2005 | return (0); | |
2006 | ||
428870ff BB |
2007 | ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn)); |
2008 | if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) { | |
2009 | zio_flags |= ZIO_FLAG_SCRUB; | |
428870ff | 2010 | needs_io = B_TRUE; |
572e2857 | 2011 | scan_delay = zfs_scrub_delay; |
a117a6d6 GW |
2012 | } else { |
2013 | ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER); | |
428870ff | 2014 | zio_flags |= ZIO_FLAG_RESILVER; |
428870ff | 2015 | needs_io = B_FALSE; |
572e2857 | 2016 | scan_delay = zfs_resilver_delay; |
428870ff BB |
2017 | } |
2018 | ||
2019 | /* If it's an intent log block, failure is expected. */ | |
2020 | if (zb->zb_level == ZB_ZIL_LEVEL) | |
2021 | zio_flags |= ZIO_FLAG_SPECULATIVE; | |
2022 | ||
d6320ddb | 2023 | for (d = 0; d < BP_GET_NDVAS(bp); d++) { |
3d6da72d | 2024 | const dva_t *dva = &bp->blk_dva[d]; |
428870ff BB |
2025 | |
2026 | /* | |
2027 | * Keep track of how much data we've examined so that | |
2028 | * zpool(1M) status can make useful progress reports. | |
2029 | */ | |
3d6da72d IH |
2030 | scn->scn_phys.scn_examined += DVA_GET_ASIZE(dva); |
2031 | spa->spa_scan_pass_exam += DVA_GET_ASIZE(dva); | |
428870ff BB |
2032 | |
2033 | /* if it's a resilver, this may not be in the target range */ | |
3d6da72d IH |
2034 | if (!needs_io) |
2035 | needs_io = dsl_scan_need_resilver(spa, dva, psize, | |
2036 | phys_birth); | |
428870ff BB |
2037 | } |
2038 | ||
2039 | if (needs_io && !zfs_no_scrub_io) { | |
572e2857 BB |
2040 | vdev_t *rvd = spa->spa_root_vdev; |
2041 | uint64_t maxinflight = rvd->vdev_children * zfs_top_maxinflight; | |
428870ff BB |
2042 | |
2043 | mutex_enter(&spa->spa_scrub_lock); | |
572e2857 | 2044 | while (spa->spa_scrub_inflight >= maxinflight) |
428870ff BB |
2045 | cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock); |
2046 | spa->spa_scrub_inflight++; | |
2047 | mutex_exit(&spa->spa_scrub_lock); | |
2048 | ||
572e2857 BB |
2049 | /* |
2050 | * If we're seeing recent (zfs_scan_idle) "important" I/Os | |
2051 | * then throttle our workload to limit the impact of a scan. | |
2052 | */ | |
2053 | if (ddi_get_lbolt64() - spa->spa_last_io <= zfs_scan_idle) | |
2054 | delay(scan_delay); | |
2055 | ||
a6255b7f | 2056 | zio_nowait(zio_read(NULL, spa, bp, |
3d6da72d IH |
2057 | abd_alloc_for_io(psize, B_FALSE), |
2058 | psize, dsl_scan_scrub_done, NULL, | |
2059 | ZIO_PRIORITY_SCRUB, zio_flags, zb)); | |
428870ff BB |
2060 | } |
2061 | ||
2062 | /* do not relocate this block */ | |
2063 | return (0); | |
2064 | } | |
2065 | ||
0ea05c64 AP |
2066 | /* |
2067 | * Called by the ZFS_IOC_POOL_SCAN ioctl to start a scrub or resilver. | |
2068 | * Can also be called to resume a paused scrub. | |
2069 | */ | |
428870ff BB |
2070 | int |
2071 | dsl_scan(dsl_pool_t *dp, pool_scan_func_t func) | |
2072 | { | |
2073 | spa_t *spa = dp->dp_spa; | |
0ea05c64 | 2074 | dsl_scan_t *scn = dp->dp_scan; |
428870ff BB |
2075 | |
2076 | /* | |
2077 | * Purge all vdev caches and probe all devices. We do this here | |
2078 | * rather than in sync context because this requires a writer lock | |
2079 | * on the spa_config lock, which we can't do from sync context. The | |
2080 | * spa_scrub_reopen flag indicates that vdev_open() should not | |
2081 | * attempt to start another scrub. | |
2082 | */ | |
2083 | spa_vdev_state_enter(spa, SCL_NONE); | |
2084 | spa->spa_scrub_reopen = B_TRUE; | |
2085 | vdev_reopen(spa->spa_root_vdev); | |
2086 | spa->spa_scrub_reopen = B_FALSE; | |
2087 | (void) spa_vdev_state_exit(spa, NULL, 0); | |
2088 | ||
0ea05c64 AP |
2089 | if (func == POOL_SCAN_SCRUB && dsl_scan_is_paused_scrub(scn)) { |
2090 | /* got scrub start cmd, resume paused scrub */ | |
2091 | int err = dsl_scrub_set_pause_resume(scn->scn_dp, | |
2092 | POOL_SCRUB_NORMAL); | |
2093 | if (err == 0) | |
ecb2b7dc | 2094 | return (SET_ERROR(ECANCELED)); |
0ea05c64 AP |
2095 | |
2096 | return (SET_ERROR(err)); | |
2097 | } | |
2098 | ||
13fe0198 | 2099 | return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check, |
3d45fdd6 | 2100 | dsl_scan_setup_sync, &func, 0, ZFS_SPACE_CHECK_NONE)); |
428870ff | 2101 | } |
c409e464 | 2102 | |
784d15c1 NR |
2103 | static boolean_t |
2104 | dsl_scan_restarting(dsl_scan_t *scn, dmu_tx_t *tx) | |
2105 | { | |
2106 | return (scn->scn_restart_txg != 0 && | |
2107 | scn->scn_restart_txg <= tx->tx_txg); | |
2108 | } | |
2109 | ||
c409e464 BB |
2110 | #if defined(_KERNEL) && defined(HAVE_SPL) |
2111 | module_param(zfs_top_maxinflight, int, 0644); | |
2112 | MODULE_PARM_DESC(zfs_top_maxinflight, "Max I/Os per top-level"); | |
2113 | ||
2114 | module_param(zfs_resilver_delay, int, 0644); | |
2115 | MODULE_PARM_DESC(zfs_resilver_delay, "Number of ticks to delay resilver"); | |
2116 | ||
2117 | module_param(zfs_scrub_delay, int, 0644); | |
2118 | MODULE_PARM_DESC(zfs_scrub_delay, "Number of ticks to delay scrub"); | |
2119 | ||
2120 | module_param(zfs_scan_idle, int, 0644); | |
2121 | MODULE_PARM_DESC(zfs_scan_idle, "Idle window in clock ticks"); | |
2122 | ||
2123 | module_param(zfs_scan_min_time_ms, int, 0644); | |
2124 | MODULE_PARM_DESC(zfs_scan_min_time_ms, "Min millisecs to scrub per txg"); | |
2125 | ||
2126 | module_param(zfs_free_min_time_ms, int, 0644); | |
2127 | MODULE_PARM_DESC(zfs_free_min_time_ms, "Min millisecs to free per txg"); | |
2128 | ||
2129 | module_param(zfs_resilver_min_time_ms, int, 0644); | |
2130 | MODULE_PARM_DESC(zfs_resilver_min_time_ms, "Min millisecs to resilver per txg"); | |
2131 | ||
2132 | module_param(zfs_no_scrub_io, int, 0644); | |
2133 | MODULE_PARM_DESC(zfs_no_scrub_io, "Set to disable scrub I/O"); | |
2134 | ||
2135 | module_param(zfs_no_scrub_prefetch, int, 0644); | |
2136 | MODULE_PARM_DESC(zfs_no_scrub_prefetch, "Set to disable scrub prefetching"); | |
36283ca2 | 2137 | |
02730c33 | 2138 | /* CSTYLED */ |
36283ca2 MG |
2139 | module_param(zfs_free_max_blocks, ulong, 0644); |
2140 | MODULE_PARM_DESC(zfs_free_max_blocks, "Max number of blocks freed in one txg"); | |
ba5ad9a4 GW |
2141 | |
2142 | module_param(zfs_free_bpobj_enabled, int, 0644); | |
2143 | MODULE_PARM_DESC(zfs_free_bpobj_enabled, "Enable processing of the free_bpobj"); | |
c409e464 | 2144 | #endif |