xref: /titanic_41/usr/src/uts/common/fs/zfs/dsl_pool.c (revision b7d3956b92a285d8dac2c7f5f7e28d2ef5347ef8)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/dsl_pool.h>
29 #include <sys/dsl_dataset.h>
30 #include <sys/dsl_dir.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_tx.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/arc.h>
35 #include <sys/zap.h>
36 #include <sys/zio.h>
37 #include <sys/zfs_context.h>
38 #include <sys/fs/zfs.h>
39 #include <sys/zfs_znode.h>
40 #include <sys/spa_impl.h>
41 
42 int zfs_no_write_throttle = 0;
43 uint64_t zfs_write_limit_override = 0;
44 
45 
46 static int
47 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
48 {
49 	uint64_t obj;
50 	int err;
51 
52 	err = zap_lookup(dp->dp_meta_objset,
53 	    dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
54 	    name, sizeof (obj), 1, &obj);
55 	if (err)
56 		return (err);
57 
58 	return (dsl_dir_open_obj(dp, obj, name, dp, ddp));
59 }
60 
61 static dsl_pool_t *
62 dsl_pool_open_impl(spa_t *spa, uint64_t txg)
63 {
64 	dsl_pool_t *dp;
65 	blkptr_t *bp = spa_get_rootblkptr(spa);
66 	extern uint64_t zfs_write_limit_min;
67 
68 	dp = kmem_zalloc(sizeof (dsl_pool_t), KM_SLEEP);
69 	dp->dp_spa = spa;
70 	dp->dp_meta_rootbp = *bp;
71 	rw_init(&dp->dp_config_rwlock, NULL, RW_DEFAULT, NULL);
72 	dp->dp_write_limit = zfs_write_limit_min;
73 	txg_init(dp, txg);
74 
75 	txg_list_create(&dp->dp_dirty_datasets,
76 	    offsetof(dsl_dataset_t, ds_dirty_link));
77 	txg_list_create(&dp->dp_dirty_dirs,
78 	    offsetof(dsl_dir_t, dd_dirty_link));
79 	txg_list_create(&dp->dp_sync_tasks,
80 	    offsetof(dsl_sync_task_group_t, dstg_node));
81 	list_create(&dp->dp_synced_datasets, sizeof (dsl_dataset_t),
82 	    offsetof(dsl_dataset_t, ds_synced_link));
83 
84 	mutex_init(&dp->dp_lock, NULL, MUTEX_DEFAULT, NULL);
85 	mutex_init(&dp->dp_scrub_cancel_lock, NULL, MUTEX_DEFAULT, NULL);
86 
87 	return (dp);
88 }
89 
90 int
91 dsl_pool_open(spa_t *spa, uint64_t txg, dsl_pool_t **dpp)
92 {
93 	int err;
94 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
95 	dsl_dir_t *dd;
96 	dsl_dataset_t *ds;
97 	objset_impl_t *osi;
98 
99 	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
100 	err = dmu_objset_open_impl(spa, NULL, &dp->dp_meta_rootbp, &osi);
101 	if (err)
102 		goto out;
103 	dp->dp_meta_objset = &osi->os;
104 
105 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
106 	    DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
107 	    &dp->dp_root_dir_obj);
108 	if (err)
109 		goto out;
110 
111 	err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
112 	    NULL, dp, &dp->dp_root_dir);
113 	if (err)
114 		goto out;
115 
116 	err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
117 	if (err)
118 		goto out;
119 
120 	if (spa_version(spa) >= SPA_VERSION_ORIGIN) {
121 		err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
122 		if (err)
123 			goto out;
124 		err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
125 		    FTAG, &ds);
126 		if (err)
127 			goto out;
128 		err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
129 		    dp, &dp->dp_origin_snap);
130 		if (err)
131 			goto out;
132 		dsl_dataset_rele(ds, FTAG);
133 		dsl_dir_close(dd, dp);
134 	}
135 
136 	/* get scrub status */
137 	err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
138 	    DMU_POOL_SCRUB_FUNC, sizeof (uint32_t), 1,
139 	    &dp->dp_scrub_func);
140 	if (err == 0) {
141 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
142 		    DMU_POOL_SCRUB_QUEUE, sizeof (uint64_t), 1,
143 		    &dp->dp_scrub_queue_obj);
144 		if (err)
145 			goto out;
146 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
147 		    DMU_POOL_SCRUB_MIN_TXG, sizeof (uint64_t), 1,
148 		    &dp->dp_scrub_min_txg);
149 		if (err)
150 			goto out;
151 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
152 		    DMU_POOL_SCRUB_MAX_TXG, sizeof (uint64_t), 1,
153 		    &dp->dp_scrub_max_txg);
154 		if (err)
155 			goto out;
156 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
157 		    DMU_POOL_SCRUB_BOOKMARK, sizeof (uint64_t), 4,
158 		    &dp->dp_scrub_bookmark);
159 		if (err)
160 			goto out;
161 		err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
162 		    DMU_POOL_SCRUB_ERRORS, sizeof (uint64_t), 1,
163 		    &spa->spa_scrub_errors);
164 		if (err)
165 			goto out;
166 		if (spa_version(spa) < SPA_VERSION_DSL_SCRUB) {
167 			/*
168 			 * A new-type scrub was in progress on an old
169 			 * pool.  Restart from the beginning, since the
170 			 * old software may have changed the pool in the
171 			 * meantime.
172 			 */
173 			dsl_pool_scrub_restart(dp);
174 		}
175 	} else {
176 		/*
177 		 * It's OK if there is no scrub in progress (and if
178 		 * there was an I/O error, ignore it).
179 		 */
180 		err = 0;
181 	}
182 
183 out:
184 	rw_exit(&dp->dp_config_rwlock);
185 	if (err)
186 		dsl_pool_close(dp);
187 	else
188 		*dpp = dp;
189 
190 	return (err);
191 }
192 
193 void
194 dsl_pool_close(dsl_pool_t *dp)
195 {
196 	/* drop our references from dsl_pool_open() */
197 
198 	/*
199 	 * Since we held the origin_snap from "syncing" context (which
200 	 * includes pool-opening context), it actually only got a "ref"
201 	 * and not a hold, so just drop that here.
202 	 */
203 	if (dp->dp_origin_snap)
204 		dsl_dataset_drop_ref(dp->dp_origin_snap, dp);
205 	if (dp->dp_mos_dir)
206 		dsl_dir_close(dp->dp_mos_dir, dp);
207 	if (dp->dp_root_dir)
208 		dsl_dir_close(dp->dp_root_dir, dp);
209 
210 	/* undo the dmu_objset_open_impl(mos) from dsl_pool_open() */
211 	if (dp->dp_meta_objset)
212 		dmu_objset_evict(NULL, dp->dp_meta_objset->os);
213 
214 	txg_list_destroy(&dp->dp_dirty_datasets);
215 	txg_list_destroy(&dp->dp_dirty_dirs);
216 	list_destroy(&dp->dp_synced_datasets);
217 
218 	arc_flush(dp->dp_spa);
219 	txg_fini(dp);
220 	rw_destroy(&dp->dp_config_rwlock);
221 	mutex_destroy(&dp->dp_lock);
222 	mutex_destroy(&dp->dp_scrub_cancel_lock);
223 	kmem_free(dp, sizeof (dsl_pool_t));
224 }
225 
226 dsl_pool_t *
227 dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
228 {
229 	int err;
230 	dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
231 	dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
232 	objset_impl_t *osip;
233 	dsl_dataset_t *ds;
234 	uint64_t dsobj;
235 
236 	/* create and open the MOS (meta-objset) */
237 	dp->dp_meta_objset = &dmu_objset_create_impl(spa,
238 	    NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx)->os;
239 
240 	/* create the pool directory */
241 	err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
242 	    DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
243 	ASSERT3U(err, ==, 0);
244 
245 	/* create and open the root dir */
246 	dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
247 	VERIFY(0 == dsl_dir_open_obj(dp, dp->dp_root_dir_obj,
248 	    NULL, dp, &dp->dp_root_dir));
249 
250 	/* create and open the meta-objset dir */
251 	(void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
252 	VERIFY(0 == dsl_pool_open_special_dir(dp,
253 	    MOS_DIR_NAME, &dp->dp_mos_dir));
254 
255 	if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
256 		dsl_pool_create_origin(dp, tx);
257 
258 	/* create the root dataset */
259 	dsobj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
260 
261 	/* create the root objset */
262 	VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
263 	osip = dmu_objset_create_impl(dp->dp_spa, ds,
264 	    dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
265 #ifdef _KERNEL
266 	zfs_create_fs(&osip->os, kcred, zplprops, tx);
267 #endif
268 	dsl_dataset_rele(ds, FTAG);
269 
270 	dmu_tx_commit(tx);
271 
272 	return (dp);
273 }
274 
275 void
276 dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
277 {
278 	zio_t *zio;
279 	dmu_tx_t *tx;
280 	dsl_dir_t *dd;
281 	dsl_dataset_t *ds;
282 	dsl_sync_task_group_t *dstg;
283 	objset_impl_t *mosi = dp->dp_meta_objset->os;
284 	int err;
285 
286 	tx = dmu_tx_create_assigned(dp, txg);
287 
288 	zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
289 	while (ds = txg_list_remove(&dp->dp_dirty_datasets, txg)) {
290 		if (!list_link_active(&ds->ds_synced_link))
291 			list_insert_tail(&dp->dp_synced_datasets, ds);
292 		else
293 			dmu_buf_rele(ds->ds_dbuf, ds);
294 		dsl_dataset_sync(ds, zio, tx);
295 	}
296 	err = zio_wait(zio);
297 	ASSERT(err == 0);
298 
299 	while (dstg = txg_list_remove(&dp->dp_sync_tasks, txg))
300 		dsl_sync_task_group_sync(dstg, tx);
301 	while (dd = txg_list_remove(&dp->dp_dirty_dirs, txg))
302 		dsl_dir_sync(dd, tx);
303 
304 	if (spa_sync_pass(dp->dp_spa) == 1)
305 		dsl_pool_scrub_sync(dp, tx);
306 
307 	if (list_head(&mosi->os_dirty_dnodes[txg & TXG_MASK]) != NULL ||
308 	    list_head(&mosi->os_free_dnodes[txg & TXG_MASK]) != NULL) {
309 		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
310 		dmu_objset_sync(mosi, zio, tx);
311 		err = zio_wait(zio);
312 		ASSERT(err == 0);
313 		dprintf_bp(&dp->dp_meta_rootbp, "meta objset rootbp is %s", "");
314 		spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
315 	}
316 
317 	dmu_tx_commit(tx);
318 }
319 
320 void
321 dsl_pool_zil_clean(dsl_pool_t *dp)
322 {
323 	dsl_dataset_t *ds;
324 
325 	while (ds = list_head(&dp->dp_synced_datasets)) {
326 		list_remove(&dp->dp_synced_datasets, ds);
327 		ASSERT(ds->ds_user_ptr != NULL);
328 		zil_clean(((objset_impl_t *)ds->ds_user_ptr)->os_zil);
329 		dmu_buf_rele(ds->ds_dbuf, ds);
330 	}
331 }
332 
333 /*
334  * TRUE if the current thread is the tx_sync_thread or if we
335  * are being called from SPA context during pool initialization.
336  */
337 int
338 dsl_pool_sync_context(dsl_pool_t *dp)
339 {
340 	return (curthread == dp->dp_tx.tx_sync_thread ||
341 	    spa_get_dsl(dp->dp_spa) == NULL);
342 }
343 
344 uint64_t
345 dsl_pool_adjustedsize(dsl_pool_t *dp, boolean_t netfree)
346 {
347 	uint64_t space, resv;
348 
349 	/*
350 	 * Reserve about 1.6% (1/64), or at least 32MB, for allocation
351 	 * efficiency.
352 	 * XXX The intent log is not accounted for, so it must fit
353 	 * within this slop.
354 	 *
355 	 * If we're trying to assess whether it's OK to do a free,
356 	 * cut the reservation in half to allow forward progress
357 	 * (e.g. make it possible to rm(1) files from a full pool).
358 	 */
359 	space = spa_get_dspace(dp->dp_spa);
360 	resv = MAX(space >> 6, SPA_MINDEVSIZE >> 1);
361 	if (netfree)
362 		resv >>= 1;
363 
364 	return (space - resv);
365 }
366 
367 int
368 dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
369 {
370 	uint64_t reserved = 0;
371 	uint64_t write_limit = (zfs_write_limit_override ?
372 	    zfs_write_limit_override : dp->dp_write_limit);
373 
374 	if (zfs_no_write_throttle) {
375 		atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK],
376 		    space);
377 		return (0);
378 	}
379 
380 	/*
381 	 * Check to see if we have exceeded the maximum allowed IO for
382 	 * this transaction group.  We can do this without locks since
383 	 * a little slop here is ok.  Note that we do the reserved check
384 	 * with only half the requested reserve: this is because the
385 	 * reserve requests are worst-case, and we really don't want to
386 	 * throttle based off of worst-case estimates.
387 	 */
388 	if (write_limit > 0) {
389 		reserved = dp->dp_space_towrite[tx->tx_txg & TXG_MASK]
390 		    + dp->dp_tempreserved[tx->tx_txg & TXG_MASK] / 2;
391 
392 		if (reserved && reserved > write_limit)
393 			return (ERESTART);
394 	}
395 
396 	atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], space);
397 
398 	/*
399 	 * If this transaction group is over 7/8ths capacity, delay
400 	 * the caller 1 clock tick.  This will slow down the "fill"
401 	 * rate until the sync process can catch up with us.
402 	 */
403 	if (reserved && reserved > (write_limit - (write_limit >> 3)))
404 		txg_delay(dp, tx->tx_txg, 1);
405 
406 	return (0);
407 }
408 
409 void
410 dsl_pool_tempreserve_clear(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
411 {
412 	ASSERT(dp->dp_tempreserved[tx->tx_txg & TXG_MASK] >= space);
413 	atomic_add_64(&dp->dp_tempreserved[tx->tx_txg & TXG_MASK], -space);
414 }
415 
416 void
417 dsl_pool_memory_pressure(dsl_pool_t *dp)
418 {
419 	extern uint64_t zfs_write_limit_min;
420 	uint64_t space_inuse = 0;
421 	int i;
422 
423 	if (dp->dp_write_limit == zfs_write_limit_min)
424 		return;
425 
426 	for (i = 0; i < TXG_SIZE; i++) {
427 		space_inuse += dp->dp_space_towrite[i];
428 		space_inuse += dp->dp_tempreserved[i];
429 	}
430 	dp->dp_write_limit = MAX(zfs_write_limit_min,
431 	    MIN(dp->dp_write_limit, space_inuse / 4));
432 }
433 
434 void
435 dsl_pool_willuse_space(dsl_pool_t *dp, int64_t space, dmu_tx_t *tx)
436 {
437 	if (space > 0) {
438 		mutex_enter(&dp->dp_lock);
439 		dp->dp_space_towrite[tx->tx_txg & TXG_MASK] += space;
440 		mutex_exit(&dp->dp_lock);
441 	}
442 }
443 
444 /* ARGSUSED */
445 static int
446 upgrade_clones_cb(spa_t *spa, uint64_t dsobj, const char *dsname, void *arg)
447 {
448 	dmu_tx_t *tx = arg;
449 	dsl_dataset_t *ds, *prev = NULL;
450 	int err;
451 	dsl_pool_t *dp = spa_get_dsl(spa);
452 
453 	err = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
454 	if (err)
455 		return (err);
456 
457 	while (ds->ds_phys->ds_prev_snap_obj != 0) {
458 		err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
459 		    FTAG, &prev);
460 		if (err) {
461 			dsl_dataset_rele(ds, FTAG);
462 			return (err);
463 		}
464 
465 		if (prev->ds_phys->ds_next_snap_obj != ds->ds_object)
466 			break;
467 		dsl_dataset_rele(ds, FTAG);
468 		ds = prev;
469 		prev = NULL;
470 	}
471 
472 	if (prev == NULL) {
473 		prev = dp->dp_origin_snap;
474 
475 		/*
476 		 * The $ORIGIN can't have any data, or the accounting
477 		 * will be wrong.
478 		 */
479 		ASSERT(prev->ds_phys->ds_bp.blk_birth == 0);
480 
481 		/* The origin doesn't get attached to itself */
482 		if (ds->ds_object == prev->ds_object) {
483 			dsl_dataset_rele(ds, FTAG);
484 			return (0);
485 		}
486 
487 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
488 		ds->ds_phys->ds_prev_snap_obj = prev->ds_object;
489 		ds->ds_phys->ds_prev_snap_txg = prev->ds_phys->ds_creation_txg;
490 
491 		dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
492 		ds->ds_dir->dd_phys->dd_origin_obj = prev->ds_object;
493 
494 		dmu_buf_will_dirty(prev->ds_dbuf, tx);
495 		prev->ds_phys->ds_num_children++;
496 
497 		if (ds->ds_phys->ds_next_snap_obj == 0) {
498 			ASSERT(ds->ds_prev == NULL);
499 			VERIFY(0 == dsl_dataset_hold_obj(dp,
500 			    ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
501 		}
502 	}
503 
504 	ASSERT(ds->ds_dir->dd_phys->dd_origin_obj == prev->ds_object);
505 	ASSERT(ds->ds_phys->ds_prev_snap_obj == prev->ds_object);
506 
507 	if (prev->ds_phys->ds_next_clones_obj == 0) {
508 		prev->ds_phys->ds_next_clones_obj =
509 		    zap_create(dp->dp_meta_objset,
510 		    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
511 	}
512 	VERIFY(0 == zap_add_int(dp->dp_meta_objset,
513 	    prev->ds_phys->ds_next_clones_obj, ds->ds_object, tx));
514 
515 	dsl_dataset_rele(ds, FTAG);
516 	if (prev != dp->dp_origin_snap)
517 		dsl_dataset_rele(prev, FTAG);
518 	return (0);
519 }
520 
521 void
522 dsl_pool_upgrade_clones(dsl_pool_t *dp, dmu_tx_t *tx)
523 {
524 	ASSERT(dmu_tx_is_syncing(tx));
525 	ASSERT(dp->dp_origin_snap != NULL);
526 
527 	(void) dmu_objset_find_spa(dp->dp_spa, NULL, upgrade_clones_cb,
528 	    tx, DS_FIND_CHILDREN);
529 }
530 
531 void
532 dsl_pool_create_origin(dsl_pool_t *dp, dmu_tx_t *tx)
533 {
534 	uint64_t dsobj;
535 	dsl_dataset_t *ds;
536 
537 	ASSERT(dmu_tx_is_syncing(tx));
538 	ASSERT(dp->dp_origin_snap == NULL);
539 
540 	/* create the origin dir, ds, & snap-ds */
541 	rw_enter(&dp->dp_config_rwlock, RW_WRITER);
542 	dsobj = dsl_dataset_create_sync(dp->dp_root_dir, ORIGIN_DIR_NAME,
543 	    NULL, 0, kcred, tx);
544 	VERIFY(0 == dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
545 	dsl_dataset_snapshot_sync(ds, ORIGIN_DIR_NAME, kcred, tx);
546 	VERIFY(0 == dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
547 	    dp, &dp->dp_origin_snap));
548 	dsl_dataset_rele(ds, FTAG);
549 	rw_exit(&dp->dp_config_rwlock);
550 }
551