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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 */
27
28 #include <sys/zfs_context.h>
29 #include <sys/dmu.h>
30 #include <sys/dmu_tx.h>
31 #include <sys/space_map.h>
32 #include <sys/metaslab_impl.h>
33 #include <sys/vdev_impl.h>
34 #include <sys/zio.h>
35 #include <sys/spa_impl.h>
36 #include <sys/zfeature.h>
37
38 #define GANG_ALLOCATION(flags) \
39 ((flags) & (METASLAB_GANG_CHILD | METASLAB_GANG_HEADER))
40
41 uint64_t metaslab_aliquot = 512ULL << 10;
42 uint64_t metaslab_gang_bang = SPA_MAXBLOCKSIZE + 1; /* force gang blocks */
43
44 /*
45 * The in-core space map representation is more compact than its on-disk form.
46 * The zfs_condense_pct determines how much more compact the in-core
47 * space map representation must be before we compact it on-disk.
48 * Values should be greater than or equal to 100.
49 */
50 int zfs_condense_pct = 200;
51
52 /*
53 * Condensing a metaslab is not guaranteed to actually reduce the amount of
54 * space used on disk. In particular, a space map uses data in increments of
55 * MAX(1 << ashift, space_map_blksize), so a metaslab might use the
56 * same number of blocks after condensing. Since the goal of condensing is to
57 * reduce the number of IOPs required to read the space map, we only want to
58 * condense when we can be sure we will reduce the number of blocks used by the
59 * space map. Unfortunately, we cannot precisely compute whether or not this is
60 * the case in metaslab_should_condense since we are holding ms_lock. Instead,
61 * we apply the following heuristic: do not condense a spacemap unless the
62 * uncondensed size consumes greater than zfs_metaslab_condense_block_threshold
63 * blocks.
64 */
65 int zfs_metaslab_condense_block_threshold = 4;
66
67 /*
68 * The zfs_mg_noalloc_threshold defines which metaslab groups should
69 * be eligible for allocation. The value is defined as a percentage of
70 * free space. Metaslab groups that have more free space than
71 * zfs_mg_noalloc_threshold are always eligible for allocations. Once
72 * a metaslab group's free space is less than or equal to the
73 * zfs_mg_noalloc_threshold the allocator will avoid allocating to that
74 * group unless all groups in the pool have reached zfs_mg_noalloc_threshold.
75 * Once all groups in the pool reach zfs_mg_noalloc_threshold then all
76 * groups are allowed to accept allocations. Gang blocks are always
77 * eligible to allocate on any metaslab group. The default value of 0 means
78 * no metaslab group will be excluded based on this criterion.
79 */
80 int zfs_mg_noalloc_threshold = 0;
81
82 /*
83 * Metaslab groups are considered eligible for allocations if their
84 * fragmenation metric (measured as a percentage) is less than or equal to
85 * zfs_mg_fragmentation_threshold. If a metaslab group exceeds this threshold
86 * then it will be skipped unless all metaslab groups within the metaslab
87 * class have also crossed this threshold.
88 */
89 int zfs_mg_fragmentation_threshold = 85;
90
91 /*
92 * Allow metaslabs to keep their active state as long as their fragmentation
93 * percentage is less than or equal to zfs_metaslab_fragmentation_threshold. An
94 * active metaslab that exceeds this threshold will no longer keep its active
95 * status allowing better metaslabs to be selected.
96 */
97 int zfs_metaslab_fragmentation_threshold = 70;
98
99 /*
100 * When set will load all metaslabs when pool is first opened.
101 */
102 int metaslab_debug_load = 0;
103
104 /*
105 * When set will prevent metaslabs from being unloaded.
106 */
107 int metaslab_debug_unload = 0;
108
109 /*
110 * Minimum size which forces the dynamic allocator to change
111 * it's allocation strategy. Once the space map cannot satisfy
112 * an allocation of this size then it switches to using more
113 * aggressive strategy (i.e search by size rather than offset).
114 */
115 uint64_t metaslab_df_alloc_threshold = SPA_OLD_MAXBLOCKSIZE;
116
117 /*
118 * The minimum free space, in percent, which must be available
119 * in a space map to continue allocations in a first-fit fashion.
120 * Once the space map's free space drops below this level we dynamically
121 * switch to using best-fit allocations.
122 */
123 int metaslab_df_free_pct = 4;
124
125 /*
126 * A metaslab is considered "free" if it contains a contiguous
127 * segment which is greater than metaslab_min_alloc_size.
128 */
129 uint64_t metaslab_min_alloc_size = DMU_MAX_ACCESS;
130
131 /*
132 * Percentage of all cpus that can be used by the metaslab taskq.
133 */
134 int metaslab_load_pct = 50;
135
136 /*
137 * Determines how many txgs a metaslab may remain loaded without having any
138 * allocations from it. As long as a metaslab continues to be used we will
139 * keep it loaded.
140 */
141 int metaslab_unload_delay = TXG_SIZE * 2;
142
143 /*
144 * Max number of metaslabs per group to preload.
145 */
146 int metaslab_preload_limit = SPA_DVAS_PER_BP;
147
148 /*
149 * Enable/disable preloading of metaslab.
150 */
151 boolean_t metaslab_preload_enabled = B_TRUE;
152
153 /*
154 * Enable/disable fragmentation weighting on metaslabs.
155 */
156 boolean_t metaslab_fragmentation_factor_enabled = B_TRUE;
157
158 /*
159 * Enable/disable lba weighting (i.e. outer tracks are given preference).
160 */
161 boolean_t metaslab_lba_weighting_enabled = B_TRUE;
162
163 /*
164 * Enable/disable metaslab group biasing.
165 */
166 boolean_t metaslab_bias_enabled = B_TRUE;
167
168 /*
169 * Enable/disable segment-based metaslab selection.
170 */
171 boolean_t zfs_metaslab_segment_weight_enabled = B_TRUE;
172
173 /*
174 * When using segment-based metaslab selection, we will continue
175 * allocating from the active metaslab until we have exhausted
176 * zfs_metaslab_switch_threshold of its buckets.
177 */
178 int zfs_metaslab_switch_threshold = 2;
179
180 /*
181 * Internal switch to enable/disable the metaslab allocation tracing
182 * facility.
183 */
184 boolean_t metaslab_trace_enabled = B_TRUE;
185
186 /*
187 * Maximum entries that the metaslab allocation tracing facility will keep
188 * in a given list when running in non-debug mode. We limit the number
189 * of entries in non-debug mode to prevent us from using up too much memory.
190 * The limit should be sufficiently large that we don't expect any allocation
191 * to every exceed this value. In debug mode, the system will panic if this
192 * limit is ever reached allowing for further investigation.
193 */
194 uint64_t metaslab_trace_max_entries = 5000;
195
196 static uint64_t metaslab_weight(metaslab_t *);
197 static void metaslab_set_fragmentation(metaslab_t *);
198
199 kmem_cache_t *metaslab_alloc_trace_cache;
200
201 /*
202 * ==========================================================================
203 * Metaslab classes
204 * ==========================================================================
205 */
206 metaslab_class_t *
207 metaslab_class_create(spa_t *spa, metaslab_ops_t *ops)
208 {
209 metaslab_class_t *mc;
210
211 mc = kmem_zalloc(sizeof (metaslab_class_t), KM_SLEEP);
212
213 mc->mc_spa = spa;
214 mc->mc_rotor = NULL;
215 mc->mc_ops = ops;
216 mutex_init(&mc->mc_lock, NULL, MUTEX_DEFAULT, NULL);
217 refcount_create_tracked(&mc->mc_alloc_slots);
218
219 return (mc);
220 }
221
222 void
223 metaslab_class_destroy(metaslab_class_t *mc)
224 {
225 ASSERT(mc->mc_rotor == NULL);
226 ASSERT(mc->mc_alloc == 0);
227 ASSERT(mc->mc_deferred == 0);
228 ASSERT(mc->mc_space == 0);
229 ASSERT(mc->mc_dspace == 0);
230
231 refcount_destroy(&mc->mc_alloc_slots);
232 mutex_destroy(&mc->mc_lock);
233 kmem_free(mc, sizeof (metaslab_class_t));
234 }
235
236 int
237 metaslab_class_validate(metaslab_class_t *mc)
238 {
239 metaslab_group_t *mg;
240 vdev_t *vd;
241
242 /*
243 * Must hold one of the spa_config locks.
244 */
245 ASSERT(spa_config_held(mc->mc_spa, SCL_ALL, RW_READER) ||
246 spa_config_held(mc->mc_spa, SCL_ALL, RW_WRITER));
247
248 if ((mg = mc->mc_rotor) == NULL)
249 return (0);
250
251 do {
252 vd = mg->mg_vd;
253 ASSERT(vd->vdev_mg != NULL);
254 ASSERT3P(vd->vdev_top, ==, vd);
255 ASSERT3P(mg->mg_class, ==, mc);
256 ASSERT3P(vd->vdev_ops, !=, &vdev_hole_ops);
257 } while ((mg = mg->mg_next) != mc->mc_rotor);
258
259 return (0);
260 }
261
262 void
263 metaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta,
264 int64_t defer_delta, int64_t space_delta, int64_t dspace_delta)
265 {
266 atomic_add_64(&mc->mc_alloc, alloc_delta);
267 atomic_add_64(&mc->mc_deferred, defer_delta);
268 atomic_add_64(&mc->mc_space, space_delta);
269 atomic_add_64(&mc->mc_dspace, dspace_delta);
270 }
271
272 uint64_t
273 metaslab_class_get_alloc(metaslab_class_t *mc)
274 {
275 return (mc->mc_alloc);
276 }
277
278 uint64_t
279 metaslab_class_get_deferred(metaslab_class_t *mc)
280 {
281 return (mc->mc_deferred);
282 }
283
284 uint64_t
285 metaslab_class_get_space(metaslab_class_t *mc)
286 {
287 return (mc->mc_space);
288 }
289
290 uint64_t
291 metaslab_class_get_dspace(metaslab_class_t *mc)
292 {
293 return (spa_deflate(mc->mc_spa) ? mc->mc_dspace : mc->mc_space);
294 }
295
296 void
297 metaslab_class_histogram_verify(metaslab_class_t *mc)
298 {
299 vdev_t *rvd = mc->mc_spa->spa_root_vdev;
300 uint64_t *mc_hist;
301 int i;
302
303 if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0)
304 return;
305
306 mc_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE,
307 KM_SLEEP);
308
309 for (int c = 0; c < rvd->vdev_children; c++) {
310 vdev_t *tvd = rvd->vdev_child[c];
311 metaslab_group_t *mg = tvd->vdev_mg;
312
313 /*
314 * Skip any holes, uninitialized top-levels, or
315 * vdevs that are not in this metalab class.
316 */
317 if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 ||
318 mg->mg_class != mc) {
319 continue;
320 }
321
322 for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
323 mc_hist[i] += mg->mg_histogram[i];
324 }
325
326 for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
327 VERIFY3U(mc_hist[i], ==, mc->mc_histogram[i]);
328
329 kmem_free(mc_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE);
330 }
331
332 /*
333 * Calculate the metaslab class's fragmentation metric. The metric
334 * is weighted based on the space contribution of each metaslab group.
335 * The return value will be a number between 0 and 100 (inclusive), or
336 * ZFS_FRAG_INVALID if the metric has not been set. See comment above the
337 * zfs_frag_table for more information about the metric.
338 */
339 uint64_t
340 metaslab_class_fragmentation(metaslab_class_t *mc)
341 {
342 vdev_t *rvd = mc->mc_spa->spa_root_vdev;
343 uint64_t fragmentation = 0;
344
345 spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER);
346
347 for (int c = 0; c < rvd->vdev_children; c++) {
348 vdev_t *tvd = rvd->vdev_child[c];
349 metaslab_group_t *mg = tvd->vdev_mg;
350
351 /*
352 * Skip any holes, uninitialized top-levels, or
353 * vdevs that are not in this metalab class.
354 */
355 if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 ||
356 mg->mg_class != mc) {
357 continue;
358 }
359
360 /*
361 * If a metaslab group does not contain a fragmentation
362 * metric then just bail out.
363 */
364 if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
365 spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
366 return (ZFS_FRAG_INVALID);
367 }
368
369 /*
370 * Determine how much this metaslab_group is contributing
371 * to the overall pool fragmentation metric.
372 */
373 fragmentation += mg->mg_fragmentation *
374 metaslab_group_get_space(mg);
375 }
376 fragmentation /= metaslab_class_get_space(mc);
377
378 ASSERT3U(fragmentation, <=, 100);
379 spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
380 return (fragmentation);
381 }
382
383 /*
384 * Calculate the amount of expandable space that is available in
385 * this metaslab class. If a device is expanded then its expandable
386 * space will be the amount of allocatable space that is currently not
387 * part of this metaslab class.
388 */
389 uint64_t
390 metaslab_class_expandable_space(metaslab_class_t *mc)
391 {
392 vdev_t *rvd = mc->mc_spa->spa_root_vdev;
393 uint64_t space = 0;
394
395 spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER);
396 for (int c = 0; c < rvd->vdev_children; c++) {
397 vdev_t *tvd = rvd->vdev_child[c];
398 metaslab_group_t *mg = tvd->vdev_mg;
399
400 if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 ||
401 mg->mg_class != mc) {
402 continue;
403 }
404
405 /*
406 * Calculate if we have enough space to add additional
407 * metaslabs. We report the expandable space in terms
408 * of the metaslab size since that's the unit of expansion.
409 */
410 space += P2ALIGN(tvd->vdev_max_asize - tvd->vdev_asize,
411 1ULL << tvd->vdev_ms_shift);
412 }
413 spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
414 return (space);
415 }
416
417 static int
418 metaslab_compare(const void *x1, const void *x2)
419 {
420 const metaslab_t *m1 = x1;
421 const metaslab_t *m2 = x2;
422
423 if (m1->ms_weight < m2->ms_weight)
424 return (1);
425 if (m1->ms_weight > m2->ms_weight)
426 return (-1);
427
428 /*
429 * If the weights are identical, use the offset to force uniqueness.
430 */
431 if (m1->ms_start < m2->ms_start)
432 return (-1);
433 if (m1->ms_start > m2->ms_start)
434 return (1);
435
436 ASSERT3P(m1, ==, m2);
437
438 return (0);
439 }
440
441 /*
442 * Verify that the space accounting on disk matches the in-core range_trees.
443 */
444 void
445 metaslab_verify_space(metaslab_t *msp, uint64_t txg)
446 {
447 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
448 uint64_t allocated = 0;
449 uint64_t sm_free_space, msp_free_space;
450
451 ASSERT(MUTEX_HELD(&msp->ms_lock));
452
453 if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0)
454 return;
455
456 /*
457 * We can only verify the metaslab space when we're called
458 * from syncing context with a loaded metaslab that has an allocated
459 * space map. Calling this in non-syncing context does not
460 * provide a consistent view of the metaslab since we're performing
461 * allocations in the future.
462 */
463 if (txg != spa_syncing_txg(spa) || msp->ms_sm == NULL ||
464 !msp->ms_loaded)
465 return;
466
467 sm_free_space = msp->ms_size - space_map_allocated(msp->ms_sm) -
468 space_map_alloc_delta(msp->ms_sm);
469
470 /*
471 * Account for future allocations since we would have already
472 * deducted that space from the ms_freetree.
473 */
474 for (int t = 0; t < TXG_CONCURRENT_STATES; t++) {
475 allocated +=
476 range_tree_space(msp->ms_alloctree[(txg + t) & TXG_MASK]);
477 }
478
479 msp_free_space = range_tree_space(msp->ms_tree) + allocated +
480 msp->ms_deferspace + range_tree_space(msp->ms_freedtree);
481
482 VERIFY3U(sm_free_space, ==, msp_free_space);
483 }
484
485 /*
486 * ==========================================================================
487 * Metaslab groups
488 * ==========================================================================
489 */
490 /*
491 * Update the allocatable flag and the metaslab group's capacity.
492 * The allocatable flag is set to true if the capacity is below
493 * the zfs_mg_noalloc_threshold or has a fragmentation value that is
494 * greater than zfs_mg_fragmentation_threshold. If a metaslab group
495 * transitions from allocatable to non-allocatable or vice versa then the
496 * metaslab group's class is updated to reflect the transition.
497 */
498 static void
499 metaslab_group_alloc_update(metaslab_group_t *mg)
500 {
501 vdev_t *vd = mg->mg_vd;
502 metaslab_class_t *mc = mg->mg_class;
503 vdev_stat_t *vs = &vd->vdev_stat;
504 boolean_t was_allocatable;
505 boolean_t was_initialized;
506
507 ASSERT(vd == vd->vdev_top);
508
509 mutex_enter(&mg->mg_lock);
510 was_allocatable = mg->mg_allocatable;
511 was_initialized = mg->mg_initialized;
512
513 mg->mg_free_capacity = ((vs->vs_space - vs->vs_alloc) * 100) /
514 (vs->vs_space + 1);
515
516 mutex_enter(&mc->mc_lock);
517
518 /*
519 * If the metaslab group was just added then it won't
520 * have any space until we finish syncing out this txg.
521 * At that point we will consider it initialized and available
522 * for allocations. We also don't consider non-activated
523 * metaslab groups (e.g. vdevs that are in the middle of being removed)
524 * to be initialized, because they can't be used for allocation.
525 */
526 mg->mg_initialized = metaslab_group_initialized(mg);
527 if (!was_initialized && mg->mg_initialized) {
528 mc->mc_groups++;
529 } else if (was_initialized && !mg->mg_initialized) {
530 ASSERT3U(mc->mc_groups, >, 0);
531 mc->mc_groups--;
532 }
533 if (mg->mg_initialized)
534 mg->mg_no_free_space = B_FALSE;
535
536 /*
537 * A metaslab group is considered allocatable if it has plenty
538 * of free space or is not heavily fragmented. We only take
539 * fragmentation into account if the metaslab group has a valid
540 * fragmentation metric (i.e. a value between 0 and 100).
541 */
542 mg->mg_allocatable = (mg->mg_activation_count > 0 &&
543 mg->mg_free_capacity > zfs_mg_noalloc_threshold &&
544 (mg->mg_fragmentation == ZFS_FRAG_INVALID ||
545 mg->mg_fragmentation <= zfs_mg_fragmentation_threshold));
546
547 /*
548 * The mc_alloc_groups maintains a count of the number of
549 * groups in this metaslab class that are still above the
550 * zfs_mg_noalloc_threshold. This is used by the allocating
551 * threads to determine if they should avoid allocations to
552 * a given group. The allocator will avoid allocations to a group
553 * if that group has reached or is below the zfs_mg_noalloc_threshold
554 * and there are still other groups that are above the threshold.
555 * When a group transitions from allocatable to non-allocatable or
556 * vice versa we update the metaslab class to reflect that change.
557 * When the mc_alloc_groups value drops to 0 that means that all
558 * groups have reached the zfs_mg_noalloc_threshold making all groups
559 * eligible for allocations. This effectively means that all devices
560 * are balanced again.
561 */
562 if (was_allocatable && !mg->mg_allocatable)
563 mc->mc_alloc_groups--;
564 else if (!was_allocatable && mg->mg_allocatable)
565 mc->mc_alloc_groups++;
566 mutex_exit(&mc->mc_lock);
567
568 mutex_exit(&mg->mg_lock);
569 }
570
571 metaslab_group_t *
572 metaslab_group_create(metaslab_class_t *mc, vdev_t *vd)
573 {
574 metaslab_group_t *mg;
575
576 mg = kmem_zalloc(sizeof (metaslab_group_t), KM_SLEEP);
577 mutex_init(&mg->mg_lock, NULL, MUTEX_DEFAULT, NULL);
578 avl_create(&mg->mg_metaslab_tree, metaslab_compare,
579 sizeof (metaslab_t), offsetof(struct metaslab, ms_group_node));
580 mg->mg_vd = vd;
581 mg->mg_class = mc;
582 mg->mg_activation_count = 0;
583 mg->mg_initialized = B_FALSE;
584 mg->mg_no_free_space = B_TRUE;
585 refcount_create_tracked(&mg->mg_alloc_queue_depth);
586
587 mg->mg_taskq = taskq_create("metaslab_group_taskq", metaslab_load_pct,
588 minclsyspri, 10, INT_MAX, TASKQ_THREADS_CPU_PCT);
589
590 return (mg);
591 }
592
593 void
594 metaslab_group_destroy(metaslab_group_t *mg)
595 {
596 ASSERT(mg->mg_prev == NULL);
597 ASSERT(mg->mg_next == NULL);
598 /*
599 * We may have gone below zero with the activation count
600 * either because we never activated in the first place or
601 * because we're done, and possibly removing the vdev.
602 */
603 ASSERT(mg->mg_activation_count <= 0);
604
605 taskq_destroy(mg->mg_taskq);
606 avl_destroy(&mg->mg_metaslab_tree);
607 mutex_destroy(&mg->mg_lock);
608 refcount_destroy(&mg->mg_alloc_queue_depth);
609 kmem_free(mg, sizeof (metaslab_group_t));
610 }
611
612 void
613 metaslab_group_activate(metaslab_group_t *mg)
614 {
615 metaslab_class_t *mc = mg->mg_class;
616 metaslab_group_t *mgprev, *mgnext;
617
618 ASSERT(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER));
619
620 ASSERT(mc->mc_rotor != mg);
621 ASSERT(mg->mg_prev == NULL);
622 ASSERT(mg->mg_next == NULL);
623 ASSERT(mg->mg_activation_count <= 0);
624
625 if (++mg->mg_activation_count <= 0)
626 return;
627
628 mg->mg_aliquot = metaslab_aliquot * MAX(1, mg->mg_vd->vdev_children);
629 metaslab_group_alloc_update(mg);
630
631 if ((mgprev = mc->mc_rotor) == NULL) {
632 mg->mg_prev = mg;
633 mg->mg_next = mg;
634 } else {
635 mgnext = mgprev->mg_next;
636 mg->mg_prev = mgprev;
637 mg->mg_next = mgnext;
638 mgprev->mg_next = mg;
639 mgnext->mg_prev = mg;
640 }
641 mc->mc_rotor = mg;
642 }
643
644 void
645 metaslab_group_passivate(metaslab_group_t *mg)
646 {
647 metaslab_class_t *mc = mg->mg_class;
648 metaslab_group_t *mgprev, *mgnext;
649
650 ASSERT(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER));
651
652 if (--mg->mg_activation_count != 0) {
653 ASSERT(mc->mc_rotor != mg);
654 ASSERT(mg->mg_prev == NULL);
655 ASSERT(mg->mg_next == NULL);
656 ASSERT(mg->mg_activation_count < 0);
657 return;
658 }
659
660 taskq_wait(mg->mg_taskq);
661 metaslab_group_alloc_update(mg);
662
663 mgprev = mg->mg_prev;
664 mgnext = mg->mg_next;
665
666 if (mg == mgnext) {
667 mc->mc_rotor = NULL;
668 } else {
669 mc->mc_rotor = mgnext;
670 mgprev->mg_next = mgnext;
671 mgnext->mg_prev = mgprev;
672 }
673
674 mg->mg_prev = NULL;
675 mg->mg_next = NULL;
676 }
677
678 boolean_t
679 metaslab_group_initialized(metaslab_group_t *mg)
680 {
681 vdev_t *vd = mg->mg_vd;
682 vdev_stat_t *vs = &vd->vdev_stat;
683
684 return (vs->vs_space != 0 && mg->mg_activation_count > 0);
685 }
686
687 uint64_t
688 metaslab_group_get_space(metaslab_group_t *mg)
689 {
690 return ((1ULL << mg->mg_vd->vdev_ms_shift) * mg->mg_vd->vdev_ms_count);
691 }
692
693 void
694 metaslab_group_histogram_verify(metaslab_group_t *mg)
695 {
696 uint64_t *mg_hist;
697 vdev_t *vd = mg->mg_vd;
698 uint64_t ashift = vd->vdev_ashift;
699 int i;
700
701 if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0)
702 return;
703
704 mg_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE,
705 KM_SLEEP);
706
707 ASSERT3U(RANGE_TREE_HISTOGRAM_SIZE, >=,
708 SPACE_MAP_HISTOGRAM_SIZE + ashift);
709
710 for (int m = 0; m < vd->vdev_ms_count; m++) {
711 metaslab_t *msp = vd->vdev_ms[m];
712
713 if (msp->ms_sm == NULL)
714 continue;
715
716 for (i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++)
717 mg_hist[i + ashift] +=
718 msp->ms_sm->sm_phys->smp_histogram[i];
719 }
720
721 for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i ++)
722 VERIFY3U(mg_hist[i], ==, mg->mg_histogram[i]);
723
724 kmem_free(mg_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE);
725 }
726
727 static void
728 metaslab_group_histogram_add(metaslab_group_t *mg, metaslab_t *msp)
729 {
730 metaslab_class_t *mc = mg->mg_class;
731 uint64_t ashift = mg->mg_vd->vdev_ashift;
732
733 ASSERT(MUTEX_HELD(&msp->ms_lock));
734 if (msp->ms_sm == NULL)
735 return;
736
737 mutex_enter(&mg->mg_lock);
738 for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
739 mg->mg_histogram[i + ashift] +=
740 msp->ms_sm->sm_phys->smp_histogram[i];
741 mc->mc_histogram[i + ashift] +=
742 msp->ms_sm->sm_phys->smp_histogram[i];
743 }
744 mutex_exit(&mg->mg_lock);
745 }
746
747 void
748 metaslab_group_histogram_remove(metaslab_group_t *mg, metaslab_t *msp)
749 {
750 metaslab_class_t *mc = mg->mg_class;
751 uint64_t ashift = mg->mg_vd->vdev_ashift;
752
753 ASSERT(MUTEX_HELD(&msp->ms_lock));
754 if (msp->ms_sm == NULL)
755 return;
756
757 mutex_enter(&mg->mg_lock);
758 for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
759 ASSERT3U(mg->mg_histogram[i + ashift], >=,
760 msp->ms_sm->sm_phys->smp_histogram[i]);
761 ASSERT3U(mc->mc_histogram[i + ashift], >=,
762 msp->ms_sm->sm_phys->smp_histogram[i]);
763
764 mg->mg_histogram[i + ashift] -=
765 msp->ms_sm->sm_phys->smp_histogram[i];
766 mc->mc_histogram[i + ashift] -=
767 msp->ms_sm->sm_phys->smp_histogram[i];
768 }
769 mutex_exit(&mg->mg_lock);
770 }
771
772 static void
773 metaslab_group_add(metaslab_group_t *mg, metaslab_t *msp)
774 {
775 ASSERT(msp->ms_group == NULL);
776 mutex_enter(&mg->mg_lock);
777 msp->ms_group = mg;
778 msp->ms_weight = 0;
779 avl_add(&mg->mg_metaslab_tree, msp);
780 mutex_exit(&mg->mg_lock);
781
782 mutex_enter(&msp->ms_lock);
783 metaslab_group_histogram_add(mg, msp);
784 mutex_exit(&msp->ms_lock);
785 }
786
787 static void
788 metaslab_group_remove(metaslab_group_t *mg, metaslab_t *msp)
789 {
790 mutex_enter(&msp->ms_lock);
791 metaslab_group_histogram_remove(mg, msp);
792 mutex_exit(&msp->ms_lock);
793
794 mutex_enter(&mg->mg_lock);
795 ASSERT(msp->ms_group == mg);
796 avl_remove(&mg->mg_metaslab_tree, msp);
797 msp->ms_group = NULL;
798 mutex_exit(&mg->mg_lock);
799 }
800
801 static void
802 metaslab_group_sort(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight)
803 {
804 /*
805 * Although in principle the weight can be any value, in
806 * practice we do not use values in the range [1, 511].
807 */
808 ASSERT(weight >= SPA_MINBLOCKSIZE || weight == 0);
809 ASSERT(MUTEX_HELD(&msp->ms_lock));
810
811 mutex_enter(&mg->mg_lock);
812 ASSERT(msp->ms_group == mg);
813 avl_remove(&mg->mg_metaslab_tree, msp);
814 msp->ms_weight = weight;
815 avl_add(&mg->mg_metaslab_tree, msp);
816 mutex_exit(&mg->mg_lock);
817 }
818
819 /*
820 * Calculate the fragmentation for a given metaslab group. We can use
821 * a simple average here since all metaslabs within the group must have
822 * the same size. The return value will be a value between 0 and 100
823 * (inclusive), or ZFS_FRAG_INVALID if less than half of the metaslab in this
824 * group have a fragmentation metric.
825 */
826 uint64_t
827 metaslab_group_fragmentation(metaslab_group_t *mg)
828 {
829 vdev_t *vd = mg->mg_vd;
830 uint64_t fragmentation = 0;
831 uint64_t valid_ms = 0;
832
833 for (int m = 0; m < vd->vdev_ms_count; m++) {
834 metaslab_t *msp = vd->vdev_ms[m];
835
836 if (msp->ms_fragmentation == ZFS_FRAG_INVALID)
837 continue;
838
839 valid_ms++;
840 fragmentation += msp->ms_fragmentation;
841 }
842
843 if (valid_ms <= vd->vdev_ms_count / 2)
844 return (ZFS_FRAG_INVALID);
845
846 fragmentation /= valid_ms;
847 ASSERT3U(fragmentation, <=, 100);
848 return (fragmentation);
849 }
850
851 /*
852 * Determine if a given metaslab group should skip allocations. A metaslab
853 * group should avoid allocations if its free capacity is less than the
854 * zfs_mg_noalloc_threshold or its fragmentation metric is greater than
855 * zfs_mg_fragmentation_threshold and there is at least one metaslab group
856 * that can still handle allocations. If the allocation throttle is enabled
857 * then we skip allocations to devices that have reached their maximum
858 * allocation queue depth unless the selected metaslab group is the only
859 * eligible group remaining.
860 */
861 static boolean_t
862 metaslab_group_allocatable(metaslab_group_t *mg, metaslab_group_t *rotor,
863 uint64_t psize)
864 {
865 spa_t *spa = mg->mg_vd->vdev_spa;
866 metaslab_class_t *mc = mg->mg_class;
867
868 /*
869 * We can only consider skipping this metaslab group if it's
870 * in the normal metaslab class and there are other metaslab
871 * groups to select from. Otherwise, we always consider it eligible
872 * for allocations.
873 */
874 if (mc != spa_normal_class(spa) || mc->mc_groups <= 1)
875 return (B_TRUE);
876
877 /*
878 * If the metaslab group's mg_allocatable flag is set (see comments
879 * in metaslab_group_alloc_update() for more information) and
880 * the allocation throttle is disabled then allow allocations to this
881 * device. However, if the allocation throttle is enabled then
882 * check if we have reached our allocation limit (mg_alloc_queue_depth)
883 * to determine if we should allow allocations to this metaslab group.
884 * If all metaslab groups are no longer considered allocatable
885 * (mc_alloc_groups == 0) or we're trying to allocate the smallest
886 * gang block size then we allow allocations on this metaslab group
887 * regardless of the mg_allocatable or throttle settings.
888 */
889 if (mg->mg_allocatable) {
890 metaslab_group_t *mgp;
891 int64_t qdepth;
892 uint64_t qmax = mg->mg_max_alloc_queue_depth;
893
894 if (!mc->mc_alloc_throttle_enabled)
895 return (B_TRUE);
896
897 /*
898 * If this metaslab group does not have any free space, then
899 * there is no point in looking further.
900 */
901 if (mg->mg_no_free_space)
902 return (B_FALSE);
903
904 qdepth = refcount_count(&mg->mg_alloc_queue_depth);
905
906 /*
907 * If this metaslab group is below its qmax or it's
908 * the only allocatable metasable group, then attempt
909 * to allocate from it.
910 */
911 if (qdepth < qmax || mc->mc_alloc_groups == 1)
912 return (B_TRUE);
913 ASSERT3U(mc->mc_alloc_groups, >, 1);
914
915 /*
916 * Since this metaslab group is at or over its qmax, we
917 * need to determine if there are metaslab groups after this
918 * one that might be able to handle this allocation. This is
919 * racy since we can't hold the locks for all metaslab
920 * groups at the same time when we make this check.
921 */
922 for (mgp = mg->mg_next; mgp != rotor; mgp = mgp->mg_next) {
923 qmax = mgp->mg_max_alloc_queue_depth;
924
925 qdepth = refcount_count(&mgp->mg_alloc_queue_depth);
926
927 /*
928 * If there is another metaslab group that
929 * might be able to handle the allocation, then
930 * we return false so that we skip this group.
931 */
932 if (qdepth < qmax && !mgp->mg_no_free_space)
933 return (B_FALSE);
934 }
935
936 /*
937 * We didn't find another group to handle the allocation
938 * so we can't skip this metaslab group even though
939 * we are at or over our qmax.
940 */
941 return (B_TRUE);
942
943 } else if (mc->mc_alloc_groups == 0 || psize == SPA_MINBLOCKSIZE) {
944 return (B_TRUE);
945 }
946 return (B_FALSE);
947 }
948
949 /*
950 * ==========================================================================
951 * Range tree callbacks
952 * ==========================================================================
953 */
954
955 /*
956 * Comparison function for the private size-ordered tree. Tree is sorted
957 * by size, larger sizes at the end of the tree.
958 */
959 static int
960 metaslab_rangesize_compare(const void *x1, const void *x2)
961 {
962 const range_seg_t *r1 = x1;
963 const range_seg_t *r2 = x2;
964 uint64_t rs_size1 = r1->rs_end - r1->rs_start;
965 uint64_t rs_size2 = r2->rs_end - r2->rs_start;
966
967 if (rs_size1 < rs_size2)
968 return (-1);
969 if (rs_size1 > rs_size2)
970 return (1);
971
972 if (r1->rs_start < r2->rs_start)
973 return (-1);
974
975 if (r1->rs_start > r2->rs_start)
976 return (1);
977
978 return (0);
979 }
980
981 /*
982 * Create any block allocator specific components. The current allocators
983 * rely on using both a size-ordered range_tree_t and an array of uint64_t's.
984 */
985 static void
986 metaslab_rt_create(range_tree_t *rt, void *arg)
987 {
988 metaslab_t *msp = arg;
989
990 ASSERT3P(rt->rt_arg, ==, msp);
991 ASSERT(msp->ms_tree == NULL);
992
993 avl_create(&msp->ms_size_tree, metaslab_rangesize_compare,
994 sizeof (range_seg_t), offsetof(range_seg_t, rs_pp_node));
995 }
996
997 /*
998 * Destroy the block allocator specific components.
999 */
1000 static void
1001 metaslab_rt_destroy(range_tree_t *rt, void *arg)
1002 {
1003 metaslab_t *msp = arg;
1004
1005 ASSERT3P(rt->rt_arg, ==, msp);
1006 ASSERT3P(msp->ms_tree, ==, rt);
1007 ASSERT0(avl_numnodes(&msp->ms_size_tree));
1008
1009 avl_destroy(&msp->ms_size_tree);
1010 }
1011
1012 static void
1013 metaslab_rt_add(range_tree_t *rt, range_seg_t *rs, void *arg)
1014 {
1015 metaslab_t *msp = arg;
1016
1017 ASSERT3P(rt->rt_arg, ==, msp);
1018 ASSERT3P(msp->ms_tree, ==, rt);
1019 VERIFY(!msp->ms_condensing);
1020 avl_add(&msp->ms_size_tree, rs);
1021 }
1022
1023 static void
1024 metaslab_rt_remove(range_tree_t *rt, range_seg_t *rs, void *arg)
1025 {
1026 metaslab_t *msp = arg;
1027
1028 ASSERT3P(rt->rt_arg, ==, msp);
1029 ASSERT3P(msp->ms_tree, ==, rt);
1030 VERIFY(!msp->ms_condensing);
1031 avl_remove(&msp->ms_size_tree, rs);
1032 }
1033
1034 static void
1035 metaslab_rt_vacate(range_tree_t *rt, void *arg)
1036 {
1037 metaslab_t *msp = arg;
1038
1039 ASSERT3P(rt->rt_arg, ==, msp);
1040 ASSERT3P(msp->ms_tree, ==, rt);
1041
1042 /*
1043 * Normally one would walk the tree freeing nodes along the way.
1044 * Since the nodes are shared with the range trees we can avoid
1045 * walking all nodes and just reinitialize the avl tree. The nodes
1046 * will be freed by the range tree, so we don't want to free them here.
1047 */
1048 avl_create(&msp->ms_size_tree, metaslab_rangesize_compare,
1049 sizeof (range_seg_t), offsetof(range_seg_t, rs_pp_node));
1050 }
1051
1052 static range_tree_ops_t metaslab_rt_ops = {
1053 metaslab_rt_create,
1054 metaslab_rt_destroy,
1055 metaslab_rt_add,
1056 metaslab_rt_remove,
1057 metaslab_rt_vacate
1058 };
1059
1060 /*
1061 * ==========================================================================
1062 * Common allocator routines
1063 * ==========================================================================
1064 */
1065
1066 /*
1067 * Return the maximum contiguous segment within the metaslab.
1068 */
1069 uint64_t
1070 metaslab_block_maxsize(metaslab_t *msp)
1071 {
1072 avl_tree_t *t = &msp->ms_size_tree;
1073 range_seg_t *rs;
1074
1075 if (t == NULL || (rs = avl_last(t)) == NULL)
1076 return (0ULL);
1077
1078 return (rs->rs_end - rs->rs_start);
1079 }
1080
1081 static range_seg_t *
1082 metaslab_block_find(avl_tree_t *t, uint64_t start, uint64_t size)
1083 {
1084 range_seg_t *rs, rsearch;
1085 avl_index_t where;
1086
1087 rsearch.rs_start = start;
1088 rsearch.rs_end = start + size;
1089
1090 rs = avl_find(t, &rsearch, &where);
1091 if (rs == NULL) {
1092 rs = avl_nearest(t, where, AVL_AFTER);
1093 }
1094
1095 return (rs);
1096 }
1097
1098 /*
1099 * This is a helper function that can be used by the allocator to find
1100 * a suitable block to allocate. This will search the specified AVL
1101 * tree looking for a block that matches the specified criteria.
1102 */
1103 static uint64_t
1104 metaslab_block_picker(avl_tree_t *t, uint64_t *cursor, uint64_t size,
1105 uint64_t align)
1106 {
1107 range_seg_t *rs = metaslab_block_find(t, *cursor, size);
1108
1109 while (rs != NULL) {
1110 uint64_t offset = P2ROUNDUP(rs->rs_start, align);
1111
1112 if (offset + size <= rs->rs_end) {
1113 *cursor = offset + size;
1114 return (offset);
1115 }
1116 rs = AVL_NEXT(t, rs);
1117 }
1118
1119 /*
1120 * If we know we've searched the whole map (*cursor == 0), give up.
1121 * Otherwise, reset the cursor to the beginning and try again.
1122 */
1123 if (*cursor == 0)
1124 return (-1ULL);
1125
1126 *cursor = 0;
1127 return (metaslab_block_picker(t, cursor, size, align));
1128 }
1129
1130 /*
1131 * ==========================================================================
1132 * The first-fit block allocator
1133 * ==========================================================================
1134 */
1135 static uint64_t
1136 metaslab_ff_alloc(metaslab_t *msp, uint64_t size)
1137 {
1138 /*
1139 * Find the largest power of 2 block size that evenly divides the
1140 * requested size. This is used to try to allocate blocks with similar
1141 * alignment from the same area of the metaslab (i.e. same cursor
1142 * bucket) but it does not guarantee that other allocations sizes
1143 * may exist in the same region.
1144 */
1145 uint64_t align = size & -size;
1146 uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1];
1147 avl_tree_t *t = &msp->ms_tree->rt_root;
1148
1149 return (metaslab_block_picker(t, cursor, size, align));
1150 }
1151
1152 static metaslab_ops_t metaslab_ff_ops = {
1153 metaslab_ff_alloc
1154 };
1155
1156 /*
1157 * ==========================================================================
1158 * Dynamic block allocator -
1159 * Uses the first fit allocation scheme until space get low and then
1160 * adjusts to a best fit allocation method. Uses metaslab_df_alloc_threshold
1161 * and metaslab_df_free_pct to determine when to switch the allocation scheme.
1162 * ==========================================================================
1163 */
1164 static uint64_t
1165 metaslab_df_alloc(metaslab_t *msp, uint64_t size)
1166 {
1167 /*
1168 * Find the largest power of 2 block size that evenly divides the
1169 * requested size. This is used to try to allocate blocks with similar
1170 * alignment from the same area of the metaslab (i.e. same cursor
1171 * bucket) but it does not guarantee that other allocations sizes
1172 * may exist in the same region.
1173 */
1174 uint64_t align = size & -size;
1175 uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1];
1176 range_tree_t *rt = msp->ms_tree;
1177 avl_tree_t *t = &rt->rt_root;
1178 uint64_t max_size = metaslab_block_maxsize(msp);
1179 int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
1180
1181 ASSERT(MUTEX_HELD(&msp->ms_lock));
1182 ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_size_tree));
1183
1184 if (max_size < size)
1185 return (-1ULL);
1186
1187 /*
1188 * If we're running low on space switch to using the size
1189 * sorted AVL tree (best-fit).
1190 */
1191 if (max_size < metaslab_df_alloc_threshold ||
1192 free_pct < metaslab_df_free_pct) {
1193 t = &msp->ms_size_tree;
1194 *cursor = 0;
1195 }
1196
1197 return (metaslab_block_picker(t, cursor, size, 1ULL));
1198 }
1199
1200 static metaslab_ops_t metaslab_df_ops = {
1201 metaslab_df_alloc
1202 };
1203
1204 /*
1205 * ==========================================================================
1206 * Cursor fit block allocator -
1207 * Select the largest region in the metaslab, set the cursor to the beginning
1208 * of the range and the cursor_end to the end of the range. As allocations
1209 * are made advance the cursor. Continue allocating from the cursor until
1210 * the range is exhausted and then find a new range.
1211 * ==========================================================================
1212 */
1213 static uint64_t
1214 metaslab_cf_alloc(metaslab_t *msp, uint64_t size)
1215 {
1216 range_tree_t *rt = msp->ms_tree;
1217 avl_tree_t *t = &msp->ms_size_tree;
1218 uint64_t *cursor = &msp->ms_lbas[0];
1219 uint64_t *cursor_end = &msp->ms_lbas[1];
1220 uint64_t offset = 0;
1221
1222 ASSERT(MUTEX_HELD(&msp->ms_lock));
1223 ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&rt->rt_root));
1224
1225 ASSERT3U(*cursor_end, >=, *cursor);
1226
1227 if ((*cursor + size) > *cursor_end) {
1228 range_seg_t *rs;
1229
1230 rs = avl_last(&msp->ms_size_tree);
1231 if (rs == NULL || (rs->rs_end - rs->rs_start) < size)
1232 return (-1ULL);
1233
1234 *cursor = rs->rs_start;
1235 *cursor_end = rs->rs_end;
1236 }
1237
1238 offset = *cursor;
1239 *cursor += size;
1240
1241 return (offset);
1242 }
1243
1244 static metaslab_ops_t metaslab_cf_ops = {
1245 metaslab_cf_alloc
1246 };
1247
1248 /*
1249 * ==========================================================================
1250 * New dynamic fit allocator -
1251 * Select a region that is large enough to allocate 2^metaslab_ndf_clump_shift
1252 * contiguous blocks. If no region is found then just use the largest segment
1253 * that remains.
1254 * ==========================================================================
1255 */
1256
1257 /*
1258 * Determines desired number of contiguous blocks (2^metaslab_ndf_clump_shift)
1259 * to request from the allocator.
1260 */
1261 uint64_t metaslab_ndf_clump_shift = 4;
1262
1263 static uint64_t
1264 metaslab_ndf_alloc(metaslab_t *msp, uint64_t size)
1265 {
1266 avl_tree_t *t = &msp->ms_tree->rt_root;
1267 avl_index_t where;
1268 range_seg_t *rs, rsearch;
1269 uint64_t hbit = highbit64(size);
1270 uint64_t *cursor = &msp->ms_lbas[hbit - 1];
1271 uint64_t max_size = metaslab_block_maxsize(msp);
1272
1273 ASSERT(MUTEX_HELD(&msp->ms_lock));
1274 ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_size_tree));
1275
1276 if (max_size < size)
1277 return (-1ULL);
1278
1279 rsearch.rs_start = *cursor;
1280 rsearch.rs_end = *cursor + size;
1281
1282 rs = avl_find(t, &rsearch, &where);
1283 if (rs == NULL || (rs->rs_end - rs->rs_start) < size) {
1284 t = &msp->ms_size_tree;
1285
1286 rsearch.rs_start = 0;
1287 rsearch.rs_end = MIN(max_size,
1288 1ULL << (hbit + metaslab_ndf_clump_shift));
1289 rs = avl_find(t, &rsearch, &where);
1290 if (rs == NULL)
1291 rs = avl_nearest(t, where, AVL_AFTER);
1292 ASSERT(rs != NULL);
1293 }
1294
1295 if ((rs->rs_end - rs->rs_start) >= size) {
1296 *cursor = rs->rs_start + size;
1297 return (rs->rs_start);
1298 }
1299 return (-1ULL);
1300 }
1301
1302 static metaslab_ops_t metaslab_ndf_ops = {
1303 metaslab_ndf_alloc
1304 };
1305
1306 metaslab_ops_t *zfs_metaslab_ops = &metaslab_df_ops;
1307
1308 /*
1309 * ==========================================================================
1310 * Metaslabs
1311 * ==========================================================================
1312 */
1313
1314 /*
1315 * Wait for any in-progress metaslab loads to complete.
1316 */
1317 void
1318 metaslab_load_wait(metaslab_t *msp)
1319 {
1320 ASSERT(MUTEX_HELD(&msp->ms_lock));
1321
1322 while (msp->ms_loading) {
1323 ASSERT(!msp->ms_loaded);
1324 cv_wait(&msp->ms_load_cv, &msp->ms_lock);
1325 }
1326 }
1327
1328 int
1329 metaslab_load(metaslab_t *msp)
1330 {
1331 int error = 0;
1332 boolean_t success = B_FALSE;
1333
1334 ASSERT(MUTEX_HELD(&msp->ms_lock));
1335 ASSERT(!msp->ms_loaded);
1336 ASSERT(!msp->ms_loading);
1337
1338 msp->ms_loading = B_TRUE;
1339
1340 /*
1341 * If the space map has not been allocated yet, then treat
1342 * all the space in the metaslab as free and add it to the
1343 * ms_tree.
1344 */
1345 if (msp->ms_sm != NULL)
1346 error = space_map_load(msp->ms_sm, msp->ms_tree, SM_FREE);
1347 else
1348 range_tree_add(msp->ms_tree, msp->ms_start, msp->ms_size);
1349
1350 success = (error == 0);
1351 msp->ms_loading = B_FALSE;
1352
1353 if (success) {
1354 ASSERT3P(msp->ms_group, !=, NULL);
1355 msp->ms_loaded = B_TRUE;
1356
1357 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
1358 range_tree_walk(msp->ms_defertree[t],
1359 range_tree_remove, msp->ms_tree);
1360 }
1361 msp->ms_max_size = metaslab_block_maxsize(msp);
1362 }
1363 cv_broadcast(&msp->ms_load_cv);
1364 return (error);
1365 }
1366
1367 void
1368 metaslab_unload(metaslab_t *msp)
1369 {
1370 ASSERT(MUTEX_HELD(&msp->ms_lock));
1371 range_tree_vacate(msp->ms_tree, NULL, NULL);
1372 msp->ms_loaded = B_FALSE;
1373 msp->ms_weight &= ~METASLAB_ACTIVE_MASK;
1374 msp->ms_max_size = 0;
1375 }
1376
1377 int
1378 metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg,
1379 metaslab_t **msp)
1380 {
1381 vdev_t *vd = mg->mg_vd;
1382 objset_t *mos = vd->vdev_spa->spa_meta_objset;
1383 metaslab_t *ms;
1384 int error;
1385
1386 ms = kmem_zalloc(sizeof (metaslab_t), KM_SLEEP);
1387 mutex_init(&ms->ms_lock, NULL, MUTEX_DEFAULT, NULL);
1388 cv_init(&ms->ms_load_cv, NULL, CV_DEFAULT, NULL);
1389 ms->ms_id = id;
1390 ms->ms_start = id << vd->vdev_ms_shift;
1391 ms->ms_size = 1ULL << vd->vdev_ms_shift;
1392
1393 /*
1394 * We only open space map objects that already exist. All others
1395 * will be opened when we finally allocate an object for it.
1396 */
1397 if (object != 0) {
1398 error = space_map_open(&ms->ms_sm, mos, object, ms->ms_start,
1399 ms->ms_size, vd->vdev_ashift, &ms->ms_lock);
1400
1401 if (error != 0) {
1402 kmem_free(ms, sizeof (metaslab_t));
1403 return (error);
1404 }
1405
1406 ASSERT(ms->ms_sm != NULL);
1407 }
1408
1409 /*
1410 * We create the main range tree here, but we don't create the
1411 * other range trees until metaslab_sync_done(). This serves
1412 * two purposes: it allows metaslab_sync_done() to detect the
1413 * addition of new space; and for debugging, it ensures that we'd
1414 * data fault on any attempt to use this metaslab before it's ready.
1415 */
1416 ms->ms_tree = range_tree_create(&metaslab_rt_ops, ms, &ms->ms_lock);
1417 metaslab_group_add(mg, ms);
1418
1419 metaslab_set_fragmentation(ms);
1420
1421 /*
1422 * If we're opening an existing pool (txg == 0) or creating
1423 * a new one (txg == TXG_INITIAL), all space is available now.
1424 * If we're adding space to an existing pool, the new space
1425 * does not become available until after this txg has synced.
1426 * The metaslab's weight will also be initialized when we sync
1427 * out this txg. This ensures that we don't attempt to allocate
1428 * from it before we have initialized it completely.
1429 */
1430 if (txg <= TXG_INITIAL)
1431 metaslab_sync_done(ms, 0);
1432
1433 /*
1434 * If metaslab_debug_load is set and we're initializing a metaslab
1435 * that has an allocated space map object then load the its space
1436 * map so that can verify frees.
1437 */
1438 if (metaslab_debug_load && ms->ms_sm != NULL) {
1439 mutex_enter(&ms->ms_lock);
1440 VERIFY0(metaslab_load(ms));
1441 mutex_exit(&ms->ms_lock);
1442 }
1443
1444 if (txg != 0) {
1445 vdev_dirty(vd, 0, NULL, txg);
1446 vdev_dirty(vd, VDD_METASLAB, ms, txg);
1447 }
1448
1449 *msp = ms;
1450
1451 return (0);
1452 }
1453
1454 void
1455 metaslab_fini(metaslab_t *msp)
1456 {
1457 metaslab_group_t *mg = msp->ms_group;
1458
1459 metaslab_group_remove(mg, msp);
1460
1461 mutex_enter(&msp->ms_lock);
1462 VERIFY(msp->ms_group == NULL);
1463 vdev_space_update(mg->mg_vd, -space_map_allocated(msp->ms_sm),
1464 0, -msp->ms_size);
1465 space_map_close(msp->ms_sm);
1466
1467 metaslab_unload(msp);
1468 range_tree_destroy(msp->ms_tree);
1469 range_tree_destroy(msp->ms_freeingtree);
1470 range_tree_destroy(msp->ms_freedtree);
1471
1472 for (int t = 0; t < TXG_SIZE; t++) {
1473 range_tree_destroy(msp->ms_alloctree[t]);
1474 }
1475
1476 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
1477 range_tree_destroy(msp->ms_defertree[t]);
1478 }
1479
1480 ASSERT0(msp->ms_deferspace);
1481
1482 mutex_exit(&msp->ms_lock);
1483 cv_destroy(&msp->ms_load_cv);
1484 mutex_destroy(&msp->ms_lock);
1485
1486 kmem_free(msp, sizeof (metaslab_t));
1487 }
1488
1489 #define FRAGMENTATION_TABLE_SIZE 17
1490
1491 /*
1492 * This table defines a segment size based fragmentation metric that will
1493 * allow each metaslab to derive its own fragmentation value. This is done
1494 * by calculating the space in each bucket of the spacemap histogram and
1495 * multiplying that by the fragmetation metric in this table. Doing
1496 * this for all buckets and dividing it by the total amount of free
1497 * space in this metaslab (i.e. the total free space in all buckets) gives
1498 * us the fragmentation metric. This means that a high fragmentation metric
1499 * equates to most of the free space being comprised of small segments.
1500 * Conversely, if the metric is low, then most of the free space is in
1501 * large segments. A 10% change in fragmentation equates to approximately
1502 * double the number of segments.
1503 *
1504 * This table defines 0% fragmented space using 16MB segments. Testing has
1505 * shown that segments that are greater than or equal to 16MB do not suffer
1506 * from drastic performance problems. Using this value, we derive the rest
1507 * of the table. Since the fragmentation value is never stored on disk, it
1508 * is possible to change these calculations in the future.
1509 */
1510 int zfs_frag_table[FRAGMENTATION_TABLE_SIZE] = {
1511 100, /* 512B */
1512 100, /* 1K */
1513 98, /* 2K */
1514 95, /* 4K */
1515 90, /* 8K */
1516 80, /* 16K */
1517 70, /* 32K */
1518 60, /* 64K */
1519 50, /* 128K */
1520 40, /* 256K */
1521 30, /* 512K */
1522 20, /* 1M */
1523 15, /* 2M */
1524 10, /* 4M */
1525 5, /* 8M */
1526 0 /* 16M */
1527 };
1528
1529 /*
1530 * Calclate the metaslab's fragmentation metric. A return value
1531 * of ZFS_FRAG_INVALID means that the metaslab has not been upgraded and does
1532 * not support this metric. Otherwise, the return value should be in the
1533 * range [0, 100].
1534 */
1535 static void
1536 metaslab_set_fragmentation(metaslab_t *msp)
1537 {
1538 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
1539 uint64_t fragmentation = 0;
1540 uint64_t total = 0;
1541 boolean_t feature_enabled = spa_feature_is_enabled(spa,
1542 SPA_FEATURE_SPACEMAP_HISTOGRAM);
1543
1544 if (!feature_enabled) {
1545 msp->ms_fragmentation = ZFS_FRAG_INVALID;
1546 return;
1547 }
1548
1549 /*
1550 * A null space map means that the entire metaslab is free
1551 * and thus is not fragmented.
1552 */
1553 if (msp->ms_sm == NULL) {
1554 msp->ms_fragmentation = 0;
1555 return;
1556 }
1557
1558 /*
1559 * If this metaslab's space map has not been upgraded, flag it
1560 * so that we upgrade next time we encounter it.
1561 */
1562 if (msp->ms_sm->sm_dbuf->db_size != sizeof (space_map_phys_t)) {
1563 uint64_t txg = spa_syncing_txg(spa);
1564 vdev_t *vd = msp->ms_group->mg_vd;
1565
1566 if (spa_writeable(spa)) {
1567 msp->ms_condense_wanted = B_TRUE;
1568 vdev_dirty(vd, VDD_METASLAB, msp, txg + 1);
1569 spa_dbgmsg(spa, "txg %llu, requesting force condense: "
1570 "msp %p, vd %p", txg, msp, vd);
1571 }
1572 msp->ms_fragmentation = ZFS_FRAG_INVALID;
1573 return;
1574 }
1575
1576 for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
1577 uint64_t space = 0;
1578 uint8_t shift = msp->ms_sm->sm_shift;
1579
1580 int idx = MIN(shift - SPA_MINBLOCKSHIFT + i,
1581 FRAGMENTATION_TABLE_SIZE - 1);
1582
1583 if (msp->ms_sm->sm_phys->smp_histogram[i] == 0)
1584 continue;
1585
1586 space = msp->ms_sm->sm_phys->smp_histogram[i] << (i + shift);
1587 total += space;
1588
1589 ASSERT3U(idx, <, FRAGMENTATION_TABLE_SIZE);
1590 fragmentation += space * zfs_frag_table[idx];
1591 }
1592
1593 if (total > 0)
1594 fragmentation /= total;
1595 ASSERT3U(fragmentation, <=, 100);
1596
1597 msp->ms_fragmentation = fragmentation;
1598 }
1599
1600 /*
1601 * Compute a weight -- a selection preference value -- for the given metaslab.
1602 * This is based on the amount of free space, the level of fragmentation,
1603 * the LBA range, and whether the metaslab is loaded.
1604 */
1605 static uint64_t
1606 metaslab_space_weight(metaslab_t *msp)
1607 {
1608 metaslab_group_t *mg = msp->ms_group;
1609 vdev_t *vd = mg->mg_vd;
1610 uint64_t weight, space;
1611
1612 ASSERT(MUTEX_HELD(&msp->ms_lock));
1613 ASSERT(!vd->vdev_removing);
1614
1615 /*
1616 * The baseline weight is the metaslab's free space.
1617 */
1618 space = msp->ms_size - space_map_allocated(msp->ms_sm);
1619
1620 if (metaslab_fragmentation_factor_enabled &&
1621 msp->ms_fragmentation != ZFS_FRAG_INVALID) {
1622 /*
1623 * Use the fragmentation information to inversely scale
1624 * down the baseline weight. We need to ensure that we
1625 * don't exclude this metaslab completely when it's 100%
1626 * fragmented. To avoid this we reduce the fragmented value
1627 * by 1.
1628 */
1629 space = (space * (100 - (msp->ms_fragmentation - 1))) / 100;
1630
1631 /*
1632 * If space < SPA_MINBLOCKSIZE, then we will not allocate from
1633 * this metaslab again. The fragmentation metric may have
1634 * decreased the space to something smaller than
1635 * SPA_MINBLOCKSIZE, so reset the space to SPA_MINBLOCKSIZE
1636 * so that we can consume any remaining space.
1637 */
1638 if (space > 0 && space < SPA_MINBLOCKSIZE)
1639 space = SPA_MINBLOCKSIZE;
1640 }
1641 weight = space;
1642
1643 /*
1644 * Modern disks have uniform bit density and constant angular velocity.
1645 * Therefore, the outer recording zones are faster (higher bandwidth)
1646 * than the inner zones by the ratio of outer to inner track diameter,
1647 * which is typically around 2:1. We account for this by assigning
1648 * higher weight to lower metaslabs (multiplier ranging from 2x to 1x).
1649 * In effect, this means that we'll select the metaslab with the most
1650 * free bandwidth rather than simply the one with the most free space.
1651 */
1652 if (metaslab_lba_weighting_enabled) {
1653 weight = 2 * weight - (msp->ms_id * weight) / vd->vdev_ms_count;
1654 ASSERT(weight >= space && weight <= 2 * space);
1655 }
1656
1657 /*
1658 * If this metaslab is one we're actively using, adjust its
1659 * weight to make it preferable to any inactive metaslab so
1660 * we'll polish it off. If the fragmentation on this metaslab
1661 * has exceed our threshold, then don't mark it active.
1662 */
1663 if (msp->ms_loaded && msp->ms_fragmentation != ZFS_FRAG_INVALID &&
1664 msp->ms_fragmentation <= zfs_metaslab_fragmentation_threshold) {
1665 weight |= (msp->ms_weight & METASLAB_ACTIVE_MASK);
1666 }
1667
1668 WEIGHT_SET_SPACEBASED(weight);
1669 return (weight);
1670 }
1671
1672 /*
1673 * Return the weight of the specified metaslab, according to the segment-based
1674 * weighting algorithm. The metaslab must be loaded. This function can
1675 * be called within a sync pass since it relies only on the metaslab's
1676 * range tree which is always accurate when the metaslab is loaded.
1677 */
1678 static uint64_t
1679 metaslab_weight_from_range_tree(metaslab_t *msp)
1680 {
1681 uint64_t weight = 0;
1682 uint32_t segments = 0;
1683
1684 ASSERT(msp->ms_loaded);
1685
1686 for (int i = RANGE_TREE_HISTOGRAM_SIZE - 1; i >= SPA_MINBLOCKSHIFT;
1687 i--) {
1688 uint8_t shift = msp->ms_group->mg_vd->vdev_ashift;
1689 int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1;
1690
1691 segments <<= 1;
1692 segments += msp->ms_tree->rt_histogram[i];
1693
1694 /*
1695 * The range tree provides more precision than the space map
1696 * and must be downgraded so that all values fit within the
1697 * space map's histogram. This allows us to compare loaded
1698 * vs. unloaded metaslabs to determine which metaslab is
1699 * considered "best".
1700 */
1701 if (i > max_idx)
1702 continue;
1703
1704 if (segments != 0) {
1705 WEIGHT_SET_COUNT(weight, segments);
1706 WEIGHT_SET_INDEX(weight, i);
1707 WEIGHT_SET_ACTIVE(weight, 0);
1708 break;
1709 }
1710 }
1711 return (weight);
1712 }
1713
1714 /*
1715 * Calculate the weight based on the on-disk histogram. This should only
1716 * be called after a sync pass has completely finished since the on-disk
1717 * information is updated in metaslab_sync().
1718 */
1719 static uint64_t
1720 metaslab_weight_from_spacemap(metaslab_t *msp)
1721 {
1722 uint64_t weight = 0;
1723
1724 for (int i = SPACE_MAP_HISTOGRAM_SIZE - 1; i >= 0; i--) {
1725 if (msp->ms_sm->sm_phys->smp_histogram[i] != 0) {
1726 WEIGHT_SET_COUNT(weight,
1727 msp->ms_sm->sm_phys->smp_histogram[i]);
1728 WEIGHT_SET_INDEX(weight, i +
1729 msp->ms_sm->sm_shift);
1730 WEIGHT_SET_ACTIVE(weight, 0);
1731 break;
1732 }
1733 }
1734 return (weight);
1735 }
1736
1737 /*
1738 * Compute a segment-based weight for the specified metaslab. The weight
1739 * is determined by highest bucket in the histogram. The information
1740 * for the highest bucket is encoded into the weight value.
1741 */
1742 static uint64_t
1743 metaslab_segment_weight(metaslab_t *msp)
1744 {
1745 metaslab_group_t *mg = msp->ms_group;
1746 uint64_t weight = 0;
1747 uint8_t shift = mg->mg_vd->vdev_ashift;
1748
1749 ASSERT(MUTEX_HELD(&msp->ms_lock));
1750
1751 /*
1752 * The metaslab is completely free.
1753 */
1754 if (space_map_allocated(msp->ms_sm) == 0) {
1755 int idx = highbit64(msp->ms_size) - 1;
1756 int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1;
1757
1758 if (idx < max_idx) {
1759 WEIGHT_SET_COUNT(weight, 1ULL);
1760 WEIGHT_SET_INDEX(weight, idx);
1761 } else {
1762 WEIGHT_SET_COUNT(weight, 1ULL << (idx - max_idx));
1763 WEIGHT_SET_INDEX(weight, max_idx);
1764 }
1765 WEIGHT_SET_ACTIVE(weight, 0);
1766 ASSERT(!WEIGHT_IS_SPACEBASED(weight));
1767
1768 return (weight);
1769 }
1770
1771 ASSERT3U(msp->ms_sm->sm_dbuf->db_size, ==, sizeof (space_map_phys_t));
1772
1773 /*
1774 * If the metaslab is fully allocated then just make the weight 0.
1775 */
1776 if (space_map_allocated(msp->ms_sm) == msp->ms_size)
1777 return (0);
1778 /*
1779 * If the metaslab is already loaded, then use the range tree to
1780 * determine the weight. Otherwise, we rely on the space map information
1781 * to generate the weight.
1782 */
1783 if (msp->ms_loaded) {
1784 weight = metaslab_weight_from_range_tree(msp);
1785 } else {
1786 weight = metaslab_weight_from_spacemap(msp);
1787 }
1788
1789 /*
1790 * If the metaslab was active the last time we calculated its weight
1791 * then keep it active. We want to consume the entire region that
1792 * is associated with this weight.
1793 */
1794 if (msp->ms_activation_weight != 0 && weight != 0)
1795 WEIGHT_SET_ACTIVE(weight, WEIGHT_GET_ACTIVE(msp->ms_weight));
1796 return (weight);
1797 }
1798
1799 /*
1800 * Determine if we should attempt to allocate from this metaslab. If the
1801 * metaslab has a maximum size then we can quickly determine if the desired
1802 * allocation size can be satisfied. Otherwise, if we're using segment-based
1803 * weighting then we can determine the maximum allocation that this metaslab
1804 * can accommodate based on the index encoded in the weight. If we're using
1805 * space-based weights then rely on the entire weight (excluding the weight
1806 * type bit).
1807 */
1808 boolean_t
1809 metaslab_should_allocate(metaslab_t *msp, uint64_t asize)
1810 {
1811 boolean_t should_allocate;
1812
1813 if (msp->ms_max_size != 0)
1814 return (msp->ms_max_size >= asize);
1815
1816 if (!WEIGHT_IS_SPACEBASED(msp->ms_weight)) {
1817 /*
1818 * The metaslab segment weight indicates segments in the
1819 * range [2^i, 2^(i+1)), where i is the index in the weight.
1820 * Since the asize might be in the middle of the range, we
1821 * should attempt the allocation if asize < 2^(i+1).
1822 */
1823 should_allocate = (asize <
1824 1ULL << (WEIGHT_GET_INDEX(msp->ms_weight) + 1));
1825 } else {
1826 should_allocate = (asize <=
1827 (msp->ms_weight & ~METASLAB_WEIGHT_TYPE));
1828 }
1829 return (should_allocate);
1830 }
1831
1832 static uint64_t
1833 metaslab_weight(metaslab_t *msp)
1834 {
1835 vdev_t *vd = msp->ms_group->mg_vd;
1836 spa_t *spa = vd->vdev_spa;
1837 uint64_t weight;
1838
1839 ASSERT(MUTEX_HELD(&msp->ms_lock));
1840
1841 /*
1842 * This vdev is in the process of being removed so there is nothing
1843 * for us to do here.
1844 */
1845 if (vd->vdev_removing) {
1846 ASSERT0(space_map_allocated(msp->ms_sm));
1847 ASSERT0(vd->vdev_ms_shift);
1848 return (0);
1849 }
1850
1851 metaslab_set_fragmentation(msp);
1852
1853 /*
1854 * Update the maximum size if the metaslab is loaded. This will
1855 * ensure that we get an accurate maximum size if newly freed space
1856 * has been added back into the free tree.
1857 */
1858 if (msp->ms_loaded)
1859 msp->ms_max_size = metaslab_block_maxsize(msp);
1860
1861 /*
1862 * Segment-based weighting requires space map histogram support.
1863 */
1864 if (zfs_metaslab_segment_weight_enabled &&
1865 spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM) &&
1866 (msp->ms_sm == NULL || msp->ms_sm->sm_dbuf->db_size ==
1867 sizeof (space_map_phys_t))) {
1868 weight = metaslab_segment_weight(msp);
1869 } else {
1870 weight = metaslab_space_weight(msp);
1871 }
1872 return (weight);
1873 }
1874
1875 static int
1876 metaslab_activate(metaslab_t *msp, uint64_t activation_weight)
1877 {
1878 ASSERT(MUTEX_HELD(&msp->ms_lock));
1879
1880 if ((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0) {
1881 metaslab_load_wait(msp);
1882 if (!msp->ms_loaded) {
1883 int error = metaslab_load(msp);
1884 if (error) {
1885 metaslab_group_sort(msp->ms_group, msp, 0);
1886 return (error);
1887 }
1888 }
1889
1890 msp->ms_activation_weight = msp->ms_weight;
1891 metaslab_group_sort(msp->ms_group, msp,
1892 msp->ms_weight | activation_weight);
1893 }
1894 ASSERT(msp->ms_loaded);
1895 ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK);
1896
1897 return (0);
1898 }
1899
1900 static void
1901 metaslab_passivate(metaslab_t *msp, uint64_t weight)
1902 {
1903 uint64_t size = weight & ~METASLAB_WEIGHT_TYPE;
1904
1905 /*
1906 * If size < SPA_MINBLOCKSIZE, then we will not allocate from
1907 * this metaslab again. In that case, it had better be empty,
1908 * or we would be leaving space on the table.
1909 */
1910 ASSERT(size >= SPA_MINBLOCKSIZE ||
1911 range_tree_space(msp->ms_tree) == 0);
1912 ASSERT0(weight & METASLAB_ACTIVE_MASK);
1913
1914 msp->ms_activation_weight = 0;
1915 metaslab_group_sort(msp->ms_group, msp, weight);
1916 ASSERT((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0);
1917 }
1918
1919 /*
1920 * Segment-based metaslabs are activated once and remain active until
1921 * we either fail an allocation attempt (similar to space-based metaslabs)
1922 * or have exhausted the free space in zfs_metaslab_switch_threshold
1923 * buckets since the metaslab was activated. This function checks to see
1924 * if we've exhaused the zfs_metaslab_switch_threshold buckets in the
1925 * metaslab and passivates it proactively. This will allow us to select a
1926 * metaslabs with larger contiguous region if any remaining within this
1927 * metaslab group. If we're in sync pass > 1, then we continue using this
1928 * metaslab so that we don't dirty more block and cause more sync passes.
1929 */
1930 void
1931 metaslab_segment_may_passivate(metaslab_t *msp)
1932 {
1933 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
1934
1935 if (WEIGHT_IS_SPACEBASED(msp->ms_weight) || spa_sync_pass(spa) > 1)
1936 return;
1937
1938 /*
1939 * Since we are in the middle of a sync pass, the most accurate
1940 * information that is accessible to us is the in-core range tree
1941 * histogram; calculate the new weight based on that information.
1942 */
1943 uint64_t weight = metaslab_weight_from_range_tree(msp);
1944 int activation_idx = WEIGHT_GET_INDEX(msp->ms_activation_weight);
1945 int current_idx = WEIGHT_GET_INDEX(weight);
1946
1947 if (current_idx <= activation_idx - zfs_metaslab_switch_threshold)
1948 metaslab_passivate(msp, weight);
1949 }
1950
1951 static void
1952 metaslab_preload(void *arg)
1953 {
1954 metaslab_t *msp = arg;
1955 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
1956
1957 ASSERT(!MUTEX_HELD(&msp->ms_group->mg_lock));
1958
1959 mutex_enter(&msp->ms_lock);
1960 metaslab_load_wait(msp);
1961 if (!msp->ms_loaded)
1962 (void) metaslab_load(msp);
1963 msp->ms_selected_txg = spa_syncing_txg(spa);
1964 mutex_exit(&msp->ms_lock);
1965 }
1966
1967 static void
1968 metaslab_group_preload(metaslab_group_t *mg)
1969 {
1970 spa_t *spa = mg->mg_vd->vdev_spa;
1971 metaslab_t *msp;
1972 avl_tree_t *t = &mg->mg_metaslab_tree;
1973 int m = 0;
1974
1975 if (spa_shutting_down(spa) || !metaslab_preload_enabled) {
1976 taskq_wait(mg->mg_taskq);
1977 return;
1978 }
1979
1980 mutex_enter(&mg->mg_lock);
1981 /*
1982 * Load the next potential metaslabs
1983 */
1984 for (msp = avl_first(t); msp != NULL; msp = AVL_NEXT(t, msp)) {
1985 /*
1986 * We preload only the maximum number of metaslabs specified
1987 * by metaslab_preload_limit. If a metaslab is being forced
1988 * to condense then we preload it too. This will ensure
1989 * that force condensing happens in the next txg.
1990 */
1991 if (++m > metaslab_preload_limit && !msp->ms_condense_wanted) {
1992 continue;
1993 }
1994
1995 VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload,
1996 msp, TQ_SLEEP) != NULL);
1997 }
1998 mutex_exit(&mg->mg_lock);
1999 }
2000
2001 /*
2002 * Determine if the space map's on-disk footprint is past our tolerance
2003 * for inefficiency. We would like to use the following criteria to make
2004 * our decision:
2005 *
2006 * 1. The size of the space map object should not dramatically increase as a
2007 * result of writing out the free space range tree.
2008 *
2009 * 2. The minimal on-disk space map representation is zfs_condense_pct/100
2010 * times the size than the free space range tree representation
2011 * (i.e. zfs_condense_pct = 110 and in-core = 1MB, minimal = 1.1.MB).
2012 *
2013 * 3. The on-disk size of the space map should actually decrease.
2014 *
2015 * Checking the first condition is tricky since we don't want to walk
2016 * the entire AVL tree calculating the estimated on-disk size. Instead we
2017 * use the size-ordered range tree in the metaslab and calculate the
2018 * size required to write out the largest segment in our free tree. If the
2019 * size required to represent that segment on disk is larger than the space
2020 * map object then we avoid condensing this map.
2021 *
2022 * To determine the second criterion we use a best-case estimate and assume
2023 * each segment can be represented on-disk as a single 64-bit entry. We refer
2024 * to this best-case estimate as the space map's minimal form.
2025 *
2026 * Unfortunately, we cannot compute the on-disk size of the space map in this
2027 * context because we cannot accurately compute the effects of compression, etc.
2028 * Instead, we apply the heuristic described in the block comment for
2029 * zfs_metaslab_condense_block_threshold - we only condense if the space used
2030 * is greater than a threshold number of blocks.
2031 */
2032 static boolean_t
2033 metaslab_should_condense(metaslab_t *msp)
2034 {
2035 space_map_t *sm = msp->ms_sm;
2036 range_seg_t *rs;
2037 uint64_t size, entries, segsz, object_size, optimal_size, record_size;
2038 dmu_object_info_t doi;
2039 uint64_t vdev_blocksize = 1 << msp->ms_group->mg_vd->vdev_ashift;
2040
2041 ASSERT(MUTEX_HELD(&msp->ms_lock));
2042 ASSERT(msp->ms_loaded);
2043
2044 /*
2045 * Use the ms_size_tree range tree, which is ordered by size, to
2046 * obtain the largest segment in the free tree. We always condense
2047 * metaslabs that are empty and metaslabs for which a condense
2048 * request has been made.
2049 */
2050 rs = avl_last(&msp->ms_size_tree);
2051 if (rs == NULL || msp->ms_condense_wanted)
2052 return (B_TRUE);
2053
2054 /*
2055 * Calculate the number of 64-bit entries this segment would
2056 * require when written to disk. If this single segment would be
2057 * larger on-disk than the entire current on-disk structure, then
2058 * clearly condensing will increase the on-disk structure size.
2059 */
2060 size = (rs->rs_end - rs->rs_start) >> sm->sm_shift;
2061 entries = size / (MIN(size, SM_RUN_MAX));
2062 segsz = entries * sizeof (uint64_t);
2063
2064 optimal_size = sizeof (uint64_t) * avl_numnodes(&msp->ms_tree->rt_root);
2065 object_size = space_map_length(msp->ms_sm);
2066
2067 dmu_object_info_from_db(sm->sm_dbuf, &doi);
2068 record_size = MAX(doi.doi_data_block_size, vdev_blocksize);
2069
2070 return (segsz <= object_size &&
2071 object_size >= (optimal_size * zfs_condense_pct / 100) &&
2072 object_size > zfs_metaslab_condense_block_threshold * record_size);
2073 }
2074
2075 /*
2076 * Condense the on-disk space map representation to its minimized form.
2077 * The minimized form consists of a small number of allocations followed by
2078 * the entries of the free range tree.
2079 */
2080 static void
2081 metaslab_condense(metaslab_t *msp, uint64_t txg, dmu_tx_t *tx)
2082 {
2083 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
2084 range_tree_t *condense_tree;
2085 space_map_t *sm = msp->ms_sm;
2086
2087 ASSERT(MUTEX_HELD(&msp->ms_lock));
2088 ASSERT3U(spa_sync_pass(spa), ==, 1);
2089 ASSERT(msp->ms_loaded);
2090
2091
2092 spa_dbgmsg(spa, "condensing: txg %llu, msp[%llu] %p, vdev id %llu, "
2093 "spa %s, smp size %llu, segments %lu, forcing condense=%s", txg,
2094 msp->ms_id, msp, msp->ms_group->mg_vd->vdev_id,
2095 msp->ms_group->mg_vd->vdev_spa->spa_name,
2096 space_map_length(msp->ms_sm), avl_numnodes(&msp->ms_tree->rt_root),
2097 msp->ms_condense_wanted ? "TRUE" : "FALSE");
2098
2099 msp->ms_condense_wanted = B_FALSE;
2100
2101 /*
2102 * Create an range tree that is 100% allocated. We remove segments
2103 * that have been freed in this txg, any deferred frees that exist,
2104 * and any allocation in the future. Removing segments should be
2105 * a relatively inexpensive operation since we expect these trees to
2106 * have a small number of nodes.
2107 */
2108 condense_tree = range_tree_create(NULL, NULL, &msp->ms_lock);
2109 range_tree_add(condense_tree, msp->ms_start, msp->ms_size);
2110
2111 /*
2112 * Remove what's been freed in this txg from the condense_tree.
2113 * Since we're in sync_pass 1, we know that all the frees from
2114 * this txg are in the freeingtree.
2115 */
2116 range_tree_walk(msp->ms_freeingtree, range_tree_remove, condense_tree);
2117
2118 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
2119 range_tree_walk(msp->ms_defertree[t],
2120 range_tree_remove, condense_tree);
2121 }
2122
2123 for (int t = 1; t < TXG_CONCURRENT_STATES; t++) {
2124 range_tree_walk(msp->ms_alloctree[(txg + t) & TXG_MASK],
2125 range_tree_remove, condense_tree);
2126 }
2127
2128 /*
2129 * We're about to drop the metaslab's lock thus allowing
2130 * other consumers to change it's content. Set the
2131 * metaslab's ms_condensing flag to ensure that
2132 * allocations on this metaslab do not occur while we're
2133 * in the middle of committing it to disk. This is only critical
2134 * for the ms_tree as all other range trees use per txg
2135 * views of their content.
2136 */
2137 msp->ms_condensing = B_TRUE;
2138
2139 mutex_exit(&msp->ms_lock);
2140 space_map_truncate(sm, tx);
2141 mutex_enter(&msp->ms_lock);
2142
2143 /*
2144 * While we would ideally like to create a space map representation
2145 * that consists only of allocation records, doing so can be
2146 * prohibitively expensive because the in-core free tree can be
2147 * large, and therefore computationally expensive to subtract
2148 * from the condense_tree. Instead we sync out two trees, a cheap
2149 * allocation only tree followed by the in-core free tree. While not
2150 * optimal, this is typically close to optimal, and much cheaper to
2151 * compute.
2152 */
2153 space_map_write(sm, condense_tree, SM_ALLOC, tx);
2154 range_tree_vacate(condense_tree, NULL, NULL);
2155 range_tree_destroy(condense_tree);
2156
2157 space_map_write(sm, msp->ms_tree, SM_FREE, tx);
2158 msp->ms_condensing = B_FALSE;
2159 }
2160
2161 /*
2162 * Write a metaslab to disk in the context of the specified transaction group.
2163 */
2164 void
2165 metaslab_sync(metaslab_t *msp, uint64_t txg)
2166 {
2167 metaslab_group_t *mg = msp->ms_group;
2168 vdev_t *vd = mg->mg_vd;
2169 spa_t *spa = vd->vdev_spa;
2170 objset_t *mos = spa_meta_objset(spa);
2171 range_tree_t *alloctree = msp->ms_alloctree[txg & TXG_MASK];
2172 dmu_tx_t *tx;
2173 uint64_t object = space_map_object(msp->ms_sm);
2174
2175 ASSERT(!vd->vdev_ishole);
2176
2177 /*
2178 * This metaslab has just been added so there's no work to do now.
2179 */
2180 if (msp->ms_freeingtree == NULL) {
2181 ASSERT3P(alloctree, ==, NULL);
2182 return;
2183 }
2184
2185 ASSERT3P(alloctree, !=, NULL);
2186 ASSERT3P(msp->ms_freeingtree, !=, NULL);
2187 ASSERT3P(msp->ms_freedtree, !=, NULL);
2188
2189 /*
2190 * Normally, we don't want to process a metaslab if there
2191 * are no allocations or frees to perform. However, if the metaslab
2192 * is being forced to condense we need to let it through.
2193 */
2194 if (range_tree_space(alloctree) == 0 &&
2195 range_tree_space(msp->ms_freeingtree) == 0 &&
2196 !msp->ms_condense_wanted)
2197 return;
2198
2199 /*
2200 * The only state that can actually be changing concurrently with
2201 * metaslab_sync() is the metaslab's ms_tree. No other thread can
2202 * be modifying this txg's alloctree, freeingtree, freedtree, or
2203 * space_map_phys_t. Therefore, we only hold ms_lock to satify
2204 * space map ASSERTs. We drop it whenever we call into the DMU,
2205 * because the DMU can call down to us (e.g. via zio_free()) at
2206 * any time.
2207 */
2208
2209 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2210
2211 if (msp->ms_sm == NULL) {
2212 uint64_t new_object;
2213
2214 new_object = space_map_alloc(mos, tx);
2215 VERIFY3U(new_object, !=, 0);
2216
2217 VERIFY0(space_map_open(&msp->ms_sm, mos, new_object,
2218 msp->ms_start, msp->ms_size, vd->vdev_ashift,
2219 &msp->ms_lock));
2220 ASSERT(msp->ms_sm != NULL);
2221 }
2222
2223 mutex_enter(&msp->ms_lock);
2224
2225 /*
2226 * Note: metaslab_condense() clears the space map's histogram.
2227 * Therefore we must verify and remove this histogram before
2228 * condensing.
2229 */
2230 metaslab_group_histogram_verify(mg);
2231 metaslab_class_histogram_verify(mg->mg_class);
2232 metaslab_group_histogram_remove(mg, msp);
2233
2234 if (msp->ms_loaded && spa_sync_pass(spa) == 1 &&
2235 metaslab_should_condense(msp)) {
2236 metaslab_condense(msp, txg, tx);
2237 } else {
2238 space_map_write(msp->ms_sm, alloctree, SM_ALLOC, tx);
2239 space_map_write(msp->ms_sm, msp->ms_freeingtree, SM_FREE, tx);
2240 }
2241
2242 if (msp->ms_loaded) {
2243 /*
2244 * When the space map is loaded, we have an accruate
2245 * histogram in the range tree. This gives us an opportunity
2246 * to bring the space map's histogram up-to-date so we clear
2247 * it first before updating it.
2248 */
2249 space_map_histogram_clear(msp->ms_sm);
2250 space_map_histogram_add(msp->ms_sm, msp->ms_tree, tx);
2251
2252 /*
2253 * Since we've cleared the histogram we need to add back
2254 * any free space that has already been processed, plus
2255 * any deferred space. This allows the on-disk histogram
2256 * to accurately reflect all free space even if some space
2257 * is not yet available for allocation (i.e. deferred).
2258 */
2259 space_map_histogram_add(msp->ms_sm, msp->ms_freedtree, tx);
2260
2261 /*
2262 * Add back any deferred free space that has not been
2263 * added back into the in-core free tree yet. This will
2264 * ensure that we don't end up with a space map histogram
2265 * that is completely empty unless the metaslab is fully
2266 * allocated.
2267 */
2268 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
2269 space_map_histogram_add(msp->ms_sm,
2270 msp->ms_defertree[t], tx);
2271 }
2272 }
2273
2274 /*
2275 * Always add the free space from this sync pass to the space
2276 * map histogram. We want to make sure that the on-disk histogram
2277 * accounts for all free space. If the space map is not loaded,
2278 * then we will lose some accuracy but will correct it the next
2279 * time we load the space map.
2280 */
2281 space_map_histogram_add(msp->ms_sm, msp->ms_freeingtree, tx);
2282
2283 metaslab_group_histogram_add(mg, msp);
2284 metaslab_group_histogram_verify(mg);
2285 metaslab_class_histogram_verify(mg->mg_class);
2286
2287 /*
2288 * For sync pass 1, we avoid traversing this txg's free range tree
2289 * and instead will just swap the pointers for freeingtree and
2290 * freedtree. We can safely do this since the freed_tree is
2291 * guaranteed to be empty on the initial pass.
2292 */
2293 if (spa_sync_pass(spa) == 1) {
2294 range_tree_swap(&msp->ms_freeingtree, &msp->ms_freedtree);
2295 } else {
2296 range_tree_vacate(msp->ms_freeingtree,
2297 range_tree_add, msp->ms_freedtree);
2298 }
2299 range_tree_vacate(alloctree, NULL, NULL);
2300
2301 ASSERT0(range_tree_space(msp->ms_alloctree[txg & TXG_MASK]));
2302 ASSERT0(range_tree_space(msp->ms_alloctree[TXG_CLEAN(txg) & TXG_MASK]));
2303 ASSERT0(range_tree_space(msp->ms_freeingtree));
2304
2305 mutex_exit(&msp->ms_lock);
2306
2307 if (object != space_map_object(msp->ms_sm)) {
2308 object = space_map_object(msp->ms_sm);
2309 dmu_write(mos, vd->vdev_ms_array, sizeof (uint64_t) *
2310 msp->ms_id, sizeof (uint64_t), &object, tx);
2311 }
2312 dmu_tx_commit(tx);
2313 }
2314
2315 /*
2316 * Called after a transaction group has completely synced to mark
2317 * all of the metaslab's free space as usable.
2318 */
2319 void
2320 metaslab_sync_done(metaslab_t *msp, uint64_t txg)
2321 {
2322 metaslab_group_t *mg = msp->ms_group;
2323 vdev_t *vd = mg->mg_vd;
2324 spa_t *spa = vd->vdev_spa;
2325 range_tree_t **defer_tree;
2326 int64_t alloc_delta, defer_delta;
2327 boolean_t defer_allowed = B_TRUE;
2328
2329 ASSERT(!vd->vdev_ishole);
2330
2331 mutex_enter(&msp->ms_lock);
2332
2333 /*
2334 * If this metaslab is just becoming available, initialize its
2335 * range trees and add its capacity to the vdev.
2336 */
2337 if (msp->ms_freedtree == NULL) {
2338 for (int t = 0; t < TXG_SIZE; t++) {
2339 ASSERT(msp->ms_alloctree[t] == NULL);
2340
2341 msp->ms_alloctree[t] = range_tree_create(NULL, msp,
2342 &msp->ms_lock);
2343 }
2344
2345 ASSERT3P(msp->ms_freeingtree, ==, NULL);
2346 msp->ms_freeingtree = range_tree_create(NULL, msp,
2347 &msp->ms_lock);
2348
2349 ASSERT3P(msp->ms_freedtree, ==, NULL);
2350 msp->ms_freedtree = range_tree_create(NULL, msp,
2351 &msp->ms_lock);
2352
2353 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
2354 ASSERT(msp->ms_defertree[t] == NULL);
2355
2356 msp->ms_defertree[t] = range_tree_create(NULL, msp,
2357 &msp->ms_lock);
2358 }
2359
2360 vdev_space_update(vd, 0, 0, msp->ms_size);
2361 }
2362
2363 defer_tree = &msp->ms_defertree[txg % TXG_DEFER_SIZE];
2364
2365 uint64_t free_space = metaslab_class_get_space(spa_normal_class(spa)) -
2366 metaslab_class_get_alloc(spa_normal_class(spa));
2367 if (free_space <= spa_get_slop_space(spa)) {
2368 defer_allowed = B_FALSE;
2369 }
2370
2371 defer_delta = 0;
2372 alloc_delta = space_map_alloc_delta(msp->ms_sm);
2373 if (defer_allowed) {
2374 defer_delta = range_tree_space(msp->ms_freedtree) -
2375 range_tree_space(*defer_tree);
2376 } else {
2377 defer_delta -= range_tree_space(*defer_tree);
2378 }
2379
2380 vdev_space_update(vd, alloc_delta + defer_delta, defer_delta, 0);
2381
2382 /*
2383 * If there's a metaslab_load() in progress, wait for it to complete
2384 * so that we have a consistent view of the in-core space map.
2385 */
2386 metaslab_load_wait(msp);
2387
2388 /*
2389 * Move the frees from the defer_tree back to the free
2390 * range tree (if it's loaded). Swap the freed_tree and the
2391 * defer_tree -- this is safe to do because we've just emptied out
2392 * the defer_tree.
2393 */
2394 range_tree_vacate(*defer_tree,
2395 msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree);
2396 if (defer_allowed) {
2397 range_tree_swap(&msp->ms_freedtree, defer_tree);
2398 } else {
2399 range_tree_vacate(msp->ms_freedtree,
2400 msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree);
2401 }
2402
2403 space_map_update(msp->ms_sm);
2404
2405 msp->ms_deferspace += defer_delta;
2406 ASSERT3S(msp->ms_deferspace, >=, 0);
2407 ASSERT3S(msp->ms_deferspace, <=, msp->ms_size);
2408 if (msp->ms_deferspace != 0) {
2409 /*
2410 * Keep syncing this metaslab until all deferred frees
2411 * are back in circulation.
2412 */
2413 vdev_dirty(vd, VDD_METASLAB, msp, txg + 1);
2414 }
2415
2416 /*
2417 * Calculate the new weights before unloading any metaslabs.
2418 * This will give us the most accurate weighting.
2419 */
2420 metaslab_group_sort(mg, msp, metaslab_weight(msp));
2421
2422 /*
2423 * If the metaslab is loaded and we've not tried to load or allocate
2424 * from it in 'metaslab_unload_delay' txgs, then unload it.
2425 */
2426 if (msp->ms_loaded &&
2427 msp->ms_selected_txg + metaslab_unload_delay < txg) {
2428 for (int t = 1; t < TXG_CONCURRENT_STATES; t++) {
2429 VERIFY0(range_tree_space(
2430 msp->ms_alloctree[(txg + t) & TXG_MASK]));
2431 }
2432
2433 if (!metaslab_debug_unload)
2434 metaslab_unload(msp);
2435 }
2436
2437 mutex_exit(&msp->ms_lock);
2438 }
2439
2440 void
2441 metaslab_sync_reassess(metaslab_group_t *mg)
2442 {
2443 metaslab_group_alloc_update(mg);
2444 mg->mg_fragmentation = metaslab_group_fragmentation(mg);
2445
2446 /*
2447 * Preload the next potential metaslabs
2448 */
2449 metaslab_group_preload(mg);
2450 }
2451
2452 static uint64_t
2453 metaslab_distance(metaslab_t *msp, dva_t *dva)
2454 {
2455 uint64_t ms_shift = msp->ms_group->mg_vd->vdev_ms_shift;
2456 uint64_t offset = DVA_GET_OFFSET(dva) >> ms_shift;
2457 uint64_t start = msp->ms_id;
2458
2459 if (msp->ms_group->mg_vd->vdev_id != DVA_GET_VDEV(dva))
2460 return (1ULL << 63);
2461
2462 if (offset < start)
2463 return ((start - offset) << ms_shift);
2464 if (offset > start)
2465 return ((offset - start) << ms_shift);
2466 return (0);
2467 }
2468
2469 /*
2470 * ==========================================================================
2471 * Metaslab allocation tracing facility
2472 * ==========================================================================
2473 */
2474 kstat_t *metaslab_trace_ksp;
2475 kstat_named_t metaslab_trace_over_limit;
2476
2477 void
2478 metaslab_alloc_trace_init(void)
2479 {
2480 ASSERT(metaslab_alloc_trace_cache == NULL);
2481 metaslab_alloc_trace_cache = kmem_cache_create(
2482 "metaslab_alloc_trace_cache", sizeof (metaslab_alloc_trace_t),
2483 0, NULL, NULL, NULL, NULL, NULL, 0);
2484 metaslab_trace_ksp = kstat_create("zfs", 0, "metaslab_trace_stats",
2485 "misc", KSTAT_TYPE_NAMED, 1, KSTAT_FLAG_VIRTUAL);
2486 if (metaslab_trace_ksp != NULL) {
2487 metaslab_trace_ksp->ks_data = &metaslab_trace_over_limit;
2488 kstat_named_init(&metaslab_trace_over_limit,
2489 "metaslab_trace_over_limit", KSTAT_DATA_UINT64);
2490 kstat_install(metaslab_trace_ksp);
2491 }
2492 }
2493
2494 void
2495 metaslab_alloc_trace_fini(void)
2496 {
2497 if (metaslab_trace_ksp != NULL) {
2498 kstat_delete(metaslab_trace_ksp);
2499 metaslab_trace_ksp = NULL;
2500 }
2501 kmem_cache_destroy(metaslab_alloc_trace_cache);
2502 metaslab_alloc_trace_cache = NULL;
2503 }
2504
2505 /*
2506 * Add an allocation trace element to the allocation tracing list.
2507 */
2508 static void
2509 metaslab_trace_add(zio_alloc_list_t *zal, metaslab_group_t *mg,
2510 metaslab_t *msp, uint64_t psize, uint32_t dva_id, uint64_t offset)
2511 {
2512 if (!metaslab_trace_enabled)
2513 return;
2514
2515 /*
2516 * When the tracing list reaches its maximum we remove
2517 * the second element in the list before adding a new one.
2518 * By removing the second element we preserve the original
2519 * entry as a clue to what allocations steps have already been
2520 * performed.
2521 */
2522 if (zal->zal_size == metaslab_trace_max_entries) {
2523 metaslab_alloc_trace_t *mat_next;
2524 #ifdef DEBUG
2525 panic("too many entries in allocation list");
2526 #endif
2527 atomic_inc_64(&metaslab_trace_over_limit.value.ui64);
2528 zal->zal_size--;
2529 mat_next = list_next(&zal->zal_list, list_head(&zal->zal_list));
2530 list_remove(&zal->zal_list, mat_next);
2531 kmem_cache_free(metaslab_alloc_trace_cache, mat_next);
2532 }
2533
2534 metaslab_alloc_trace_t *mat =
2535 kmem_cache_alloc(metaslab_alloc_trace_cache, KM_SLEEP);
2536 list_link_init(&mat->mat_list_node);
2537 mat->mat_mg = mg;
2538 mat->mat_msp = msp;
2539 mat->mat_size = psize;
2540 mat->mat_dva_id = dva_id;
2541 mat->mat_offset = offset;
2542 mat->mat_weight = 0;
2543
2544 if (msp != NULL)
2545 mat->mat_weight = msp->ms_weight;
2546
2547 /*
2548 * The list is part of the zio so locking is not required. Only
2549 * a single thread will perform allocations for a given zio.
2550 */
2551 list_insert_tail(&zal->zal_list, mat);
2552 zal->zal_size++;
2553
2554 ASSERT3U(zal->zal_size, <=, metaslab_trace_max_entries);
2555 }
2556
2557 void
2558 metaslab_trace_init(zio_alloc_list_t *zal)
2559 {
2560 list_create(&zal->zal_list, sizeof (metaslab_alloc_trace_t),
2561 offsetof(metaslab_alloc_trace_t, mat_list_node));
2562 zal->zal_size = 0;
2563 }
2564
2565 void
2566 metaslab_trace_fini(zio_alloc_list_t *zal)
2567 {
2568 metaslab_alloc_trace_t *mat;
2569
2570 while ((mat = list_remove_head(&zal->zal_list)) != NULL)
2571 kmem_cache_free(metaslab_alloc_trace_cache, mat);
2572 list_destroy(&zal->zal_list);
2573 zal->zal_size = 0;
2574 }
2575
2576 /*
2577 * ==========================================================================
2578 * Metaslab block operations
2579 * ==========================================================================
2580 */
2581
2582 static void
2583 metaslab_group_alloc_increment(spa_t *spa, uint64_t vdev, void *tag, int flags)
2584 {
2585 if (!(flags & METASLAB_ASYNC_ALLOC) ||
2586 flags & METASLAB_DONT_THROTTLE)
2587 return;
2588
2589 metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
2590 if (!mg->mg_class->mc_alloc_throttle_enabled)
2591 return;
2592
2593 (void) refcount_add(&mg->mg_alloc_queue_depth, tag);
2594 }
2595
2596 void
2597 metaslab_group_alloc_decrement(spa_t *spa, uint64_t vdev, void *tag, int flags)
2598 {
2599 if (!(flags & METASLAB_ASYNC_ALLOC) ||
2600 flags & METASLAB_DONT_THROTTLE)
2601 return;
2602
2603 metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
2604 if (!mg->mg_class->mc_alloc_throttle_enabled)
2605 return;
2606
2607 (void) refcount_remove(&mg->mg_alloc_queue_depth, tag);
2608 }
2609
2610 void
2611 metaslab_group_alloc_verify(spa_t *spa, const blkptr_t *bp, void *tag)
2612 {
2613 #ifdef ZFS_DEBUG
2614 const dva_t *dva = bp->blk_dva;
2615 int ndvas = BP_GET_NDVAS(bp);
2616
2617 for (int d = 0; d < ndvas; d++) {
2618 uint64_t vdev = DVA_GET_VDEV(&dva[d]);
2619 metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
2620 VERIFY(refcount_not_held(&mg->mg_alloc_queue_depth, tag));
2621 }
2622 #endif
2623 }
2624
2625 static uint64_t
2626 metaslab_block_alloc(metaslab_t *msp, uint64_t size, uint64_t txg)
2627 {
2628 uint64_t start;
2629 range_tree_t *rt = msp->ms_tree;
2630 metaslab_class_t *mc = msp->ms_group->mg_class;
2631
2632 VERIFY(!msp->ms_condensing);
2633
2634 start = mc->mc_ops->msop_alloc(msp, size);
2635 if (start != -1ULL) {
2636 metaslab_group_t *mg = msp->ms_group;
2637 vdev_t *vd = mg->mg_vd;
2638
2639 VERIFY0(P2PHASE(start, 1ULL << vd->vdev_ashift));
2640 VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
2641 VERIFY3U(range_tree_space(rt) - size, <=, msp->ms_size);
2642 range_tree_remove(rt, start, size);
2643
2644 if (range_tree_space(msp->ms_alloctree[txg & TXG_MASK]) == 0)
2645 vdev_dirty(mg->mg_vd, VDD_METASLAB, msp, txg);
2646
2647 range_tree_add(msp->ms_alloctree[txg & TXG_MASK], start, size);
2648
2649 /* Track the last successful allocation */
2650 msp->ms_alloc_txg = txg;
2651 metaslab_verify_space(msp, txg);
2652 }
2653
2654 /*
2655 * Now that we've attempted the allocation we need to update the
2656 * metaslab's maximum block size since it may have changed.
2657 */
2658 msp->ms_max_size = metaslab_block_maxsize(msp);
2659 return (start);
2660 }
2661
2662 static uint64_t
2663 metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal,
2664 uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d)
2665 {
2666 metaslab_t *msp = NULL;
2667 uint64_t offset = -1ULL;
2668 uint64_t activation_weight;
2669 uint64_t target_distance;
2670 int i;
2671
2672 activation_weight = METASLAB_WEIGHT_PRIMARY;
2673 for (i = 0; i < d; i++) {
2674 if (DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) {
2675 activation_weight = METASLAB_WEIGHT_SECONDARY;
2676 break;
2677 }
2678 }
2679
2680 metaslab_t *search = kmem_alloc(sizeof (*search), KM_SLEEP);
2681 search->ms_weight = UINT64_MAX;
2682 search->ms_start = 0;
2683 for (;;) {
2684 boolean_t was_active;
2685 avl_tree_t *t = &mg->mg_metaslab_tree;
2686 avl_index_t idx;
2687
2688 mutex_enter(&mg->mg_lock);
2689
2690 /*
2691 * Find the metaslab with the highest weight that is less
2692 * than what we've already tried. In the common case, this
2693 * means that we will examine each metaslab at most once.
2694 * Note that concurrent callers could reorder metaslabs
2695 * by activation/passivation once we have dropped the mg_lock.
2696 * If a metaslab is activated by another thread, and we fail
2697 * to allocate from the metaslab we have selected, we may
2698 * not try the newly-activated metaslab, and instead activate
2699 * another metaslab. This is not optimal, but generally
2700 * does not cause any problems (a possible exception being
2701 * if every metaslab is completely full except for the
2702 * the newly-activated metaslab which we fail to examine).
2703 */
2704 msp = avl_find(t, search, &idx);
2705 if (msp == NULL)
2706 msp = avl_nearest(t, idx, AVL_AFTER);
2707 for (; msp != NULL; msp = AVL_NEXT(t, msp)) {
2708
2709 if (!metaslab_should_allocate(msp, asize)) {
2710 metaslab_trace_add(zal, mg, msp, asize, d,
2711 TRACE_TOO_SMALL);
2712 continue;
2713 }
2714
2715 /*
2716 * If the selected metaslab is condensing, skip it.
2717 */
2718 if (msp->ms_condensing)
2719 continue;
2720
2721 was_active = msp->ms_weight & METASLAB_ACTIVE_MASK;
2722 if (activation_weight == METASLAB_WEIGHT_PRIMARY)
2723 break;
2724
2725 target_distance = min_distance +
2726 (space_map_allocated(msp->ms_sm) != 0 ? 0 :
2727 min_distance >> 1);
2728
2729 for (i = 0; i < d; i++) {
2730 if (metaslab_distance(msp, &dva[i]) <
2731 target_distance)
2732 break;
2733 }
2734 if (i == d)
2735 break;
2736 }
2737 mutex_exit(&mg->mg_lock);
2738 if (msp == NULL) {
2739 kmem_free(search, sizeof (*search));
2740 return (-1ULL);
2741 }
2742 search->ms_weight = msp->ms_weight;
2743 search->ms_start = msp->ms_start + 1;
2744
2745 mutex_enter(&msp->ms_lock);
2746
2747 /*
2748 * Ensure that the metaslab we have selected is still
2749 * capable of handling our request. It's possible that
2750 * another thread may have changed the weight while we
2751 * were blocked on the metaslab lock. We check the
2752 * active status first to see if we need to reselect
2753 * a new metaslab.
2754 */
2755 if (was_active && !(msp->ms_weight & METASLAB_ACTIVE_MASK)) {
2756 mutex_exit(&msp->ms_lock);
2757 continue;
2758 }
2759
2760 if ((msp->ms_weight & METASLAB_WEIGHT_SECONDARY) &&
2761 activation_weight == METASLAB_WEIGHT_PRIMARY) {
2762 metaslab_passivate(msp,
2763 msp->ms_weight & ~METASLAB_ACTIVE_MASK);
2764 mutex_exit(&msp->ms_lock);
2765 continue;
2766 }
2767
2768 if (metaslab_activate(msp, activation_weight) != 0) {
2769 mutex_exit(&msp->ms_lock);
2770 continue;
2771 }
2772 msp->ms_selected_txg = txg;
2773
2774 /*
2775 * Now that we have the lock, recheck to see if we should
2776 * continue to use this metaslab for this allocation. The
2777 * the metaslab is now loaded so metaslab_should_allocate() can
2778 * accurately determine if the allocation attempt should
2779 * proceed.
2780 */
2781 if (!metaslab_should_allocate(msp, asize)) {
2782 /* Passivate this metaslab and select a new one. */
2783 metaslab_trace_add(zal, mg, msp, asize, d,
2784 TRACE_TOO_SMALL);
2785 goto next;
2786 }
2787
2788 /*
2789 * If this metaslab is currently condensing then pick again as
2790 * we can't manipulate this metaslab until it's committed
2791 * to disk.
2792 */
2793 if (msp->ms_condensing) {
2794 metaslab_trace_add(zal, mg, msp, asize, d,
2795 TRACE_CONDENSING);
2796 mutex_exit(&msp->ms_lock);
2797 continue;
2798 }
2799
2800 offset = metaslab_block_alloc(msp, asize, txg);
2801 metaslab_trace_add(zal, mg, msp, asize, d, offset);
2802
2803 if (offset != -1ULL) {
2804 /* Proactively passivate the metaslab, if needed */
2805 metaslab_segment_may_passivate(msp);
2806 break;
2807 }
2808 next:
2809 ASSERT(msp->ms_loaded);
2810
2811 /*
2812 * We were unable to allocate from this metaslab so determine
2813 * a new weight for this metaslab. Now that we have loaded
2814 * the metaslab we can provide a better hint to the metaslab
2815 * selector.
2816 *
2817 * For space-based metaslabs, we use the maximum block size.
2818 * This information is only available when the metaslab
2819 * is loaded and is more accurate than the generic free
2820 * space weight that was calculated by metaslab_weight().
2821 * This information allows us to quickly compare the maximum
2822 * available allocation in the metaslab to the allocation
2823 * size being requested.
2824 *
2825 * For segment-based metaslabs, determine the new weight
2826 * based on the highest bucket in the range tree. We
2827 * explicitly use the loaded segment weight (i.e. the range
2828 * tree histogram) since it contains the space that is
2829 * currently available for allocation and is accurate
2830 * even within a sync pass.
2831 */
2832 if (WEIGHT_IS_SPACEBASED(msp->ms_weight)) {
2833 uint64_t weight = metaslab_block_maxsize(msp);
2834 WEIGHT_SET_SPACEBASED(weight);
2835 metaslab_passivate(msp, weight);
2836 } else {
2837 metaslab_passivate(msp,
2838 metaslab_weight_from_range_tree(msp));
2839 }
2840
2841 /*
2842 * We have just failed an allocation attempt, check
2843 * that metaslab_should_allocate() agrees. Otherwise,
2844 * we may end up in an infinite loop retrying the same
2845 * metaslab.
2846 */
2847 ASSERT(!metaslab_should_allocate(msp, asize));
2848 mutex_exit(&msp->ms_lock);
2849 }
2850 mutex_exit(&msp->ms_lock);
2851 kmem_free(search, sizeof (*search));
2852 return (offset);
2853 }
2854
2855 static uint64_t
2856 metaslab_group_alloc(metaslab_group_t *mg, zio_alloc_list_t *zal,
2857 uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d)
2858 {
2859 uint64_t offset;
2860 ASSERT(mg->mg_initialized);
2861
2862 offset = metaslab_group_alloc_normal(mg, zal, asize, txg,
2863 min_distance, dva, d);
2864
2865 mutex_enter(&mg->mg_lock);
2866 if (offset == -1ULL) {
2867 mg->mg_failed_allocations++;
2868 metaslab_trace_add(zal, mg, NULL, asize, d,
2869 TRACE_GROUP_FAILURE);
2870 if (asize == SPA_GANGBLOCKSIZE) {
2871 /*
2872 * This metaslab group was unable to allocate
2873 * the minimum gang block size so it must be out of
2874 * space. We must notify the allocation throttle
2875 * to start skipping allocation attempts to this
2876 * metaslab group until more space becomes available.
2877 * Note: this failure cannot be caused by the
2878 * allocation throttle since the allocation throttle
2879 * is only responsible for skipping devices and
2880 * not failing block allocations.
2881 */
2882 mg->mg_no_free_space = B_TRUE;
2883 }
2884 }
2885 mg->mg_allocations++;
2886 mutex_exit(&mg->mg_lock);
2887 return (offset);
2888 }
2889
2890 /*
2891 * If we have to write a ditto block (i.e. more than one DVA for a given BP)
2892 * on the same vdev as an existing DVA of this BP, then try to allocate it
2893 * at least (vdev_asize / (2 ^ ditto_same_vdev_distance_shift)) away from the
2894 * existing DVAs.
2895 */
2896 int ditto_same_vdev_distance_shift = 3;
2897
2898 /*
2899 * Allocate a block for the specified i/o.
2900 */
2901 static int
2902 metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
2903 dva_t *dva, int d, dva_t *hintdva, uint64_t txg, int flags,
2904 zio_alloc_list_t *zal)
2905 {
2906 metaslab_group_t *mg, *rotor;
2907 vdev_t *vd;
2908 boolean_t try_hard = B_FALSE;
2909
2910 ASSERT(!DVA_IS_VALID(&dva[d]));
2911
2912 /*
2913 * For testing, make some blocks above a certain size be gang blocks.
2914 */
2915 if (psize >= metaslab_gang_bang && (ddi_get_lbolt() & 3) == 0) {
2916 metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG);
2917 return (SET_ERROR(ENOSPC));
2918 }
2919
2920 /*
2921 * Start at the rotor and loop through all mgs until we find something.
2922 * Note that there's no locking on mc_rotor or mc_aliquot because
2923 * nothing actually breaks if we miss a few updates -- we just won't
2924 * allocate quite as evenly. It all balances out over time.
2925 *
2926 * If we are doing ditto or log blocks, try to spread them across
2927 * consecutive vdevs. If we're forced to reuse a vdev before we've
2928 * allocated all of our ditto blocks, then try and spread them out on
2929 * that vdev as much as possible. If it turns out to not be possible,
2930 * gradually lower our standards until anything becomes acceptable.
2931 * Also, allocating on consecutive vdevs (as opposed to random vdevs)
2932 * gives us hope of containing our fault domains to something we're
2933 * able to reason about. Otherwise, any two top-level vdev failures
2934 * will guarantee the loss of data. With consecutive allocation,
2935 * only two adjacent top-level vdev failures will result in data loss.
2936 *
2937 * If we are doing gang blocks (hintdva is non-NULL), try to keep
2938 * ourselves on the same vdev as our gang block header. That
2939 * way, we can hope for locality in vdev_cache, plus it makes our
2940 * fault domains something tractable.
2941 */
2942 if (hintdva) {
2943 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&hintdva[d]));
2944
2945 /*
2946 * It's possible the vdev we're using as the hint no
2947 * longer exists (i.e. removed). Consult the rotor when
2948 * all else fails.
2949 */
2950 if (vd != NULL) {
2951 mg = vd->vdev_mg;
2952
2953 if (flags & METASLAB_HINTBP_AVOID &&
2954 mg->mg_next != NULL)
2955 mg = mg->mg_next;
2956 } else {
2957 mg = mc->mc_rotor;
2958 }
2959 } else if (d != 0) {
2960 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d - 1]));
2961 mg = vd->vdev_mg->mg_next;
2962 } else {
2963 mg = mc->mc_rotor;
2964 }
2965
2966 /*
2967 * If the hint put us into the wrong metaslab class, or into a
2968 * metaslab group that has been passivated, just follow the rotor.
2969 */
2970 if (mg->mg_class != mc || mg->mg_activation_count <= 0)
2971 mg = mc->mc_rotor;
2972
2973 rotor = mg;
2974 top:
2975 do {
2976 boolean_t allocatable;
2977
2978 ASSERT(mg->mg_activation_count == 1);
2979 vd = mg->mg_vd;
2980
2981 /*
2982 * Don't allocate from faulted devices.
2983 */
2984 if (try_hard) {
2985 spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER);
2986 allocatable = vdev_allocatable(vd);
2987 spa_config_exit(spa, SCL_ZIO, FTAG);
2988 } else {
2989 allocatable = vdev_allocatable(vd);
2990 }
2991
2992 /*
2993 * Determine if the selected metaslab group is eligible
2994 * for allocations. If we're ganging then don't allow
2995 * this metaslab group to skip allocations since that would
2996 * inadvertently return ENOSPC and suspend the pool
2997 * even though space is still available.
2998 */
2999 if (allocatable && !GANG_ALLOCATION(flags) && !try_hard) {
3000 allocatable = metaslab_group_allocatable(mg, rotor,
3001 psize);
3002 }
3003
3004 if (!allocatable) {
3005 metaslab_trace_add(zal, mg, NULL, psize, d,
3006 TRACE_NOT_ALLOCATABLE);
3007 goto next;
3008 }
3009
3010 ASSERT(mg->mg_initialized);
3011
3012 /*
3013 * Avoid writing single-copy data to a failing,
3014 * non-redundant vdev, unless we've already tried all
3015 * other vdevs.
3016 */
3017 if ((vd->vdev_stat.vs_write_errors > 0 ||
3018 vd->vdev_state < VDEV_STATE_HEALTHY) &&
3019 d == 0 && !try_hard && vd->vdev_children == 0) {
3020 metaslab_trace_add(zal, mg, NULL, psize, d,
3021 TRACE_VDEV_ERROR);
3022 goto next;
3023 }
3024
3025 ASSERT(mg->mg_class == mc);
3026
3027 /*
3028 * If we don't need to try hard, then require that the
3029 * block be 1/8th of the device away from any other DVAs
3030 * in this BP. If we are trying hard, allow any offset
3031 * to be used (distance=0).
3032 */
3033 uint64_t distance = 0;
3034 if (!try_hard) {
3035 distance = vd->vdev_asize >>
3036 ditto_same_vdev_distance_shift;
3037 if (distance <= (1ULL << vd->vdev_ms_shift))
3038 distance = 0;
3039 }
3040
3041 uint64_t asize = vdev_psize_to_asize(vd, psize);
3042 ASSERT(P2PHASE(asize, 1ULL << vd->vdev_ashift) == 0);
3043
3044 uint64_t offset = metaslab_group_alloc(mg, zal, asize, txg,
3045 distance, dva, d);
3046
3047 if (offset != -1ULL) {
3048 /*
3049 * If we've just selected this metaslab group,
3050 * figure out whether the corresponding vdev is
3051 * over- or under-used relative to the pool,
3052 * and set an allocation bias to even it out.
3053 */
3054 if (mc->mc_aliquot == 0 && metaslab_bias_enabled) {
3055 vdev_stat_t *vs = &vd->vdev_stat;
3056 int64_t vu, cu;
3057
3058 vu = (vs->vs_alloc * 100) / (vs->vs_space + 1);
3059 cu = (mc->mc_alloc * 100) / (mc->mc_space + 1);
3060
3061 /*
3062 * Calculate how much more or less we should
3063 * try to allocate from this device during
3064 * this iteration around the rotor.
3065 * For example, if a device is 80% full
3066 * and the pool is 20% full then we should
3067 * reduce allocations by 60% on this device.
3068 *
3069 * mg_bias = (20 - 80) * 512K / 100 = -307K
3070 *
3071 * This reduces allocations by 307K for this
3072 * iteration.
3073 */
3074 mg->mg_bias = ((cu - vu) *
3075 (int64_t)mg->mg_aliquot) / 100;
3076 } else if (!metaslab_bias_enabled) {
3077 mg->mg_bias = 0;
3078 }
3079
3080 if (atomic_add_64_nv(&mc->mc_aliquot, asize) >=
3081 mg->mg_aliquot + mg->mg_bias) {
3082 mc->mc_rotor = mg->mg_next;
3083 mc->mc_aliquot = 0;
3084 }
3085
3086 DVA_SET_VDEV(&dva[d], vd->vdev_id);
3087 DVA_SET_OFFSET(&dva[d], offset);
3088 DVA_SET_GANG(&dva[d], !!(flags & METASLAB_GANG_HEADER));
3089 DVA_SET_ASIZE(&dva[d], asize);
3090
3091 return (0);
3092 }
3093 next:
3094 mc->mc_rotor = mg->mg_next;
3095 mc->mc_aliquot = 0;
3096 } while ((mg = mg->mg_next) != rotor);
3097
3098 /*
3099 * If we haven't tried hard, do so now.
3100 */
3101 if (!try_hard) {
3102 try_hard = B_TRUE;
3103 goto top;
3104 }
3105
3106 bzero(&dva[d], sizeof (dva_t));
3107
3108 metaslab_trace_add(zal, rotor, NULL, psize, d, TRACE_ENOSPC);
3109 return (SET_ERROR(ENOSPC));
3110 }
3111
3112 /*
3113 * Free the block represented by DVA in the context of the specified
3114 * transaction group.
3115 */
3116 static void
3117 metaslab_free_dva(spa_t *spa, const dva_t *dva, uint64_t txg, boolean_t now)
3118 {
3119 uint64_t vdev = DVA_GET_VDEV(dva);
3120 uint64_t offset = DVA_GET_OFFSET(dva);
3121 uint64_t size = DVA_GET_ASIZE(dva);
3122 vdev_t *vd;
3123 metaslab_t *msp;
3124
3125 ASSERT(DVA_IS_VALID(dva));
3126
3127 if (txg > spa_freeze_txg(spa))
3128 return;
3129
3130 if ((vd = vdev_lookup_top(spa, vdev)) == NULL ||
3131 (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count) {
3132 cmn_err(CE_WARN, "metaslab_free_dva(): bad DVA %llu:%llu",
3133 (u_longlong_t)vdev, (u_longlong_t)offset);
3134 ASSERT(0);
3135 return;
3136 }
3137
3138 msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
3139
3140 if (DVA_GET_GANG(dva))
3141 size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE);
3142
3143 mutex_enter(&msp->ms_lock);
3144
3145 if (now) {
3146 range_tree_remove(msp->ms_alloctree[txg & TXG_MASK],
3147 offset, size);
3148
3149 VERIFY(!msp->ms_condensing);
3150 VERIFY3U(offset, >=, msp->ms_start);
3151 VERIFY3U(offset + size, <=, msp->ms_start + msp->ms_size);
3152 VERIFY3U(range_tree_space(msp->ms_tree) + size, <=,
3153 msp->ms_size);
3154 VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
3155 VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
3156 range_tree_add(msp->ms_tree, offset, size);
3157 msp->ms_max_size = metaslab_block_maxsize(msp);
3158 } else {
3159 VERIFY3U(txg, ==, spa->spa_syncing_txg);
3160 if (range_tree_space(msp->ms_freeingtree) == 0)
3161 vdev_dirty(vd, VDD_METASLAB, msp, txg);
3162 range_tree_add(msp->ms_freeingtree, offset, size);
3163 }
3164
3165 mutex_exit(&msp->ms_lock);
3166 }
3167
3168 /*
3169 * Intent log support: upon opening the pool after a crash, notify the SPA
3170 * of blocks that the intent log has allocated for immediate write, but
3171 * which are still considered free by the SPA because the last transaction
3172 * group didn't commit yet.
3173 */
3174 static int
3175 metaslab_claim_dva(spa_t *spa, const dva_t *dva, uint64_t txg)
3176 {
3177 uint64_t vdev = DVA_GET_VDEV(dva);
3178 uint64_t offset = DVA_GET_OFFSET(dva);
3179 uint64_t size = DVA_GET_ASIZE(dva);
3180 vdev_t *vd;
3181 metaslab_t *msp;
3182 int error = 0;
3183
3184 ASSERT(DVA_IS_VALID(dva));
3185
3186 if ((vd = vdev_lookup_top(spa, vdev)) == NULL ||
3187 (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count)
3188 return (SET_ERROR(ENXIO));
3189
3190 msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
3191
3192 if (DVA_GET_GANG(dva))
3193 size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE);
3194
3195 mutex_enter(&msp->ms_lock);
3196
3197 if ((txg != 0 && spa_writeable(spa)) || !msp->ms_loaded)
3198 error = metaslab_activate(msp, METASLAB_WEIGHT_SECONDARY);
3199
3200 if (error == 0 && !range_tree_contains(msp->ms_tree, offset, size))
3201 error = SET_ERROR(ENOENT);
3202
3203 if (error || txg == 0) { /* txg == 0 indicates dry run */
3204 mutex_exit(&msp->ms_lock);
3205 return (error);
3206 }
3207
3208 VERIFY(!msp->ms_condensing);
3209 VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
3210 VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
3211 VERIFY3U(range_tree_space(msp->ms_tree) - size, <=, msp->ms_size);
3212 range_tree_remove(msp->ms_tree, offset, size);
3213
3214 if (spa_writeable(spa)) { /* don't dirty if we're zdb(1M) */
3215 if (range_tree_space(msp->ms_alloctree[txg & TXG_MASK]) == 0)
3216 vdev_dirty(vd, VDD_METASLAB, msp, txg);
3217 range_tree_add(msp->ms_alloctree[txg & TXG_MASK], offset, size);
3218 }
3219
3220 mutex_exit(&msp->ms_lock);
3221
3222 return (0);
3223 }
3224
3225 /*
3226 * Reserve some allocation slots. The reservation system must be called
3227 * before we call into the allocator. If there aren't any available slots
3228 * then the I/O will be throttled until an I/O completes and its slots are
3229 * freed up. The function returns true if it was successful in placing
3230 * the reservation.
3231 */
3232 boolean_t
3233 metaslab_class_throttle_reserve(metaslab_class_t *mc, int slots, zio_t *zio,
3234 int flags)
3235 {
3236 uint64_t available_slots = 0;
3237 boolean_t slot_reserved = B_FALSE;
3238
3239 ASSERT(mc->mc_alloc_throttle_enabled);
3240 mutex_enter(&mc->mc_lock);
3241
3242 uint64_t reserved_slots = refcount_count(&mc->mc_alloc_slots);
3243 if (reserved_slots < mc->mc_alloc_max_slots)
3244 available_slots = mc->mc_alloc_max_slots - reserved_slots;
3245
3246 if (slots <= available_slots || GANG_ALLOCATION(flags)) {
3247 /*
3248 * We reserve the slots individually so that we can unreserve
3249 * them individually when an I/O completes.
3250 */
3251 for (int d = 0; d < slots; d++) {
3252 reserved_slots = refcount_add(&mc->mc_alloc_slots, zio);
3253 }
3254 zio->io_flags |= ZIO_FLAG_IO_ALLOCATING;
3255 slot_reserved = B_TRUE;
3256 }
3257
3258 mutex_exit(&mc->mc_lock);
3259 return (slot_reserved);
3260 }
3261
3262 void
3263 metaslab_class_throttle_unreserve(metaslab_class_t *mc, int slots, zio_t *zio)
3264 {
3265 ASSERT(mc->mc_alloc_throttle_enabled);
3266 mutex_enter(&mc->mc_lock);
3267 for (int d = 0; d < slots; d++) {
3268 (void) refcount_remove(&mc->mc_alloc_slots, zio);
3269 }
3270 mutex_exit(&mc->mc_lock);
3271 }
3272
3273 int
3274 metaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize, blkptr_t *bp,
3275 int ndvas, uint64_t txg, blkptr_t *hintbp, int flags,
3276 zio_alloc_list_t *zal, zio_t *zio)
3277 {
3278 dva_t *dva = bp->blk_dva;
3279 dva_t *hintdva = hintbp->blk_dva;
3280 int error = 0;
3281
3282 ASSERT(bp->blk_birth == 0);
3283 ASSERT(BP_PHYSICAL_BIRTH(bp) == 0);
3284
3285 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
3286
3287 if (mc->mc_rotor == NULL) { /* no vdevs in this class */
3288 spa_config_exit(spa, SCL_ALLOC, FTAG);
3289 return (SET_ERROR(ENOSPC));
3290 }
3291
3292 ASSERT(ndvas > 0 && ndvas <= spa_max_replication(spa));
3293 ASSERT(BP_GET_NDVAS(bp) == 0);
3294 ASSERT(hintbp == NULL || ndvas <= BP_GET_NDVAS(hintbp));
3295 ASSERT3P(zal, !=, NULL);
3296
3297 for (int d = 0; d < ndvas; d++) {
3298 error = metaslab_alloc_dva(spa, mc, psize, dva, d, hintdva,
3299 txg, flags, zal);
3300 if (error != 0) {
3301 for (d--; d >= 0; d--) {
3302 metaslab_free_dva(spa, &dva[d], txg, B_TRUE);
3303 metaslab_group_alloc_decrement(spa,
3304 DVA_GET_VDEV(&dva[d]), zio, flags);
3305 bzero(&dva[d], sizeof (dva_t));
3306 }
3307 spa_config_exit(spa, SCL_ALLOC, FTAG);
3308 return (error);
3309 } else {
3310 /*
3311 * Update the metaslab group's queue depth
3312 * based on the newly allocated dva.
3313 */
3314 metaslab_group_alloc_increment(spa,
3315 DVA_GET_VDEV(&dva[d]), zio, flags);
3316 }
3317
3318 }
3319 ASSERT(error == 0);
3320 ASSERT(BP_GET_NDVAS(bp) == ndvas);
3321
3322 spa_config_exit(spa, SCL_ALLOC, FTAG);
3323
3324 BP_SET_BIRTH(bp, txg, txg);
3325
3326 return (0);
3327 }
3328
3329 void
3330 metaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg, boolean_t now)
3331 {
3332 const dva_t *dva = bp->blk_dva;
3333 int ndvas = BP_GET_NDVAS(bp);
3334
3335 ASSERT(!BP_IS_HOLE(bp));
3336 ASSERT(!now || bp->blk_birth >= spa_syncing_txg(spa));
3337
3338 spa_config_enter(spa, SCL_FREE, FTAG, RW_READER);
3339
3340 for (int d = 0; d < ndvas; d++)
3341 metaslab_free_dva(spa, &dva[d], txg, now);
3342
3343 spa_config_exit(spa, SCL_FREE, FTAG);
3344 }
3345
3346 int
3347 metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg)
3348 {
3349 const dva_t *dva = bp->blk_dva;
3350 int ndvas = BP_GET_NDVAS(bp);
3351 int error = 0;
3352
3353 ASSERT(!BP_IS_HOLE(bp));
3354
3355 if (txg != 0) {
3356 /*
3357 * First do a dry run to make sure all DVAs are claimable,
3358 * so we don't have to unwind from partial failures below.
3359 */
3360 if ((error = metaslab_claim(spa, bp, 0)) != 0)
3361 return (error);
3362 }
3363
3364 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
3365
3366 for (int d = 0; d < ndvas; d++)
3367 if ((error = metaslab_claim_dva(spa, &dva[d], txg)) != 0)
3368 break;
3369
3370 spa_config_exit(spa, SCL_ALLOC, FTAG);
3371
3372 ASSERT(error == 0 || txg == 0);
3373
3374 return (error);
3375 }
3376
3377 void
3378 metaslab_check_free(spa_t *spa, const blkptr_t *bp)
3379 {
3380 if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0)
3381 return;
3382
3383 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3384 for (int i = 0; i < BP_GET_NDVAS(bp); i++) {
3385 uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[i]);
3386 vdev_t *vd = vdev_lookup_top(spa, vdev);
3387 uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
3388 uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]);
3389 metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
3390
3391 if (msp->ms_loaded)
3392 range_tree_verify(msp->ms_tree, offset, size);
3393
3394 range_tree_verify(msp->ms_freeingtree, offset, size);
3395 range_tree_verify(msp->ms_freedtree, offset, size);
3396 for (int j = 0; j < TXG_DEFER_SIZE; j++)
3397 range_tree_verify(msp->ms_defertree[j], offset, size);
3398 }
3399 spa_config_exit(spa, SCL_VDEV, FTAG);
3400 }