Print this page
7938 Port ZOL #3712 disable LBA weighting on files and SSDs
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/metaslab.c
+++ new/usr/src/uts/common/fs/zfs/metaslab.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24 24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 25 * Copyright (c) 2014 Integros [integros.com]
26 26 */
27 27
28 28 #include <sys/zfs_context.h>
29 29 #include <sys/dmu.h>
30 30 #include <sys/dmu_tx.h>
31 31 #include <sys/space_map.h>
32 32 #include <sys/metaslab_impl.h>
33 33 #include <sys/vdev_impl.h>
34 34 #include <sys/zio.h>
35 35 #include <sys/spa_impl.h>
36 36 #include <sys/zfeature.h>
37 37
38 38 #define GANG_ALLOCATION(flags) \
39 39 ((flags) & (METASLAB_GANG_CHILD | METASLAB_GANG_HEADER))
40 40
41 41 uint64_t metaslab_aliquot = 512ULL << 10;
42 42 uint64_t metaslab_gang_bang = SPA_MAXBLOCKSIZE + 1; /* force gang blocks */
43 43
44 44 /*
45 45 * The in-core space map representation is more compact than its on-disk form.
46 46 * The zfs_condense_pct determines how much more compact the in-core
47 47 * space map representation must be before we compact it on-disk.
48 48 * Values should be greater than or equal to 100.
49 49 */
50 50 int zfs_condense_pct = 200;
51 51
52 52 /*
53 53 * Condensing a metaslab is not guaranteed to actually reduce the amount of
54 54 * space used on disk. In particular, a space map uses data in increments of
55 55 * MAX(1 << ashift, space_map_blksize), so a metaslab might use the
56 56 * same number of blocks after condensing. Since the goal of condensing is to
57 57 * reduce the number of IOPs required to read the space map, we only want to
58 58 * condense when we can be sure we will reduce the number of blocks used by the
59 59 * space map. Unfortunately, we cannot precisely compute whether or not this is
60 60 * the case in metaslab_should_condense since we are holding ms_lock. Instead,
61 61 * we apply the following heuristic: do not condense a spacemap unless the
62 62 * uncondensed size consumes greater than zfs_metaslab_condense_block_threshold
63 63 * blocks.
64 64 */
65 65 int zfs_metaslab_condense_block_threshold = 4;
66 66
67 67 /*
68 68 * The zfs_mg_noalloc_threshold defines which metaslab groups should
69 69 * be eligible for allocation. The value is defined as a percentage of
70 70 * free space. Metaslab groups that have more free space than
71 71 * zfs_mg_noalloc_threshold are always eligible for allocations. Once
72 72 * a metaslab group's free space is less than or equal to the
73 73 * zfs_mg_noalloc_threshold the allocator will avoid allocating to that
74 74 * group unless all groups in the pool have reached zfs_mg_noalloc_threshold.
75 75 * Once all groups in the pool reach zfs_mg_noalloc_threshold then all
76 76 * groups are allowed to accept allocations. Gang blocks are always
77 77 * eligible to allocate on any metaslab group. The default value of 0 means
78 78 * no metaslab group will be excluded based on this criterion.
79 79 */
80 80 int zfs_mg_noalloc_threshold = 0;
81 81
82 82 /*
83 83 * Metaslab groups are considered eligible for allocations if their
84 84 * fragmenation metric (measured as a percentage) is less than or equal to
85 85 * zfs_mg_fragmentation_threshold. If a metaslab group exceeds this threshold
86 86 * then it will be skipped unless all metaslab groups within the metaslab
87 87 * class have also crossed this threshold.
88 88 */
89 89 int zfs_mg_fragmentation_threshold = 85;
90 90
91 91 /*
92 92 * Allow metaslabs to keep their active state as long as their fragmentation
93 93 * percentage is less than or equal to zfs_metaslab_fragmentation_threshold. An
94 94 * active metaslab that exceeds this threshold will no longer keep its active
95 95 * status allowing better metaslabs to be selected.
96 96 */
97 97 int zfs_metaslab_fragmentation_threshold = 70;
98 98
99 99 /*
100 100 * When set will load all metaslabs when pool is first opened.
101 101 */
102 102 int metaslab_debug_load = 0;
103 103
104 104 /*
105 105 * When set will prevent metaslabs from being unloaded.
106 106 */
107 107 int metaslab_debug_unload = 0;
108 108
109 109 /*
110 110 * Minimum size which forces the dynamic allocator to change
111 111 * it's allocation strategy. Once the space map cannot satisfy
112 112 * an allocation of this size then it switches to using more
113 113 * aggressive strategy (i.e search by size rather than offset).
114 114 */
115 115 uint64_t metaslab_df_alloc_threshold = SPA_OLD_MAXBLOCKSIZE;
116 116
117 117 /*
118 118 * The minimum free space, in percent, which must be available
119 119 * in a space map to continue allocations in a first-fit fashion.
120 120 * Once the space map's free space drops below this level we dynamically
121 121 * switch to using best-fit allocations.
122 122 */
123 123 int metaslab_df_free_pct = 4;
124 124
125 125 /*
126 126 * A metaslab is considered "free" if it contains a contiguous
127 127 * segment which is greater than metaslab_min_alloc_size.
128 128 */
129 129 uint64_t metaslab_min_alloc_size = DMU_MAX_ACCESS;
130 130
131 131 /*
132 132 * Percentage of all cpus that can be used by the metaslab taskq.
133 133 */
134 134 int metaslab_load_pct = 50;
135 135
136 136 /*
137 137 * Determines how many txgs a metaslab may remain loaded without having any
138 138 * allocations from it. As long as a metaslab continues to be used we will
139 139 * keep it loaded.
140 140 */
141 141 int metaslab_unload_delay = TXG_SIZE * 2;
142 142
143 143 /*
144 144 * Max number of metaslabs per group to preload.
145 145 */
146 146 int metaslab_preload_limit = SPA_DVAS_PER_BP;
147 147
148 148 /*
149 149 * Enable/disable preloading of metaslab.
150 150 */
151 151 boolean_t metaslab_preload_enabled = B_TRUE;
152 152
153 153 /*
154 154 * Enable/disable fragmentation weighting on metaslabs.
155 155 */
156 156 boolean_t metaslab_fragmentation_factor_enabled = B_TRUE;
157 157
158 158 /*
159 159 * Enable/disable lba weighting (i.e. outer tracks are given preference).
160 160 */
161 161 boolean_t metaslab_lba_weighting_enabled = B_TRUE;
162 162
163 163 /*
164 164 * Enable/disable metaslab group biasing.
165 165 */
166 166 boolean_t metaslab_bias_enabled = B_TRUE;
167 167
168 168 /*
169 169 * Enable/disable segment-based metaslab selection.
170 170 */
171 171 boolean_t zfs_metaslab_segment_weight_enabled = B_TRUE;
172 172
173 173 /*
174 174 * When using segment-based metaslab selection, we will continue
175 175 * allocating from the active metaslab until we have exhausted
176 176 * zfs_metaslab_switch_threshold of its buckets.
177 177 */
178 178 int zfs_metaslab_switch_threshold = 2;
179 179
180 180 /*
181 181 * Internal switch to enable/disable the metaslab allocation tracing
182 182 * facility.
183 183 */
184 184 boolean_t metaslab_trace_enabled = B_TRUE;
185 185
186 186 /*
187 187 * Maximum entries that the metaslab allocation tracing facility will keep
188 188 * in a given list when running in non-debug mode. We limit the number
189 189 * of entries in non-debug mode to prevent us from using up too much memory.
190 190 * The limit should be sufficiently large that we don't expect any allocation
191 191 * to every exceed this value. In debug mode, the system will panic if this
192 192 * limit is ever reached allowing for further investigation.
193 193 */
194 194 uint64_t metaslab_trace_max_entries = 5000;
195 195
196 196 static uint64_t metaslab_weight(metaslab_t *);
197 197 static void metaslab_set_fragmentation(metaslab_t *);
198 198
199 199 kmem_cache_t *metaslab_alloc_trace_cache;
200 200
201 201 /*
202 202 * ==========================================================================
203 203 * Metaslab classes
204 204 * ==========================================================================
205 205 */
206 206 metaslab_class_t *
207 207 metaslab_class_create(spa_t *spa, metaslab_ops_t *ops)
208 208 {
209 209 metaslab_class_t *mc;
210 210
211 211 mc = kmem_zalloc(sizeof (metaslab_class_t), KM_SLEEP);
212 212
213 213 mc->mc_spa = spa;
214 214 mc->mc_rotor = NULL;
215 215 mc->mc_ops = ops;
216 216 mutex_init(&mc->mc_lock, NULL, MUTEX_DEFAULT, NULL);
217 217 refcount_create_tracked(&mc->mc_alloc_slots);
218 218
219 219 return (mc);
220 220 }
221 221
222 222 void
223 223 metaslab_class_destroy(metaslab_class_t *mc)
224 224 {
225 225 ASSERT(mc->mc_rotor == NULL);
226 226 ASSERT(mc->mc_alloc == 0);
227 227 ASSERT(mc->mc_deferred == 0);
228 228 ASSERT(mc->mc_space == 0);
229 229 ASSERT(mc->mc_dspace == 0);
230 230
231 231 refcount_destroy(&mc->mc_alloc_slots);
232 232 mutex_destroy(&mc->mc_lock);
233 233 kmem_free(mc, sizeof (metaslab_class_t));
234 234 }
235 235
236 236 int
237 237 metaslab_class_validate(metaslab_class_t *mc)
238 238 {
239 239 metaslab_group_t *mg;
240 240 vdev_t *vd;
241 241
242 242 /*
243 243 * Must hold one of the spa_config locks.
244 244 */
245 245 ASSERT(spa_config_held(mc->mc_spa, SCL_ALL, RW_READER) ||
246 246 spa_config_held(mc->mc_spa, SCL_ALL, RW_WRITER));
247 247
248 248 if ((mg = mc->mc_rotor) == NULL)
249 249 return (0);
250 250
251 251 do {
252 252 vd = mg->mg_vd;
253 253 ASSERT(vd->vdev_mg != NULL);
254 254 ASSERT3P(vd->vdev_top, ==, vd);
255 255 ASSERT3P(mg->mg_class, ==, mc);
256 256 ASSERT3P(vd->vdev_ops, !=, &vdev_hole_ops);
257 257 } while ((mg = mg->mg_next) != mc->mc_rotor);
258 258
259 259 return (0);
260 260 }
261 261
262 262 void
263 263 metaslab_class_space_update(metaslab_class_t *mc, int64_t alloc_delta,
264 264 int64_t defer_delta, int64_t space_delta, int64_t dspace_delta)
265 265 {
266 266 atomic_add_64(&mc->mc_alloc, alloc_delta);
267 267 atomic_add_64(&mc->mc_deferred, defer_delta);
268 268 atomic_add_64(&mc->mc_space, space_delta);
269 269 atomic_add_64(&mc->mc_dspace, dspace_delta);
270 270 }
271 271
272 272 uint64_t
273 273 metaslab_class_get_alloc(metaslab_class_t *mc)
274 274 {
275 275 return (mc->mc_alloc);
276 276 }
277 277
278 278 uint64_t
279 279 metaslab_class_get_deferred(metaslab_class_t *mc)
280 280 {
281 281 return (mc->mc_deferred);
282 282 }
283 283
284 284 uint64_t
285 285 metaslab_class_get_space(metaslab_class_t *mc)
286 286 {
287 287 return (mc->mc_space);
288 288 }
289 289
290 290 uint64_t
291 291 metaslab_class_get_dspace(metaslab_class_t *mc)
292 292 {
293 293 return (spa_deflate(mc->mc_spa) ? mc->mc_dspace : mc->mc_space);
294 294 }
295 295
296 296 void
297 297 metaslab_class_histogram_verify(metaslab_class_t *mc)
298 298 {
299 299 vdev_t *rvd = mc->mc_spa->spa_root_vdev;
300 300 uint64_t *mc_hist;
301 301 int i;
302 302
303 303 if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0)
304 304 return;
305 305
306 306 mc_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE,
307 307 KM_SLEEP);
308 308
309 309 for (int c = 0; c < rvd->vdev_children; c++) {
310 310 vdev_t *tvd = rvd->vdev_child[c];
311 311 metaslab_group_t *mg = tvd->vdev_mg;
312 312
313 313 /*
314 314 * Skip any holes, uninitialized top-levels, or
315 315 * vdevs that are not in this metalab class.
316 316 */
317 317 if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 ||
318 318 mg->mg_class != mc) {
319 319 continue;
320 320 }
321 321
322 322 for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
323 323 mc_hist[i] += mg->mg_histogram[i];
324 324 }
325 325
326 326 for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++)
327 327 VERIFY3U(mc_hist[i], ==, mc->mc_histogram[i]);
328 328
329 329 kmem_free(mc_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE);
330 330 }
331 331
332 332 /*
333 333 * Calculate the metaslab class's fragmentation metric. The metric
334 334 * is weighted based on the space contribution of each metaslab group.
335 335 * The return value will be a number between 0 and 100 (inclusive), or
336 336 * ZFS_FRAG_INVALID if the metric has not been set. See comment above the
337 337 * zfs_frag_table for more information about the metric.
338 338 */
339 339 uint64_t
340 340 metaslab_class_fragmentation(metaslab_class_t *mc)
341 341 {
342 342 vdev_t *rvd = mc->mc_spa->spa_root_vdev;
343 343 uint64_t fragmentation = 0;
344 344
345 345 spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER);
346 346
347 347 for (int c = 0; c < rvd->vdev_children; c++) {
348 348 vdev_t *tvd = rvd->vdev_child[c];
349 349 metaslab_group_t *mg = tvd->vdev_mg;
350 350
351 351 /*
352 352 * Skip any holes, uninitialized top-levels, or
353 353 * vdevs that are not in this metalab class.
354 354 */
355 355 if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 ||
356 356 mg->mg_class != mc) {
357 357 continue;
358 358 }
359 359
360 360 /*
361 361 * If a metaslab group does not contain a fragmentation
362 362 * metric then just bail out.
363 363 */
364 364 if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
365 365 spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
366 366 return (ZFS_FRAG_INVALID);
367 367 }
368 368
369 369 /*
370 370 * Determine how much this metaslab_group is contributing
371 371 * to the overall pool fragmentation metric.
372 372 */
373 373 fragmentation += mg->mg_fragmentation *
374 374 metaslab_group_get_space(mg);
375 375 }
376 376 fragmentation /= metaslab_class_get_space(mc);
377 377
378 378 ASSERT3U(fragmentation, <=, 100);
379 379 spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
380 380 return (fragmentation);
381 381 }
382 382
383 383 /*
384 384 * Calculate the amount of expandable space that is available in
385 385 * this metaslab class. If a device is expanded then its expandable
386 386 * space will be the amount of allocatable space that is currently not
387 387 * part of this metaslab class.
388 388 */
389 389 uint64_t
390 390 metaslab_class_expandable_space(metaslab_class_t *mc)
391 391 {
392 392 vdev_t *rvd = mc->mc_spa->spa_root_vdev;
393 393 uint64_t space = 0;
394 394
395 395 spa_config_enter(mc->mc_spa, SCL_VDEV, FTAG, RW_READER);
396 396 for (int c = 0; c < rvd->vdev_children; c++) {
397 397 vdev_t *tvd = rvd->vdev_child[c];
398 398 metaslab_group_t *mg = tvd->vdev_mg;
399 399
400 400 if (tvd->vdev_ishole || tvd->vdev_ms_shift == 0 ||
401 401 mg->mg_class != mc) {
402 402 continue;
403 403 }
404 404
405 405 /*
406 406 * Calculate if we have enough space to add additional
407 407 * metaslabs. We report the expandable space in terms
408 408 * of the metaslab size since that's the unit of expansion.
409 409 */
410 410 space += P2ALIGN(tvd->vdev_max_asize - tvd->vdev_asize,
411 411 1ULL << tvd->vdev_ms_shift);
412 412 }
413 413 spa_config_exit(mc->mc_spa, SCL_VDEV, FTAG);
414 414 return (space);
415 415 }
416 416
417 417 static int
418 418 metaslab_compare(const void *x1, const void *x2)
419 419 {
420 420 const metaslab_t *m1 = x1;
421 421 const metaslab_t *m2 = x2;
422 422
423 423 if (m1->ms_weight < m2->ms_weight)
424 424 return (1);
425 425 if (m1->ms_weight > m2->ms_weight)
426 426 return (-1);
427 427
428 428 /*
429 429 * If the weights are identical, use the offset to force uniqueness.
430 430 */
431 431 if (m1->ms_start < m2->ms_start)
432 432 return (-1);
433 433 if (m1->ms_start > m2->ms_start)
434 434 return (1);
435 435
436 436 ASSERT3P(m1, ==, m2);
437 437
438 438 return (0);
439 439 }
440 440
441 441 /*
442 442 * Verify that the space accounting on disk matches the in-core range_trees.
443 443 */
444 444 void
445 445 metaslab_verify_space(metaslab_t *msp, uint64_t txg)
446 446 {
447 447 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
448 448 uint64_t allocated = 0;
449 449 uint64_t sm_free_space, msp_free_space;
450 450
451 451 ASSERT(MUTEX_HELD(&msp->ms_lock));
452 452
453 453 if ((zfs_flags & ZFS_DEBUG_METASLAB_VERIFY) == 0)
454 454 return;
455 455
456 456 /*
457 457 * We can only verify the metaslab space when we're called
458 458 * from syncing context with a loaded metaslab that has an allocated
459 459 * space map. Calling this in non-syncing context does not
460 460 * provide a consistent view of the metaslab since we're performing
461 461 * allocations in the future.
462 462 */
463 463 if (txg != spa_syncing_txg(spa) || msp->ms_sm == NULL ||
464 464 !msp->ms_loaded)
465 465 return;
466 466
467 467 sm_free_space = msp->ms_size - space_map_allocated(msp->ms_sm) -
468 468 space_map_alloc_delta(msp->ms_sm);
469 469
470 470 /*
471 471 * Account for future allocations since we would have already
472 472 * deducted that space from the ms_freetree.
473 473 */
474 474 for (int t = 0; t < TXG_CONCURRENT_STATES; t++) {
475 475 allocated +=
476 476 range_tree_space(msp->ms_alloctree[(txg + t) & TXG_MASK]);
477 477 }
478 478
479 479 msp_free_space = range_tree_space(msp->ms_tree) + allocated +
480 480 msp->ms_deferspace + range_tree_space(msp->ms_freedtree);
481 481
482 482 VERIFY3U(sm_free_space, ==, msp_free_space);
483 483 }
484 484
485 485 /*
486 486 * ==========================================================================
487 487 * Metaslab groups
488 488 * ==========================================================================
489 489 */
490 490 /*
491 491 * Update the allocatable flag and the metaslab group's capacity.
492 492 * The allocatable flag is set to true if the capacity is below
493 493 * the zfs_mg_noalloc_threshold or has a fragmentation value that is
494 494 * greater than zfs_mg_fragmentation_threshold. If a metaslab group
495 495 * transitions from allocatable to non-allocatable or vice versa then the
496 496 * metaslab group's class is updated to reflect the transition.
497 497 */
498 498 static void
499 499 metaslab_group_alloc_update(metaslab_group_t *mg)
500 500 {
501 501 vdev_t *vd = mg->mg_vd;
502 502 metaslab_class_t *mc = mg->mg_class;
503 503 vdev_stat_t *vs = &vd->vdev_stat;
504 504 boolean_t was_allocatable;
505 505 boolean_t was_initialized;
506 506
507 507 ASSERT(vd == vd->vdev_top);
508 508
509 509 mutex_enter(&mg->mg_lock);
510 510 was_allocatable = mg->mg_allocatable;
511 511 was_initialized = mg->mg_initialized;
512 512
513 513 mg->mg_free_capacity = ((vs->vs_space - vs->vs_alloc) * 100) /
514 514 (vs->vs_space + 1);
515 515
516 516 mutex_enter(&mc->mc_lock);
517 517
518 518 /*
519 519 * If the metaslab group was just added then it won't
520 520 * have any space until we finish syncing out this txg.
521 521 * At that point we will consider it initialized and available
522 522 * for allocations. We also don't consider non-activated
523 523 * metaslab groups (e.g. vdevs that are in the middle of being removed)
524 524 * to be initialized, because they can't be used for allocation.
525 525 */
526 526 mg->mg_initialized = metaslab_group_initialized(mg);
527 527 if (!was_initialized && mg->mg_initialized) {
528 528 mc->mc_groups++;
529 529 } else if (was_initialized && !mg->mg_initialized) {
530 530 ASSERT3U(mc->mc_groups, >, 0);
531 531 mc->mc_groups--;
532 532 }
533 533 if (mg->mg_initialized)
534 534 mg->mg_no_free_space = B_FALSE;
535 535
536 536 /*
537 537 * A metaslab group is considered allocatable if it has plenty
538 538 * of free space or is not heavily fragmented. We only take
539 539 * fragmentation into account if the metaslab group has a valid
540 540 * fragmentation metric (i.e. a value between 0 and 100).
541 541 */
542 542 mg->mg_allocatable = (mg->mg_activation_count > 0 &&
543 543 mg->mg_free_capacity > zfs_mg_noalloc_threshold &&
544 544 (mg->mg_fragmentation == ZFS_FRAG_INVALID ||
545 545 mg->mg_fragmentation <= zfs_mg_fragmentation_threshold));
546 546
547 547 /*
548 548 * The mc_alloc_groups maintains a count of the number of
549 549 * groups in this metaslab class that are still above the
550 550 * zfs_mg_noalloc_threshold. This is used by the allocating
551 551 * threads to determine if they should avoid allocations to
552 552 * a given group. The allocator will avoid allocations to a group
553 553 * if that group has reached or is below the zfs_mg_noalloc_threshold
554 554 * and there are still other groups that are above the threshold.
555 555 * When a group transitions from allocatable to non-allocatable or
556 556 * vice versa we update the metaslab class to reflect that change.
557 557 * When the mc_alloc_groups value drops to 0 that means that all
558 558 * groups have reached the zfs_mg_noalloc_threshold making all groups
559 559 * eligible for allocations. This effectively means that all devices
560 560 * are balanced again.
561 561 */
562 562 if (was_allocatable && !mg->mg_allocatable)
563 563 mc->mc_alloc_groups--;
564 564 else if (!was_allocatable && mg->mg_allocatable)
565 565 mc->mc_alloc_groups++;
566 566 mutex_exit(&mc->mc_lock);
567 567
568 568 mutex_exit(&mg->mg_lock);
569 569 }
570 570
571 571 metaslab_group_t *
572 572 metaslab_group_create(metaslab_class_t *mc, vdev_t *vd)
573 573 {
574 574 metaslab_group_t *mg;
575 575
576 576 mg = kmem_zalloc(sizeof (metaslab_group_t), KM_SLEEP);
577 577 mutex_init(&mg->mg_lock, NULL, MUTEX_DEFAULT, NULL);
578 578 avl_create(&mg->mg_metaslab_tree, metaslab_compare,
579 579 sizeof (metaslab_t), offsetof(struct metaslab, ms_group_node));
580 580 mg->mg_vd = vd;
581 581 mg->mg_class = mc;
582 582 mg->mg_activation_count = 0;
583 583 mg->mg_initialized = B_FALSE;
584 584 mg->mg_no_free_space = B_TRUE;
585 585 refcount_create_tracked(&mg->mg_alloc_queue_depth);
586 586
587 587 mg->mg_taskq = taskq_create("metaslab_group_taskq", metaslab_load_pct,
588 588 minclsyspri, 10, INT_MAX, TASKQ_THREADS_CPU_PCT);
589 589
590 590 return (mg);
591 591 }
592 592
593 593 void
594 594 metaslab_group_destroy(metaslab_group_t *mg)
595 595 {
596 596 ASSERT(mg->mg_prev == NULL);
597 597 ASSERT(mg->mg_next == NULL);
598 598 /*
599 599 * We may have gone below zero with the activation count
600 600 * either because we never activated in the first place or
601 601 * because we're done, and possibly removing the vdev.
602 602 */
603 603 ASSERT(mg->mg_activation_count <= 0);
604 604
605 605 taskq_destroy(mg->mg_taskq);
606 606 avl_destroy(&mg->mg_metaslab_tree);
607 607 mutex_destroy(&mg->mg_lock);
608 608 refcount_destroy(&mg->mg_alloc_queue_depth);
609 609 kmem_free(mg, sizeof (metaslab_group_t));
610 610 }
611 611
612 612 void
613 613 metaslab_group_activate(metaslab_group_t *mg)
614 614 {
615 615 metaslab_class_t *mc = mg->mg_class;
616 616 metaslab_group_t *mgprev, *mgnext;
617 617
618 618 ASSERT(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER));
619 619
620 620 ASSERT(mc->mc_rotor != mg);
621 621 ASSERT(mg->mg_prev == NULL);
622 622 ASSERT(mg->mg_next == NULL);
623 623 ASSERT(mg->mg_activation_count <= 0);
624 624
625 625 if (++mg->mg_activation_count <= 0)
626 626 return;
627 627
628 628 mg->mg_aliquot = metaslab_aliquot * MAX(1, mg->mg_vd->vdev_children);
629 629 metaslab_group_alloc_update(mg);
630 630
631 631 if ((mgprev = mc->mc_rotor) == NULL) {
632 632 mg->mg_prev = mg;
633 633 mg->mg_next = mg;
634 634 } else {
635 635 mgnext = mgprev->mg_next;
636 636 mg->mg_prev = mgprev;
637 637 mg->mg_next = mgnext;
638 638 mgprev->mg_next = mg;
639 639 mgnext->mg_prev = mg;
640 640 }
641 641 mc->mc_rotor = mg;
642 642 }
643 643
644 644 void
645 645 metaslab_group_passivate(metaslab_group_t *mg)
646 646 {
647 647 metaslab_class_t *mc = mg->mg_class;
648 648 metaslab_group_t *mgprev, *mgnext;
649 649
650 650 ASSERT(spa_config_held(mc->mc_spa, SCL_ALLOC, RW_WRITER));
651 651
652 652 if (--mg->mg_activation_count != 0) {
653 653 ASSERT(mc->mc_rotor != mg);
654 654 ASSERT(mg->mg_prev == NULL);
655 655 ASSERT(mg->mg_next == NULL);
656 656 ASSERT(mg->mg_activation_count < 0);
657 657 return;
658 658 }
659 659
660 660 taskq_wait(mg->mg_taskq);
661 661 metaslab_group_alloc_update(mg);
662 662
663 663 mgprev = mg->mg_prev;
664 664 mgnext = mg->mg_next;
665 665
666 666 if (mg == mgnext) {
667 667 mc->mc_rotor = NULL;
668 668 } else {
669 669 mc->mc_rotor = mgnext;
670 670 mgprev->mg_next = mgnext;
671 671 mgnext->mg_prev = mgprev;
672 672 }
673 673
674 674 mg->mg_prev = NULL;
675 675 mg->mg_next = NULL;
676 676 }
677 677
678 678 boolean_t
679 679 metaslab_group_initialized(metaslab_group_t *mg)
680 680 {
681 681 vdev_t *vd = mg->mg_vd;
682 682 vdev_stat_t *vs = &vd->vdev_stat;
683 683
684 684 return (vs->vs_space != 0 && mg->mg_activation_count > 0);
685 685 }
686 686
687 687 uint64_t
688 688 metaslab_group_get_space(metaslab_group_t *mg)
689 689 {
690 690 return ((1ULL << mg->mg_vd->vdev_ms_shift) * mg->mg_vd->vdev_ms_count);
691 691 }
692 692
693 693 void
694 694 metaslab_group_histogram_verify(metaslab_group_t *mg)
695 695 {
696 696 uint64_t *mg_hist;
697 697 vdev_t *vd = mg->mg_vd;
698 698 uint64_t ashift = vd->vdev_ashift;
699 699 int i;
700 700
701 701 if ((zfs_flags & ZFS_DEBUG_HISTOGRAM_VERIFY) == 0)
702 702 return;
703 703
704 704 mg_hist = kmem_zalloc(sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE,
705 705 KM_SLEEP);
706 706
707 707 ASSERT3U(RANGE_TREE_HISTOGRAM_SIZE, >=,
708 708 SPACE_MAP_HISTOGRAM_SIZE + ashift);
709 709
710 710 for (int m = 0; m < vd->vdev_ms_count; m++) {
711 711 metaslab_t *msp = vd->vdev_ms[m];
712 712
713 713 if (msp->ms_sm == NULL)
714 714 continue;
715 715
716 716 for (i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++)
717 717 mg_hist[i + ashift] +=
718 718 msp->ms_sm->sm_phys->smp_histogram[i];
719 719 }
720 720
721 721 for (i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i ++)
722 722 VERIFY3U(mg_hist[i], ==, mg->mg_histogram[i]);
723 723
724 724 kmem_free(mg_hist, sizeof (uint64_t) * RANGE_TREE_HISTOGRAM_SIZE);
725 725 }
726 726
727 727 static void
728 728 metaslab_group_histogram_add(metaslab_group_t *mg, metaslab_t *msp)
729 729 {
730 730 metaslab_class_t *mc = mg->mg_class;
731 731 uint64_t ashift = mg->mg_vd->vdev_ashift;
732 732
733 733 ASSERT(MUTEX_HELD(&msp->ms_lock));
734 734 if (msp->ms_sm == NULL)
735 735 return;
736 736
737 737 mutex_enter(&mg->mg_lock);
738 738 for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
739 739 mg->mg_histogram[i + ashift] +=
740 740 msp->ms_sm->sm_phys->smp_histogram[i];
741 741 mc->mc_histogram[i + ashift] +=
742 742 msp->ms_sm->sm_phys->smp_histogram[i];
743 743 }
744 744 mutex_exit(&mg->mg_lock);
745 745 }
746 746
747 747 void
748 748 metaslab_group_histogram_remove(metaslab_group_t *mg, metaslab_t *msp)
749 749 {
750 750 metaslab_class_t *mc = mg->mg_class;
751 751 uint64_t ashift = mg->mg_vd->vdev_ashift;
752 752
753 753 ASSERT(MUTEX_HELD(&msp->ms_lock));
754 754 if (msp->ms_sm == NULL)
755 755 return;
756 756
757 757 mutex_enter(&mg->mg_lock);
758 758 for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
759 759 ASSERT3U(mg->mg_histogram[i + ashift], >=,
760 760 msp->ms_sm->sm_phys->smp_histogram[i]);
761 761 ASSERT3U(mc->mc_histogram[i + ashift], >=,
762 762 msp->ms_sm->sm_phys->smp_histogram[i]);
763 763
764 764 mg->mg_histogram[i + ashift] -=
765 765 msp->ms_sm->sm_phys->smp_histogram[i];
766 766 mc->mc_histogram[i + ashift] -=
767 767 msp->ms_sm->sm_phys->smp_histogram[i];
768 768 }
769 769 mutex_exit(&mg->mg_lock);
770 770 }
771 771
772 772 static void
773 773 metaslab_group_add(metaslab_group_t *mg, metaslab_t *msp)
774 774 {
775 775 ASSERT(msp->ms_group == NULL);
776 776 mutex_enter(&mg->mg_lock);
777 777 msp->ms_group = mg;
778 778 msp->ms_weight = 0;
779 779 avl_add(&mg->mg_metaslab_tree, msp);
780 780 mutex_exit(&mg->mg_lock);
781 781
782 782 mutex_enter(&msp->ms_lock);
783 783 metaslab_group_histogram_add(mg, msp);
784 784 mutex_exit(&msp->ms_lock);
785 785 }
786 786
787 787 static void
788 788 metaslab_group_remove(metaslab_group_t *mg, metaslab_t *msp)
789 789 {
790 790 mutex_enter(&msp->ms_lock);
791 791 metaslab_group_histogram_remove(mg, msp);
792 792 mutex_exit(&msp->ms_lock);
793 793
794 794 mutex_enter(&mg->mg_lock);
795 795 ASSERT(msp->ms_group == mg);
796 796 avl_remove(&mg->mg_metaslab_tree, msp);
797 797 msp->ms_group = NULL;
798 798 mutex_exit(&mg->mg_lock);
799 799 }
800 800
801 801 static void
802 802 metaslab_group_sort(metaslab_group_t *mg, metaslab_t *msp, uint64_t weight)
803 803 {
804 804 /*
805 805 * Although in principle the weight can be any value, in
806 806 * practice we do not use values in the range [1, 511].
807 807 */
808 808 ASSERT(weight >= SPA_MINBLOCKSIZE || weight == 0);
809 809 ASSERT(MUTEX_HELD(&msp->ms_lock));
810 810
811 811 mutex_enter(&mg->mg_lock);
812 812 ASSERT(msp->ms_group == mg);
813 813 avl_remove(&mg->mg_metaslab_tree, msp);
814 814 msp->ms_weight = weight;
815 815 avl_add(&mg->mg_metaslab_tree, msp);
816 816 mutex_exit(&mg->mg_lock);
817 817 }
818 818
819 819 /*
820 820 * Calculate the fragmentation for a given metaslab group. We can use
821 821 * a simple average here since all metaslabs within the group must have
822 822 * the same size. The return value will be a value between 0 and 100
823 823 * (inclusive), or ZFS_FRAG_INVALID if less than half of the metaslab in this
824 824 * group have a fragmentation metric.
825 825 */
826 826 uint64_t
827 827 metaslab_group_fragmentation(metaslab_group_t *mg)
828 828 {
829 829 vdev_t *vd = mg->mg_vd;
830 830 uint64_t fragmentation = 0;
831 831 uint64_t valid_ms = 0;
832 832
833 833 for (int m = 0; m < vd->vdev_ms_count; m++) {
834 834 metaslab_t *msp = vd->vdev_ms[m];
835 835
836 836 if (msp->ms_fragmentation == ZFS_FRAG_INVALID)
837 837 continue;
838 838
839 839 valid_ms++;
840 840 fragmentation += msp->ms_fragmentation;
841 841 }
842 842
843 843 if (valid_ms <= vd->vdev_ms_count / 2)
844 844 return (ZFS_FRAG_INVALID);
845 845
846 846 fragmentation /= valid_ms;
847 847 ASSERT3U(fragmentation, <=, 100);
848 848 return (fragmentation);
849 849 }
850 850
851 851 /*
852 852 * Determine if a given metaslab group should skip allocations. A metaslab
853 853 * group should avoid allocations if its free capacity is less than the
854 854 * zfs_mg_noalloc_threshold or its fragmentation metric is greater than
855 855 * zfs_mg_fragmentation_threshold and there is at least one metaslab group
856 856 * that can still handle allocations. If the allocation throttle is enabled
857 857 * then we skip allocations to devices that have reached their maximum
858 858 * allocation queue depth unless the selected metaslab group is the only
859 859 * eligible group remaining.
860 860 */
861 861 static boolean_t
862 862 metaslab_group_allocatable(metaslab_group_t *mg, metaslab_group_t *rotor,
863 863 uint64_t psize)
864 864 {
865 865 spa_t *spa = mg->mg_vd->vdev_spa;
866 866 metaslab_class_t *mc = mg->mg_class;
867 867
868 868 /*
869 869 * We can only consider skipping this metaslab group if it's
870 870 * in the normal metaslab class and there are other metaslab
871 871 * groups to select from. Otherwise, we always consider it eligible
872 872 * for allocations.
873 873 */
874 874 if (mc != spa_normal_class(spa) || mc->mc_groups <= 1)
875 875 return (B_TRUE);
876 876
877 877 /*
878 878 * If the metaslab group's mg_allocatable flag is set (see comments
879 879 * in metaslab_group_alloc_update() for more information) and
880 880 * the allocation throttle is disabled then allow allocations to this
881 881 * device. However, if the allocation throttle is enabled then
882 882 * check if we have reached our allocation limit (mg_alloc_queue_depth)
883 883 * to determine if we should allow allocations to this metaslab group.
884 884 * If all metaslab groups are no longer considered allocatable
885 885 * (mc_alloc_groups == 0) or we're trying to allocate the smallest
886 886 * gang block size then we allow allocations on this metaslab group
887 887 * regardless of the mg_allocatable or throttle settings.
888 888 */
889 889 if (mg->mg_allocatable) {
890 890 metaslab_group_t *mgp;
891 891 int64_t qdepth;
892 892 uint64_t qmax = mg->mg_max_alloc_queue_depth;
893 893
894 894 if (!mc->mc_alloc_throttle_enabled)
895 895 return (B_TRUE);
896 896
897 897 /*
898 898 * If this metaslab group does not have any free space, then
899 899 * there is no point in looking further.
900 900 */
901 901 if (mg->mg_no_free_space)
902 902 return (B_FALSE);
903 903
904 904 qdepth = refcount_count(&mg->mg_alloc_queue_depth);
905 905
906 906 /*
907 907 * If this metaslab group is below its qmax or it's
908 908 * the only allocatable metasable group, then attempt
909 909 * to allocate from it.
910 910 */
911 911 if (qdepth < qmax || mc->mc_alloc_groups == 1)
912 912 return (B_TRUE);
913 913 ASSERT3U(mc->mc_alloc_groups, >, 1);
914 914
915 915 /*
916 916 * Since this metaslab group is at or over its qmax, we
917 917 * need to determine if there are metaslab groups after this
918 918 * one that might be able to handle this allocation. This is
919 919 * racy since we can't hold the locks for all metaslab
920 920 * groups at the same time when we make this check.
921 921 */
922 922 for (mgp = mg->mg_next; mgp != rotor; mgp = mgp->mg_next) {
923 923 qmax = mgp->mg_max_alloc_queue_depth;
924 924
925 925 qdepth = refcount_count(&mgp->mg_alloc_queue_depth);
926 926
927 927 /*
928 928 * If there is another metaslab group that
929 929 * might be able to handle the allocation, then
930 930 * we return false so that we skip this group.
931 931 */
932 932 if (qdepth < qmax && !mgp->mg_no_free_space)
933 933 return (B_FALSE);
934 934 }
935 935
936 936 /*
937 937 * We didn't find another group to handle the allocation
938 938 * so we can't skip this metaslab group even though
939 939 * we are at or over our qmax.
940 940 */
941 941 return (B_TRUE);
942 942
943 943 } else if (mc->mc_alloc_groups == 0 || psize == SPA_MINBLOCKSIZE) {
944 944 return (B_TRUE);
945 945 }
946 946 return (B_FALSE);
947 947 }
948 948
949 949 /*
950 950 * ==========================================================================
951 951 * Range tree callbacks
952 952 * ==========================================================================
953 953 */
954 954
955 955 /*
956 956 * Comparison function for the private size-ordered tree. Tree is sorted
957 957 * by size, larger sizes at the end of the tree.
958 958 */
959 959 static int
960 960 metaslab_rangesize_compare(const void *x1, const void *x2)
961 961 {
962 962 const range_seg_t *r1 = x1;
963 963 const range_seg_t *r2 = x2;
964 964 uint64_t rs_size1 = r1->rs_end - r1->rs_start;
965 965 uint64_t rs_size2 = r2->rs_end - r2->rs_start;
966 966
967 967 if (rs_size1 < rs_size2)
968 968 return (-1);
969 969 if (rs_size1 > rs_size2)
970 970 return (1);
971 971
972 972 if (r1->rs_start < r2->rs_start)
973 973 return (-1);
974 974
975 975 if (r1->rs_start > r2->rs_start)
976 976 return (1);
977 977
978 978 return (0);
979 979 }
980 980
981 981 /*
982 982 * Create any block allocator specific components. The current allocators
983 983 * rely on using both a size-ordered range_tree_t and an array of uint64_t's.
984 984 */
985 985 static void
986 986 metaslab_rt_create(range_tree_t *rt, void *arg)
987 987 {
988 988 metaslab_t *msp = arg;
989 989
990 990 ASSERT3P(rt->rt_arg, ==, msp);
991 991 ASSERT(msp->ms_tree == NULL);
992 992
993 993 avl_create(&msp->ms_size_tree, metaslab_rangesize_compare,
994 994 sizeof (range_seg_t), offsetof(range_seg_t, rs_pp_node));
995 995 }
996 996
997 997 /*
998 998 * Destroy the block allocator specific components.
999 999 */
1000 1000 static void
1001 1001 metaslab_rt_destroy(range_tree_t *rt, void *arg)
1002 1002 {
1003 1003 metaslab_t *msp = arg;
1004 1004
1005 1005 ASSERT3P(rt->rt_arg, ==, msp);
1006 1006 ASSERT3P(msp->ms_tree, ==, rt);
1007 1007 ASSERT0(avl_numnodes(&msp->ms_size_tree));
1008 1008
1009 1009 avl_destroy(&msp->ms_size_tree);
1010 1010 }
1011 1011
1012 1012 static void
1013 1013 metaslab_rt_add(range_tree_t *rt, range_seg_t *rs, void *arg)
1014 1014 {
1015 1015 metaslab_t *msp = arg;
1016 1016
1017 1017 ASSERT3P(rt->rt_arg, ==, msp);
1018 1018 ASSERT3P(msp->ms_tree, ==, rt);
1019 1019 VERIFY(!msp->ms_condensing);
1020 1020 avl_add(&msp->ms_size_tree, rs);
1021 1021 }
1022 1022
1023 1023 static void
1024 1024 metaslab_rt_remove(range_tree_t *rt, range_seg_t *rs, void *arg)
1025 1025 {
1026 1026 metaslab_t *msp = arg;
1027 1027
1028 1028 ASSERT3P(rt->rt_arg, ==, msp);
1029 1029 ASSERT3P(msp->ms_tree, ==, rt);
1030 1030 VERIFY(!msp->ms_condensing);
1031 1031 avl_remove(&msp->ms_size_tree, rs);
1032 1032 }
1033 1033
1034 1034 static void
1035 1035 metaslab_rt_vacate(range_tree_t *rt, void *arg)
1036 1036 {
1037 1037 metaslab_t *msp = arg;
1038 1038
1039 1039 ASSERT3P(rt->rt_arg, ==, msp);
1040 1040 ASSERT3P(msp->ms_tree, ==, rt);
1041 1041
1042 1042 /*
1043 1043 * Normally one would walk the tree freeing nodes along the way.
1044 1044 * Since the nodes are shared with the range trees we can avoid
1045 1045 * walking all nodes and just reinitialize the avl tree. The nodes
1046 1046 * will be freed by the range tree, so we don't want to free them here.
1047 1047 */
1048 1048 avl_create(&msp->ms_size_tree, metaslab_rangesize_compare,
1049 1049 sizeof (range_seg_t), offsetof(range_seg_t, rs_pp_node));
1050 1050 }
1051 1051
1052 1052 static range_tree_ops_t metaslab_rt_ops = {
1053 1053 metaslab_rt_create,
1054 1054 metaslab_rt_destroy,
1055 1055 metaslab_rt_add,
1056 1056 metaslab_rt_remove,
1057 1057 metaslab_rt_vacate
1058 1058 };
1059 1059
1060 1060 /*
1061 1061 * ==========================================================================
1062 1062 * Common allocator routines
1063 1063 * ==========================================================================
1064 1064 */
1065 1065
1066 1066 /*
1067 1067 * Return the maximum contiguous segment within the metaslab.
1068 1068 */
1069 1069 uint64_t
1070 1070 metaslab_block_maxsize(metaslab_t *msp)
1071 1071 {
1072 1072 avl_tree_t *t = &msp->ms_size_tree;
1073 1073 range_seg_t *rs;
1074 1074
1075 1075 if (t == NULL || (rs = avl_last(t)) == NULL)
1076 1076 return (0ULL);
1077 1077
1078 1078 return (rs->rs_end - rs->rs_start);
1079 1079 }
1080 1080
1081 1081 static range_seg_t *
1082 1082 metaslab_block_find(avl_tree_t *t, uint64_t start, uint64_t size)
1083 1083 {
1084 1084 range_seg_t *rs, rsearch;
1085 1085 avl_index_t where;
1086 1086
1087 1087 rsearch.rs_start = start;
1088 1088 rsearch.rs_end = start + size;
1089 1089
1090 1090 rs = avl_find(t, &rsearch, &where);
1091 1091 if (rs == NULL) {
1092 1092 rs = avl_nearest(t, where, AVL_AFTER);
1093 1093 }
1094 1094
1095 1095 return (rs);
1096 1096 }
1097 1097
1098 1098 /*
1099 1099 * This is a helper function that can be used by the allocator to find
1100 1100 * a suitable block to allocate. This will search the specified AVL
1101 1101 * tree looking for a block that matches the specified criteria.
1102 1102 */
1103 1103 static uint64_t
1104 1104 metaslab_block_picker(avl_tree_t *t, uint64_t *cursor, uint64_t size,
1105 1105 uint64_t align)
1106 1106 {
1107 1107 range_seg_t *rs = metaslab_block_find(t, *cursor, size);
1108 1108
1109 1109 while (rs != NULL) {
1110 1110 uint64_t offset = P2ROUNDUP(rs->rs_start, align);
1111 1111
1112 1112 if (offset + size <= rs->rs_end) {
1113 1113 *cursor = offset + size;
1114 1114 return (offset);
1115 1115 }
1116 1116 rs = AVL_NEXT(t, rs);
1117 1117 }
1118 1118
1119 1119 /*
1120 1120 * If we know we've searched the whole map (*cursor == 0), give up.
1121 1121 * Otherwise, reset the cursor to the beginning and try again.
1122 1122 */
1123 1123 if (*cursor == 0)
1124 1124 return (-1ULL);
1125 1125
1126 1126 *cursor = 0;
1127 1127 return (metaslab_block_picker(t, cursor, size, align));
1128 1128 }
1129 1129
1130 1130 /*
1131 1131 * ==========================================================================
1132 1132 * The first-fit block allocator
1133 1133 * ==========================================================================
1134 1134 */
1135 1135 static uint64_t
1136 1136 metaslab_ff_alloc(metaslab_t *msp, uint64_t size)
1137 1137 {
1138 1138 /*
1139 1139 * Find the largest power of 2 block size that evenly divides the
1140 1140 * requested size. This is used to try to allocate blocks with similar
1141 1141 * alignment from the same area of the metaslab (i.e. same cursor
1142 1142 * bucket) but it does not guarantee that other allocations sizes
1143 1143 * may exist in the same region.
1144 1144 */
1145 1145 uint64_t align = size & -size;
1146 1146 uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1];
1147 1147 avl_tree_t *t = &msp->ms_tree->rt_root;
1148 1148
1149 1149 return (metaslab_block_picker(t, cursor, size, align));
1150 1150 }
1151 1151
1152 1152 static metaslab_ops_t metaslab_ff_ops = {
1153 1153 metaslab_ff_alloc
1154 1154 };
1155 1155
1156 1156 /*
1157 1157 * ==========================================================================
1158 1158 * Dynamic block allocator -
1159 1159 * Uses the first fit allocation scheme until space get low and then
1160 1160 * adjusts to a best fit allocation method. Uses metaslab_df_alloc_threshold
1161 1161 * and metaslab_df_free_pct to determine when to switch the allocation scheme.
1162 1162 * ==========================================================================
1163 1163 */
1164 1164 static uint64_t
1165 1165 metaslab_df_alloc(metaslab_t *msp, uint64_t size)
1166 1166 {
1167 1167 /*
1168 1168 * Find the largest power of 2 block size that evenly divides the
1169 1169 * requested size. This is used to try to allocate blocks with similar
1170 1170 * alignment from the same area of the metaslab (i.e. same cursor
1171 1171 * bucket) but it does not guarantee that other allocations sizes
1172 1172 * may exist in the same region.
1173 1173 */
1174 1174 uint64_t align = size & -size;
1175 1175 uint64_t *cursor = &msp->ms_lbas[highbit64(align) - 1];
1176 1176 range_tree_t *rt = msp->ms_tree;
1177 1177 avl_tree_t *t = &rt->rt_root;
1178 1178 uint64_t max_size = metaslab_block_maxsize(msp);
1179 1179 int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
1180 1180
1181 1181 ASSERT(MUTEX_HELD(&msp->ms_lock));
1182 1182 ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_size_tree));
1183 1183
1184 1184 if (max_size < size)
1185 1185 return (-1ULL);
1186 1186
1187 1187 /*
1188 1188 * If we're running low on space switch to using the size
1189 1189 * sorted AVL tree (best-fit).
1190 1190 */
1191 1191 if (max_size < metaslab_df_alloc_threshold ||
1192 1192 free_pct < metaslab_df_free_pct) {
1193 1193 t = &msp->ms_size_tree;
1194 1194 *cursor = 0;
1195 1195 }
1196 1196
1197 1197 return (metaslab_block_picker(t, cursor, size, 1ULL));
1198 1198 }
1199 1199
1200 1200 static metaslab_ops_t metaslab_df_ops = {
1201 1201 metaslab_df_alloc
1202 1202 };
1203 1203
1204 1204 /*
1205 1205 * ==========================================================================
1206 1206 * Cursor fit block allocator -
1207 1207 * Select the largest region in the metaslab, set the cursor to the beginning
1208 1208 * of the range and the cursor_end to the end of the range. As allocations
1209 1209 * are made advance the cursor. Continue allocating from the cursor until
1210 1210 * the range is exhausted and then find a new range.
1211 1211 * ==========================================================================
1212 1212 */
1213 1213 static uint64_t
1214 1214 metaslab_cf_alloc(metaslab_t *msp, uint64_t size)
1215 1215 {
1216 1216 range_tree_t *rt = msp->ms_tree;
1217 1217 avl_tree_t *t = &msp->ms_size_tree;
1218 1218 uint64_t *cursor = &msp->ms_lbas[0];
1219 1219 uint64_t *cursor_end = &msp->ms_lbas[1];
1220 1220 uint64_t offset = 0;
1221 1221
1222 1222 ASSERT(MUTEX_HELD(&msp->ms_lock));
1223 1223 ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&rt->rt_root));
1224 1224
1225 1225 ASSERT3U(*cursor_end, >=, *cursor);
1226 1226
1227 1227 if ((*cursor + size) > *cursor_end) {
1228 1228 range_seg_t *rs;
1229 1229
1230 1230 rs = avl_last(&msp->ms_size_tree);
1231 1231 if (rs == NULL || (rs->rs_end - rs->rs_start) < size)
1232 1232 return (-1ULL);
1233 1233
1234 1234 *cursor = rs->rs_start;
1235 1235 *cursor_end = rs->rs_end;
1236 1236 }
1237 1237
1238 1238 offset = *cursor;
1239 1239 *cursor += size;
1240 1240
1241 1241 return (offset);
1242 1242 }
1243 1243
1244 1244 static metaslab_ops_t metaslab_cf_ops = {
1245 1245 metaslab_cf_alloc
1246 1246 };
1247 1247
1248 1248 /*
1249 1249 * ==========================================================================
1250 1250 * New dynamic fit allocator -
1251 1251 * Select a region that is large enough to allocate 2^metaslab_ndf_clump_shift
1252 1252 * contiguous blocks. If no region is found then just use the largest segment
1253 1253 * that remains.
1254 1254 * ==========================================================================
1255 1255 */
1256 1256
1257 1257 /*
1258 1258 * Determines desired number of contiguous blocks (2^metaslab_ndf_clump_shift)
1259 1259 * to request from the allocator.
1260 1260 */
1261 1261 uint64_t metaslab_ndf_clump_shift = 4;
1262 1262
1263 1263 static uint64_t
1264 1264 metaslab_ndf_alloc(metaslab_t *msp, uint64_t size)
1265 1265 {
1266 1266 avl_tree_t *t = &msp->ms_tree->rt_root;
1267 1267 avl_index_t where;
1268 1268 range_seg_t *rs, rsearch;
1269 1269 uint64_t hbit = highbit64(size);
1270 1270 uint64_t *cursor = &msp->ms_lbas[hbit - 1];
1271 1271 uint64_t max_size = metaslab_block_maxsize(msp);
1272 1272
1273 1273 ASSERT(MUTEX_HELD(&msp->ms_lock));
1274 1274 ASSERT3U(avl_numnodes(t), ==, avl_numnodes(&msp->ms_size_tree));
1275 1275
1276 1276 if (max_size < size)
1277 1277 return (-1ULL);
1278 1278
1279 1279 rsearch.rs_start = *cursor;
1280 1280 rsearch.rs_end = *cursor + size;
1281 1281
1282 1282 rs = avl_find(t, &rsearch, &where);
1283 1283 if (rs == NULL || (rs->rs_end - rs->rs_start) < size) {
1284 1284 t = &msp->ms_size_tree;
1285 1285
1286 1286 rsearch.rs_start = 0;
1287 1287 rsearch.rs_end = MIN(max_size,
1288 1288 1ULL << (hbit + metaslab_ndf_clump_shift));
1289 1289 rs = avl_find(t, &rsearch, &where);
1290 1290 if (rs == NULL)
1291 1291 rs = avl_nearest(t, where, AVL_AFTER);
1292 1292 ASSERT(rs != NULL);
1293 1293 }
1294 1294
1295 1295 if ((rs->rs_end - rs->rs_start) >= size) {
1296 1296 *cursor = rs->rs_start + size;
1297 1297 return (rs->rs_start);
1298 1298 }
1299 1299 return (-1ULL);
1300 1300 }
1301 1301
1302 1302 static metaslab_ops_t metaslab_ndf_ops = {
1303 1303 metaslab_ndf_alloc
1304 1304 };
1305 1305
1306 1306 metaslab_ops_t *zfs_metaslab_ops = &metaslab_df_ops;
1307 1307
1308 1308 /*
1309 1309 * ==========================================================================
1310 1310 * Metaslabs
1311 1311 * ==========================================================================
1312 1312 */
1313 1313
1314 1314 /*
1315 1315 * Wait for any in-progress metaslab loads to complete.
1316 1316 */
1317 1317 void
1318 1318 metaslab_load_wait(metaslab_t *msp)
1319 1319 {
1320 1320 ASSERT(MUTEX_HELD(&msp->ms_lock));
1321 1321
1322 1322 while (msp->ms_loading) {
1323 1323 ASSERT(!msp->ms_loaded);
1324 1324 cv_wait(&msp->ms_load_cv, &msp->ms_lock);
1325 1325 }
1326 1326 }
1327 1327
1328 1328 int
1329 1329 metaslab_load(metaslab_t *msp)
1330 1330 {
1331 1331 int error = 0;
1332 1332 boolean_t success = B_FALSE;
1333 1333
1334 1334 ASSERT(MUTEX_HELD(&msp->ms_lock));
1335 1335 ASSERT(!msp->ms_loaded);
1336 1336 ASSERT(!msp->ms_loading);
1337 1337
1338 1338 msp->ms_loading = B_TRUE;
1339 1339
1340 1340 /*
1341 1341 * If the space map has not been allocated yet, then treat
1342 1342 * all the space in the metaslab as free and add it to the
1343 1343 * ms_tree.
1344 1344 */
1345 1345 if (msp->ms_sm != NULL)
1346 1346 error = space_map_load(msp->ms_sm, msp->ms_tree, SM_FREE);
1347 1347 else
1348 1348 range_tree_add(msp->ms_tree, msp->ms_start, msp->ms_size);
1349 1349
1350 1350 success = (error == 0);
1351 1351 msp->ms_loading = B_FALSE;
1352 1352
1353 1353 if (success) {
1354 1354 ASSERT3P(msp->ms_group, !=, NULL);
1355 1355 msp->ms_loaded = B_TRUE;
1356 1356
1357 1357 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
1358 1358 range_tree_walk(msp->ms_defertree[t],
1359 1359 range_tree_remove, msp->ms_tree);
1360 1360 }
1361 1361 msp->ms_max_size = metaslab_block_maxsize(msp);
1362 1362 }
1363 1363 cv_broadcast(&msp->ms_load_cv);
1364 1364 return (error);
1365 1365 }
1366 1366
1367 1367 void
1368 1368 metaslab_unload(metaslab_t *msp)
1369 1369 {
1370 1370 ASSERT(MUTEX_HELD(&msp->ms_lock));
1371 1371 range_tree_vacate(msp->ms_tree, NULL, NULL);
1372 1372 msp->ms_loaded = B_FALSE;
1373 1373 msp->ms_weight &= ~METASLAB_ACTIVE_MASK;
1374 1374 msp->ms_max_size = 0;
1375 1375 }
1376 1376
1377 1377 int
1378 1378 metaslab_init(metaslab_group_t *mg, uint64_t id, uint64_t object, uint64_t txg,
1379 1379 metaslab_t **msp)
1380 1380 {
1381 1381 vdev_t *vd = mg->mg_vd;
1382 1382 objset_t *mos = vd->vdev_spa->spa_meta_objset;
1383 1383 metaslab_t *ms;
1384 1384 int error;
1385 1385
1386 1386 ms = kmem_zalloc(sizeof (metaslab_t), KM_SLEEP);
1387 1387 mutex_init(&ms->ms_lock, NULL, MUTEX_DEFAULT, NULL);
1388 1388 cv_init(&ms->ms_load_cv, NULL, CV_DEFAULT, NULL);
1389 1389 ms->ms_id = id;
1390 1390 ms->ms_start = id << vd->vdev_ms_shift;
1391 1391 ms->ms_size = 1ULL << vd->vdev_ms_shift;
1392 1392
1393 1393 /*
1394 1394 * We only open space map objects that already exist. All others
1395 1395 * will be opened when we finally allocate an object for it.
1396 1396 */
1397 1397 if (object != 0) {
1398 1398 error = space_map_open(&ms->ms_sm, mos, object, ms->ms_start,
1399 1399 ms->ms_size, vd->vdev_ashift, &ms->ms_lock);
1400 1400
1401 1401 if (error != 0) {
1402 1402 kmem_free(ms, sizeof (metaslab_t));
1403 1403 return (error);
1404 1404 }
1405 1405
1406 1406 ASSERT(ms->ms_sm != NULL);
1407 1407 }
1408 1408
1409 1409 /*
1410 1410 * We create the main range tree here, but we don't create the
1411 1411 * other range trees until metaslab_sync_done(). This serves
1412 1412 * two purposes: it allows metaslab_sync_done() to detect the
1413 1413 * addition of new space; and for debugging, it ensures that we'd
1414 1414 * data fault on any attempt to use this metaslab before it's ready.
1415 1415 */
1416 1416 ms->ms_tree = range_tree_create(&metaslab_rt_ops, ms, &ms->ms_lock);
1417 1417 metaslab_group_add(mg, ms);
1418 1418
1419 1419 metaslab_set_fragmentation(ms);
1420 1420
1421 1421 /*
1422 1422 * If we're opening an existing pool (txg == 0) or creating
1423 1423 * a new one (txg == TXG_INITIAL), all space is available now.
1424 1424 * If we're adding space to an existing pool, the new space
1425 1425 * does not become available until after this txg has synced.
1426 1426 * The metaslab's weight will also be initialized when we sync
1427 1427 * out this txg. This ensures that we don't attempt to allocate
1428 1428 * from it before we have initialized it completely.
1429 1429 */
1430 1430 if (txg <= TXG_INITIAL)
1431 1431 metaslab_sync_done(ms, 0);
1432 1432
1433 1433 /*
1434 1434 * If metaslab_debug_load is set and we're initializing a metaslab
1435 1435 * that has an allocated space map object then load the its space
1436 1436 * map so that can verify frees.
1437 1437 */
1438 1438 if (metaslab_debug_load && ms->ms_sm != NULL) {
1439 1439 mutex_enter(&ms->ms_lock);
1440 1440 VERIFY0(metaslab_load(ms));
1441 1441 mutex_exit(&ms->ms_lock);
1442 1442 }
1443 1443
1444 1444 if (txg != 0) {
1445 1445 vdev_dirty(vd, 0, NULL, txg);
1446 1446 vdev_dirty(vd, VDD_METASLAB, ms, txg);
1447 1447 }
1448 1448
1449 1449 *msp = ms;
1450 1450
1451 1451 return (0);
1452 1452 }
1453 1453
1454 1454 void
1455 1455 metaslab_fini(metaslab_t *msp)
1456 1456 {
1457 1457 metaslab_group_t *mg = msp->ms_group;
1458 1458
1459 1459 metaslab_group_remove(mg, msp);
1460 1460
1461 1461 mutex_enter(&msp->ms_lock);
1462 1462 VERIFY(msp->ms_group == NULL);
1463 1463 vdev_space_update(mg->mg_vd, -space_map_allocated(msp->ms_sm),
1464 1464 0, -msp->ms_size);
1465 1465 space_map_close(msp->ms_sm);
1466 1466
1467 1467 metaslab_unload(msp);
1468 1468 range_tree_destroy(msp->ms_tree);
1469 1469 range_tree_destroy(msp->ms_freeingtree);
1470 1470 range_tree_destroy(msp->ms_freedtree);
1471 1471
1472 1472 for (int t = 0; t < TXG_SIZE; t++) {
1473 1473 range_tree_destroy(msp->ms_alloctree[t]);
1474 1474 }
1475 1475
1476 1476 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
1477 1477 range_tree_destroy(msp->ms_defertree[t]);
1478 1478 }
1479 1479
1480 1480 ASSERT0(msp->ms_deferspace);
1481 1481
1482 1482 mutex_exit(&msp->ms_lock);
1483 1483 cv_destroy(&msp->ms_load_cv);
1484 1484 mutex_destroy(&msp->ms_lock);
1485 1485
1486 1486 kmem_free(msp, sizeof (metaslab_t));
1487 1487 }
1488 1488
1489 1489 #define FRAGMENTATION_TABLE_SIZE 17
1490 1490
1491 1491 /*
1492 1492 * This table defines a segment size based fragmentation metric that will
1493 1493 * allow each metaslab to derive its own fragmentation value. This is done
1494 1494 * by calculating the space in each bucket of the spacemap histogram and
1495 1495 * multiplying that by the fragmetation metric in this table. Doing
1496 1496 * this for all buckets and dividing it by the total amount of free
1497 1497 * space in this metaslab (i.e. the total free space in all buckets) gives
1498 1498 * us the fragmentation metric. This means that a high fragmentation metric
1499 1499 * equates to most of the free space being comprised of small segments.
1500 1500 * Conversely, if the metric is low, then most of the free space is in
1501 1501 * large segments. A 10% change in fragmentation equates to approximately
1502 1502 * double the number of segments.
1503 1503 *
1504 1504 * This table defines 0% fragmented space using 16MB segments. Testing has
1505 1505 * shown that segments that are greater than or equal to 16MB do not suffer
1506 1506 * from drastic performance problems. Using this value, we derive the rest
1507 1507 * of the table. Since the fragmentation value is never stored on disk, it
1508 1508 * is possible to change these calculations in the future.
1509 1509 */
1510 1510 int zfs_frag_table[FRAGMENTATION_TABLE_SIZE] = {
1511 1511 100, /* 512B */
1512 1512 100, /* 1K */
1513 1513 98, /* 2K */
1514 1514 95, /* 4K */
1515 1515 90, /* 8K */
1516 1516 80, /* 16K */
1517 1517 70, /* 32K */
1518 1518 60, /* 64K */
1519 1519 50, /* 128K */
1520 1520 40, /* 256K */
1521 1521 30, /* 512K */
1522 1522 20, /* 1M */
1523 1523 15, /* 2M */
1524 1524 10, /* 4M */
1525 1525 5, /* 8M */
1526 1526 0 /* 16M */
1527 1527 };
1528 1528
1529 1529 /*
1530 1530 * Calclate the metaslab's fragmentation metric. A return value
1531 1531 * of ZFS_FRAG_INVALID means that the metaslab has not been upgraded and does
1532 1532 * not support this metric. Otherwise, the return value should be in the
1533 1533 * range [0, 100].
1534 1534 */
1535 1535 static void
1536 1536 metaslab_set_fragmentation(metaslab_t *msp)
1537 1537 {
1538 1538 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
1539 1539 uint64_t fragmentation = 0;
1540 1540 uint64_t total = 0;
1541 1541 boolean_t feature_enabled = spa_feature_is_enabled(spa,
1542 1542 SPA_FEATURE_SPACEMAP_HISTOGRAM);
1543 1543
1544 1544 if (!feature_enabled) {
1545 1545 msp->ms_fragmentation = ZFS_FRAG_INVALID;
1546 1546 return;
1547 1547 }
1548 1548
1549 1549 /*
1550 1550 * A null space map means that the entire metaslab is free
1551 1551 * and thus is not fragmented.
1552 1552 */
1553 1553 if (msp->ms_sm == NULL) {
1554 1554 msp->ms_fragmentation = 0;
1555 1555 return;
1556 1556 }
1557 1557
1558 1558 /*
1559 1559 * If this metaslab's space map has not been upgraded, flag it
1560 1560 * so that we upgrade next time we encounter it.
1561 1561 */
1562 1562 if (msp->ms_sm->sm_dbuf->db_size != sizeof (space_map_phys_t)) {
1563 1563 uint64_t txg = spa_syncing_txg(spa);
1564 1564 vdev_t *vd = msp->ms_group->mg_vd;
1565 1565
1566 1566 if (spa_writeable(spa)) {
1567 1567 msp->ms_condense_wanted = B_TRUE;
1568 1568 vdev_dirty(vd, VDD_METASLAB, msp, txg + 1);
1569 1569 spa_dbgmsg(spa, "txg %llu, requesting force condense: "
1570 1570 "msp %p, vd %p", txg, msp, vd);
1571 1571 }
1572 1572 msp->ms_fragmentation = ZFS_FRAG_INVALID;
1573 1573 return;
1574 1574 }
1575 1575
1576 1576 for (int i = 0; i < SPACE_MAP_HISTOGRAM_SIZE; i++) {
1577 1577 uint64_t space = 0;
1578 1578 uint8_t shift = msp->ms_sm->sm_shift;
1579 1579
1580 1580 int idx = MIN(shift - SPA_MINBLOCKSHIFT + i,
1581 1581 FRAGMENTATION_TABLE_SIZE - 1);
1582 1582
1583 1583 if (msp->ms_sm->sm_phys->smp_histogram[i] == 0)
1584 1584 continue;
1585 1585
1586 1586 space = msp->ms_sm->sm_phys->smp_histogram[i] << (i + shift);
1587 1587 total += space;
1588 1588
1589 1589 ASSERT3U(idx, <, FRAGMENTATION_TABLE_SIZE);
1590 1590 fragmentation += space * zfs_frag_table[idx];
1591 1591 }
1592 1592
1593 1593 if (total > 0)
1594 1594 fragmentation /= total;
1595 1595 ASSERT3U(fragmentation, <=, 100);
1596 1596
1597 1597 msp->ms_fragmentation = fragmentation;
1598 1598 }
1599 1599
1600 1600 /*
1601 1601 * Compute a weight -- a selection preference value -- for the given metaslab.
1602 1602 * This is based on the amount of free space, the level of fragmentation,
1603 1603 * the LBA range, and whether the metaslab is loaded.
1604 1604 */
1605 1605 static uint64_t
1606 1606 metaslab_space_weight(metaslab_t *msp)
1607 1607 {
1608 1608 metaslab_group_t *mg = msp->ms_group;
1609 1609 vdev_t *vd = mg->mg_vd;
1610 1610 uint64_t weight, space;
1611 1611
1612 1612 ASSERT(MUTEX_HELD(&msp->ms_lock));
1613 1613 ASSERT(!vd->vdev_removing);
1614 1614
1615 1615 /*
1616 1616 * The baseline weight is the metaslab's free space.
1617 1617 */
1618 1618 space = msp->ms_size - space_map_allocated(msp->ms_sm);
1619 1619
1620 1620 if (metaslab_fragmentation_factor_enabled &&
1621 1621 msp->ms_fragmentation != ZFS_FRAG_INVALID) {
1622 1622 /*
1623 1623 * Use the fragmentation information to inversely scale
1624 1624 * down the baseline weight. We need to ensure that we
1625 1625 * don't exclude this metaslab completely when it's 100%
1626 1626 * fragmented. To avoid this we reduce the fragmented value
1627 1627 * by 1.
1628 1628 */
1629 1629 space = (space * (100 - (msp->ms_fragmentation - 1))) / 100;
1630 1630
1631 1631 /*
1632 1632 * If space < SPA_MINBLOCKSIZE, then we will not allocate from
1633 1633 * this metaslab again. The fragmentation metric may have
1634 1634 * decreased the space to something smaller than
1635 1635 * SPA_MINBLOCKSIZE, so reset the space to SPA_MINBLOCKSIZE
1636 1636 * so that we can consume any remaining space.
1637 1637 */
1638 1638 if (space > 0 && space < SPA_MINBLOCKSIZE)
1639 1639 space = SPA_MINBLOCKSIZE;
1640 1640 }
1641 1641 weight = space;
|
↓ open down ↓ |
1641 lines elided |
↑ open up ↑ |
1642 1642
1643 1643 /*
1644 1644 * Modern disks have uniform bit density and constant angular velocity.
1645 1645 * Therefore, the outer recording zones are faster (higher bandwidth)
1646 1646 * than the inner zones by the ratio of outer to inner track diameter,
1647 1647 * which is typically around 2:1. We account for this by assigning
1648 1648 * higher weight to lower metaslabs (multiplier ranging from 2x to 1x).
1649 1649 * In effect, this means that we'll select the metaslab with the most
1650 1650 * free bandwidth rather than simply the one with the most free space.
1651 1651 */
1652 - if (metaslab_lba_weighting_enabled) {
1652 + if (!vd->vdev_nonrot && metaslab_lba_weighting_enabled) {
1653 1653 weight = 2 * weight - (msp->ms_id * weight) / vd->vdev_ms_count;
1654 1654 ASSERT(weight >= space && weight <= 2 * space);
1655 1655 }
1656 1656
1657 1657 /*
1658 1658 * If this metaslab is one we're actively using, adjust its
1659 1659 * weight to make it preferable to any inactive metaslab so
1660 1660 * we'll polish it off. If the fragmentation on this metaslab
1661 1661 * has exceed our threshold, then don't mark it active.
1662 1662 */
1663 1663 if (msp->ms_loaded && msp->ms_fragmentation != ZFS_FRAG_INVALID &&
1664 1664 msp->ms_fragmentation <= zfs_metaslab_fragmentation_threshold) {
1665 1665 weight |= (msp->ms_weight & METASLAB_ACTIVE_MASK);
1666 1666 }
1667 1667
1668 1668 WEIGHT_SET_SPACEBASED(weight);
1669 1669 return (weight);
1670 1670 }
1671 1671
1672 1672 /*
1673 1673 * Return the weight of the specified metaslab, according to the segment-based
1674 1674 * weighting algorithm. The metaslab must be loaded. This function can
1675 1675 * be called within a sync pass since it relies only on the metaslab's
1676 1676 * range tree which is always accurate when the metaslab is loaded.
1677 1677 */
1678 1678 static uint64_t
1679 1679 metaslab_weight_from_range_tree(metaslab_t *msp)
1680 1680 {
1681 1681 uint64_t weight = 0;
1682 1682 uint32_t segments = 0;
1683 1683
1684 1684 ASSERT(msp->ms_loaded);
1685 1685
1686 1686 for (int i = RANGE_TREE_HISTOGRAM_SIZE - 1; i >= SPA_MINBLOCKSHIFT;
1687 1687 i--) {
1688 1688 uint8_t shift = msp->ms_group->mg_vd->vdev_ashift;
1689 1689 int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1;
1690 1690
1691 1691 segments <<= 1;
1692 1692 segments += msp->ms_tree->rt_histogram[i];
1693 1693
1694 1694 /*
1695 1695 * The range tree provides more precision than the space map
1696 1696 * and must be downgraded so that all values fit within the
1697 1697 * space map's histogram. This allows us to compare loaded
1698 1698 * vs. unloaded metaslabs to determine which metaslab is
1699 1699 * considered "best".
1700 1700 */
1701 1701 if (i > max_idx)
1702 1702 continue;
1703 1703
1704 1704 if (segments != 0) {
1705 1705 WEIGHT_SET_COUNT(weight, segments);
1706 1706 WEIGHT_SET_INDEX(weight, i);
1707 1707 WEIGHT_SET_ACTIVE(weight, 0);
1708 1708 break;
1709 1709 }
1710 1710 }
1711 1711 return (weight);
1712 1712 }
1713 1713
1714 1714 /*
1715 1715 * Calculate the weight based on the on-disk histogram. This should only
1716 1716 * be called after a sync pass has completely finished since the on-disk
1717 1717 * information is updated in metaslab_sync().
1718 1718 */
1719 1719 static uint64_t
1720 1720 metaslab_weight_from_spacemap(metaslab_t *msp)
1721 1721 {
1722 1722 uint64_t weight = 0;
1723 1723
1724 1724 for (int i = SPACE_MAP_HISTOGRAM_SIZE - 1; i >= 0; i--) {
1725 1725 if (msp->ms_sm->sm_phys->smp_histogram[i] != 0) {
1726 1726 WEIGHT_SET_COUNT(weight,
1727 1727 msp->ms_sm->sm_phys->smp_histogram[i]);
1728 1728 WEIGHT_SET_INDEX(weight, i +
1729 1729 msp->ms_sm->sm_shift);
1730 1730 WEIGHT_SET_ACTIVE(weight, 0);
1731 1731 break;
1732 1732 }
1733 1733 }
1734 1734 return (weight);
1735 1735 }
1736 1736
1737 1737 /*
1738 1738 * Compute a segment-based weight for the specified metaslab. The weight
1739 1739 * is determined by highest bucket in the histogram. The information
1740 1740 * for the highest bucket is encoded into the weight value.
1741 1741 */
1742 1742 static uint64_t
1743 1743 metaslab_segment_weight(metaslab_t *msp)
1744 1744 {
1745 1745 metaslab_group_t *mg = msp->ms_group;
1746 1746 uint64_t weight = 0;
1747 1747 uint8_t shift = mg->mg_vd->vdev_ashift;
1748 1748
1749 1749 ASSERT(MUTEX_HELD(&msp->ms_lock));
1750 1750
1751 1751 /*
1752 1752 * The metaslab is completely free.
1753 1753 */
1754 1754 if (space_map_allocated(msp->ms_sm) == 0) {
1755 1755 int idx = highbit64(msp->ms_size) - 1;
1756 1756 int max_idx = SPACE_MAP_HISTOGRAM_SIZE + shift - 1;
1757 1757
1758 1758 if (idx < max_idx) {
1759 1759 WEIGHT_SET_COUNT(weight, 1ULL);
1760 1760 WEIGHT_SET_INDEX(weight, idx);
1761 1761 } else {
1762 1762 WEIGHT_SET_COUNT(weight, 1ULL << (idx - max_idx));
1763 1763 WEIGHT_SET_INDEX(weight, max_idx);
1764 1764 }
1765 1765 WEIGHT_SET_ACTIVE(weight, 0);
1766 1766 ASSERT(!WEIGHT_IS_SPACEBASED(weight));
1767 1767
1768 1768 return (weight);
1769 1769 }
1770 1770
1771 1771 ASSERT3U(msp->ms_sm->sm_dbuf->db_size, ==, sizeof (space_map_phys_t));
1772 1772
1773 1773 /*
1774 1774 * If the metaslab is fully allocated then just make the weight 0.
1775 1775 */
1776 1776 if (space_map_allocated(msp->ms_sm) == msp->ms_size)
1777 1777 return (0);
1778 1778 /*
1779 1779 * If the metaslab is already loaded, then use the range tree to
1780 1780 * determine the weight. Otherwise, we rely on the space map information
1781 1781 * to generate the weight.
1782 1782 */
1783 1783 if (msp->ms_loaded) {
1784 1784 weight = metaslab_weight_from_range_tree(msp);
1785 1785 } else {
1786 1786 weight = metaslab_weight_from_spacemap(msp);
1787 1787 }
1788 1788
1789 1789 /*
1790 1790 * If the metaslab was active the last time we calculated its weight
1791 1791 * then keep it active. We want to consume the entire region that
1792 1792 * is associated with this weight.
1793 1793 */
1794 1794 if (msp->ms_activation_weight != 0 && weight != 0)
1795 1795 WEIGHT_SET_ACTIVE(weight, WEIGHT_GET_ACTIVE(msp->ms_weight));
1796 1796 return (weight);
1797 1797 }
1798 1798
1799 1799 /*
1800 1800 * Determine if we should attempt to allocate from this metaslab. If the
1801 1801 * metaslab has a maximum size then we can quickly determine if the desired
1802 1802 * allocation size can be satisfied. Otherwise, if we're using segment-based
1803 1803 * weighting then we can determine the maximum allocation that this metaslab
1804 1804 * can accommodate based on the index encoded in the weight. If we're using
1805 1805 * space-based weights then rely on the entire weight (excluding the weight
1806 1806 * type bit).
1807 1807 */
1808 1808 boolean_t
1809 1809 metaslab_should_allocate(metaslab_t *msp, uint64_t asize)
1810 1810 {
1811 1811 boolean_t should_allocate;
1812 1812
1813 1813 if (msp->ms_max_size != 0)
1814 1814 return (msp->ms_max_size >= asize);
1815 1815
1816 1816 if (!WEIGHT_IS_SPACEBASED(msp->ms_weight)) {
1817 1817 /*
1818 1818 * The metaslab segment weight indicates segments in the
1819 1819 * range [2^i, 2^(i+1)), where i is the index in the weight.
1820 1820 * Since the asize might be in the middle of the range, we
1821 1821 * should attempt the allocation if asize < 2^(i+1).
1822 1822 */
1823 1823 should_allocate = (asize <
1824 1824 1ULL << (WEIGHT_GET_INDEX(msp->ms_weight) + 1));
1825 1825 } else {
1826 1826 should_allocate = (asize <=
1827 1827 (msp->ms_weight & ~METASLAB_WEIGHT_TYPE));
1828 1828 }
1829 1829 return (should_allocate);
1830 1830 }
1831 1831
1832 1832 static uint64_t
1833 1833 metaslab_weight(metaslab_t *msp)
1834 1834 {
1835 1835 vdev_t *vd = msp->ms_group->mg_vd;
1836 1836 spa_t *spa = vd->vdev_spa;
1837 1837 uint64_t weight;
1838 1838
1839 1839 ASSERT(MUTEX_HELD(&msp->ms_lock));
1840 1840
1841 1841 /*
1842 1842 * This vdev is in the process of being removed so there is nothing
1843 1843 * for us to do here.
1844 1844 */
1845 1845 if (vd->vdev_removing) {
1846 1846 ASSERT0(space_map_allocated(msp->ms_sm));
1847 1847 ASSERT0(vd->vdev_ms_shift);
1848 1848 return (0);
1849 1849 }
1850 1850
1851 1851 metaslab_set_fragmentation(msp);
1852 1852
1853 1853 /*
1854 1854 * Update the maximum size if the metaslab is loaded. This will
1855 1855 * ensure that we get an accurate maximum size if newly freed space
1856 1856 * has been added back into the free tree.
1857 1857 */
1858 1858 if (msp->ms_loaded)
1859 1859 msp->ms_max_size = metaslab_block_maxsize(msp);
1860 1860
1861 1861 /*
1862 1862 * Segment-based weighting requires space map histogram support.
1863 1863 */
1864 1864 if (zfs_metaslab_segment_weight_enabled &&
1865 1865 spa_feature_is_enabled(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM) &&
1866 1866 (msp->ms_sm == NULL || msp->ms_sm->sm_dbuf->db_size ==
1867 1867 sizeof (space_map_phys_t))) {
1868 1868 weight = metaslab_segment_weight(msp);
1869 1869 } else {
1870 1870 weight = metaslab_space_weight(msp);
1871 1871 }
1872 1872 return (weight);
1873 1873 }
1874 1874
1875 1875 static int
1876 1876 metaslab_activate(metaslab_t *msp, uint64_t activation_weight)
1877 1877 {
1878 1878 ASSERT(MUTEX_HELD(&msp->ms_lock));
1879 1879
1880 1880 if ((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0) {
1881 1881 metaslab_load_wait(msp);
1882 1882 if (!msp->ms_loaded) {
1883 1883 int error = metaslab_load(msp);
1884 1884 if (error) {
1885 1885 metaslab_group_sort(msp->ms_group, msp, 0);
1886 1886 return (error);
1887 1887 }
1888 1888 }
1889 1889
1890 1890 msp->ms_activation_weight = msp->ms_weight;
1891 1891 metaslab_group_sort(msp->ms_group, msp,
1892 1892 msp->ms_weight | activation_weight);
1893 1893 }
1894 1894 ASSERT(msp->ms_loaded);
1895 1895 ASSERT(msp->ms_weight & METASLAB_ACTIVE_MASK);
1896 1896
1897 1897 return (0);
1898 1898 }
1899 1899
1900 1900 static void
1901 1901 metaslab_passivate(metaslab_t *msp, uint64_t weight)
1902 1902 {
1903 1903 uint64_t size = weight & ~METASLAB_WEIGHT_TYPE;
1904 1904
1905 1905 /*
1906 1906 * If size < SPA_MINBLOCKSIZE, then we will not allocate from
1907 1907 * this metaslab again. In that case, it had better be empty,
1908 1908 * or we would be leaving space on the table.
1909 1909 */
1910 1910 ASSERT(size >= SPA_MINBLOCKSIZE ||
1911 1911 range_tree_space(msp->ms_tree) == 0);
1912 1912 ASSERT0(weight & METASLAB_ACTIVE_MASK);
1913 1913
1914 1914 msp->ms_activation_weight = 0;
1915 1915 metaslab_group_sort(msp->ms_group, msp, weight);
1916 1916 ASSERT((msp->ms_weight & METASLAB_ACTIVE_MASK) == 0);
1917 1917 }
1918 1918
1919 1919 /*
1920 1920 * Segment-based metaslabs are activated once and remain active until
1921 1921 * we either fail an allocation attempt (similar to space-based metaslabs)
1922 1922 * or have exhausted the free space in zfs_metaslab_switch_threshold
1923 1923 * buckets since the metaslab was activated. This function checks to see
1924 1924 * if we've exhaused the zfs_metaslab_switch_threshold buckets in the
1925 1925 * metaslab and passivates it proactively. This will allow us to select a
1926 1926 * metaslabs with larger contiguous region if any remaining within this
1927 1927 * metaslab group. If we're in sync pass > 1, then we continue using this
1928 1928 * metaslab so that we don't dirty more block and cause more sync passes.
1929 1929 */
1930 1930 void
1931 1931 metaslab_segment_may_passivate(metaslab_t *msp)
1932 1932 {
1933 1933 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
1934 1934
1935 1935 if (WEIGHT_IS_SPACEBASED(msp->ms_weight) || spa_sync_pass(spa) > 1)
1936 1936 return;
1937 1937
1938 1938 /*
1939 1939 * Since we are in the middle of a sync pass, the most accurate
1940 1940 * information that is accessible to us is the in-core range tree
1941 1941 * histogram; calculate the new weight based on that information.
1942 1942 */
1943 1943 uint64_t weight = metaslab_weight_from_range_tree(msp);
1944 1944 int activation_idx = WEIGHT_GET_INDEX(msp->ms_activation_weight);
1945 1945 int current_idx = WEIGHT_GET_INDEX(weight);
1946 1946
1947 1947 if (current_idx <= activation_idx - zfs_metaslab_switch_threshold)
1948 1948 metaslab_passivate(msp, weight);
1949 1949 }
1950 1950
1951 1951 static void
1952 1952 metaslab_preload(void *arg)
1953 1953 {
1954 1954 metaslab_t *msp = arg;
1955 1955 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
1956 1956
1957 1957 ASSERT(!MUTEX_HELD(&msp->ms_group->mg_lock));
1958 1958
1959 1959 mutex_enter(&msp->ms_lock);
1960 1960 metaslab_load_wait(msp);
1961 1961 if (!msp->ms_loaded)
1962 1962 (void) metaslab_load(msp);
1963 1963 msp->ms_selected_txg = spa_syncing_txg(spa);
1964 1964 mutex_exit(&msp->ms_lock);
1965 1965 }
1966 1966
1967 1967 static void
1968 1968 metaslab_group_preload(metaslab_group_t *mg)
1969 1969 {
1970 1970 spa_t *spa = mg->mg_vd->vdev_spa;
1971 1971 metaslab_t *msp;
1972 1972 avl_tree_t *t = &mg->mg_metaslab_tree;
1973 1973 int m = 0;
1974 1974
1975 1975 if (spa_shutting_down(spa) || !metaslab_preload_enabled) {
1976 1976 taskq_wait(mg->mg_taskq);
1977 1977 return;
1978 1978 }
1979 1979
1980 1980 mutex_enter(&mg->mg_lock);
1981 1981 /*
1982 1982 * Load the next potential metaslabs
1983 1983 */
1984 1984 for (msp = avl_first(t); msp != NULL; msp = AVL_NEXT(t, msp)) {
1985 1985 /*
1986 1986 * We preload only the maximum number of metaslabs specified
1987 1987 * by metaslab_preload_limit. If a metaslab is being forced
1988 1988 * to condense then we preload it too. This will ensure
1989 1989 * that force condensing happens in the next txg.
1990 1990 */
1991 1991 if (++m > metaslab_preload_limit && !msp->ms_condense_wanted) {
1992 1992 continue;
1993 1993 }
1994 1994
1995 1995 VERIFY(taskq_dispatch(mg->mg_taskq, metaslab_preload,
1996 1996 msp, TQ_SLEEP) != NULL);
1997 1997 }
1998 1998 mutex_exit(&mg->mg_lock);
1999 1999 }
2000 2000
2001 2001 /*
2002 2002 * Determine if the space map's on-disk footprint is past our tolerance
2003 2003 * for inefficiency. We would like to use the following criteria to make
2004 2004 * our decision:
2005 2005 *
2006 2006 * 1. The size of the space map object should not dramatically increase as a
2007 2007 * result of writing out the free space range tree.
2008 2008 *
2009 2009 * 2. The minimal on-disk space map representation is zfs_condense_pct/100
2010 2010 * times the size than the free space range tree representation
2011 2011 * (i.e. zfs_condense_pct = 110 and in-core = 1MB, minimal = 1.1.MB).
2012 2012 *
2013 2013 * 3. The on-disk size of the space map should actually decrease.
2014 2014 *
2015 2015 * Checking the first condition is tricky since we don't want to walk
2016 2016 * the entire AVL tree calculating the estimated on-disk size. Instead we
2017 2017 * use the size-ordered range tree in the metaslab and calculate the
2018 2018 * size required to write out the largest segment in our free tree. If the
2019 2019 * size required to represent that segment on disk is larger than the space
2020 2020 * map object then we avoid condensing this map.
2021 2021 *
2022 2022 * To determine the second criterion we use a best-case estimate and assume
2023 2023 * each segment can be represented on-disk as a single 64-bit entry. We refer
2024 2024 * to this best-case estimate as the space map's minimal form.
2025 2025 *
2026 2026 * Unfortunately, we cannot compute the on-disk size of the space map in this
2027 2027 * context because we cannot accurately compute the effects of compression, etc.
2028 2028 * Instead, we apply the heuristic described in the block comment for
2029 2029 * zfs_metaslab_condense_block_threshold - we only condense if the space used
2030 2030 * is greater than a threshold number of blocks.
2031 2031 */
2032 2032 static boolean_t
2033 2033 metaslab_should_condense(metaslab_t *msp)
2034 2034 {
2035 2035 space_map_t *sm = msp->ms_sm;
2036 2036 range_seg_t *rs;
2037 2037 uint64_t size, entries, segsz, object_size, optimal_size, record_size;
2038 2038 dmu_object_info_t doi;
2039 2039 uint64_t vdev_blocksize = 1 << msp->ms_group->mg_vd->vdev_ashift;
2040 2040
2041 2041 ASSERT(MUTEX_HELD(&msp->ms_lock));
2042 2042 ASSERT(msp->ms_loaded);
2043 2043
2044 2044 /*
2045 2045 * Use the ms_size_tree range tree, which is ordered by size, to
2046 2046 * obtain the largest segment in the free tree. We always condense
2047 2047 * metaslabs that are empty and metaslabs for which a condense
2048 2048 * request has been made.
2049 2049 */
2050 2050 rs = avl_last(&msp->ms_size_tree);
2051 2051 if (rs == NULL || msp->ms_condense_wanted)
2052 2052 return (B_TRUE);
2053 2053
2054 2054 /*
2055 2055 * Calculate the number of 64-bit entries this segment would
2056 2056 * require when written to disk. If this single segment would be
2057 2057 * larger on-disk than the entire current on-disk structure, then
2058 2058 * clearly condensing will increase the on-disk structure size.
2059 2059 */
2060 2060 size = (rs->rs_end - rs->rs_start) >> sm->sm_shift;
2061 2061 entries = size / (MIN(size, SM_RUN_MAX));
2062 2062 segsz = entries * sizeof (uint64_t);
2063 2063
2064 2064 optimal_size = sizeof (uint64_t) * avl_numnodes(&msp->ms_tree->rt_root);
2065 2065 object_size = space_map_length(msp->ms_sm);
2066 2066
2067 2067 dmu_object_info_from_db(sm->sm_dbuf, &doi);
2068 2068 record_size = MAX(doi.doi_data_block_size, vdev_blocksize);
2069 2069
2070 2070 return (segsz <= object_size &&
2071 2071 object_size >= (optimal_size * zfs_condense_pct / 100) &&
2072 2072 object_size > zfs_metaslab_condense_block_threshold * record_size);
2073 2073 }
2074 2074
2075 2075 /*
2076 2076 * Condense the on-disk space map representation to its minimized form.
2077 2077 * The minimized form consists of a small number of allocations followed by
2078 2078 * the entries of the free range tree.
2079 2079 */
2080 2080 static void
2081 2081 metaslab_condense(metaslab_t *msp, uint64_t txg, dmu_tx_t *tx)
2082 2082 {
2083 2083 spa_t *spa = msp->ms_group->mg_vd->vdev_spa;
2084 2084 range_tree_t *condense_tree;
2085 2085 space_map_t *sm = msp->ms_sm;
2086 2086
2087 2087 ASSERT(MUTEX_HELD(&msp->ms_lock));
2088 2088 ASSERT3U(spa_sync_pass(spa), ==, 1);
2089 2089 ASSERT(msp->ms_loaded);
2090 2090
2091 2091
2092 2092 spa_dbgmsg(spa, "condensing: txg %llu, msp[%llu] %p, vdev id %llu, "
2093 2093 "spa %s, smp size %llu, segments %lu, forcing condense=%s", txg,
2094 2094 msp->ms_id, msp, msp->ms_group->mg_vd->vdev_id,
2095 2095 msp->ms_group->mg_vd->vdev_spa->spa_name,
2096 2096 space_map_length(msp->ms_sm), avl_numnodes(&msp->ms_tree->rt_root),
2097 2097 msp->ms_condense_wanted ? "TRUE" : "FALSE");
2098 2098
2099 2099 msp->ms_condense_wanted = B_FALSE;
2100 2100
2101 2101 /*
2102 2102 * Create an range tree that is 100% allocated. We remove segments
2103 2103 * that have been freed in this txg, any deferred frees that exist,
2104 2104 * and any allocation in the future. Removing segments should be
2105 2105 * a relatively inexpensive operation since we expect these trees to
2106 2106 * have a small number of nodes.
2107 2107 */
2108 2108 condense_tree = range_tree_create(NULL, NULL, &msp->ms_lock);
2109 2109 range_tree_add(condense_tree, msp->ms_start, msp->ms_size);
2110 2110
2111 2111 /*
2112 2112 * Remove what's been freed in this txg from the condense_tree.
2113 2113 * Since we're in sync_pass 1, we know that all the frees from
2114 2114 * this txg are in the freeingtree.
2115 2115 */
2116 2116 range_tree_walk(msp->ms_freeingtree, range_tree_remove, condense_tree);
2117 2117
2118 2118 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
2119 2119 range_tree_walk(msp->ms_defertree[t],
2120 2120 range_tree_remove, condense_tree);
2121 2121 }
2122 2122
2123 2123 for (int t = 1; t < TXG_CONCURRENT_STATES; t++) {
2124 2124 range_tree_walk(msp->ms_alloctree[(txg + t) & TXG_MASK],
2125 2125 range_tree_remove, condense_tree);
2126 2126 }
2127 2127
2128 2128 /*
2129 2129 * We're about to drop the metaslab's lock thus allowing
2130 2130 * other consumers to change it's content. Set the
2131 2131 * metaslab's ms_condensing flag to ensure that
2132 2132 * allocations on this metaslab do not occur while we're
2133 2133 * in the middle of committing it to disk. This is only critical
2134 2134 * for the ms_tree as all other range trees use per txg
2135 2135 * views of their content.
2136 2136 */
2137 2137 msp->ms_condensing = B_TRUE;
2138 2138
2139 2139 mutex_exit(&msp->ms_lock);
2140 2140 space_map_truncate(sm, tx);
2141 2141 mutex_enter(&msp->ms_lock);
2142 2142
2143 2143 /*
2144 2144 * While we would ideally like to create a space map representation
2145 2145 * that consists only of allocation records, doing so can be
2146 2146 * prohibitively expensive because the in-core free tree can be
2147 2147 * large, and therefore computationally expensive to subtract
2148 2148 * from the condense_tree. Instead we sync out two trees, a cheap
2149 2149 * allocation only tree followed by the in-core free tree. While not
2150 2150 * optimal, this is typically close to optimal, and much cheaper to
2151 2151 * compute.
2152 2152 */
2153 2153 space_map_write(sm, condense_tree, SM_ALLOC, tx);
2154 2154 range_tree_vacate(condense_tree, NULL, NULL);
2155 2155 range_tree_destroy(condense_tree);
2156 2156
2157 2157 space_map_write(sm, msp->ms_tree, SM_FREE, tx);
2158 2158 msp->ms_condensing = B_FALSE;
2159 2159 }
2160 2160
2161 2161 /*
2162 2162 * Write a metaslab to disk in the context of the specified transaction group.
2163 2163 */
2164 2164 void
2165 2165 metaslab_sync(metaslab_t *msp, uint64_t txg)
2166 2166 {
2167 2167 metaslab_group_t *mg = msp->ms_group;
2168 2168 vdev_t *vd = mg->mg_vd;
2169 2169 spa_t *spa = vd->vdev_spa;
2170 2170 objset_t *mos = spa_meta_objset(spa);
2171 2171 range_tree_t *alloctree = msp->ms_alloctree[txg & TXG_MASK];
2172 2172 dmu_tx_t *tx;
2173 2173 uint64_t object = space_map_object(msp->ms_sm);
2174 2174
2175 2175 ASSERT(!vd->vdev_ishole);
2176 2176
2177 2177 /*
2178 2178 * This metaslab has just been added so there's no work to do now.
2179 2179 */
2180 2180 if (msp->ms_freeingtree == NULL) {
2181 2181 ASSERT3P(alloctree, ==, NULL);
2182 2182 return;
2183 2183 }
2184 2184
2185 2185 ASSERT3P(alloctree, !=, NULL);
2186 2186 ASSERT3P(msp->ms_freeingtree, !=, NULL);
2187 2187 ASSERT3P(msp->ms_freedtree, !=, NULL);
2188 2188
2189 2189 /*
2190 2190 * Normally, we don't want to process a metaslab if there
2191 2191 * are no allocations or frees to perform. However, if the metaslab
2192 2192 * is being forced to condense we need to let it through.
2193 2193 */
2194 2194 if (range_tree_space(alloctree) == 0 &&
2195 2195 range_tree_space(msp->ms_freeingtree) == 0 &&
2196 2196 !msp->ms_condense_wanted)
2197 2197 return;
2198 2198
2199 2199 /*
2200 2200 * The only state that can actually be changing concurrently with
2201 2201 * metaslab_sync() is the metaslab's ms_tree. No other thread can
2202 2202 * be modifying this txg's alloctree, freeingtree, freedtree, or
2203 2203 * space_map_phys_t. Therefore, we only hold ms_lock to satify
2204 2204 * space map ASSERTs. We drop it whenever we call into the DMU,
2205 2205 * because the DMU can call down to us (e.g. via zio_free()) at
2206 2206 * any time.
2207 2207 */
2208 2208
2209 2209 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2210 2210
2211 2211 if (msp->ms_sm == NULL) {
2212 2212 uint64_t new_object;
2213 2213
2214 2214 new_object = space_map_alloc(mos, tx);
2215 2215 VERIFY3U(new_object, !=, 0);
2216 2216
2217 2217 VERIFY0(space_map_open(&msp->ms_sm, mos, new_object,
2218 2218 msp->ms_start, msp->ms_size, vd->vdev_ashift,
2219 2219 &msp->ms_lock));
2220 2220 ASSERT(msp->ms_sm != NULL);
2221 2221 }
2222 2222
2223 2223 mutex_enter(&msp->ms_lock);
2224 2224
2225 2225 /*
2226 2226 * Note: metaslab_condense() clears the space map's histogram.
2227 2227 * Therefore we must verify and remove this histogram before
2228 2228 * condensing.
2229 2229 */
2230 2230 metaslab_group_histogram_verify(mg);
2231 2231 metaslab_class_histogram_verify(mg->mg_class);
2232 2232 metaslab_group_histogram_remove(mg, msp);
2233 2233
2234 2234 if (msp->ms_loaded && spa_sync_pass(spa) == 1 &&
2235 2235 metaslab_should_condense(msp)) {
2236 2236 metaslab_condense(msp, txg, tx);
2237 2237 } else {
2238 2238 space_map_write(msp->ms_sm, alloctree, SM_ALLOC, tx);
2239 2239 space_map_write(msp->ms_sm, msp->ms_freeingtree, SM_FREE, tx);
2240 2240 }
2241 2241
2242 2242 if (msp->ms_loaded) {
2243 2243 /*
2244 2244 * When the space map is loaded, we have an accruate
2245 2245 * histogram in the range tree. This gives us an opportunity
2246 2246 * to bring the space map's histogram up-to-date so we clear
2247 2247 * it first before updating it.
2248 2248 */
2249 2249 space_map_histogram_clear(msp->ms_sm);
2250 2250 space_map_histogram_add(msp->ms_sm, msp->ms_tree, tx);
2251 2251
2252 2252 /*
2253 2253 * Since we've cleared the histogram we need to add back
2254 2254 * any free space that has already been processed, plus
2255 2255 * any deferred space. This allows the on-disk histogram
2256 2256 * to accurately reflect all free space even if some space
2257 2257 * is not yet available for allocation (i.e. deferred).
2258 2258 */
2259 2259 space_map_histogram_add(msp->ms_sm, msp->ms_freedtree, tx);
2260 2260
2261 2261 /*
2262 2262 * Add back any deferred free space that has not been
2263 2263 * added back into the in-core free tree yet. This will
2264 2264 * ensure that we don't end up with a space map histogram
2265 2265 * that is completely empty unless the metaslab is fully
2266 2266 * allocated.
2267 2267 */
2268 2268 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
2269 2269 space_map_histogram_add(msp->ms_sm,
2270 2270 msp->ms_defertree[t], tx);
2271 2271 }
2272 2272 }
2273 2273
2274 2274 /*
2275 2275 * Always add the free space from this sync pass to the space
2276 2276 * map histogram. We want to make sure that the on-disk histogram
2277 2277 * accounts for all free space. If the space map is not loaded,
2278 2278 * then we will lose some accuracy but will correct it the next
2279 2279 * time we load the space map.
2280 2280 */
2281 2281 space_map_histogram_add(msp->ms_sm, msp->ms_freeingtree, tx);
2282 2282
2283 2283 metaslab_group_histogram_add(mg, msp);
2284 2284 metaslab_group_histogram_verify(mg);
2285 2285 metaslab_class_histogram_verify(mg->mg_class);
2286 2286
2287 2287 /*
2288 2288 * For sync pass 1, we avoid traversing this txg's free range tree
2289 2289 * and instead will just swap the pointers for freeingtree and
2290 2290 * freedtree. We can safely do this since the freed_tree is
2291 2291 * guaranteed to be empty on the initial pass.
2292 2292 */
2293 2293 if (spa_sync_pass(spa) == 1) {
2294 2294 range_tree_swap(&msp->ms_freeingtree, &msp->ms_freedtree);
2295 2295 } else {
2296 2296 range_tree_vacate(msp->ms_freeingtree,
2297 2297 range_tree_add, msp->ms_freedtree);
2298 2298 }
2299 2299 range_tree_vacate(alloctree, NULL, NULL);
2300 2300
2301 2301 ASSERT0(range_tree_space(msp->ms_alloctree[txg & TXG_MASK]));
2302 2302 ASSERT0(range_tree_space(msp->ms_alloctree[TXG_CLEAN(txg) & TXG_MASK]));
2303 2303 ASSERT0(range_tree_space(msp->ms_freeingtree));
2304 2304
2305 2305 mutex_exit(&msp->ms_lock);
2306 2306
2307 2307 if (object != space_map_object(msp->ms_sm)) {
2308 2308 object = space_map_object(msp->ms_sm);
2309 2309 dmu_write(mos, vd->vdev_ms_array, sizeof (uint64_t) *
2310 2310 msp->ms_id, sizeof (uint64_t), &object, tx);
2311 2311 }
2312 2312 dmu_tx_commit(tx);
2313 2313 }
2314 2314
2315 2315 /*
2316 2316 * Called after a transaction group has completely synced to mark
2317 2317 * all of the metaslab's free space as usable.
2318 2318 */
2319 2319 void
2320 2320 metaslab_sync_done(metaslab_t *msp, uint64_t txg)
2321 2321 {
2322 2322 metaslab_group_t *mg = msp->ms_group;
2323 2323 vdev_t *vd = mg->mg_vd;
2324 2324 spa_t *spa = vd->vdev_spa;
2325 2325 range_tree_t **defer_tree;
2326 2326 int64_t alloc_delta, defer_delta;
2327 2327 boolean_t defer_allowed = B_TRUE;
2328 2328
2329 2329 ASSERT(!vd->vdev_ishole);
2330 2330
2331 2331 mutex_enter(&msp->ms_lock);
2332 2332
2333 2333 /*
2334 2334 * If this metaslab is just becoming available, initialize its
2335 2335 * range trees and add its capacity to the vdev.
2336 2336 */
2337 2337 if (msp->ms_freedtree == NULL) {
2338 2338 for (int t = 0; t < TXG_SIZE; t++) {
2339 2339 ASSERT(msp->ms_alloctree[t] == NULL);
2340 2340
2341 2341 msp->ms_alloctree[t] = range_tree_create(NULL, msp,
2342 2342 &msp->ms_lock);
2343 2343 }
2344 2344
2345 2345 ASSERT3P(msp->ms_freeingtree, ==, NULL);
2346 2346 msp->ms_freeingtree = range_tree_create(NULL, msp,
2347 2347 &msp->ms_lock);
2348 2348
2349 2349 ASSERT3P(msp->ms_freedtree, ==, NULL);
2350 2350 msp->ms_freedtree = range_tree_create(NULL, msp,
2351 2351 &msp->ms_lock);
2352 2352
2353 2353 for (int t = 0; t < TXG_DEFER_SIZE; t++) {
2354 2354 ASSERT(msp->ms_defertree[t] == NULL);
2355 2355
2356 2356 msp->ms_defertree[t] = range_tree_create(NULL, msp,
2357 2357 &msp->ms_lock);
2358 2358 }
2359 2359
2360 2360 vdev_space_update(vd, 0, 0, msp->ms_size);
2361 2361 }
2362 2362
2363 2363 defer_tree = &msp->ms_defertree[txg % TXG_DEFER_SIZE];
2364 2364
2365 2365 uint64_t free_space = metaslab_class_get_space(spa_normal_class(spa)) -
2366 2366 metaslab_class_get_alloc(spa_normal_class(spa));
2367 2367 if (free_space <= spa_get_slop_space(spa)) {
2368 2368 defer_allowed = B_FALSE;
2369 2369 }
2370 2370
2371 2371 defer_delta = 0;
2372 2372 alloc_delta = space_map_alloc_delta(msp->ms_sm);
2373 2373 if (defer_allowed) {
2374 2374 defer_delta = range_tree_space(msp->ms_freedtree) -
2375 2375 range_tree_space(*defer_tree);
2376 2376 } else {
2377 2377 defer_delta -= range_tree_space(*defer_tree);
2378 2378 }
2379 2379
2380 2380 vdev_space_update(vd, alloc_delta + defer_delta, defer_delta, 0);
2381 2381
2382 2382 /*
2383 2383 * If there's a metaslab_load() in progress, wait for it to complete
2384 2384 * so that we have a consistent view of the in-core space map.
2385 2385 */
2386 2386 metaslab_load_wait(msp);
2387 2387
2388 2388 /*
2389 2389 * Move the frees from the defer_tree back to the free
2390 2390 * range tree (if it's loaded). Swap the freed_tree and the
2391 2391 * defer_tree -- this is safe to do because we've just emptied out
2392 2392 * the defer_tree.
2393 2393 */
2394 2394 range_tree_vacate(*defer_tree,
2395 2395 msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree);
2396 2396 if (defer_allowed) {
2397 2397 range_tree_swap(&msp->ms_freedtree, defer_tree);
2398 2398 } else {
2399 2399 range_tree_vacate(msp->ms_freedtree,
2400 2400 msp->ms_loaded ? range_tree_add : NULL, msp->ms_tree);
2401 2401 }
2402 2402
2403 2403 space_map_update(msp->ms_sm);
2404 2404
2405 2405 msp->ms_deferspace += defer_delta;
2406 2406 ASSERT3S(msp->ms_deferspace, >=, 0);
2407 2407 ASSERT3S(msp->ms_deferspace, <=, msp->ms_size);
2408 2408 if (msp->ms_deferspace != 0) {
2409 2409 /*
2410 2410 * Keep syncing this metaslab until all deferred frees
2411 2411 * are back in circulation.
2412 2412 */
2413 2413 vdev_dirty(vd, VDD_METASLAB, msp, txg + 1);
2414 2414 }
2415 2415
2416 2416 /*
2417 2417 * Calculate the new weights before unloading any metaslabs.
2418 2418 * This will give us the most accurate weighting.
2419 2419 */
2420 2420 metaslab_group_sort(mg, msp, metaslab_weight(msp));
2421 2421
2422 2422 /*
2423 2423 * If the metaslab is loaded and we've not tried to load or allocate
2424 2424 * from it in 'metaslab_unload_delay' txgs, then unload it.
2425 2425 */
2426 2426 if (msp->ms_loaded &&
2427 2427 msp->ms_selected_txg + metaslab_unload_delay < txg) {
2428 2428 for (int t = 1; t < TXG_CONCURRENT_STATES; t++) {
2429 2429 VERIFY0(range_tree_space(
2430 2430 msp->ms_alloctree[(txg + t) & TXG_MASK]));
2431 2431 }
2432 2432
2433 2433 if (!metaslab_debug_unload)
2434 2434 metaslab_unload(msp);
2435 2435 }
2436 2436
2437 2437 mutex_exit(&msp->ms_lock);
2438 2438 }
2439 2439
2440 2440 void
2441 2441 metaslab_sync_reassess(metaslab_group_t *mg)
2442 2442 {
2443 2443 metaslab_group_alloc_update(mg);
2444 2444 mg->mg_fragmentation = metaslab_group_fragmentation(mg);
2445 2445
2446 2446 /*
2447 2447 * Preload the next potential metaslabs
2448 2448 */
2449 2449 metaslab_group_preload(mg);
2450 2450 }
2451 2451
2452 2452 static uint64_t
2453 2453 metaslab_distance(metaslab_t *msp, dva_t *dva)
2454 2454 {
2455 2455 uint64_t ms_shift = msp->ms_group->mg_vd->vdev_ms_shift;
2456 2456 uint64_t offset = DVA_GET_OFFSET(dva) >> ms_shift;
2457 2457 uint64_t start = msp->ms_id;
2458 2458
2459 2459 if (msp->ms_group->mg_vd->vdev_id != DVA_GET_VDEV(dva))
2460 2460 return (1ULL << 63);
2461 2461
2462 2462 if (offset < start)
2463 2463 return ((start - offset) << ms_shift);
2464 2464 if (offset > start)
2465 2465 return ((offset - start) << ms_shift);
2466 2466 return (0);
2467 2467 }
2468 2468
2469 2469 /*
2470 2470 * ==========================================================================
2471 2471 * Metaslab allocation tracing facility
2472 2472 * ==========================================================================
2473 2473 */
2474 2474 kstat_t *metaslab_trace_ksp;
2475 2475 kstat_named_t metaslab_trace_over_limit;
2476 2476
2477 2477 void
2478 2478 metaslab_alloc_trace_init(void)
2479 2479 {
2480 2480 ASSERT(metaslab_alloc_trace_cache == NULL);
2481 2481 metaslab_alloc_trace_cache = kmem_cache_create(
2482 2482 "metaslab_alloc_trace_cache", sizeof (metaslab_alloc_trace_t),
2483 2483 0, NULL, NULL, NULL, NULL, NULL, 0);
2484 2484 metaslab_trace_ksp = kstat_create("zfs", 0, "metaslab_trace_stats",
2485 2485 "misc", KSTAT_TYPE_NAMED, 1, KSTAT_FLAG_VIRTUAL);
2486 2486 if (metaslab_trace_ksp != NULL) {
2487 2487 metaslab_trace_ksp->ks_data = &metaslab_trace_over_limit;
2488 2488 kstat_named_init(&metaslab_trace_over_limit,
2489 2489 "metaslab_trace_over_limit", KSTAT_DATA_UINT64);
2490 2490 kstat_install(metaslab_trace_ksp);
2491 2491 }
2492 2492 }
2493 2493
2494 2494 void
2495 2495 metaslab_alloc_trace_fini(void)
2496 2496 {
2497 2497 if (metaslab_trace_ksp != NULL) {
2498 2498 kstat_delete(metaslab_trace_ksp);
2499 2499 metaslab_trace_ksp = NULL;
2500 2500 }
2501 2501 kmem_cache_destroy(metaslab_alloc_trace_cache);
2502 2502 metaslab_alloc_trace_cache = NULL;
2503 2503 }
2504 2504
2505 2505 /*
2506 2506 * Add an allocation trace element to the allocation tracing list.
2507 2507 */
2508 2508 static void
2509 2509 metaslab_trace_add(zio_alloc_list_t *zal, metaslab_group_t *mg,
2510 2510 metaslab_t *msp, uint64_t psize, uint32_t dva_id, uint64_t offset)
2511 2511 {
2512 2512 if (!metaslab_trace_enabled)
2513 2513 return;
2514 2514
2515 2515 /*
2516 2516 * When the tracing list reaches its maximum we remove
2517 2517 * the second element in the list before adding a new one.
2518 2518 * By removing the second element we preserve the original
2519 2519 * entry as a clue to what allocations steps have already been
2520 2520 * performed.
2521 2521 */
2522 2522 if (zal->zal_size == metaslab_trace_max_entries) {
2523 2523 metaslab_alloc_trace_t *mat_next;
2524 2524 #ifdef DEBUG
2525 2525 panic("too many entries in allocation list");
2526 2526 #endif
2527 2527 atomic_inc_64(&metaslab_trace_over_limit.value.ui64);
2528 2528 zal->zal_size--;
2529 2529 mat_next = list_next(&zal->zal_list, list_head(&zal->zal_list));
2530 2530 list_remove(&zal->zal_list, mat_next);
2531 2531 kmem_cache_free(metaslab_alloc_trace_cache, mat_next);
2532 2532 }
2533 2533
2534 2534 metaslab_alloc_trace_t *mat =
2535 2535 kmem_cache_alloc(metaslab_alloc_trace_cache, KM_SLEEP);
2536 2536 list_link_init(&mat->mat_list_node);
2537 2537 mat->mat_mg = mg;
2538 2538 mat->mat_msp = msp;
2539 2539 mat->mat_size = psize;
2540 2540 mat->mat_dva_id = dva_id;
2541 2541 mat->mat_offset = offset;
2542 2542 mat->mat_weight = 0;
2543 2543
2544 2544 if (msp != NULL)
2545 2545 mat->mat_weight = msp->ms_weight;
2546 2546
2547 2547 /*
2548 2548 * The list is part of the zio so locking is not required. Only
2549 2549 * a single thread will perform allocations for a given zio.
2550 2550 */
2551 2551 list_insert_tail(&zal->zal_list, mat);
2552 2552 zal->zal_size++;
2553 2553
2554 2554 ASSERT3U(zal->zal_size, <=, metaslab_trace_max_entries);
2555 2555 }
2556 2556
2557 2557 void
2558 2558 metaslab_trace_init(zio_alloc_list_t *zal)
2559 2559 {
2560 2560 list_create(&zal->zal_list, sizeof (metaslab_alloc_trace_t),
2561 2561 offsetof(metaslab_alloc_trace_t, mat_list_node));
2562 2562 zal->zal_size = 0;
2563 2563 }
2564 2564
2565 2565 void
2566 2566 metaslab_trace_fini(zio_alloc_list_t *zal)
2567 2567 {
2568 2568 metaslab_alloc_trace_t *mat;
2569 2569
2570 2570 while ((mat = list_remove_head(&zal->zal_list)) != NULL)
2571 2571 kmem_cache_free(metaslab_alloc_trace_cache, mat);
2572 2572 list_destroy(&zal->zal_list);
2573 2573 zal->zal_size = 0;
2574 2574 }
2575 2575
2576 2576 /*
2577 2577 * ==========================================================================
2578 2578 * Metaslab block operations
2579 2579 * ==========================================================================
2580 2580 */
2581 2581
2582 2582 static void
2583 2583 metaslab_group_alloc_increment(spa_t *spa, uint64_t vdev, void *tag, int flags)
2584 2584 {
2585 2585 if (!(flags & METASLAB_ASYNC_ALLOC) ||
2586 2586 flags & METASLAB_DONT_THROTTLE)
2587 2587 return;
2588 2588
2589 2589 metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
2590 2590 if (!mg->mg_class->mc_alloc_throttle_enabled)
2591 2591 return;
2592 2592
2593 2593 (void) refcount_add(&mg->mg_alloc_queue_depth, tag);
2594 2594 }
2595 2595
2596 2596 void
2597 2597 metaslab_group_alloc_decrement(spa_t *spa, uint64_t vdev, void *tag, int flags)
2598 2598 {
2599 2599 if (!(flags & METASLAB_ASYNC_ALLOC) ||
2600 2600 flags & METASLAB_DONT_THROTTLE)
2601 2601 return;
2602 2602
2603 2603 metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
2604 2604 if (!mg->mg_class->mc_alloc_throttle_enabled)
2605 2605 return;
2606 2606
2607 2607 (void) refcount_remove(&mg->mg_alloc_queue_depth, tag);
2608 2608 }
2609 2609
2610 2610 void
2611 2611 metaslab_group_alloc_verify(spa_t *spa, const blkptr_t *bp, void *tag)
2612 2612 {
2613 2613 #ifdef ZFS_DEBUG
2614 2614 const dva_t *dva = bp->blk_dva;
2615 2615 int ndvas = BP_GET_NDVAS(bp);
2616 2616
2617 2617 for (int d = 0; d < ndvas; d++) {
2618 2618 uint64_t vdev = DVA_GET_VDEV(&dva[d]);
2619 2619 metaslab_group_t *mg = vdev_lookup_top(spa, vdev)->vdev_mg;
2620 2620 VERIFY(refcount_not_held(&mg->mg_alloc_queue_depth, tag));
2621 2621 }
2622 2622 #endif
2623 2623 }
2624 2624
2625 2625 static uint64_t
2626 2626 metaslab_block_alloc(metaslab_t *msp, uint64_t size, uint64_t txg)
2627 2627 {
2628 2628 uint64_t start;
2629 2629 range_tree_t *rt = msp->ms_tree;
2630 2630 metaslab_class_t *mc = msp->ms_group->mg_class;
2631 2631
2632 2632 VERIFY(!msp->ms_condensing);
2633 2633
2634 2634 start = mc->mc_ops->msop_alloc(msp, size);
2635 2635 if (start != -1ULL) {
2636 2636 metaslab_group_t *mg = msp->ms_group;
2637 2637 vdev_t *vd = mg->mg_vd;
2638 2638
2639 2639 VERIFY0(P2PHASE(start, 1ULL << vd->vdev_ashift));
2640 2640 VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
2641 2641 VERIFY3U(range_tree_space(rt) - size, <=, msp->ms_size);
2642 2642 range_tree_remove(rt, start, size);
2643 2643
2644 2644 if (range_tree_space(msp->ms_alloctree[txg & TXG_MASK]) == 0)
2645 2645 vdev_dirty(mg->mg_vd, VDD_METASLAB, msp, txg);
2646 2646
2647 2647 range_tree_add(msp->ms_alloctree[txg & TXG_MASK], start, size);
2648 2648
2649 2649 /* Track the last successful allocation */
2650 2650 msp->ms_alloc_txg = txg;
2651 2651 metaslab_verify_space(msp, txg);
2652 2652 }
2653 2653
2654 2654 /*
2655 2655 * Now that we've attempted the allocation we need to update the
2656 2656 * metaslab's maximum block size since it may have changed.
2657 2657 */
2658 2658 msp->ms_max_size = metaslab_block_maxsize(msp);
2659 2659 return (start);
2660 2660 }
2661 2661
2662 2662 static uint64_t
2663 2663 metaslab_group_alloc_normal(metaslab_group_t *mg, zio_alloc_list_t *zal,
2664 2664 uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d)
2665 2665 {
2666 2666 metaslab_t *msp = NULL;
2667 2667 uint64_t offset = -1ULL;
2668 2668 uint64_t activation_weight;
2669 2669 uint64_t target_distance;
2670 2670 int i;
2671 2671
2672 2672 activation_weight = METASLAB_WEIGHT_PRIMARY;
2673 2673 for (i = 0; i < d; i++) {
2674 2674 if (DVA_GET_VDEV(&dva[i]) == mg->mg_vd->vdev_id) {
2675 2675 activation_weight = METASLAB_WEIGHT_SECONDARY;
2676 2676 break;
2677 2677 }
2678 2678 }
2679 2679
2680 2680 metaslab_t *search = kmem_alloc(sizeof (*search), KM_SLEEP);
2681 2681 search->ms_weight = UINT64_MAX;
2682 2682 search->ms_start = 0;
2683 2683 for (;;) {
2684 2684 boolean_t was_active;
2685 2685 avl_tree_t *t = &mg->mg_metaslab_tree;
2686 2686 avl_index_t idx;
2687 2687
2688 2688 mutex_enter(&mg->mg_lock);
2689 2689
2690 2690 /*
2691 2691 * Find the metaslab with the highest weight that is less
2692 2692 * than what we've already tried. In the common case, this
2693 2693 * means that we will examine each metaslab at most once.
2694 2694 * Note that concurrent callers could reorder metaslabs
2695 2695 * by activation/passivation once we have dropped the mg_lock.
2696 2696 * If a metaslab is activated by another thread, and we fail
2697 2697 * to allocate from the metaslab we have selected, we may
2698 2698 * not try the newly-activated metaslab, and instead activate
2699 2699 * another metaslab. This is not optimal, but generally
2700 2700 * does not cause any problems (a possible exception being
2701 2701 * if every metaslab is completely full except for the
2702 2702 * the newly-activated metaslab which we fail to examine).
2703 2703 */
2704 2704 msp = avl_find(t, search, &idx);
2705 2705 if (msp == NULL)
2706 2706 msp = avl_nearest(t, idx, AVL_AFTER);
2707 2707 for (; msp != NULL; msp = AVL_NEXT(t, msp)) {
2708 2708
2709 2709 if (!metaslab_should_allocate(msp, asize)) {
2710 2710 metaslab_trace_add(zal, mg, msp, asize, d,
2711 2711 TRACE_TOO_SMALL);
2712 2712 continue;
2713 2713 }
2714 2714
2715 2715 /*
2716 2716 * If the selected metaslab is condensing, skip it.
2717 2717 */
2718 2718 if (msp->ms_condensing)
2719 2719 continue;
2720 2720
2721 2721 was_active = msp->ms_weight & METASLAB_ACTIVE_MASK;
2722 2722 if (activation_weight == METASLAB_WEIGHT_PRIMARY)
2723 2723 break;
2724 2724
2725 2725 target_distance = min_distance +
2726 2726 (space_map_allocated(msp->ms_sm) != 0 ? 0 :
2727 2727 min_distance >> 1);
2728 2728
2729 2729 for (i = 0; i < d; i++) {
2730 2730 if (metaslab_distance(msp, &dva[i]) <
2731 2731 target_distance)
2732 2732 break;
2733 2733 }
2734 2734 if (i == d)
2735 2735 break;
2736 2736 }
2737 2737 mutex_exit(&mg->mg_lock);
2738 2738 if (msp == NULL) {
2739 2739 kmem_free(search, sizeof (*search));
2740 2740 return (-1ULL);
2741 2741 }
2742 2742 search->ms_weight = msp->ms_weight;
2743 2743 search->ms_start = msp->ms_start + 1;
2744 2744
2745 2745 mutex_enter(&msp->ms_lock);
2746 2746
2747 2747 /*
2748 2748 * Ensure that the metaslab we have selected is still
2749 2749 * capable of handling our request. It's possible that
2750 2750 * another thread may have changed the weight while we
2751 2751 * were blocked on the metaslab lock. We check the
2752 2752 * active status first to see if we need to reselect
2753 2753 * a new metaslab.
2754 2754 */
2755 2755 if (was_active && !(msp->ms_weight & METASLAB_ACTIVE_MASK)) {
2756 2756 mutex_exit(&msp->ms_lock);
2757 2757 continue;
2758 2758 }
2759 2759
2760 2760 if ((msp->ms_weight & METASLAB_WEIGHT_SECONDARY) &&
2761 2761 activation_weight == METASLAB_WEIGHT_PRIMARY) {
2762 2762 metaslab_passivate(msp,
2763 2763 msp->ms_weight & ~METASLAB_ACTIVE_MASK);
2764 2764 mutex_exit(&msp->ms_lock);
2765 2765 continue;
2766 2766 }
2767 2767
2768 2768 if (metaslab_activate(msp, activation_weight) != 0) {
2769 2769 mutex_exit(&msp->ms_lock);
2770 2770 continue;
2771 2771 }
2772 2772 msp->ms_selected_txg = txg;
2773 2773
2774 2774 /*
2775 2775 * Now that we have the lock, recheck to see if we should
2776 2776 * continue to use this metaslab for this allocation. The
2777 2777 * the metaslab is now loaded so metaslab_should_allocate() can
2778 2778 * accurately determine if the allocation attempt should
2779 2779 * proceed.
2780 2780 */
2781 2781 if (!metaslab_should_allocate(msp, asize)) {
2782 2782 /* Passivate this metaslab and select a new one. */
2783 2783 metaslab_trace_add(zal, mg, msp, asize, d,
2784 2784 TRACE_TOO_SMALL);
2785 2785 goto next;
2786 2786 }
2787 2787
2788 2788 /*
2789 2789 * If this metaslab is currently condensing then pick again as
2790 2790 * we can't manipulate this metaslab until it's committed
2791 2791 * to disk.
2792 2792 */
2793 2793 if (msp->ms_condensing) {
2794 2794 metaslab_trace_add(zal, mg, msp, asize, d,
2795 2795 TRACE_CONDENSING);
2796 2796 mutex_exit(&msp->ms_lock);
2797 2797 continue;
2798 2798 }
2799 2799
2800 2800 offset = metaslab_block_alloc(msp, asize, txg);
2801 2801 metaslab_trace_add(zal, mg, msp, asize, d, offset);
2802 2802
2803 2803 if (offset != -1ULL) {
2804 2804 /* Proactively passivate the metaslab, if needed */
2805 2805 metaslab_segment_may_passivate(msp);
2806 2806 break;
2807 2807 }
2808 2808 next:
2809 2809 ASSERT(msp->ms_loaded);
2810 2810
2811 2811 /*
2812 2812 * We were unable to allocate from this metaslab so determine
2813 2813 * a new weight for this metaslab. Now that we have loaded
2814 2814 * the metaslab we can provide a better hint to the metaslab
2815 2815 * selector.
2816 2816 *
2817 2817 * For space-based metaslabs, we use the maximum block size.
2818 2818 * This information is only available when the metaslab
2819 2819 * is loaded and is more accurate than the generic free
2820 2820 * space weight that was calculated by metaslab_weight().
2821 2821 * This information allows us to quickly compare the maximum
2822 2822 * available allocation in the metaslab to the allocation
2823 2823 * size being requested.
2824 2824 *
2825 2825 * For segment-based metaslabs, determine the new weight
2826 2826 * based on the highest bucket in the range tree. We
2827 2827 * explicitly use the loaded segment weight (i.e. the range
2828 2828 * tree histogram) since it contains the space that is
2829 2829 * currently available for allocation and is accurate
2830 2830 * even within a sync pass.
2831 2831 */
2832 2832 if (WEIGHT_IS_SPACEBASED(msp->ms_weight)) {
2833 2833 uint64_t weight = metaslab_block_maxsize(msp);
2834 2834 WEIGHT_SET_SPACEBASED(weight);
2835 2835 metaslab_passivate(msp, weight);
2836 2836 } else {
2837 2837 metaslab_passivate(msp,
2838 2838 metaslab_weight_from_range_tree(msp));
2839 2839 }
2840 2840
2841 2841 /*
2842 2842 * We have just failed an allocation attempt, check
2843 2843 * that metaslab_should_allocate() agrees. Otherwise,
2844 2844 * we may end up in an infinite loop retrying the same
2845 2845 * metaslab.
2846 2846 */
2847 2847 ASSERT(!metaslab_should_allocate(msp, asize));
2848 2848 mutex_exit(&msp->ms_lock);
2849 2849 }
2850 2850 mutex_exit(&msp->ms_lock);
2851 2851 kmem_free(search, sizeof (*search));
2852 2852 return (offset);
2853 2853 }
2854 2854
2855 2855 static uint64_t
2856 2856 metaslab_group_alloc(metaslab_group_t *mg, zio_alloc_list_t *zal,
2857 2857 uint64_t asize, uint64_t txg, uint64_t min_distance, dva_t *dva, int d)
2858 2858 {
2859 2859 uint64_t offset;
2860 2860 ASSERT(mg->mg_initialized);
2861 2861
2862 2862 offset = metaslab_group_alloc_normal(mg, zal, asize, txg,
2863 2863 min_distance, dva, d);
2864 2864
2865 2865 mutex_enter(&mg->mg_lock);
2866 2866 if (offset == -1ULL) {
2867 2867 mg->mg_failed_allocations++;
2868 2868 metaslab_trace_add(zal, mg, NULL, asize, d,
2869 2869 TRACE_GROUP_FAILURE);
2870 2870 if (asize == SPA_GANGBLOCKSIZE) {
2871 2871 /*
2872 2872 * This metaslab group was unable to allocate
2873 2873 * the minimum gang block size so it must be out of
2874 2874 * space. We must notify the allocation throttle
2875 2875 * to start skipping allocation attempts to this
2876 2876 * metaslab group until more space becomes available.
2877 2877 * Note: this failure cannot be caused by the
2878 2878 * allocation throttle since the allocation throttle
2879 2879 * is only responsible for skipping devices and
2880 2880 * not failing block allocations.
2881 2881 */
2882 2882 mg->mg_no_free_space = B_TRUE;
2883 2883 }
2884 2884 }
2885 2885 mg->mg_allocations++;
2886 2886 mutex_exit(&mg->mg_lock);
2887 2887 return (offset);
2888 2888 }
2889 2889
2890 2890 /*
2891 2891 * If we have to write a ditto block (i.e. more than one DVA for a given BP)
2892 2892 * on the same vdev as an existing DVA of this BP, then try to allocate it
2893 2893 * at least (vdev_asize / (2 ^ ditto_same_vdev_distance_shift)) away from the
2894 2894 * existing DVAs.
2895 2895 */
2896 2896 int ditto_same_vdev_distance_shift = 3;
2897 2897
2898 2898 /*
2899 2899 * Allocate a block for the specified i/o.
2900 2900 */
2901 2901 static int
2902 2902 metaslab_alloc_dva(spa_t *spa, metaslab_class_t *mc, uint64_t psize,
2903 2903 dva_t *dva, int d, dva_t *hintdva, uint64_t txg, int flags,
2904 2904 zio_alloc_list_t *zal)
2905 2905 {
2906 2906 metaslab_group_t *mg, *rotor;
2907 2907 vdev_t *vd;
2908 2908 boolean_t try_hard = B_FALSE;
2909 2909
2910 2910 ASSERT(!DVA_IS_VALID(&dva[d]));
2911 2911
2912 2912 /*
2913 2913 * For testing, make some blocks above a certain size be gang blocks.
2914 2914 */
2915 2915 if (psize >= metaslab_gang_bang && (ddi_get_lbolt() & 3) == 0) {
2916 2916 metaslab_trace_add(zal, NULL, NULL, psize, d, TRACE_FORCE_GANG);
2917 2917 return (SET_ERROR(ENOSPC));
2918 2918 }
2919 2919
2920 2920 /*
2921 2921 * Start at the rotor and loop through all mgs until we find something.
2922 2922 * Note that there's no locking on mc_rotor or mc_aliquot because
2923 2923 * nothing actually breaks if we miss a few updates -- we just won't
2924 2924 * allocate quite as evenly. It all balances out over time.
2925 2925 *
2926 2926 * If we are doing ditto or log blocks, try to spread them across
2927 2927 * consecutive vdevs. If we're forced to reuse a vdev before we've
2928 2928 * allocated all of our ditto blocks, then try and spread them out on
2929 2929 * that vdev as much as possible. If it turns out to not be possible,
2930 2930 * gradually lower our standards until anything becomes acceptable.
2931 2931 * Also, allocating on consecutive vdevs (as opposed to random vdevs)
2932 2932 * gives us hope of containing our fault domains to something we're
2933 2933 * able to reason about. Otherwise, any two top-level vdev failures
2934 2934 * will guarantee the loss of data. With consecutive allocation,
2935 2935 * only two adjacent top-level vdev failures will result in data loss.
2936 2936 *
2937 2937 * If we are doing gang blocks (hintdva is non-NULL), try to keep
2938 2938 * ourselves on the same vdev as our gang block header. That
2939 2939 * way, we can hope for locality in vdev_cache, plus it makes our
2940 2940 * fault domains something tractable.
2941 2941 */
2942 2942 if (hintdva) {
2943 2943 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&hintdva[d]));
2944 2944
2945 2945 /*
2946 2946 * It's possible the vdev we're using as the hint no
2947 2947 * longer exists (i.e. removed). Consult the rotor when
2948 2948 * all else fails.
2949 2949 */
2950 2950 if (vd != NULL) {
2951 2951 mg = vd->vdev_mg;
2952 2952
2953 2953 if (flags & METASLAB_HINTBP_AVOID &&
2954 2954 mg->mg_next != NULL)
2955 2955 mg = mg->mg_next;
2956 2956 } else {
2957 2957 mg = mc->mc_rotor;
2958 2958 }
2959 2959 } else if (d != 0) {
2960 2960 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&dva[d - 1]));
2961 2961 mg = vd->vdev_mg->mg_next;
2962 2962 } else {
2963 2963 mg = mc->mc_rotor;
2964 2964 }
2965 2965
2966 2966 /*
2967 2967 * If the hint put us into the wrong metaslab class, or into a
2968 2968 * metaslab group that has been passivated, just follow the rotor.
2969 2969 */
2970 2970 if (mg->mg_class != mc || mg->mg_activation_count <= 0)
2971 2971 mg = mc->mc_rotor;
2972 2972
2973 2973 rotor = mg;
2974 2974 top:
2975 2975 do {
2976 2976 boolean_t allocatable;
2977 2977
2978 2978 ASSERT(mg->mg_activation_count == 1);
2979 2979 vd = mg->mg_vd;
2980 2980
2981 2981 /*
2982 2982 * Don't allocate from faulted devices.
2983 2983 */
2984 2984 if (try_hard) {
2985 2985 spa_config_enter(spa, SCL_ZIO, FTAG, RW_READER);
2986 2986 allocatable = vdev_allocatable(vd);
2987 2987 spa_config_exit(spa, SCL_ZIO, FTAG);
2988 2988 } else {
2989 2989 allocatable = vdev_allocatable(vd);
2990 2990 }
2991 2991
2992 2992 /*
2993 2993 * Determine if the selected metaslab group is eligible
2994 2994 * for allocations. If we're ganging then don't allow
2995 2995 * this metaslab group to skip allocations since that would
2996 2996 * inadvertently return ENOSPC and suspend the pool
2997 2997 * even though space is still available.
2998 2998 */
2999 2999 if (allocatable && !GANG_ALLOCATION(flags) && !try_hard) {
3000 3000 allocatable = metaslab_group_allocatable(mg, rotor,
3001 3001 psize);
3002 3002 }
3003 3003
3004 3004 if (!allocatable) {
3005 3005 metaslab_trace_add(zal, mg, NULL, psize, d,
3006 3006 TRACE_NOT_ALLOCATABLE);
3007 3007 goto next;
3008 3008 }
3009 3009
3010 3010 ASSERT(mg->mg_initialized);
3011 3011
3012 3012 /*
3013 3013 * Avoid writing single-copy data to a failing,
3014 3014 * non-redundant vdev, unless we've already tried all
3015 3015 * other vdevs.
3016 3016 */
3017 3017 if ((vd->vdev_stat.vs_write_errors > 0 ||
3018 3018 vd->vdev_state < VDEV_STATE_HEALTHY) &&
3019 3019 d == 0 && !try_hard && vd->vdev_children == 0) {
3020 3020 metaslab_trace_add(zal, mg, NULL, psize, d,
3021 3021 TRACE_VDEV_ERROR);
3022 3022 goto next;
3023 3023 }
3024 3024
3025 3025 ASSERT(mg->mg_class == mc);
3026 3026
3027 3027 /*
3028 3028 * If we don't need to try hard, then require that the
3029 3029 * block be 1/8th of the device away from any other DVAs
3030 3030 * in this BP. If we are trying hard, allow any offset
3031 3031 * to be used (distance=0).
3032 3032 */
3033 3033 uint64_t distance = 0;
3034 3034 if (!try_hard) {
3035 3035 distance = vd->vdev_asize >>
3036 3036 ditto_same_vdev_distance_shift;
3037 3037 if (distance <= (1ULL << vd->vdev_ms_shift))
3038 3038 distance = 0;
3039 3039 }
3040 3040
3041 3041 uint64_t asize = vdev_psize_to_asize(vd, psize);
3042 3042 ASSERT(P2PHASE(asize, 1ULL << vd->vdev_ashift) == 0);
3043 3043
3044 3044 uint64_t offset = metaslab_group_alloc(mg, zal, asize, txg,
3045 3045 distance, dva, d);
3046 3046
3047 3047 if (offset != -1ULL) {
3048 3048 /*
3049 3049 * If we've just selected this metaslab group,
3050 3050 * figure out whether the corresponding vdev is
3051 3051 * over- or under-used relative to the pool,
3052 3052 * and set an allocation bias to even it out.
3053 3053 */
3054 3054 if (mc->mc_aliquot == 0 && metaslab_bias_enabled) {
3055 3055 vdev_stat_t *vs = &vd->vdev_stat;
3056 3056 int64_t vu, cu;
3057 3057
3058 3058 vu = (vs->vs_alloc * 100) / (vs->vs_space + 1);
3059 3059 cu = (mc->mc_alloc * 100) / (mc->mc_space + 1);
3060 3060
3061 3061 /*
3062 3062 * Calculate how much more or less we should
3063 3063 * try to allocate from this device during
3064 3064 * this iteration around the rotor.
3065 3065 * For example, if a device is 80% full
3066 3066 * and the pool is 20% full then we should
3067 3067 * reduce allocations by 60% on this device.
3068 3068 *
3069 3069 * mg_bias = (20 - 80) * 512K / 100 = -307K
3070 3070 *
3071 3071 * This reduces allocations by 307K for this
3072 3072 * iteration.
3073 3073 */
3074 3074 mg->mg_bias = ((cu - vu) *
3075 3075 (int64_t)mg->mg_aliquot) / 100;
3076 3076 } else if (!metaslab_bias_enabled) {
3077 3077 mg->mg_bias = 0;
3078 3078 }
3079 3079
3080 3080 if (atomic_add_64_nv(&mc->mc_aliquot, asize) >=
3081 3081 mg->mg_aliquot + mg->mg_bias) {
3082 3082 mc->mc_rotor = mg->mg_next;
3083 3083 mc->mc_aliquot = 0;
3084 3084 }
3085 3085
3086 3086 DVA_SET_VDEV(&dva[d], vd->vdev_id);
3087 3087 DVA_SET_OFFSET(&dva[d], offset);
3088 3088 DVA_SET_GANG(&dva[d], !!(flags & METASLAB_GANG_HEADER));
3089 3089 DVA_SET_ASIZE(&dva[d], asize);
3090 3090
3091 3091 return (0);
3092 3092 }
3093 3093 next:
3094 3094 mc->mc_rotor = mg->mg_next;
3095 3095 mc->mc_aliquot = 0;
3096 3096 } while ((mg = mg->mg_next) != rotor);
3097 3097
3098 3098 /*
3099 3099 * If we haven't tried hard, do so now.
3100 3100 */
3101 3101 if (!try_hard) {
3102 3102 try_hard = B_TRUE;
3103 3103 goto top;
3104 3104 }
3105 3105
3106 3106 bzero(&dva[d], sizeof (dva_t));
3107 3107
3108 3108 metaslab_trace_add(zal, rotor, NULL, psize, d, TRACE_ENOSPC);
3109 3109 return (SET_ERROR(ENOSPC));
3110 3110 }
3111 3111
3112 3112 /*
3113 3113 * Free the block represented by DVA in the context of the specified
3114 3114 * transaction group.
3115 3115 */
3116 3116 static void
3117 3117 metaslab_free_dva(spa_t *spa, const dva_t *dva, uint64_t txg, boolean_t now)
3118 3118 {
3119 3119 uint64_t vdev = DVA_GET_VDEV(dva);
3120 3120 uint64_t offset = DVA_GET_OFFSET(dva);
3121 3121 uint64_t size = DVA_GET_ASIZE(dva);
3122 3122 vdev_t *vd;
3123 3123 metaslab_t *msp;
3124 3124
3125 3125 ASSERT(DVA_IS_VALID(dva));
3126 3126
3127 3127 if (txg > spa_freeze_txg(spa))
3128 3128 return;
3129 3129
3130 3130 if ((vd = vdev_lookup_top(spa, vdev)) == NULL ||
3131 3131 (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count) {
3132 3132 cmn_err(CE_WARN, "metaslab_free_dva(): bad DVA %llu:%llu",
3133 3133 (u_longlong_t)vdev, (u_longlong_t)offset);
3134 3134 ASSERT(0);
3135 3135 return;
3136 3136 }
3137 3137
3138 3138 msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
3139 3139
3140 3140 if (DVA_GET_GANG(dva))
3141 3141 size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE);
3142 3142
3143 3143 mutex_enter(&msp->ms_lock);
3144 3144
3145 3145 if (now) {
3146 3146 range_tree_remove(msp->ms_alloctree[txg & TXG_MASK],
3147 3147 offset, size);
3148 3148
3149 3149 VERIFY(!msp->ms_condensing);
3150 3150 VERIFY3U(offset, >=, msp->ms_start);
3151 3151 VERIFY3U(offset + size, <=, msp->ms_start + msp->ms_size);
3152 3152 VERIFY3U(range_tree_space(msp->ms_tree) + size, <=,
3153 3153 msp->ms_size);
3154 3154 VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
3155 3155 VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
3156 3156 range_tree_add(msp->ms_tree, offset, size);
3157 3157 msp->ms_max_size = metaslab_block_maxsize(msp);
3158 3158 } else {
3159 3159 VERIFY3U(txg, ==, spa->spa_syncing_txg);
3160 3160 if (range_tree_space(msp->ms_freeingtree) == 0)
3161 3161 vdev_dirty(vd, VDD_METASLAB, msp, txg);
3162 3162 range_tree_add(msp->ms_freeingtree, offset, size);
3163 3163 }
3164 3164
3165 3165 mutex_exit(&msp->ms_lock);
3166 3166 }
3167 3167
3168 3168 /*
3169 3169 * Intent log support: upon opening the pool after a crash, notify the SPA
3170 3170 * of blocks that the intent log has allocated for immediate write, but
3171 3171 * which are still considered free by the SPA because the last transaction
3172 3172 * group didn't commit yet.
3173 3173 */
3174 3174 static int
3175 3175 metaslab_claim_dva(spa_t *spa, const dva_t *dva, uint64_t txg)
3176 3176 {
3177 3177 uint64_t vdev = DVA_GET_VDEV(dva);
3178 3178 uint64_t offset = DVA_GET_OFFSET(dva);
3179 3179 uint64_t size = DVA_GET_ASIZE(dva);
3180 3180 vdev_t *vd;
3181 3181 metaslab_t *msp;
3182 3182 int error = 0;
3183 3183
3184 3184 ASSERT(DVA_IS_VALID(dva));
3185 3185
3186 3186 if ((vd = vdev_lookup_top(spa, vdev)) == NULL ||
3187 3187 (offset >> vd->vdev_ms_shift) >= vd->vdev_ms_count)
3188 3188 return (SET_ERROR(ENXIO));
3189 3189
3190 3190 msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
3191 3191
3192 3192 if (DVA_GET_GANG(dva))
3193 3193 size = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE);
3194 3194
3195 3195 mutex_enter(&msp->ms_lock);
3196 3196
3197 3197 if ((txg != 0 && spa_writeable(spa)) || !msp->ms_loaded)
3198 3198 error = metaslab_activate(msp, METASLAB_WEIGHT_SECONDARY);
3199 3199
3200 3200 if (error == 0 && !range_tree_contains(msp->ms_tree, offset, size))
3201 3201 error = SET_ERROR(ENOENT);
3202 3202
3203 3203 if (error || txg == 0) { /* txg == 0 indicates dry run */
3204 3204 mutex_exit(&msp->ms_lock);
3205 3205 return (error);
3206 3206 }
3207 3207
3208 3208 VERIFY(!msp->ms_condensing);
3209 3209 VERIFY0(P2PHASE(offset, 1ULL << vd->vdev_ashift));
3210 3210 VERIFY0(P2PHASE(size, 1ULL << vd->vdev_ashift));
3211 3211 VERIFY3U(range_tree_space(msp->ms_tree) - size, <=, msp->ms_size);
3212 3212 range_tree_remove(msp->ms_tree, offset, size);
3213 3213
3214 3214 if (spa_writeable(spa)) { /* don't dirty if we're zdb(1M) */
3215 3215 if (range_tree_space(msp->ms_alloctree[txg & TXG_MASK]) == 0)
3216 3216 vdev_dirty(vd, VDD_METASLAB, msp, txg);
3217 3217 range_tree_add(msp->ms_alloctree[txg & TXG_MASK], offset, size);
3218 3218 }
3219 3219
3220 3220 mutex_exit(&msp->ms_lock);
3221 3221
3222 3222 return (0);
3223 3223 }
3224 3224
3225 3225 /*
3226 3226 * Reserve some allocation slots. The reservation system must be called
3227 3227 * before we call into the allocator. If there aren't any available slots
3228 3228 * then the I/O will be throttled until an I/O completes and its slots are
3229 3229 * freed up. The function returns true if it was successful in placing
3230 3230 * the reservation.
3231 3231 */
3232 3232 boolean_t
3233 3233 metaslab_class_throttle_reserve(metaslab_class_t *mc, int slots, zio_t *zio,
3234 3234 int flags)
3235 3235 {
3236 3236 uint64_t available_slots = 0;
3237 3237 boolean_t slot_reserved = B_FALSE;
3238 3238
3239 3239 ASSERT(mc->mc_alloc_throttle_enabled);
3240 3240 mutex_enter(&mc->mc_lock);
3241 3241
3242 3242 uint64_t reserved_slots = refcount_count(&mc->mc_alloc_slots);
3243 3243 if (reserved_slots < mc->mc_alloc_max_slots)
3244 3244 available_slots = mc->mc_alloc_max_slots - reserved_slots;
3245 3245
3246 3246 if (slots <= available_slots || GANG_ALLOCATION(flags)) {
3247 3247 /*
3248 3248 * We reserve the slots individually so that we can unreserve
3249 3249 * them individually when an I/O completes.
3250 3250 */
3251 3251 for (int d = 0; d < slots; d++) {
3252 3252 reserved_slots = refcount_add(&mc->mc_alloc_slots, zio);
3253 3253 }
3254 3254 zio->io_flags |= ZIO_FLAG_IO_ALLOCATING;
3255 3255 slot_reserved = B_TRUE;
3256 3256 }
3257 3257
3258 3258 mutex_exit(&mc->mc_lock);
3259 3259 return (slot_reserved);
3260 3260 }
3261 3261
3262 3262 void
3263 3263 metaslab_class_throttle_unreserve(metaslab_class_t *mc, int slots, zio_t *zio)
3264 3264 {
3265 3265 ASSERT(mc->mc_alloc_throttle_enabled);
3266 3266 mutex_enter(&mc->mc_lock);
3267 3267 for (int d = 0; d < slots; d++) {
3268 3268 (void) refcount_remove(&mc->mc_alloc_slots, zio);
3269 3269 }
3270 3270 mutex_exit(&mc->mc_lock);
3271 3271 }
3272 3272
3273 3273 int
3274 3274 metaslab_alloc(spa_t *spa, metaslab_class_t *mc, uint64_t psize, blkptr_t *bp,
3275 3275 int ndvas, uint64_t txg, blkptr_t *hintbp, int flags,
3276 3276 zio_alloc_list_t *zal, zio_t *zio)
3277 3277 {
3278 3278 dva_t *dva = bp->blk_dva;
3279 3279 dva_t *hintdva = hintbp->blk_dva;
3280 3280 int error = 0;
3281 3281
3282 3282 ASSERT(bp->blk_birth == 0);
3283 3283 ASSERT(BP_PHYSICAL_BIRTH(bp) == 0);
3284 3284
3285 3285 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
3286 3286
3287 3287 if (mc->mc_rotor == NULL) { /* no vdevs in this class */
3288 3288 spa_config_exit(spa, SCL_ALLOC, FTAG);
3289 3289 return (SET_ERROR(ENOSPC));
3290 3290 }
3291 3291
3292 3292 ASSERT(ndvas > 0 && ndvas <= spa_max_replication(spa));
3293 3293 ASSERT(BP_GET_NDVAS(bp) == 0);
3294 3294 ASSERT(hintbp == NULL || ndvas <= BP_GET_NDVAS(hintbp));
3295 3295 ASSERT3P(zal, !=, NULL);
3296 3296
3297 3297 for (int d = 0; d < ndvas; d++) {
3298 3298 error = metaslab_alloc_dva(spa, mc, psize, dva, d, hintdva,
3299 3299 txg, flags, zal);
3300 3300 if (error != 0) {
3301 3301 for (d--; d >= 0; d--) {
3302 3302 metaslab_free_dva(spa, &dva[d], txg, B_TRUE);
3303 3303 metaslab_group_alloc_decrement(spa,
3304 3304 DVA_GET_VDEV(&dva[d]), zio, flags);
3305 3305 bzero(&dva[d], sizeof (dva_t));
3306 3306 }
3307 3307 spa_config_exit(spa, SCL_ALLOC, FTAG);
3308 3308 return (error);
3309 3309 } else {
3310 3310 /*
3311 3311 * Update the metaslab group's queue depth
3312 3312 * based on the newly allocated dva.
3313 3313 */
3314 3314 metaslab_group_alloc_increment(spa,
3315 3315 DVA_GET_VDEV(&dva[d]), zio, flags);
3316 3316 }
3317 3317
3318 3318 }
3319 3319 ASSERT(error == 0);
3320 3320 ASSERT(BP_GET_NDVAS(bp) == ndvas);
3321 3321
3322 3322 spa_config_exit(spa, SCL_ALLOC, FTAG);
3323 3323
3324 3324 BP_SET_BIRTH(bp, txg, txg);
3325 3325
3326 3326 return (0);
3327 3327 }
3328 3328
3329 3329 void
3330 3330 metaslab_free(spa_t *spa, const blkptr_t *bp, uint64_t txg, boolean_t now)
3331 3331 {
3332 3332 const dva_t *dva = bp->blk_dva;
3333 3333 int ndvas = BP_GET_NDVAS(bp);
3334 3334
3335 3335 ASSERT(!BP_IS_HOLE(bp));
3336 3336 ASSERT(!now || bp->blk_birth >= spa_syncing_txg(spa));
3337 3337
3338 3338 spa_config_enter(spa, SCL_FREE, FTAG, RW_READER);
3339 3339
3340 3340 for (int d = 0; d < ndvas; d++)
3341 3341 metaslab_free_dva(spa, &dva[d], txg, now);
3342 3342
3343 3343 spa_config_exit(spa, SCL_FREE, FTAG);
3344 3344 }
3345 3345
3346 3346 int
3347 3347 metaslab_claim(spa_t *spa, const blkptr_t *bp, uint64_t txg)
3348 3348 {
3349 3349 const dva_t *dva = bp->blk_dva;
3350 3350 int ndvas = BP_GET_NDVAS(bp);
3351 3351 int error = 0;
3352 3352
3353 3353 ASSERT(!BP_IS_HOLE(bp));
3354 3354
3355 3355 if (txg != 0) {
3356 3356 /*
3357 3357 * First do a dry run to make sure all DVAs are claimable,
3358 3358 * so we don't have to unwind from partial failures below.
3359 3359 */
3360 3360 if ((error = metaslab_claim(spa, bp, 0)) != 0)
3361 3361 return (error);
3362 3362 }
3363 3363
3364 3364 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_READER);
3365 3365
3366 3366 for (int d = 0; d < ndvas; d++)
3367 3367 if ((error = metaslab_claim_dva(spa, &dva[d], txg)) != 0)
3368 3368 break;
3369 3369
3370 3370 spa_config_exit(spa, SCL_ALLOC, FTAG);
3371 3371
3372 3372 ASSERT(error == 0 || txg == 0);
3373 3373
3374 3374 return (error);
3375 3375 }
3376 3376
3377 3377 void
3378 3378 metaslab_check_free(spa_t *spa, const blkptr_t *bp)
3379 3379 {
3380 3380 if ((zfs_flags & ZFS_DEBUG_ZIO_FREE) == 0)
3381 3381 return;
3382 3382
3383 3383 spa_config_enter(spa, SCL_VDEV, FTAG, RW_READER);
3384 3384 for (int i = 0; i < BP_GET_NDVAS(bp); i++) {
3385 3385 uint64_t vdev = DVA_GET_VDEV(&bp->blk_dva[i]);
3386 3386 vdev_t *vd = vdev_lookup_top(spa, vdev);
3387 3387 uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
3388 3388 uint64_t size = DVA_GET_ASIZE(&bp->blk_dva[i]);
3389 3389 metaslab_t *msp = vd->vdev_ms[offset >> vd->vdev_ms_shift];
3390 3390
3391 3391 if (msp->ms_loaded)
3392 3392 range_tree_verify(msp->ms_tree, offset, size);
3393 3393
3394 3394 range_tree_verify(msp->ms_freeingtree, offset, size);
3395 3395 range_tree_verify(msp->ms_freedtree, offset, size);
3396 3396 for (int j = 0; j < TXG_DEFER_SIZE; j++)
3397 3397 range_tree_verify(msp->ms_defertree[j], offset, size);
3398 3398 }
3399 3399 spa_config_exit(spa, SCL_VDEV, FTAG);
3400 3400 }
|
↓ open down ↓ |
1738 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX