summaryrefslogtreecommitdiffstats
path: root/fs/bcachefs/movinggc.c
blob: 6209cb51efcb375d1afe520c286c90b548abc445 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
// SPDX-License-Identifier: GPL-2.0
/*
 * Moving/copying garbage collector
 *
 * Copyright 2012 Google, Inc.
 */

#include "bcachefs.h"
#include "alloc_background.h"
#include "alloc_foreground.h"
#include "btree_iter.h"
#include "btree_update.h"
#include "buckets.h"
#include "clock.h"
#include "disk_groups.h"
#include "error.h"
#include "extents.h"
#include "eytzinger.h"
#include "io.h"
#include "keylist.h"
#include "move.h"
#include "movinggc.h"
#include "super-io.h"
#include "trace.h"

#include <linux/freezer.h>
#include <linux/kthread.h>
#include <linux/math64.h>
#include <linux/sched/task.h>
#include <linux/sort.h>
#include <linux/wait.h>

static int bucket_offset_cmp(const void *_l, const void *_r, size_t size)
{
	const struct copygc_heap_entry *l = _l;
	const struct copygc_heap_entry *r = _r;

	return  cmp_int(l->dev,    r->dev) ?:
		cmp_int(l->offset, r->offset);
}

static enum data_cmd copygc_pred(struct bch_fs *c, void *arg,
				 struct bkey_s_c k,
				 struct bch_io_opts *io_opts,
				 struct data_opts *data_opts)
{
	copygc_heap *h = &c->copygc_heap;
	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
	const union bch_extent_entry *entry;
	struct extent_ptr_decoded p = { 0 };

	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
		struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
		struct copygc_heap_entry search = {
			.dev	= p.ptr.dev,
			.offset	= p.ptr.offset,
		};
		ssize_t i;

		if (p.ptr.cached)
			continue;

		i = eytzinger0_find_le(h->data, h->used,
				       sizeof(h->data[0]),
				       bucket_offset_cmp, &search);
#if 0
		/* eytzinger search verify code: */
		ssize_t j = -1, k;

		for (k = 0; k < h->used; k++)
			if (h->data[k].offset <= ptr->offset &&
			    (j < 0 || h->data[k].offset > h->data[j].offset))
				j = k;

		BUG_ON(i != j);
#endif
		if (i >= 0 &&
		    p.ptr.dev == h->data[i].dev &&
		    p.ptr.offset < h->data[i].offset + ca->mi.bucket_size &&
		    p.ptr.gen == h->data[i].gen) {
			/*
			 * We need to use the journal reserve here, because
			 *  - journal reclaim depends on btree key cache
			 *    flushing to make forward progress,
			 *  - which has to make forward progress when the
			 *    journal is pre-reservation full,
			 *  - and depends on allocation - meaning allocator and
			 *    copygc
			 */

			data_opts->target		= io_opts->background_target;
			data_opts->nr_replicas		= 1;
			data_opts->btree_insert_flags	= BTREE_INSERT_USE_RESERVE|
				JOURNAL_WATERMARK_copygc;
			data_opts->rewrite_dev		= p.ptr.dev;

			if (p.has_ec)
				data_opts->nr_replicas += p.ec.redundancy;

			return DATA_REWRITE;
		}
	}

	return DATA_SKIP;
}

static inline int fragmentation_cmp(copygc_heap *heap,
				   struct copygc_heap_entry l,
				   struct copygc_heap_entry r)
{
	return cmp_int(l.fragmentation, r.fragmentation);
}

static int walk_buckets_to_copygc(struct bch_fs *c)
{
	copygc_heap *h = &c->copygc_heap;
	struct btree_trans trans;
	struct btree_iter iter;
	struct bkey_s_c k;
	struct bch_alloc_v4 a;
	int ret;

	bch2_trans_init(&trans, c, 0, 0);

	for_each_btree_key(&trans, iter, BTREE_ID_alloc, POS_MIN,
			   BTREE_ITER_PREFETCH, k, ret) {
		struct bch_dev *ca = bch_dev_bkey_exists(c, iter.pos.inode);
		struct copygc_heap_entry e;

		bch2_alloc_to_v4(k, &a);

		if (a.data_type != BCH_DATA_user ||
		    a.dirty_sectors >= ca->mi.bucket_size ||
		    bch2_bucket_is_open(c, iter.pos.inode, iter.pos.offset))
			continue;

		e = (struct copygc_heap_entry) {
			.dev		= iter.pos.inode,
			.gen		= a.gen,
			.replicas	= 1 + a.stripe_redundancy,
			.fragmentation	= div_u64((u64) a.dirty_sectors * (1ULL << 31),
						  ca->mi.bucket_size),
			.sectors	= a.dirty_sectors,
			.offset		= bucket_to_sector(ca, iter.pos.offset),
		};
		heap_add_or_replace(h, e, -fragmentation_cmp, NULL);

	}
	bch2_trans_iter_exit(&trans, &iter);

	bch2_trans_exit(&trans);
	return ret;
}

static int bucket_inorder_cmp(const void *_l, const void *_r)
{
	const struct copygc_heap_entry *l = _l;
	const struct copygc_heap_entry *r = _r;

	return cmp_int(l->dev, r->dev) ?: cmp_int(l->offset, r->offset);
}

static int check_copygc_was_done(struct bch_fs *c,
				 u64 *sectors_not_moved,
				 u64 *buckets_not_moved)
{
	copygc_heap *h = &c->copygc_heap;
	struct btree_trans trans;
	struct btree_iter iter;
	struct bkey_s_c k;
	struct bch_alloc_v4 a;
	struct copygc_heap_entry *i;
	int ret = 0;

	sort(h->data, h->used, sizeof(h->data[0]), bucket_inorder_cmp, NULL);

	bch2_trans_init(&trans, c, 0, 0);
	bch2_trans_iter_init(&trans, &iter, BTREE_ID_alloc, POS_MIN, 0);

	for (i = h->data; i < h->data + h->used; i++) {
		struct bch_dev *ca = bch_dev_bkey_exists(c, i->dev);

		bch2_btree_iter_set_pos(&iter, POS(i->dev, sector_to_bucket(ca, i->offset)));

		ret = lockrestart_do(&trans,
				bkey_err(k = bch2_btree_iter_peek_slot(&iter)));
		if (ret)
			break;

		bch2_alloc_to_v4(k, &a);

		if (a.gen == i->gen && a.dirty_sectors) {
			*sectors_not_moved += a.dirty_sectors;
			*buckets_not_moved += 1;
		}
	}
	bch2_trans_iter_exit(&trans, &iter);

	bch2_trans_exit(&trans);
	return ret;
}

static int bch2_copygc(struct bch_fs *c)
{
	copygc_heap *h = &c->copygc_heap;
	struct copygc_heap_entry e, *i;
	struct bch_move_stats move_stats;
	u64 sectors_to_move = 0, sectors_to_write = 0, sectors_not_moved = 0;
	u64 sectors_reserved = 0;
	u64 buckets_to_move, buckets_not_moved = 0;
	struct bch_dev *ca;
	unsigned dev_idx;
	size_t heap_size = 0;
	int ret;

	bch_move_stats_init(&move_stats, "copygc");

	/*
	 * Find buckets with lowest sector counts, skipping completely
	 * empty buckets, by building a maxheap sorted by sector count,
	 * and repeatedly replacing the maximum element until all
	 * buckets have been visited.
	 */
	h->used = 0;

	for_each_rw_member(ca, c, dev_idx)
		heap_size += ca->mi.nbuckets >> 7;

	if (h->size < heap_size) {
		free_heap(&c->copygc_heap);
		if (!init_heap(&c->copygc_heap, heap_size, GFP_KERNEL)) {
			bch_err(c, "error allocating copygc heap");
			return 0;
		}
	}

	for_each_rw_member(ca, c, dev_idx) {
		struct bch_dev_usage usage = bch2_dev_usage_read(ca);

		u64 avail = max_t(s64, 0,
				  usage.d[BCH_DATA_free].buckets +
				  usage.d[BCH_DATA_need_discard].buckets -
				  ca->nr_open_buckets -
				  bch2_dev_buckets_reserved(ca, RESERVE_movinggc));

		avail = min(avail, ca->mi.nbuckets >> 6);

		sectors_reserved += avail * ca->mi.bucket_size;
	}

	ret = walk_buckets_to_copygc(c);
	if (ret) {
		bch2_fs_fatal_error(c, "error walking buckets to copygc!");
		return ret;
	}

	if (!h->used) {
		bch_err_ratelimited(c, "copygc requested to run but found no buckets to move!");
		return 0;
	}

	/*
	 * Our btree node allocations also come out of RESERVE_movingc:
	 */
	sectors_reserved = (sectors_reserved * 3) / 4;
	if (!sectors_reserved) {
		bch2_fs_fatal_error(c, "stuck, ran out of copygc reserve!");
		return -1;
	}

	for (i = h->data; i < h->data + h->used; i++) {
		sectors_to_move += i->sectors;
		sectors_to_write += i->sectors * i->replicas;
	}

	while (sectors_to_write > sectors_reserved) {
		BUG_ON(!heap_pop(h, e, -fragmentation_cmp, NULL));
		sectors_to_write -= e.sectors * e.replicas;
	}

	buckets_to_move = h->used;

	if (!buckets_to_move) {
		bch_err_ratelimited(c, "copygc cannot run - sectors_reserved %llu!",
				    sectors_reserved);
		return 0;
	}

	eytzinger0_sort(h->data, h->used,
			sizeof(h->data[0]),
			bucket_offset_cmp, NULL);

	ret = bch2_move_data(c,
			     0,			POS_MIN,
			     BTREE_ID_NR,	POS_MAX,
			     NULL,
			     writepoint_ptr(&c->copygc_write_point),
			     copygc_pred, NULL,
			     &move_stats);
	if (ret < 0)
		bch_err(c, "error %i from bch2_move_data() in copygc", ret);
	if (ret)
		return ret;

	ret = check_copygc_was_done(c, &sectors_not_moved, &buckets_not_moved);
	if (ret) {
		bch_err(c, "error %i from check_copygc_was_done()", ret);
		return ret;
	}

	if (sectors_not_moved)
		bch_warn_ratelimited(c,
			"copygc finished but %llu/%llu sectors, %llu/%llu buckets not moved (move stats: moved %llu sectors, raced %llu keys, %llu sectors)",
			 sectors_not_moved, sectors_to_move,
			 buckets_not_moved, buckets_to_move,
			 atomic64_read(&move_stats.sectors_moved),
			 atomic64_read(&move_stats.keys_raced),
			 atomic64_read(&move_stats.sectors_raced));

	trace_copygc(c,
		     atomic64_read(&move_stats.sectors_moved), sectors_not_moved,
		     buckets_to_move, buckets_not_moved);
	return 0;
}

/*
 * Copygc runs when the amount of fragmented data is above some arbitrary
 * threshold:
 *
 * The threshold at the limit - when the device is full - is the amount of space
 * we reserved in bch2_recalc_capacity; we can't have more than that amount of
 * disk space stranded due to fragmentation and store everything we have
 * promised to store.
 *
 * But we don't want to be running copygc unnecessarily when the device still
 * has plenty of free space - rather, we want copygc to smoothly run every so
 * often and continually reduce the amount of fragmented space as the device
 * fills up. So, we increase the threshold by half the current free space.
 */
unsigned long bch2_copygc_wait_amount(struct bch_fs *c)
{
	struct bch_dev *ca;
	unsigned dev_idx;
	s64 wait = S64_MAX, fragmented_allowed, fragmented;

	for_each_rw_member(ca, c, dev_idx) {
		struct bch_dev_usage usage = bch2_dev_usage_read(ca);

		fragmented_allowed = ((__dev_buckets_available(ca, usage, RESERVE_none) *
				       ca->mi.bucket_size) >> 1);
		fragmented = usage.d[BCH_DATA_user].fragmented;

		wait = min(wait, max(0LL, fragmented_allowed - fragmented));
	}

	return wait;
}

static int bch2_copygc_thread(void *arg)
{
	struct bch_fs *c = arg;
	struct io_clock *clock = &c->io_clock[WRITE];
	u64 last, wait;

	set_freezable();

	while (!kthread_should_stop()) {
		cond_resched();

		if (kthread_wait_freezable(c->copy_gc_enabled))
			break;

		last = atomic64_read(&clock->now);
		wait = bch2_copygc_wait_amount(c);

		if (wait > clock->max_slop) {
			trace_copygc_wait(c, wait, last + wait);
			c->copygc_wait = last + wait;
			bch2_kthread_io_clock_wait(clock, last + wait,
					MAX_SCHEDULE_TIMEOUT);
			continue;
		}

		c->copygc_wait = 0;

		if (bch2_copygc(c))
			break;
	}

	return 0;
}

void bch2_copygc_stop(struct bch_fs *c)
{
	if (c->copygc_thread) {
		kthread_stop(c->copygc_thread);
		put_task_struct(c->copygc_thread);
	}
	c->copygc_thread = NULL;
}

int bch2_copygc_start(struct bch_fs *c)
{
	struct task_struct *t;

	if (c->copygc_thread)
		return 0;

	if (c->opts.nochanges)
		return 0;

	if (bch2_fs_init_fault("copygc_start"))
		return -ENOMEM;

	t = kthread_create(bch2_copygc_thread, c, "bch-copygc/%s", c->name);
	if (IS_ERR(t)) {
		bch_err(c, "error creating copygc thread: %li", PTR_ERR(t));
		return PTR_ERR(t);
	}

	get_task_struct(t);

	c->copygc_thread = t;
	wake_up_process(c->copygc_thread);

	return 0;
}

void bch2_fs_copygc_init(struct bch_fs *c)
{
}