xref: /linux/fs/bcachefs/buckets.c (revision 72bea132f3680ee51e7ed2cee62892b6f5121909)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Code for manipulating bucket marks for garbage collection.
4  *
5  * Copyright 2014 Datera, Inc.
6  */
7 
8 #include "bcachefs.h"
9 #include "alloc_background.h"
10 #include "backpointers.h"
11 #include "bset.h"
12 #include "btree_gc.h"
13 #include "btree_update.h"
14 #include "buckets.h"
15 #include "buckets_waiting_for_journal.h"
16 #include "ec.h"
17 #include "error.h"
18 #include "inode.h"
19 #include "movinggc.h"
20 #include "recovery.h"
21 #include "reflink.h"
22 #include "replicas.h"
23 #include "subvolume.h"
24 #include "trace.h"
25 
26 #include <linux/preempt.h>
27 
28 static inline void fs_usage_data_type_to_base(struct bch_fs_usage_base *fs_usage,
29 					      enum bch_data_type data_type,
30 					      s64 sectors)
31 {
32 	switch (data_type) {
33 	case BCH_DATA_btree:
34 		fs_usage->btree		+= sectors;
35 		break;
36 	case BCH_DATA_user:
37 	case BCH_DATA_parity:
38 		fs_usage->data		+= sectors;
39 		break;
40 	case BCH_DATA_cached:
41 		fs_usage->cached	+= sectors;
42 		break;
43 	default:
44 		break;
45 	}
46 }
47 
48 void bch2_fs_usage_initialize(struct bch_fs *c)
49 {
50 	percpu_down_write(&c->mark_lock);
51 	struct bch_fs_usage *usage = c->usage_base;
52 
53 	for (unsigned i = 0; i < ARRAY_SIZE(c->usage); i++)
54 		bch2_fs_usage_acc_to_base(c, i);
55 
56 	for (unsigned i = 0; i < BCH_REPLICAS_MAX; i++)
57 		usage->b.reserved += usage->persistent_reserved[i];
58 
59 	for (unsigned i = 0; i < c->replicas.nr; i++) {
60 		struct bch_replicas_entry_v1 *e =
61 			cpu_replicas_entry(&c->replicas, i);
62 
63 		fs_usage_data_type_to_base(&usage->b, e->data_type, usage->replicas[i]);
64 	}
65 
66 	for_each_member_device(c, ca) {
67 		struct bch_dev_usage dev = bch2_dev_usage_read(ca);
68 
69 		usage->b.hidden += (dev.d[BCH_DATA_sb].buckets +
70 				    dev.d[BCH_DATA_journal].buckets) *
71 			ca->mi.bucket_size;
72 	}
73 
74 	percpu_up_write(&c->mark_lock);
75 }
76 
77 static inline struct bch_dev_usage *dev_usage_ptr(struct bch_dev *ca,
78 						  unsigned journal_seq,
79 						  bool gc)
80 {
81 	BUG_ON(!gc && !journal_seq);
82 
83 	return this_cpu_ptr(gc
84 			    ? ca->usage_gc
85 			    : ca->usage[journal_seq & JOURNAL_BUF_MASK]);
86 }
87 
88 void bch2_dev_usage_read_fast(struct bch_dev *ca, struct bch_dev_usage *usage)
89 {
90 	struct bch_fs *c = ca->fs;
91 	unsigned seq, i, u64s = dev_usage_u64s();
92 
93 	do {
94 		seq = read_seqcount_begin(&c->usage_lock);
95 		memcpy(usage, ca->usage_base, u64s * sizeof(u64));
96 		for (i = 0; i < ARRAY_SIZE(ca->usage); i++)
97 			acc_u64s_percpu((u64 *) usage, (u64 __percpu *) ca->usage[i], u64s);
98 	} while (read_seqcount_retry(&c->usage_lock, seq));
99 }
100 
101 u64 bch2_fs_usage_read_one(struct bch_fs *c, u64 *v)
102 {
103 	ssize_t offset = v - (u64 *) c->usage_base;
104 	unsigned i, seq;
105 	u64 ret;
106 
107 	BUG_ON(offset < 0 || offset >= fs_usage_u64s(c));
108 	percpu_rwsem_assert_held(&c->mark_lock);
109 
110 	do {
111 		seq = read_seqcount_begin(&c->usage_lock);
112 		ret = *v;
113 
114 		for (i = 0; i < ARRAY_SIZE(c->usage); i++)
115 			ret += percpu_u64_get((u64 __percpu *) c->usage[i] + offset);
116 	} while (read_seqcount_retry(&c->usage_lock, seq));
117 
118 	return ret;
119 }
120 
121 struct bch_fs_usage_online *bch2_fs_usage_read(struct bch_fs *c)
122 {
123 	struct bch_fs_usage_online *ret;
124 	unsigned nr_replicas = READ_ONCE(c->replicas.nr);
125 	unsigned seq, i;
126 retry:
127 	ret = kmalloc(__fs_usage_online_u64s(nr_replicas) * sizeof(u64), GFP_KERNEL);
128 	if (unlikely(!ret))
129 		return NULL;
130 
131 	percpu_down_read(&c->mark_lock);
132 
133 	if (nr_replicas != c->replicas.nr) {
134 		nr_replicas = c->replicas.nr;
135 		percpu_up_read(&c->mark_lock);
136 		kfree(ret);
137 		goto retry;
138 	}
139 
140 	ret->online_reserved = percpu_u64_get(c->online_reserved);
141 
142 	do {
143 		seq = read_seqcount_begin(&c->usage_lock);
144 		unsafe_memcpy(&ret->u, c->usage_base,
145 			      __fs_usage_u64s(nr_replicas) * sizeof(u64),
146 			      "embedded variable length struct");
147 		for (i = 0; i < ARRAY_SIZE(c->usage); i++)
148 			acc_u64s_percpu((u64 *) &ret->u, (u64 __percpu *) c->usage[i],
149 					__fs_usage_u64s(nr_replicas));
150 	} while (read_seqcount_retry(&c->usage_lock, seq));
151 
152 	return ret;
153 }
154 
155 void bch2_fs_usage_acc_to_base(struct bch_fs *c, unsigned idx)
156 {
157 	unsigned u64s = fs_usage_u64s(c);
158 
159 	BUG_ON(idx >= ARRAY_SIZE(c->usage));
160 
161 	preempt_disable();
162 	write_seqcount_begin(&c->usage_lock);
163 
164 	acc_u64s_percpu((u64 *) c->usage_base,
165 			(u64 __percpu *) c->usage[idx], u64s);
166 	percpu_memset(c->usage[idx], 0, u64s * sizeof(u64));
167 
168 	rcu_read_lock();
169 	for_each_member_device_rcu(c, ca, NULL) {
170 		u64s = dev_usage_u64s();
171 
172 		acc_u64s_percpu((u64 *) ca->usage_base,
173 				(u64 __percpu *) ca->usage[idx], u64s);
174 		percpu_memset(ca->usage[idx], 0, u64s * sizeof(u64));
175 	}
176 	rcu_read_unlock();
177 
178 	write_seqcount_end(&c->usage_lock);
179 	preempt_enable();
180 }
181 
182 void bch2_fs_usage_to_text(struct printbuf *out,
183 			   struct bch_fs *c,
184 			   struct bch_fs_usage_online *fs_usage)
185 {
186 	unsigned i;
187 
188 	prt_printf(out, "capacity:\t\t\t%llu\n", c->capacity);
189 
190 	prt_printf(out, "hidden:\t\t\t\t%llu\n",
191 	       fs_usage->u.b.hidden);
192 	prt_printf(out, "data:\t\t\t\t%llu\n",
193 	       fs_usage->u.b.data);
194 	prt_printf(out, "cached:\t\t\t\t%llu\n",
195 	       fs_usage->u.b.cached);
196 	prt_printf(out, "reserved:\t\t\t%llu\n",
197 	       fs_usage->u.b.reserved);
198 	prt_printf(out, "nr_inodes:\t\t\t%llu\n",
199 	       fs_usage->u.b.nr_inodes);
200 	prt_printf(out, "online reserved:\t\t%llu\n",
201 	       fs_usage->online_reserved);
202 
203 	for (i = 0;
204 	     i < ARRAY_SIZE(fs_usage->u.persistent_reserved);
205 	     i++) {
206 		prt_printf(out, "%u replicas:\n", i + 1);
207 		prt_printf(out, "\treserved:\t\t%llu\n",
208 		       fs_usage->u.persistent_reserved[i]);
209 	}
210 
211 	for (i = 0; i < c->replicas.nr; i++) {
212 		struct bch_replicas_entry_v1 *e =
213 			cpu_replicas_entry(&c->replicas, i);
214 
215 		prt_printf(out, "\t");
216 		bch2_replicas_entry_to_text(out, e);
217 		prt_printf(out, ":\t%llu\n", fs_usage->u.replicas[i]);
218 	}
219 }
220 
221 static u64 reserve_factor(u64 r)
222 {
223 	return r + (round_up(r, (1 << RESERVE_FACTOR)) >> RESERVE_FACTOR);
224 }
225 
226 u64 bch2_fs_sectors_used(struct bch_fs *c, struct bch_fs_usage_online *fs_usage)
227 {
228 	return min(fs_usage->u.b.hidden +
229 		   fs_usage->u.b.btree +
230 		   fs_usage->u.b.data +
231 		   reserve_factor(fs_usage->u.b.reserved +
232 				  fs_usage->online_reserved),
233 		   c->capacity);
234 }
235 
236 static struct bch_fs_usage_short
237 __bch2_fs_usage_read_short(struct bch_fs *c)
238 {
239 	struct bch_fs_usage_short ret;
240 	u64 data, reserved;
241 
242 	ret.capacity = c->capacity -
243 		bch2_fs_usage_read_one(c, &c->usage_base->b.hidden);
244 
245 	data		= bch2_fs_usage_read_one(c, &c->usage_base->b.data) +
246 		bch2_fs_usage_read_one(c, &c->usage_base->b.btree);
247 	reserved	= bch2_fs_usage_read_one(c, &c->usage_base->b.reserved) +
248 		percpu_u64_get(c->online_reserved);
249 
250 	ret.used	= min(ret.capacity, data + reserve_factor(reserved));
251 	ret.free	= ret.capacity - ret.used;
252 
253 	ret.nr_inodes	= bch2_fs_usage_read_one(c, &c->usage_base->b.nr_inodes);
254 
255 	return ret;
256 }
257 
258 struct bch_fs_usage_short
259 bch2_fs_usage_read_short(struct bch_fs *c)
260 {
261 	struct bch_fs_usage_short ret;
262 
263 	percpu_down_read(&c->mark_lock);
264 	ret = __bch2_fs_usage_read_short(c);
265 	percpu_up_read(&c->mark_lock);
266 
267 	return ret;
268 }
269 
270 void bch2_dev_usage_init(struct bch_dev *ca)
271 {
272 	ca->usage_base->d[BCH_DATA_free].buckets = ca->mi.nbuckets - ca->mi.first_bucket;
273 }
274 
275 void bch2_dev_usage_to_text(struct printbuf *out, struct bch_dev_usage *usage)
276 {
277 	prt_tab(out);
278 	prt_str(out, "buckets");
279 	prt_tab_rjust(out);
280 	prt_str(out, "sectors");
281 	prt_tab_rjust(out);
282 	prt_str(out, "fragmented");
283 	prt_tab_rjust(out);
284 	prt_newline(out);
285 
286 	for (unsigned i = 0; i < BCH_DATA_NR; i++) {
287 		bch2_prt_data_type(out, i);
288 		prt_tab(out);
289 		prt_u64(out, usage->d[i].buckets);
290 		prt_tab_rjust(out);
291 		prt_u64(out, usage->d[i].sectors);
292 		prt_tab_rjust(out);
293 		prt_u64(out, usage->d[i].fragmented);
294 		prt_tab_rjust(out);
295 		prt_newline(out);
296 	}
297 }
298 
299 void bch2_dev_usage_update(struct bch_fs *c, struct bch_dev *ca,
300 			   const struct bch_alloc_v4 *old,
301 			   const struct bch_alloc_v4 *new,
302 			   u64 journal_seq, bool gc)
303 {
304 	struct bch_fs_usage *fs_usage;
305 	struct bch_dev_usage *u;
306 
307 	preempt_disable();
308 	fs_usage = fs_usage_ptr(c, journal_seq, gc);
309 
310 	if (data_type_is_hidden(old->data_type))
311 		fs_usage->b.hidden -= ca->mi.bucket_size;
312 	if (data_type_is_hidden(new->data_type))
313 		fs_usage->b.hidden += ca->mi.bucket_size;
314 
315 	u = dev_usage_ptr(ca, journal_seq, gc);
316 
317 	u->d[old->data_type].buckets--;
318 	u->d[new->data_type].buckets++;
319 
320 	u->d[old->data_type].sectors -= bch2_bucket_sectors_dirty(*old);
321 	u->d[new->data_type].sectors += bch2_bucket_sectors_dirty(*new);
322 
323 	u->d[BCH_DATA_cached].sectors += new->cached_sectors;
324 	u->d[BCH_DATA_cached].sectors -= old->cached_sectors;
325 
326 	u->d[old->data_type].fragmented -= bch2_bucket_sectors_fragmented(ca, *old);
327 	u->d[new->data_type].fragmented += bch2_bucket_sectors_fragmented(ca, *new);
328 
329 	preempt_enable();
330 }
331 
332 static inline struct bch_alloc_v4 bucket_m_to_alloc(struct bucket b)
333 {
334 	return (struct bch_alloc_v4) {
335 		.gen		= b.gen,
336 		.data_type	= b.data_type,
337 		.dirty_sectors	= b.dirty_sectors,
338 		.cached_sectors	= b.cached_sectors,
339 		.stripe		= b.stripe,
340 	};
341 }
342 
343 void bch2_dev_usage_update_m(struct bch_fs *c, struct bch_dev *ca,
344 			     struct bucket *old, struct bucket *new)
345 {
346 	struct bch_alloc_v4 old_a = bucket_m_to_alloc(*old);
347 	struct bch_alloc_v4 new_a = bucket_m_to_alloc(*new);
348 
349 	bch2_dev_usage_update(c, ca, &old_a, &new_a, 0, true);
350 }
351 
352 static inline int __update_replicas(struct bch_fs *c,
353 				    struct bch_fs_usage *fs_usage,
354 				    struct bch_replicas_entry_v1 *r,
355 				    s64 sectors)
356 {
357 	int idx = bch2_replicas_entry_idx(c, r);
358 
359 	if (idx < 0)
360 		return -1;
361 
362 	fs_usage_data_type_to_base(&fs_usage->b, r->data_type, sectors);
363 	fs_usage->replicas[idx]		+= sectors;
364 	return 0;
365 }
366 
367 int bch2_update_replicas(struct bch_fs *c, struct bkey_s_c k,
368 			 struct bch_replicas_entry_v1 *r, s64 sectors,
369 			 unsigned journal_seq, bool gc)
370 {
371 	struct bch_fs_usage *fs_usage;
372 	int idx, ret = 0;
373 	struct printbuf buf = PRINTBUF;
374 
375 	percpu_down_read(&c->mark_lock);
376 
377 	idx = bch2_replicas_entry_idx(c, r);
378 	if (idx < 0 &&
379 	    fsck_err(c, ptr_to_missing_replicas_entry,
380 		     "no replicas entry\n  while marking %s",
381 		     (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
382 		percpu_up_read(&c->mark_lock);
383 		ret = bch2_mark_replicas(c, r);
384 		percpu_down_read(&c->mark_lock);
385 
386 		if (ret)
387 			goto err;
388 		idx = bch2_replicas_entry_idx(c, r);
389 	}
390 	if (idx < 0) {
391 		ret = -1;
392 		goto err;
393 	}
394 
395 	preempt_disable();
396 	fs_usage = fs_usage_ptr(c, journal_seq, gc);
397 	fs_usage_data_type_to_base(&fs_usage->b, r->data_type, sectors);
398 	fs_usage->replicas[idx]		+= sectors;
399 	preempt_enable();
400 err:
401 fsck_err:
402 	percpu_up_read(&c->mark_lock);
403 	printbuf_exit(&buf);
404 	return ret;
405 }
406 
407 static inline int update_cached_sectors(struct bch_fs *c,
408 			struct bkey_s_c k,
409 			unsigned dev, s64 sectors,
410 			unsigned journal_seq, bool gc)
411 {
412 	struct bch_replicas_padded r;
413 
414 	bch2_replicas_entry_cached(&r.e, dev);
415 
416 	return bch2_update_replicas(c, k, &r.e, sectors, journal_seq, gc);
417 }
418 
419 static int __replicas_deltas_realloc(struct btree_trans *trans, unsigned more,
420 				     gfp_t gfp)
421 {
422 	struct replicas_delta_list *d = trans->fs_usage_deltas;
423 	unsigned new_size = d ? (d->size + more) * 2 : 128;
424 	unsigned alloc_size = sizeof(*d) + new_size;
425 
426 	WARN_ON_ONCE(alloc_size > REPLICAS_DELTA_LIST_MAX);
427 
428 	if (!d || d->used + more > d->size) {
429 		d = krealloc(d, alloc_size, gfp|__GFP_ZERO);
430 
431 		if (unlikely(!d)) {
432 			if (alloc_size > REPLICAS_DELTA_LIST_MAX)
433 				return -ENOMEM;
434 
435 			d = mempool_alloc(&trans->c->replicas_delta_pool, gfp);
436 			if (!d)
437 				return -ENOMEM;
438 
439 			memset(d, 0, REPLICAS_DELTA_LIST_MAX);
440 
441 			if (trans->fs_usage_deltas)
442 				memcpy(d, trans->fs_usage_deltas,
443 				       trans->fs_usage_deltas->size + sizeof(*d));
444 
445 			new_size = REPLICAS_DELTA_LIST_MAX - sizeof(*d);
446 			kfree(trans->fs_usage_deltas);
447 		}
448 
449 		d->size = new_size;
450 		trans->fs_usage_deltas = d;
451 	}
452 
453 	return 0;
454 }
455 
456 int bch2_replicas_deltas_realloc(struct btree_trans *trans, unsigned more)
457 {
458 	return allocate_dropping_locks_errcode(trans,
459 				__replicas_deltas_realloc(trans, more, _gfp));
460 }
461 
462 int bch2_update_replicas_list(struct btree_trans *trans,
463 			 struct bch_replicas_entry_v1 *r,
464 			 s64 sectors)
465 {
466 	struct replicas_delta_list *d;
467 	struct replicas_delta *n;
468 	unsigned b;
469 	int ret;
470 
471 	if (!sectors)
472 		return 0;
473 
474 	b = replicas_entry_bytes(r) + 8;
475 	ret = bch2_replicas_deltas_realloc(trans, b);
476 	if (ret)
477 		return ret;
478 
479 	d = trans->fs_usage_deltas;
480 	n = (void *) d->d + d->used;
481 	n->delta = sectors;
482 	unsafe_memcpy((void *) n + offsetof(struct replicas_delta, r),
483 		      r, replicas_entry_bytes(r),
484 		      "flexible array member embedded in strcuct with padding");
485 	bch2_replicas_entry_sort(&n->r);
486 	d->used += b;
487 	return 0;
488 }
489 
490 int bch2_update_cached_sectors_list(struct btree_trans *trans, unsigned dev, s64 sectors)
491 {
492 	struct bch_replicas_padded r;
493 
494 	bch2_replicas_entry_cached(&r.e, dev);
495 
496 	return bch2_update_replicas_list(trans, &r.e, sectors);
497 }
498 
499 int bch2_mark_metadata_bucket(struct bch_fs *c, struct bch_dev *ca,
500 			      size_t b, enum bch_data_type data_type,
501 			      unsigned sectors, struct gc_pos pos,
502 			      unsigned flags)
503 {
504 	struct bucket old, new, *g;
505 	int ret = 0;
506 
507 	BUG_ON(!(flags & BTREE_TRIGGER_GC));
508 	BUG_ON(data_type != BCH_DATA_sb &&
509 	       data_type != BCH_DATA_journal);
510 
511 	/*
512 	 * Backup superblock might be past the end of our normal usable space:
513 	 */
514 	if (b >= ca->mi.nbuckets)
515 		return 0;
516 
517 	percpu_down_read(&c->mark_lock);
518 	g = gc_bucket(ca, b);
519 
520 	bucket_lock(g);
521 	old = *g;
522 
523 	if (bch2_fs_inconsistent_on(g->data_type &&
524 			g->data_type != data_type, c,
525 			"different types of data in same bucket: %s, %s",
526 			bch2_data_type_str(g->data_type),
527 			bch2_data_type_str(data_type))) {
528 		ret = -EIO;
529 		goto err;
530 	}
531 
532 	if (bch2_fs_inconsistent_on((u64) g->dirty_sectors + sectors > ca->mi.bucket_size, c,
533 			"bucket %u:%zu gen %u data type %s sector count overflow: %u + %u > bucket size",
534 			ca->dev_idx, b, g->gen,
535 			bch2_data_type_str(g->data_type ?: data_type),
536 			g->dirty_sectors, sectors)) {
537 		ret = -EIO;
538 		goto err;
539 	}
540 
541 	g->data_type = data_type;
542 	g->dirty_sectors += sectors;
543 	new = *g;
544 err:
545 	bucket_unlock(g);
546 	if (!ret)
547 		bch2_dev_usage_update_m(c, ca, &old, &new);
548 	percpu_up_read(&c->mark_lock);
549 	return ret;
550 }
551 
552 int bch2_check_bucket_ref(struct btree_trans *trans,
553 			  struct bkey_s_c k,
554 			  const struct bch_extent_ptr *ptr,
555 			  s64 sectors, enum bch_data_type ptr_data_type,
556 			  u8 b_gen, u8 bucket_data_type,
557 			  u32 bucket_sectors)
558 {
559 	struct bch_fs *c = trans->c;
560 	struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
561 	size_t bucket_nr = PTR_BUCKET_NR(ca, ptr);
562 	struct printbuf buf = PRINTBUF;
563 	int ret = 0;
564 
565 	if (bucket_data_type == BCH_DATA_cached)
566 		bucket_data_type = BCH_DATA_user;
567 
568 	if ((bucket_data_type == BCH_DATA_stripe && ptr_data_type == BCH_DATA_user) ||
569 	    (bucket_data_type == BCH_DATA_user   && ptr_data_type == BCH_DATA_stripe))
570 		bucket_data_type = ptr_data_type = BCH_DATA_stripe;
571 
572 	if (gen_after(ptr->gen, b_gen)) {
573 		bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
574 			      BCH_FSCK_ERR_ptr_gen_newer_than_bucket_gen,
575 			"bucket %u:%zu gen %u data type %s: ptr gen %u newer than bucket gen\n"
576 			"while marking %s",
577 			ptr->dev, bucket_nr, b_gen,
578 			bch2_data_type_str(bucket_data_type ?: ptr_data_type),
579 			ptr->gen,
580 			(bch2_bkey_val_to_text(&buf, c, k), buf.buf));
581 		ret = -EIO;
582 		goto err;
583 	}
584 
585 	if (gen_cmp(b_gen, ptr->gen) > BUCKET_GC_GEN_MAX) {
586 		bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
587 			      BCH_FSCK_ERR_ptr_too_stale,
588 			"bucket %u:%zu gen %u data type %s: ptr gen %u too stale\n"
589 			"while marking %s",
590 			ptr->dev, bucket_nr, b_gen,
591 			bch2_data_type_str(bucket_data_type ?: ptr_data_type),
592 			ptr->gen,
593 			(printbuf_reset(&buf),
594 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
595 		ret = -EIO;
596 		goto err;
597 	}
598 
599 	if (b_gen != ptr->gen && !ptr->cached) {
600 		bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
601 			      BCH_FSCK_ERR_stale_dirty_ptr,
602 			"bucket %u:%zu gen %u (mem gen %u) data type %s: stale dirty ptr (gen %u)\n"
603 			"while marking %s",
604 			ptr->dev, bucket_nr, b_gen,
605 			*bucket_gen(ca, bucket_nr),
606 			bch2_data_type_str(bucket_data_type ?: ptr_data_type),
607 			ptr->gen,
608 			(printbuf_reset(&buf),
609 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
610 		ret = -EIO;
611 		goto err;
612 	}
613 
614 	if (b_gen != ptr->gen) {
615 		ret = 1;
616 		goto out;
617 	}
618 
619 	if (!data_type_is_empty(bucket_data_type) &&
620 	    ptr_data_type &&
621 	    bucket_data_type != ptr_data_type) {
622 		bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
623 			      BCH_FSCK_ERR_ptr_bucket_data_type_mismatch,
624 			"bucket %u:%zu gen %u different types of data in same bucket: %s, %s\n"
625 			"while marking %s",
626 			ptr->dev, bucket_nr, b_gen,
627 			bch2_data_type_str(bucket_data_type),
628 			bch2_data_type_str(ptr_data_type),
629 			(printbuf_reset(&buf),
630 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
631 		ret = -EIO;
632 		goto err;
633 	}
634 
635 	if ((u64) bucket_sectors + sectors > U32_MAX) {
636 		bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
637 			      BCH_FSCK_ERR_bucket_sector_count_overflow,
638 			"bucket %u:%zu gen %u data type %s sector count overflow: %u + %lli > U32_MAX\n"
639 			"while marking %s",
640 			ptr->dev, bucket_nr, b_gen,
641 			bch2_data_type_str(bucket_data_type ?: ptr_data_type),
642 			bucket_sectors, sectors,
643 			(printbuf_reset(&buf),
644 			 bch2_bkey_val_to_text(&buf, c, k), buf.buf));
645 		ret = -EIO;
646 		goto err;
647 	}
648 out:
649 	printbuf_exit(&buf);
650 	return ret;
651 err:
652 	bch2_dump_trans_updates(trans);
653 	goto out;
654 }
655 
656 void bch2_trans_fs_usage_revert(struct btree_trans *trans,
657 				struct replicas_delta_list *deltas)
658 {
659 	struct bch_fs *c = trans->c;
660 	struct bch_fs_usage *dst;
661 	struct replicas_delta *d, *top = (void *) deltas->d + deltas->used;
662 	s64 added = 0;
663 	unsigned i;
664 
665 	percpu_down_read(&c->mark_lock);
666 	preempt_disable();
667 	dst = fs_usage_ptr(c, trans->journal_res.seq, false);
668 
669 	/* revert changes: */
670 	for (d = deltas->d; d != top; d = replicas_delta_next(d)) {
671 		switch (d->r.data_type) {
672 		case BCH_DATA_btree:
673 		case BCH_DATA_user:
674 		case BCH_DATA_parity:
675 			added += d->delta;
676 		}
677 		BUG_ON(__update_replicas(c, dst, &d->r, -d->delta));
678 	}
679 
680 	dst->b.nr_inodes -= deltas->nr_inodes;
681 
682 	for (i = 0; i < BCH_REPLICAS_MAX; i++) {
683 		added				-= deltas->persistent_reserved[i];
684 		dst->b.reserved			-= deltas->persistent_reserved[i];
685 		dst->persistent_reserved[i]	-= deltas->persistent_reserved[i];
686 	}
687 
688 	if (added > 0) {
689 		trans->disk_res->sectors += added;
690 		this_cpu_add(*c->online_reserved, added);
691 	}
692 
693 	preempt_enable();
694 	percpu_up_read(&c->mark_lock);
695 }
696 
697 void bch2_trans_account_disk_usage_change(struct btree_trans *trans)
698 {
699 	struct bch_fs *c = trans->c;
700 	u64 disk_res_sectors = trans->disk_res ? trans->disk_res->sectors : 0;
701 	static int warned_disk_usage = 0;
702 	bool warn = false;
703 
704 	percpu_down_read(&c->mark_lock);
705 	preempt_disable();
706 	struct bch_fs_usage_base *dst = &fs_usage_ptr(c, trans->journal_res.seq, false)->b;
707 	struct bch_fs_usage_base *src = &trans->fs_usage_delta;
708 
709 	s64 added = src->btree + src->data + src->reserved;
710 
711 	/*
712 	 * Not allowed to reduce sectors_available except by getting a
713 	 * reservation:
714 	 */
715 	s64 should_not_have_added = added - (s64) disk_res_sectors;
716 	if (unlikely(should_not_have_added > 0)) {
717 		u64 old, new, v = atomic64_read(&c->sectors_available);
718 
719 		do {
720 			old = v;
721 			new = max_t(s64, 0, old - should_not_have_added);
722 		} while ((v = atomic64_cmpxchg(&c->sectors_available,
723 					       old, new)) != old);
724 
725 		added -= should_not_have_added;
726 		warn = true;
727 	}
728 
729 	if (added > 0) {
730 		trans->disk_res->sectors -= added;
731 		this_cpu_sub(*c->online_reserved, added);
732 	}
733 
734 	dst->hidden	+= src->hidden;
735 	dst->btree	+= src->btree;
736 	dst->data	+= src->data;
737 	dst->cached	+= src->cached;
738 	dst->reserved	+= src->reserved;
739 	dst->nr_inodes	+= src->nr_inodes;
740 
741 	preempt_enable();
742 	percpu_up_read(&c->mark_lock);
743 
744 	if (unlikely(warn) && !xchg(&warned_disk_usage, 1))
745 		bch2_trans_inconsistent(trans,
746 					"disk usage increased %lli more than %llu sectors reserved)",
747 					should_not_have_added, disk_res_sectors);
748 }
749 
750 int bch2_trans_fs_usage_apply(struct btree_trans *trans,
751 			      struct replicas_delta_list *deltas)
752 {
753 	struct bch_fs *c = trans->c;
754 	struct replicas_delta *d, *d2;
755 	struct replicas_delta *top = (void *) deltas->d + deltas->used;
756 	struct bch_fs_usage *dst;
757 	unsigned i;
758 
759 	percpu_down_read(&c->mark_lock);
760 	preempt_disable();
761 	dst = fs_usage_ptr(c, trans->journal_res.seq, false);
762 
763 	for (d = deltas->d; d != top; d = replicas_delta_next(d))
764 		if (__update_replicas(c, dst, &d->r, d->delta))
765 			goto need_mark;
766 
767 	dst->b.nr_inodes += deltas->nr_inodes;
768 
769 	for (i = 0; i < BCH_REPLICAS_MAX; i++) {
770 		dst->b.reserved			+= deltas->persistent_reserved[i];
771 		dst->persistent_reserved[i]	+= deltas->persistent_reserved[i];
772 	}
773 
774 	preempt_enable();
775 	percpu_up_read(&c->mark_lock);
776 	return 0;
777 need_mark:
778 	/* revert changes: */
779 	for (d2 = deltas->d; d2 != d; d2 = replicas_delta_next(d2))
780 		BUG_ON(__update_replicas(c, dst, &d2->r, -d2->delta));
781 
782 	preempt_enable();
783 	percpu_up_read(&c->mark_lock);
784 	return -1;
785 }
786 
787 /* KEY_TYPE_extent: */
788 
789 static int __mark_pointer(struct btree_trans *trans,
790 			  struct bkey_s_c k,
791 			  const struct bch_extent_ptr *ptr,
792 			  s64 sectors, enum bch_data_type ptr_data_type,
793 			  u8 bucket_gen, u8 *bucket_data_type,
794 			  u32 *dirty_sectors, u32 *cached_sectors)
795 {
796 	u32 *dst_sectors = !ptr->cached
797 		? dirty_sectors
798 		: cached_sectors;
799 	int ret = bch2_check_bucket_ref(trans, k, ptr, sectors, ptr_data_type,
800 				   bucket_gen, *bucket_data_type, *dst_sectors);
801 
802 	if (ret)
803 		return ret;
804 
805 	*dst_sectors += sectors;
806 
807 	if (!*dirty_sectors && !*cached_sectors)
808 		*bucket_data_type = 0;
809 	else if (*bucket_data_type != BCH_DATA_stripe)
810 		*bucket_data_type = ptr_data_type;
811 
812 	return 0;
813 }
814 
815 static int bch2_trigger_pointer(struct btree_trans *trans,
816 			enum btree_id btree_id, unsigned level,
817 			struct bkey_s_c k, struct extent_ptr_decoded p,
818 			s64 *sectors,
819 			unsigned flags)
820 {
821 	bool insert = !(flags & BTREE_TRIGGER_OVERWRITE);
822 	struct bpos bucket;
823 	struct bch_backpointer bp;
824 
825 	bch2_extent_ptr_to_bp(trans->c, btree_id, level, k, p, &bucket, &bp);
826 	*sectors = insert ? bp.bucket_len : -((s64) bp.bucket_len);
827 
828 	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
829 		struct btree_iter iter;
830 		struct bkey_i_alloc_v4 *a = bch2_trans_start_alloc_update(trans, &iter, bucket);
831 		int ret = PTR_ERR_OR_ZERO(a);
832 		if (ret)
833 			return ret;
834 
835 		ret = __mark_pointer(trans, k, &p.ptr, *sectors, bp.data_type,
836 				     a->v.gen, &a->v.data_type,
837 				     &a->v.dirty_sectors, &a->v.cached_sectors) ?:
838 			bch2_trans_update(trans, &iter, &a->k_i, 0);
839 		bch2_trans_iter_exit(trans, &iter);
840 
841 		if (ret)
842 			return ret;
843 
844 		if (!p.ptr.cached) {
845 			ret = bch2_bucket_backpointer_mod(trans, bucket, bp, k, insert);
846 			if (ret)
847 				return ret;
848 		}
849 	}
850 
851 	if (flags & BTREE_TRIGGER_GC) {
852 		struct bch_fs *c = trans->c;
853 		struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
854 		enum bch_data_type data_type = bkey_ptr_data_type(btree_id, level, k, p);
855 
856 		percpu_down_read(&c->mark_lock);
857 		struct bucket *g = PTR_GC_BUCKET(ca, &p.ptr);
858 		bucket_lock(g);
859 		struct bucket old = *g;
860 
861 		u8 bucket_data_type = g->data_type;
862 		int ret = __mark_pointer(trans, k, &p.ptr, *sectors,
863 				     data_type, g->gen,
864 				     &bucket_data_type,
865 				     &g->dirty_sectors,
866 				     &g->cached_sectors);
867 		if (ret) {
868 			bucket_unlock(g);
869 			percpu_up_read(&c->mark_lock);
870 			return ret;
871 		}
872 
873 		g->data_type = bucket_data_type;
874 		struct bucket new = *g;
875 		bucket_unlock(g);
876 		bch2_dev_usage_update_m(c, ca, &old, &new);
877 		percpu_up_read(&c->mark_lock);
878 	}
879 
880 	return 0;
881 }
882 
883 static int bch2_trigger_stripe_ptr(struct btree_trans *trans,
884 				struct bkey_s_c k,
885 				struct extent_ptr_decoded p,
886 				enum bch_data_type data_type,
887 				s64 sectors, unsigned flags)
888 {
889 	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
890 		struct btree_iter iter;
891 		struct bkey_i_stripe *s = bch2_bkey_get_mut_typed(trans, &iter,
892 				BTREE_ID_stripes, POS(0, p.ec.idx),
893 				BTREE_ITER_WITH_UPDATES, stripe);
894 		int ret = PTR_ERR_OR_ZERO(s);
895 		if (unlikely(ret)) {
896 			bch2_trans_inconsistent_on(bch2_err_matches(ret, ENOENT), trans,
897 				"pointer to nonexistent stripe %llu",
898 				(u64) p.ec.idx);
899 			goto err;
900 		}
901 
902 		if (!bch2_ptr_matches_stripe(&s->v, p)) {
903 			bch2_trans_inconsistent(trans,
904 				"stripe pointer doesn't match stripe %llu",
905 				(u64) p.ec.idx);
906 			ret = -EIO;
907 			goto err;
908 		}
909 
910 		stripe_blockcount_set(&s->v, p.ec.block,
911 			stripe_blockcount_get(&s->v, p.ec.block) +
912 			sectors);
913 
914 		struct bch_replicas_padded r;
915 		bch2_bkey_to_replicas(&r.e, bkey_i_to_s_c(&s->k_i));
916 		r.e.data_type = data_type;
917 		ret = bch2_update_replicas_list(trans, &r.e, sectors);
918 err:
919 		bch2_trans_iter_exit(trans, &iter);
920 		return ret;
921 	}
922 
923 	if (flags & BTREE_TRIGGER_GC) {
924 		struct bch_fs *c = trans->c;
925 
926 		BUG_ON(!(flags & BTREE_TRIGGER_GC));
927 
928 		struct gc_stripe *m = genradix_ptr_alloc(&c->gc_stripes, p.ec.idx, GFP_KERNEL);
929 		if (!m) {
930 			bch_err(c, "error allocating memory for gc_stripes, idx %llu",
931 				(u64) p.ec.idx);
932 			return -BCH_ERR_ENOMEM_mark_stripe_ptr;
933 		}
934 
935 		mutex_lock(&c->ec_stripes_heap_lock);
936 
937 		if (!m || !m->alive) {
938 			mutex_unlock(&c->ec_stripes_heap_lock);
939 			struct printbuf buf = PRINTBUF;
940 			bch2_bkey_val_to_text(&buf, c, k);
941 			bch_err_ratelimited(c, "pointer to nonexistent stripe %llu\n  while marking %s",
942 					    (u64) p.ec.idx, buf.buf);
943 			printbuf_exit(&buf);
944 			bch2_inconsistent_error(c);
945 			return -EIO;
946 		}
947 
948 		m->block_sectors[p.ec.block] += sectors;
949 
950 		struct bch_replicas_padded r = m->r;
951 		mutex_unlock(&c->ec_stripes_heap_lock);
952 
953 		r.e.data_type = data_type;
954 		bch2_update_replicas(c, k, &r.e, sectors, trans->journal_res.seq, true);
955 	}
956 
957 	return 0;
958 }
959 
960 static int __trigger_extent(struct btree_trans *trans,
961 			    enum btree_id btree_id, unsigned level,
962 			    struct bkey_s_c k, unsigned flags)
963 {
964 	bool gc = flags & BTREE_TRIGGER_GC;
965 	struct bch_fs *c = trans->c;
966 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
967 	const union bch_extent_entry *entry;
968 	struct extent_ptr_decoded p;
969 	struct bch_replicas_padded r;
970 	enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
971 		? BCH_DATA_btree
972 		: BCH_DATA_user;
973 	s64 dirty_sectors = 0;
974 	int ret = 0;
975 
976 	r.e.data_type	= data_type;
977 	r.e.nr_devs	= 0;
978 	r.e.nr_required	= 1;
979 
980 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
981 		s64 disk_sectors;
982 		ret = bch2_trigger_pointer(trans, btree_id, level, k, p, &disk_sectors, flags);
983 		if (ret < 0)
984 			return ret;
985 
986 		bool stale = ret > 0;
987 
988 		if (p.ptr.cached) {
989 			if (!stale) {
990 				ret = !gc
991 					? bch2_update_cached_sectors_list(trans, p.ptr.dev, disk_sectors)
992 					: update_cached_sectors(c, k, p.ptr.dev, disk_sectors, 0, true);
993 				bch2_fs_fatal_err_on(ret && gc, c, "%s: no replicas entry while updating cached sectors",
994 						     bch2_err_str(ret));
995 				if (ret)
996 					return ret;
997 			}
998 		} else if (!p.has_ec) {
999 			dirty_sectors	       += disk_sectors;
1000 			r.e.devs[r.e.nr_devs++]	= p.ptr.dev;
1001 		} else {
1002 			ret = bch2_trigger_stripe_ptr(trans, k, p, data_type, disk_sectors, flags);
1003 			if (ret)
1004 				return ret;
1005 
1006 			/*
1007 			 * There may be other dirty pointers in this extent, but
1008 			 * if so they're not required for mounting if we have an
1009 			 * erasure coded pointer in this extent:
1010 			 */
1011 			r.e.nr_required = 0;
1012 		}
1013 	}
1014 
1015 	if (r.e.nr_devs) {
1016 		ret = !gc
1017 			? bch2_update_replicas_list(trans, &r.e, dirty_sectors)
1018 			: bch2_update_replicas(c, k, &r.e, dirty_sectors, 0, true);
1019 		if (unlikely(ret && gc)) {
1020 			struct printbuf buf = PRINTBUF;
1021 
1022 			bch2_bkey_val_to_text(&buf, c, k);
1023 			bch2_fs_fatal_error(c, ": no replicas entry for %s", buf.buf);
1024 			printbuf_exit(&buf);
1025 		}
1026 		if (ret)
1027 			return ret;
1028 	}
1029 
1030 	return 0;
1031 }
1032 
1033 int bch2_trigger_extent(struct btree_trans *trans,
1034 			enum btree_id btree_id, unsigned level,
1035 			struct bkey_s_c old, struct bkey_s new,
1036 			unsigned flags)
1037 {
1038 	struct bkey_ptrs_c new_ptrs = bch2_bkey_ptrs_c(new.s_c);
1039 	struct bkey_ptrs_c old_ptrs = bch2_bkey_ptrs_c(old);
1040 	unsigned new_ptrs_bytes = (void *) new_ptrs.end - (void *) new_ptrs.start;
1041 	unsigned old_ptrs_bytes = (void *) old_ptrs.end - (void *) old_ptrs.start;
1042 
1043 	/* if pointers aren't changing - nothing to do: */
1044 	if (new_ptrs_bytes == old_ptrs_bytes &&
1045 	    !memcmp(new_ptrs.start,
1046 		    old_ptrs.start,
1047 		    new_ptrs_bytes))
1048 		return 0;
1049 
1050 	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
1051 		struct bch_fs *c = trans->c;
1052 		int mod = (int) bch2_bkey_needs_rebalance(c, new.s_c) -
1053 			  (int) bch2_bkey_needs_rebalance(c, old);
1054 
1055 		if (mod) {
1056 			int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_rebalance_work,
1057 							      new.k->p, mod > 0);
1058 			if (ret)
1059 				return ret;
1060 		}
1061 	}
1062 
1063 	if (flags & (BTREE_TRIGGER_TRANSACTIONAL|BTREE_TRIGGER_GC))
1064 		return trigger_run_overwrite_then_insert(__trigger_extent, trans, btree_id, level, old, new, flags);
1065 
1066 	return 0;
1067 }
1068 
1069 /* KEY_TYPE_reservation */
1070 
1071 static int __trigger_reservation(struct btree_trans *trans,
1072 				 enum btree_id btree_id, unsigned level,
1073 				 struct bkey_s_c k, unsigned flags)
1074 {
1075 	struct bch_fs *c = trans->c;
1076 	unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
1077 	s64 sectors = (s64) k.k->size * replicas;
1078 
1079 	if (flags & BTREE_TRIGGER_OVERWRITE)
1080 		sectors = -sectors;
1081 
1082 	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
1083 		int ret = bch2_replicas_deltas_realloc(trans, 0);
1084 		if (ret)
1085 			return ret;
1086 
1087 		struct replicas_delta_list *d = trans->fs_usage_deltas;
1088 		replicas = min(replicas, ARRAY_SIZE(d->persistent_reserved));
1089 
1090 		d->persistent_reserved[replicas - 1] += sectors;
1091 	}
1092 
1093 	if (flags & BTREE_TRIGGER_GC) {
1094 		percpu_down_read(&c->mark_lock);
1095 		preempt_disable();
1096 
1097 		struct bch_fs_usage *fs_usage = this_cpu_ptr(c->usage_gc);
1098 
1099 		replicas = min(replicas, ARRAY_SIZE(fs_usage->persistent_reserved));
1100 		fs_usage->b.reserved				+= sectors;
1101 		fs_usage->persistent_reserved[replicas - 1]	+= sectors;
1102 
1103 		preempt_enable();
1104 		percpu_up_read(&c->mark_lock);
1105 	}
1106 
1107 	return 0;
1108 }
1109 
1110 int bch2_trigger_reservation(struct btree_trans *trans,
1111 			  enum btree_id btree_id, unsigned level,
1112 			  struct bkey_s_c old, struct bkey_s new,
1113 			  unsigned flags)
1114 {
1115 	return trigger_run_overwrite_then_insert(__trigger_reservation, trans, btree_id, level, old, new, flags);
1116 }
1117 
1118 /* Mark superblocks: */
1119 
1120 static int __bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1121 				    struct bch_dev *ca, size_t b,
1122 				    enum bch_data_type type,
1123 				    unsigned sectors)
1124 {
1125 	struct bch_fs *c = trans->c;
1126 	struct btree_iter iter;
1127 	struct bkey_i_alloc_v4 *a;
1128 	int ret = 0;
1129 
1130 	/*
1131 	 * Backup superblock might be past the end of our normal usable space:
1132 	 */
1133 	if (b >= ca->mi.nbuckets)
1134 		return 0;
1135 
1136 	a = bch2_trans_start_alloc_update(trans, &iter, POS(ca->dev_idx, b));
1137 	if (IS_ERR(a))
1138 		return PTR_ERR(a);
1139 
1140 	if (a->v.data_type && type && a->v.data_type != type) {
1141 		bch2_fsck_err(c, FSCK_CAN_IGNORE|FSCK_NEED_FSCK,
1142 			      BCH_FSCK_ERR_bucket_metadata_type_mismatch,
1143 			"bucket %llu:%llu gen %u different types of data in same bucket: %s, %s\n"
1144 			"while marking %s",
1145 			iter.pos.inode, iter.pos.offset, a->v.gen,
1146 			bch2_data_type_str(a->v.data_type),
1147 			bch2_data_type_str(type),
1148 			bch2_data_type_str(type));
1149 		ret = -EIO;
1150 		goto err;
1151 	}
1152 
1153 	if (a->v.data_type	!= type ||
1154 	    a->v.dirty_sectors	!= sectors) {
1155 		a->v.data_type		= type;
1156 		a->v.dirty_sectors	= sectors;
1157 		ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1158 	}
1159 err:
1160 	bch2_trans_iter_exit(trans, &iter);
1161 	return ret;
1162 }
1163 
1164 int bch2_trans_mark_metadata_bucket(struct btree_trans *trans,
1165 				    struct bch_dev *ca, size_t b,
1166 				    enum bch_data_type type,
1167 				    unsigned sectors)
1168 {
1169 	return commit_do(trans, NULL, NULL, 0,
1170 			__bch2_trans_mark_metadata_bucket(trans, ca, b, type, sectors));
1171 }
1172 
1173 static int bch2_trans_mark_metadata_sectors(struct btree_trans *trans,
1174 					    struct bch_dev *ca,
1175 					    u64 start, u64 end,
1176 					    enum bch_data_type type,
1177 					    u64 *bucket, unsigned *bucket_sectors)
1178 {
1179 	do {
1180 		u64 b = sector_to_bucket(ca, start);
1181 		unsigned sectors =
1182 			min_t(u64, bucket_to_sector(ca, b + 1), end) - start;
1183 
1184 		if (b != *bucket && *bucket_sectors) {
1185 			int ret = bch2_trans_mark_metadata_bucket(trans, ca, *bucket,
1186 								  type, *bucket_sectors);
1187 			if (ret)
1188 				return ret;
1189 
1190 			*bucket_sectors = 0;
1191 		}
1192 
1193 		*bucket		= b;
1194 		*bucket_sectors	+= sectors;
1195 		start += sectors;
1196 	} while (start < end);
1197 
1198 	return 0;
1199 }
1200 
1201 static int __bch2_trans_mark_dev_sb(struct btree_trans *trans,
1202 				    struct bch_dev *ca)
1203 {
1204 	struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
1205 	u64 bucket = 0;
1206 	unsigned i, bucket_sectors = 0;
1207 	int ret;
1208 
1209 	for (i = 0; i < layout->nr_superblocks; i++) {
1210 		u64 offset = le64_to_cpu(layout->sb_offset[i]);
1211 
1212 		if (offset == BCH_SB_SECTOR) {
1213 			ret = bch2_trans_mark_metadata_sectors(trans, ca,
1214 						0, BCH_SB_SECTOR,
1215 						BCH_DATA_sb, &bucket, &bucket_sectors);
1216 			if (ret)
1217 				return ret;
1218 		}
1219 
1220 		ret = bch2_trans_mark_metadata_sectors(trans, ca, offset,
1221 				      offset + (1 << layout->sb_max_size_bits),
1222 				      BCH_DATA_sb, &bucket, &bucket_sectors);
1223 		if (ret)
1224 			return ret;
1225 	}
1226 
1227 	if (bucket_sectors) {
1228 		ret = bch2_trans_mark_metadata_bucket(trans, ca,
1229 				bucket, BCH_DATA_sb, bucket_sectors);
1230 		if (ret)
1231 			return ret;
1232 	}
1233 
1234 	for (i = 0; i < ca->journal.nr; i++) {
1235 		ret = bch2_trans_mark_metadata_bucket(trans, ca,
1236 				ca->journal.buckets[i],
1237 				BCH_DATA_journal, ca->mi.bucket_size);
1238 		if (ret)
1239 			return ret;
1240 	}
1241 
1242 	return 0;
1243 }
1244 
1245 int bch2_trans_mark_dev_sb(struct bch_fs *c, struct bch_dev *ca)
1246 {
1247 	int ret = bch2_trans_run(c, __bch2_trans_mark_dev_sb(trans, ca));
1248 
1249 	bch_err_fn(c, ret);
1250 	return ret;
1251 }
1252 
1253 int bch2_trans_mark_dev_sbs(struct bch_fs *c)
1254 {
1255 	for_each_online_member(c, ca) {
1256 		int ret = bch2_trans_mark_dev_sb(c, ca);
1257 		if (ret) {
1258 			percpu_ref_put(&ca->ref);
1259 			return ret;
1260 		}
1261 	}
1262 
1263 	return 0;
1264 }
1265 
1266 /* Disk reservations: */
1267 
1268 #define SECTORS_CACHE	1024
1269 
1270 int __bch2_disk_reservation_add(struct bch_fs *c, struct disk_reservation *res,
1271 			      u64 sectors, int flags)
1272 {
1273 	struct bch_fs_pcpu *pcpu;
1274 	u64 old, v, get;
1275 	s64 sectors_available;
1276 	int ret;
1277 
1278 	percpu_down_read(&c->mark_lock);
1279 	preempt_disable();
1280 	pcpu = this_cpu_ptr(c->pcpu);
1281 
1282 	if (sectors <= pcpu->sectors_available)
1283 		goto out;
1284 
1285 	v = atomic64_read(&c->sectors_available);
1286 	do {
1287 		old = v;
1288 		get = min((u64) sectors + SECTORS_CACHE, old);
1289 
1290 		if (get < sectors) {
1291 			preempt_enable();
1292 			goto recalculate;
1293 		}
1294 	} while ((v = atomic64_cmpxchg(&c->sectors_available,
1295 				       old, old - get)) != old);
1296 
1297 	pcpu->sectors_available		+= get;
1298 
1299 out:
1300 	pcpu->sectors_available		-= sectors;
1301 	this_cpu_add(*c->online_reserved, sectors);
1302 	res->sectors			+= sectors;
1303 
1304 	preempt_enable();
1305 	percpu_up_read(&c->mark_lock);
1306 	return 0;
1307 
1308 recalculate:
1309 	mutex_lock(&c->sectors_available_lock);
1310 
1311 	percpu_u64_set(&c->pcpu->sectors_available, 0);
1312 	sectors_available = avail_factor(__bch2_fs_usage_read_short(c).free);
1313 
1314 	if (sectors <= sectors_available ||
1315 	    (flags & BCH_DISK_RESERVATION_NOFAIL)) {
1316 		atomic64_set(&c->sectors_available,
1317 			     max_t(s64, 0, sectors_available - sectors));
1318 		this_cpu_add(*c->online_reserved, sectors);
1319 		res->sectors			+= sectors;
1320 		ret = 0;
1321 	} else {
1322 		atomic64_set(&c->sectors_available, sectors_available);
1323 		ret = -BCH_ERR_ENOSPC_disk_reservation;
1324 	}
1325 
1326 	mutex_unlock(&c->sectors_available_lock);
1327 	percpu_up_read(&c->mark_lock);
1328 
1329 	return ret;
1330 }
1331 
1332 /* Startup/shutdown: */
1333 
1334 static void bucket_gens_free_rcu(struct rcu_head *rcu)
1335 {
1336 	struct bucket_gens *buckets =
1337 		container_of(rcu, struct bucket_gens, rcu);
1338 
1339 	kvfree(buckets);
1340 }
1341 
1342 int bch2_dev_buckets_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1343 {
1344 	struct bucket_gens *bucket_gens = NULL, *old_bucket_gens = NULL;
1345 	unsigned long *buckets_nouse = NULL;
1346 	bool resize = ca->bucket_gens != NULL;
1347 	int ret;
1348 
1349 	if (!(bucket_gens	= kvmalloc(sizeof(struct bucket_gens) + nbuckets,
1350 					   GFP_KERNEL|__GFP_ZERO))) {
1351 		ret = -BCH_ERR_ENOMEM_bucket_gens;
1352 		goto err;
1353 	}
1354 
1355 	if ((c->opts.buckets_nouse &&
1356 	     !(buckets_nouse	= kvmalloc(BITS_TO_LONGS(nbuckets) *
1357 					   sizeof(unsigned long),
1358 					   GFP_KERNEL|__GFP_ZERO)))) {
1359 		ret = -BCH_ERR_ENOMEM_buckets_nouse;
1360 		goto err;
1361 	}
1362 
1363 	bucket_gens->first_bucket = ca->mi.first_bucket;
1364 	bucket_gens->nbuckets	= nbuckets;
1365 
1366 	if (resize) {
1367 		down_write(&c->gc_lock);
1368 		down_write(&ca->bucket_lock);
1369 		percpu_down_write(&c->mark_lock);
1370 	}
1371 
1372 	old_bucket_gens = rcu_dereference_protected(ca->bucket_gens, 1);
1373 
1374 	if (resize) {
1375 		size_t n = min(bucket_gens->nbuckets, old_bucket_gens->nbuckets);
1376 
1377 		memcpy(bucket_gens->b,
1378 		       old_bucket_gens->b,
1379 		       n);
1380 		if (buckets_nouse)
1381 			memcpy(buckets_nouse,
1382 			       ca->buckets_nouse,
1383 			       BITS_TO_LONGS(n) * sizeof(unsigned long));
1384 	}
1385 
1386 	rcu_assign_pointer(ca->bucket_gens, bucket_gens);
1387 	bucket_gens	= old_bucket_gens;
1388 
1389 	swap(ca->buckets_nouse, buckets_nouse);
1390 
1391 	nbuckets = ca->mi.nbuckets;
1392 
1393 	if (resize) {
1394 		percpu_up_write(&c->mark_lock);
1395 		up_write(&ca->bucket_lock);
1396 		up_write(&c->gc_lock);
1397 	}
1398 
1399 	ret = 0;
1400 err:
1401 	kvfree(buckets_nouse);
1402 	if (bucket_gens)
1403 		call_rcu(&bucket_gens->rcu, bucket_gens_free_rcu);
1404 
1405 	return ret;
1406 }
1407 
1408 void bch2_dev_buckets_free(struct bch_dev *ca)
1409 {
1410 	kvfree(ca->buckets_nouse);
1411 	kvfree(rcu_dereference_protected(ca->bucket_gens, 1));
1412 
1413 	for (unsigned i = 0; i < ARRAY_SIZE(ca->usage); i++)
1414 		free_percpu(ca->usage[i]);
1415 	kfree(ca->usage_base);
1416 }
1417 
1418 int bch2_dev_buckets_alloc(struct bch_fs *c, struct bch_dev *ca)
1419 {
1420 	ca->usage_base = kzalloc(sizeof(struct bch_dev_usage), GFP_KERNEL);
1421 	if (!ca->usage_base)
1422 		return -BCH_ERR_ENOMEM_usage_init;
1423 
1424 	for (unsigned i = 0; i < ARRAY_SIZE(ca->usage); i++) {
1425 		ca->usage[i] = alloc_percpu(struct bch_dev_usage);
1426 		if (!ca->usage[i])
1427 			return -BCH_ERR_ENOMEM_usage_init;
1428 	}
1429 
1430 	return bch2_dev_buckets_resize(c, ca, ca->mi.nbuckets);
1431 }
1432