xref: /linux/fs/bcachefs/replicas.c (revision 257ca10c7317d4a424e48bb95d14ca53a1f1dd6f)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "buckets.h"
5 #include "journal.h"
6 #include "replicas.h"
7 #include "super-io.h"
8 
9 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
10 					    struct bch_replicas_cpu *);
11 
12 /* Some (buggy!) compilers don't allow memcmp to be passed as a pointer */
13 static int bch2_memcmp(const void *l, const void *r, size_t size)
14 {
15 	return memcmp(l, r, size);
16 }
17 
18 /* Replicas tracking - in memory: */
19 
20 static void verify_replicas_entry(struct bch_replicas_entry_v1 *e)
21 {
22 #ifdef CONFIG_BCACHEFS_DEBUG
23 	unsigned i;
24 
25 	BUG_ON(e->data_type >= BCH_DATA_NR);
26 	BUG_ON(!e->nr_devs);
27 	BUG_ON(e->nr_required > 1 &&
28 	       e->nr_required >= e->nr_devs);
29 
30 	for (i = 0; i + 1 < e->nr_devs; i++)
31 		BUG_ON(e->devs[i] >= e->devs[i + 1]);
32 #endif
33 }
34 
35 void bch2_replicas_entry_sort(struct bch_replicas_entry_v1 *e)
36 {
37 	bubble_sort(e->devs, e->nr_devs, u8_cmp);
38 }
39 
40 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
41 {
42 	eytzinger0_sort(r->entries, r->nr, r->entry_size, bch2_memcmp, NULL);
43 }
44 
45 static void bch2_replicas_entry_v0_to_text(struct printbuf *out,
46 					   struct bch_replicas_entry_v0 *e)
47 {
48 	bch2_prt_data_type(out, e->data_type);
49 
50 	prt_printf(out, ": %u [", e->nr_devs);
51 	for (unsigned i = 0; i < e->nr_devs; i++)
52 		prt_printf(out, i ? " %u" : "%u", e->devs[i]);
53 	prt_printf(out, "]");
54 }
55 
56 void bch2_replicas_entry_to_text(struct printbuf *out,
57 				 struct bch_replicas_entry_v1 *e)
58 {
59 	bch2_prt_data_type(out, e->data_type);
60 
61 	prt_printf(out, ": %u/%u [", e->nr_required, e->nr_devs);
62 	for (unsigned i = 0; i < e->nr_devs; i++)
63 		prt_printf(out, i ? " %u" : "%u", e->devs[i]);
64 	prt_printf(out, "]");
65 }
66 
67 int bch2_replicas_entry_validate(struct bch_replicas_entry_v1 *r,
68 				 struct bch_sb *sb,
69 				 struct printbuf *err)
70 {
71 	if (!r->nr_devs) {
72 		prt_printf(err, "no devices in entry ");
73 		goto bad;
74 	}
75 
76 	if (r->nr_required > 1 &&
77 	    r->nr_required >= r->nr_devs) {
78 		prt_printf(err, "bad nr_required in entry ");
79 		goto bad;
80 	}
81 
82 	for (unsigned i = 0; i < r->nr_devs; i++)
83 		if (!bch2_dev_exists(sb, r->devs[i])) {
84 			prt_printf(err, "invalid device %u in entry ", r->devs[i]);
85 			goto bad;
86 		}
87 
88 	return 0;
89 bad:
90 	bch2_replicas_entry_to_text(err, r);
91 	return -BCH_ERR_invalid_replicas_entry;
92 }
93 
94 void bch2_cpu_replicas_to_text(struct printbuf *out,
95 			       struct bch_replicas_cpu *r)
96 {
97 	struct bch_replicas_entry_v1 *e;
98 	bool first = true;
99 
100 	for_each_cpu_replicas_entry(r, e) {
101 		if (!first)
102 			prt_printf(out, " ");
103 		first = false;
104 
105 		bch2_replicas_entry_to_text(out, e);
106 	}
107 }
108 
109 static void extent_to_replicas(struct bkey_s_c k,
110 			       struct bch_replicas_entry_v1 *r)
111 {
112 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
113 	const union bch_extent_entry *entry;
114 	struct extent_ptr_decoded p;
115 
116 	r->nr_required	= 1;
117 
118 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
119 		if (p.ptr.cached)
120 			continue;
121 
122 		if (!p.has_ec)
123 			r->devs[r->nr_devs++] = p.ptr.dev;
124 		else
125 			r->nr_required = 0;
126 	}
127 }
128 
129 static void stripe_to_replicas(struct bkey_s_c k,
130 			       struct bch_replicas_entry_v1 *r)
131 {
132 	struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
133 	const struct bch_extent_ptr *ptr;
134 
135 	r->nr_required	= s.v->nr_blocks - s.v->nr_redundant;
136 
137 	for (ptr = s.v->ptrs;
138 	     ptr < s.v->ptrs + s.v->nr_blocks;
139 	     ptr++)
140 		r->devs[r->nr_devs++] = ptr->dev;
141 }
142 
143 void bch2_bkey_to_replicas(struct bch_replicas_entry_v1 *e,
144 			   struct bkey_s_c k)
145 {
146 	e->nr_devs = 0;
147 
148 	switch (k.k->type) {
149 	case KEY_TYPE_btree_ptr:
150 	case KEY_TYPE_btree_ptr_v2:
151 		e->data_type = BCH_DATA_btree;
152 		extent_to_replicas(k, e);
153 		break;
154 	case KEY_TYPE_extent:
155 	case KEY_TYPE_reflink_v:
156 		e->data_type = BCH_DATA_user;
157 		extent_to_replicas(k, e);
158 		break;
159 	case KEY_TYPE_stripe:
160 		e->data_type = BCH_DATA_parity;
161 		stripe_to_replicas(k, e);
162 		break;
163 	}
164 
165 	bch2_replicas_entry_sort(e);
166 }
167 
168 void bch2_devlist_to_replicas(struct bch_replicas_entry_v1 *e,
169 			      enum bch_data_type data_type,
170 			      struct bch_devs_list devs)
171 {
172 	BUG_ON(!data_type ||
173 	       data_type == BCH_DATA_sb ||
174 	       data_type >= BCH_DATA_NR);
175 
176 	e->data_type	= data_type;
177 	e->nr_devs	= 0;
178 	e->nr_required	= 1;
179 
180 	darray_for_each(devs, i)
181 		e->devs[e->nr_devs++] = *i;
182 
183 	bch2_replicas_entry_sort(e);
184 }
185 
186 static struct bch_replicas_cpu
187 cpu_replicas_add_entry(struct bch_fs *c,
188 		       struct bch_replicas_cpu *old,
189 		       struct bch_replicas_entry_v1 *new_entry)
190 {
191 	unsigned i;
192 	struct bch_replicas_cpu new = {
193 		.nr		= old->nr + 1,
194 		.entry_size	= max_t(unsigned, old->entry_size,
195 					replicas_entry_bytes(new_entry)),
196 	};
197 
198 	for (i = 0; i < new_entry->nr_devs; i++)
199 		BUG_ON(!bch2_dev_exists2(c, new_entry->devs[i]));
200 
201 	BUG_ON(!new_entry->data_type);
202 	verify_replicas_entry(new_entry);
203 
204 	new.entries = kcalloc(new.nr, new.entry_size, GFP_KERNEL);
205 	if (!new.entries)
206 		return new;
207 
208 	for (i = 0; i < old->nr; i++)
209 		memcpy(cpu_replicas_entry(&new, i),
210 		       cpu_replicas_entry(old, i),
211 		       old->entry_size);
212 
213 	memcpy(cpu_replicas_entry(&new, old->nr),
214 	       new_entry,
215 	       replicas_entry_bytes(new_entry));
216 
217 	bch2_cpu_replicas_sort(&new);
218 	return new;
219 }
220 
221 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
222 				       struct bch_replicas_entry_v1 *search)
223 {
224 	int idx, entry_size = replicas_entry_bytes(search);
225 
226 	if (unlikely(entry_size > r->entry_size))
227 		return -1;
228 
229 	verify_replicas_entry(search);
230 
231 #define entry_cmp(_l, _r, size)	memcmp(_l, _r, entry_size)
232 	idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
233 			      entry_cmp, search);
234 #undef entry_cmp
235 
236 	return idx < r->nr ? idx : -1;
237 }
238 
239 int bch2_replicas_entry_idx(struct bch_fs *c,
240 			    struct bch_replicas_entry_v1 *search)
241 {
242 	bch2_replicas_entry_sort(search);
243 
244 	return __replicas_entry_idx(&c->replicas, search);
245 }
246 
247 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
248 				 struct bch_replicas_entry_v1 *search)
249 {
250 	return __replicas_entry_idx(r, search) >= 0;
251 }
252 
253 bool bch2_replicas_marked(struct bch_fs *c,
254 			  struct bch_replicas_entry_v1 *search)
255 {
256 	bool marked;
257 
258 	if (!search->nr_devs)
259 		return true;
260 
261 	verify_replicas_entry(search);
262 
263 	percpu_down_read(&c->mark_lock);
264 	marked = __replicas_has_entry(&c->replicas, search) &&
265 		(likely((!c->replicas_gc.entries)) ||
266 		 __replicas_has_entry(&c->replicas_gc, search));
267 	percpu_up_read(&c->mark_lock);
268 
269 	return marked;
270 }
271 
272 static void __replicas_table_update(struct bch_fs_usage *dst,
273 				    struct bch_replicas_cpu *dst_r,
274 				    struct bch_fs_usage *src,
275 				    struct bch_replicas_cpu *src_r)
276 {
277 	int src_idx, dst_idx;
278 
279 	*dst = *src;
280 
281 	for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
282 		if (!src->replicas[src_idx])
283 			continue;
284 
285 		dst_idx = __replicas_entry_idx(dst_r,
286 				cpu_replicas_entry(src_r, src_idx));
287 		BUG_ON(dst_idx < 0);
288 
289 		dst->replicas[dst_idx] = src->replicas[src_idx];
290 	}
291 }
292 
293 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
294 				    struct bch_replicas_cpu *dst_r,
295 				    struct bch_fs_usage __percpu *src_p,
296 				    struct bch_replicas_cpu *src_r)
297 {
298 	unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
299 	struct bch_fs_usage *dst, *src = (void *)
300 		bch2_acc_percpu_u64s((u64 __percpu *) src_p, src_nr);
301 
302 	preempt_disable();
303 	dst = this_cpu_ptr(dst_p);
304 	preempt_enable();
305 
306 	__replicas_table_update(dst, dst_r, src, src_r);
307 }
308 
309 /*
310  * Resize filesystem accounting:
311  */
312 static int replicas_table_update(struct bch_fs *c,
313 				 struct bch_replicas_cpu *new_r)
314 {
315 	struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
316 	struct bch_fs_usage_online *new_scratch = NULL;
317 	struct bch_fs_usage __percpu *new_gc = NULL;
318 	struct bch_fs_usage *new_base = NULL;
319 	unsigned i, bytes = sizeof(struct bch_fs_usage) +
320 		sizeof(u64) * new_r->nr;
321 	unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
322 		sizeof(u64) * new_r->nr;
323 	int ret = 0;
324 
325 	memset(new_usage, 0, sizeof(new_usage));
326 
327 	for (i = 0; i < ARRAY_SIZE(new_usage); i++)
328 		if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
329 					sizeof(u64), GFP_KERNEL)))
330 			goto err;
331 
332 	if (!(new_base = kzalloc(bytes, GFP_KERNEL)) ||
333 	    !(new_scratch  = kmalloc(scratch_bytes, GFP_KERNEL)) ||
334 	    (c->usage_gc &&
335 	     !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_KERNEL))))
336 		goto err;
337 
338 	for (i = 0; i < ARRAY_SIZE(new_usage); i++)
339 		if (c->usage[i])
340 			__replicas_table_update_pcpu(new_usage[i], new_r,
341 						     c->usage[i], &c->replicas);
342 	if (c->usage_base)
343 		__replicas_table_update(new_base,		new_r,
344 					c->usage_base,		&c->replicas);
345 	if (c->usage_gc)
346 		__replicas_table_update_pcpu(new_gc,		new_r,
347 					     c->usage_gc,	&c->replicas);
348 
349 	for (i = 0; i < ARRAY_SIZE(new_usage); i++)
350 		swap(c->usage[i],	new_usage[i]);
351 	swap(c->usage_base,	new_base);
352 	swap(c->usage_scratch,	new_scratch);
353 	swap(c->usage_gc,	new_gc);
354 	swap(c->replicas,	*new_r);
355 out:
356 	free_percpu(new_gc);
357 	kfree(new_scratch);
358 	for (i = 0; i < ARRAY_SIZE(new_usage); i++)
359 		free_percpu(new_usage[i]);
360 	kfree(new_base);
361 	return ret;
362 err:
363 	bch_err(c, "error updating replicas table: memory allocation failure");
364 	ret = -BCH_ERR_ENOMEM_replicas_table;
365 	goto out;
366 }
367 
368 static unsigned reserve_journal_replicas(struct bch_fs *c,
369 				     struct bch_replicas_cpu *r)
370 {
371 	struct bch_replicas_entry_v1 *e;
372 	unsigned journal_res_u64s = 0;
373 
374 	/* nr_inodes: */
375 	journal_res_u64s +=
376 		DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
377 
378 	/* key_version: */
379 	journal_res_u64s +=
380 		DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
381 
382 	/* persistent_reserved: */
383 	journal_res_u64s +=
384 		DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
385 		BCH_REPLICAS_MAX;
386 
387 	for_each_cpu_replicas_entry(r, e)
388 		journal_res_u64s +=
389 			DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
390 				     e->nr_devs, sizeof(u64));
391 	return journal_res_u64s;
392 }
393 
394 noinline
395 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
396 				struct bch_replicas_entry_v1 *new_entry)
397 {
398 	struct bch_replicas_cpu new_r, new_gc;
399 	int ret = 0;
400 
401 	verify_replicas_entry(new_entry);
402 
403 	memset(&new_r, 0, sizeof(new_r));
404 	memset(&new_gc, 0, sizeof(new_gc));
405 
406 	mutex_lock(&c->sb_lock);
407 
408 	if (c->replicas_gc.entries &&
409 	    !__replicas_has_entry(&c->replicas_gc, new_entry)) {
410 		new_gc = cpu_replicas_add_entry(c, &c->replicas_gc, new_entry);
411 		if (!new_gc.entries) {
412 			ret = -BCH_ERR_ENOMEM_cpu_replicas;
413 			goto err;
414 		}
415 	}
416 
417 	if (!__replicas_has_entry(&c->replicas, new_entry)) {
418 		new_r = cpu_replicas_add_entry(c, &c->replicas, new_entry);
419 		if (!new_r.entries) {
420 			ret = -BCH_ERR_ENOMEM_cpu_replicas;
421 			goto err;
422 		}
423 
424 		ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
425 		if (ret)
426 			goto err;
427 
428 		bch2_journal_entry_res_resize(&c->journal,
429 				&c->replicas_journal_res,
430 				reserve_journal_replicas(c, &new_r));
431 	}
432 
433 	if (!new_r.entries &&
434 	    !new_gc.entries)
435 		goto out;
436 
437 	/* allocations done, now commit: */
438 
439 	if (new_r.entries)
440 		bch2_write_super(c);
441 
442 	/* don't update in memory replicas until changes are persistent */
443 	percpu_down_write(&c->mark_lock);
444 	if (new_r.entries)
445 		ret = replicas_table_update(c, &new_r);
446 	if (new_gc.entries)
447 		swap(new_gc, c->replicas_gc);
448 	percpu_up_write(&c->mark_lock);
449 out:
450 	mutex_unlock(&c->sb_lock);
451 
452 	kfree(new_r.entries);
453 	kfree(new_gc.entries);
454 
455 	return ret;
456 err:
457 	bch_err_msg(c, ret, "adding replicas entry");
458 	goto out;
459 }
460 
461 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry_v1 *r)
462 {
463 	return likely(bch2_replicas_marked(c, r))
464 		? 0 : bch2_mark_replicas_slowpath(c, r);
465 }
466 
467 /* replicas delta list: */
468 
469 int bch2_replicas_delta_list_mark(struct bch_fs *c,
470 				  struct replicas_delta_list *r)
471 {
472 	struct replicas_delta *d = r->d;
473 	struct replicas_delta *top = (void *) r->d + r->used;
474 	int ret = 0;
475 
476 	for (d = r->d; !ret && d != top; d = replicas_delta_next(d))
477 		ret = bch2_mark_replicas(c, &d->r);
478 	return ret;
479 }
480 
481 /*
482  * Old replicas_gc mechanism: only used for journal replicas entries now, should
483  * die at some point:
484  */
485 
486 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
487 {
488 	lockdep_assert_held(&c->replicas_gc_lock);
489 
490 	mutex_lock(&c->sb_lock);
491 	percpu_down_write(&c->mark_lock);
492 
493 	ret =   ret ?:
494 		bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc) ?:
495 		replicas_table_update(c, &c->replicas_gc);
496 
497 	kfree(c->replicas_gc.entries);
498 	c->replicas_gc.entries = NULL;
499 
500 	percpu_up_write(&c->mark_lock);
501 
502 	if (!ret)
503 		bch2_write_super(c);
504 
505 	mutex_unlock(&c->sb_lock);
506 
507 	return ret;
508 }
509 
510 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
511 {
512 	struct bch_replicas_entry_v1 *e;
513 	unsigned i = 0;
514 
515 	lockdep_assert_held(&c->replicas_gc_lock);
516 
517 	mutex_lock(&c->sb_lock);
518 	BUG_ON(c->replicas_gc.entries);
519 
520 	c->replicas_gc.nr		= 0;
521 	c->replicas_gc.entry_size	= 0;
522 
523 	for_each_cpu_replicas_entry(&c->replicas, e)
524 		if (!((1 << e->data_type) & typemask)) {
525 			c->replicas_gc.nr++;
526 			c->replicas_gc.entry_size =
527 				max_t(unsigned, c->replicas_gc.entry_size,
528 				      replicas_entry_bytes(e));
529 		}
530 
531 	c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
532 					 c->replicas_gc.entry_size,
533 					 GFP_KERNEL);
534 	if (!c->replicas_gc.entries) {
535 		mutex_unlock(&c->sb_lock);
536 		bch_err(c, "error allocating c->replicas_gc");
537 		return -BCH_ERR_ENOMEM_replicas_gc;
538 	}
539 
540 	for_each_cpu_replicas_entry(&c->replicas, e)
541 		if (!((1 << e->data_type) & typemask))
542 			memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
543 			       e, c->replicas_gc.entry_size);
544 
545 	bch2_cpu_replicas_sort(&c->replicas_gc);
546 	mutex_unlock(&c->sb_lock);
547 
548 	return 0;
549 }
550 
551 /*
552  * New much simpler mechanism for clearing out unneeded replicas entries - drop
553  * replicas entries that have 0 sectors used.
554  *
555  * However, we don't track sector counts for journal usage, so this doesn't drop
556  * any BCH_DATA_journal entries; the old bch2_replicas_gc_(start|end) mechanism
557  * is retained for that.
558  */
559 int bch2_replicas_gc2(struct bch_fs *c)
560 {
561 	struct bch_replicas_cpu new = { 0 };
562 	unsigned i, nr;
563 	int ret = 0;
564 
565 	bch2_journal_meta(&c->journal);
566 retry:
567 	nr		= READ_ONCE(c->replicas.nr);
568 	new.entry_size	= READ_ONCE(c->replicas.entry_size);
569 	new.entries	= kcalloc(nr, new.entry_size, GFP_KERNEL);
570 	if (!new.entries) {
571 		bch_err(c, "error allocating c->replicas_gc");
572 		return -BCH_ERR_ENOMEM_replicas_gc;
573 	}
574 
575 	mutex_lock(&c->sb_lock);
576 	percpu_down_write(&c->mark_lock);
577 
578 	if (nr			!= c->replicas.nr ||
579 	    new.entry_size	!= c->replicas.entry_size) {
580 		percpu_up_write(&c->mark_lock);
581 		mutex_unlock(&c->sb_lock);
582 		kfree(new.entries);
583 		goto retry;
584 	}
585 
586 	for (i = 0; i < c->replicas.nr; i++) {
587 		struct bch_replicas_entry_v1 *e =
588 			cpu_replicas_entry(&c->replicas, i);
589 
590 		if (e->data_type == BCH_DATA_journal ||
591 		    c->usage_base->replicas[i] ||
592 		    percpu_u64_get(&c->usage[0]->replicas[i]) ||
593 		    percpu_u64_get(&c->usage[1]->replicas[i]) ||
594 		    percpu_u64_get(&c->usage[2]->replicas[i]) ||
595 		    percpu_u64_get(&c->usage[3]->replicas[i]))
596 			memcpy(cpu_replicas_entry(&new, new.nr++),
597 			       e, new.entry_size);
598 	}
599 
600 	bch2_cpu_replicas_sort(&new);
601 
602 	ret =   bch2_cpu_replicas_to_sb_replicas(c, &new) ?:
603 		replicas_table_update(c, &new);
604 
605 	kfree(new.entries);
606 
607 	percpu_up_write(&c->mark_lock);
608 
609 	if (!ret)
610 		bch2_write_super(c);
611 
612 	mutex_unlock(&c->sb_lock);
613 
614 	return ret;
615 }
616 
617 int bch2_replicas_set_usage(struct bch_fs *c,
618 			    struct bch_replicas_entry_v1 *r,
619 			    u64 sectors)
620 {
621 	int ret, idx = bch2_replicas_entry_idx(c, r);
622 
623 	if (idx < 0) {
624 		struct bch_replicas_cpu n;
625 
626 		n = cpu_replicas_add_entry(c, &c->replicas, r);
627 		if (!n.entries)
628 			return -BCH_ERR_ENOMEM_cpu_replicas;
629 
630 		ret = replicas_table_update(c, &n);
631 		if (ret)
632 			return ret;
633 
634 		kfree(n.entries);
635 
636 		idx = bch2_replicas_entry_idx(c, r);
637 		BUG_ON(ret < 0);
638 	}
639 
640 	c->usage_base->replicas[idx] = sectors;
641 
642 	return 0;
643 }
644 
645 /* Replicas tracking - superblock: */
646 
647 static int
648 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
649 				   struct bch_replicas_cpu *cpu_r)
650 {
651 	struct bch_replicas_entry_v1 *e, *dst;
652 	unsigned nr = 0, entry_size = 0, idx = 0;
653 
654 	for_each_replicas_entry(sb_r, e) {
655 		entry_size = max_t(unsigned, entry_size,
656 				   replicas_entry_bytes(e));
657 		nr++;
658 	}
659 
660 	cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
661 	if (!cpu_r->entries)
662 		return -BCH_ERR_ENOMEM_cpu_replicas;
663 
664 	cpu_r->nr		= nr;
665 	cpu_r->entry_size	= entry_size;
666 
667 	for_each_replicas_entry(sb_r, e) {
668 		dst = cpu_replicas_entry(cpu_r, idx++);
669 		memcpy(dst, e, replicas_entry_bytes(e));
670 		bch2_replicas_entry_sort(dst);
671 	}
672 
673 	return 0;
674 }
675 
676 static int
677 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
678 				      struct bch_replicas_cpu *cpu_r)
679 {
680 	struct bch_replicas_entry_v0 *e;
681 	unsigned nr = 0, entry_size = 0, idx = 0;
682 
683 	for_each_replicas_entry(sb_r, e) {
684 		entry_size = max_t(unsigned, entry_size,
685 				   replicas_entry_bytes(e));
686 		nr++;
687 	}
688 
689 	entry_size += sizeof(struct bch_replicas_entry_v1) -
690 		sizeof(struct bch_replicas_entry_v0);
691 
692 	cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
693 	if (!cpu_r->entries)
694 		return -BCH_ERR_ENOMEM_cpu_replicas;
695 
696 	cpu_r->nr		= nr;
697 	cpu_r->entry_size	= entry_size;
698 
699 	for_each_replicas_entry(sb_r, e) {
700 		struct bch_replicas_entry_v1 *dst =
701 			cpu_replicas_entry(cpu_r, idx++);
702 
703 		dst->data_type	= e->data_type;
704 		dst->nr_devs	= e->nr_devs;
705 		dst->nr_required = 1;
706 		memcpy(dst->devs, e->devs, e->nr_devs);
707 		bch2_replicas_entry_sort(dst);
708 	}
709 
710 	return 0;
711 }
712 
713 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
714 {
715 	struct bch_sb_field_replicas *sb_v1;
716 	struct bch_sb_field_replicas_v0 *sb_v0;
717 	struct bch_replicas_cpu new_r = { 0, 0, NULL };
718 	int ret = 0;
719 
720 	if ((sb_v1 = bch2_sb_field_get(c->disk_sb.sb, replicas)))
721 		ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
722 	else if ((sb_v0 = bch2_sb_field_get(c->disk_sb.sb, replicas_v0)))
723 		ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
724 	if (ret)
725 		return ret;
726 
727 	bch2_cpu_replicas_sort(&new_r);
728 
729 	percpu_down_write(&c->mark_lock);
730 
731 	ret = replicas_table_update(c, &new_r);
732 	percpu_up_write(&c->mark_lock);
733 
734 	kfree(new_r.entries);
735 
736 	return 0;
737 }
738 
739 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
740 					       struct bch_replicas_cpu *r)
741 {
742 	struct bch_sb_field_replicas_v0 *sb_r;
743 	struct bch_replicas_entry_v0 *dst;
744 	struct bch_replicas_entry_v1 *src;
745 	size_t bytes;
746 
747 	bytes = sizeof(struct bch_sb_field_replicas);
748 
749 	for_each_cpu_replicas_entry(r, src)
750 		bytes += replicas_entry_bytes(src) - 1;
751 
752 	sb_r = bch2_sb_field_resize(&c->disk_sb, replicas_v0,
753 			DIV_ROUND_UP(bytes, sizeof(u64)));
754 	if (!sb_r)
755 		return -BCH_ERR_ENOSPC_sb_replicas;
756 
757 	bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
758 	sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas_v0);
759 
760 	memset(&sb_r->entries, 0,
761 	       vstruct_end(&sb_r->field) -
762 	       (void *) &sb_r->entries);
763 
764 	dst = sb_r->entries;
765 	for_each_cpu_replicas_entry(r, src) {
766 		dst->data_type	= src->data_type;
767 		dst->nr_devs	= src->nr_devs;
768 		memcpy(dst->devs, src->devs, src->nr_devs);
769 
770 		dst = replicas_entry_next(dst);
771 
772 		BUG_ON((void *) dst > vstruct_end(&sb_r->field));
773 	}
774 
775 	return 0;
776 }
777 
778 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
779 					    struct bch_replicas_cpu *r)
780 {
781 	struct bch_sb_field_replicas *sb_r;
782 	struct bch_replicas_entry_v1 *dst, *src;
783 	bool need_v1 = false;
784 	size_t bytes;
785 
786 	bytes = sizeof(struct bch_sb_field_replicas);
787 
788 	for_each_cpu_replicas_entry(r, src) {
789 		bytes += replicas_entry_bytes(src);
790 		if (src->nr_required != 1)
791 			need_v1 = true;
792 	}
793 
794 	if (!need_v1)
795 		return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
796 
797 	sb_r = bch2_sb_field_resize(&c->disk_sb, replicas,
798 			DIV_ROUND_UP(bytes, sizeof(u64)));
799 	if (!sb_r)
800 		return -BCH_ERR_ENOSPC_sb_replicas;
801 
802 	bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
803 	sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas);
804 
805 	memset(&sb_r->entries, 0,
806 	       vstruct_end(&sb_r->field) -
807 	       (void *) &sb_r->entries);
808 
809 	dst = sb_r->entries;
810 	for_each_cpu_replicas_entry(r, src) {
811 		memcpy(dst, src, replicas_entry_bytes(src));
812 
813 		dst = replicas_entry_next(dst);
814 
815 		BUG_ON((void *) dst > vstruct_end(&sb_r->field));
816 	}
817 
818 	return 0;
819 }
820 
821 static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r,
822 				      struct bch_sb *sb,
823 				      struct printbuf *err)
824 {
825 	unsigned i;
826 
827 	sort_cmp_size(cpu_r->entries,
828 		      cpu_r->nr,
829 		      cpu_r->entry_size,
830 		      bch2_memcmp, NULL);
831 
832 	for (i = 0; i < cpu_r->nr; i++) {
833 		struct bch_replicas_entry_v1 *e =
834 			cpu_replicas_entry(cpu_r, i);
835 
836 		int ret = bch2_replicas_entry_validate(e, sb, err);
837 		if (ret)
838 			return ret;
839 
840 		if (i + 1 < cpu_r->nr) {
841 			struct bch_replicas_entry_v1 *n =
842 				cpu_replicas_entry(cpu_r, i + 1);
843 
844 			BUG_ON(memcmp(e, n, cpu_r->entry_size) > 0);
845 
846 			if (!memcmp(e, n, cpu_r->entry_size)) {
847 				prt_printf(err, "duplicate replicas entry ");
848 				bch2_replicas_entry_to_text(err, e);
849 				return -BCH_ERR_invalid_sb_replicas;
850 			}
851 		}
852 	}
853 
854 	return 0;
855 }
856 
857 static int bch2_sb_replicas_validate(struct bch_sb *sb, struct bch_sb_field *f,
858 				     struct printbuf *err)
859 {
860 	struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
861 	struct bch_replicas_cpu cpu_r;
862 	int ret;
863 
864 	ret = __bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r);
865 	if (ret)
866 		return ret;
867 
868 	ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
869 	kfree(cpu_r.entries);
870 	return ret;
871 }
872 
873 static void bch2_sb_replicas_to_text(struct printbuf *out,
874 				     struct bch_sb *sb,
875 				     struct bch_sb_field *f)
876 {
877 	struct bch_sb_field_replicas *r = field_to_type(f, replicas);
878 	struct bch_replicas_entry_v1 *e;
879 	bool first = true;
880 
881 	for_each_replicas_entry(r, e) {
882 		if (!first)
883 			prt_printf(out, " ");
884 		first = false;
885 
886 		bch2_replicas_entry_to_text(out, e);
887 	}
888 	prt_newline(out);
889 }
890 
891 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
892 	.validate	= bch2_sb_replicas_validate,
893 	.to_text	= bch2_sb_replicas_to_text,
894 };
895 
896 static int bch2_sb_replicas_v0_validate(struct bch_sb *sb, struct bch_sb_field *f,
897 					struct printbuf *err)
898 {
899 	struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
900 	struct bch_replicas_cpu cpu_r;
901 	int ret;
902 
903 	ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r);
904 	if (ret)
905 		return ret;
906 
907 	ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
908 	kfree(cpu_r.entries);
909 	return ret;
910 }
911 
912 static void bch2_sb_replicas_v0_to_text(struct printbuf *out,
913 					struct bch_sb *sb,
914 					struct bch_sb_field *f)
915 {
916 	struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
917 	struct bch_replicas_entry_v0 *e;
918 	bool first = true;
919 
920 	for_each_replicas_entry(sb_r, e) {
921 		if (!first)
922 			prt_printf(out, " ");
923 		first = false;
924 
925 		bch2_replicas_entry_v0_to_text(out, e);
926 	}
927 	prt_newline(out);
928 }
929 
930 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
931 	.validate	= bch2_sb_replicas_v0_validate,
932 	.to_text	= bch2_sb_replicas_v0_to_text,
933 };
934 
935 /* Query replicas: */
936 
937 bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
938 			   unsigned flags, bool print)
939 {
940 	struct bch_replicas_entry_v1 *e;
941 	bool ret = true;
942 
943 	percpu_down_read(&c->mark_lock);
944 	for_each_cpu_replicas_entry(&c->replicas, e) {
945 		unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
946 		bool metadata = e->data_type < BCH_DATA_user;
947 
948 		if (e->data_type == BCH_DATA_cached)
949 			continue;
950 
951 		for (i = 0; i < e->nr_devs; i++) {
952 			struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
953 
954 			nr_online += test_bit(e->devs[i], devs.d);
955 			nr_failed += ca->mi.state == BCH_MEMBER_STATE_failed;
956 		}
957 
958 		if (nr_failed == e->nr_devs)
959 			continue;
960 
961 		if (nr_online < e->nr_required)
962 			dflags |= metadata
963 				? BCH_FORCE_IF_METADATA_LOST
964 				: BCH_FORCE_IF_DATA_LOST;
965 
966 		if (nr_online < e->nr_devs)
967 			dflags |= metadata
968 				? BCH_FORCE_IF_METADATA_DEGRADED
969 				: BCH_FORCE_IF_DATA_DEGRADED;
970 
971 		if (dflags & ~flags) {
972 			if (print) {
973 				struct printbuf buf = PRINTBUF;
974 
975 				bch2_replicas_entry_to_text(&buf, e);
976 				bch_err(c, "insufficient devices online (%u) for replicas entry %s",
977 					nr_online, buf.buf);
978 				printbuf_exit(&buf);
979 			}
980 			ret = false;
981 			break;
982 		}
983 
984 	}
985 	percpu_up_read(&c->mark_lock);
986 
987 	return ret;
988 }
989 
990 unsigned bch2_sb_dev_has_data(struct bch_sb *sb, unsigned dev)
991 {
992 	struct bch_sb_field_replicas *replicas;
993 	struct bch_sb_field_replicas_v0 *replicas_v0;
994 	unsigned i, data_has = 0;
995 
996 	replicas = bch2_sb_field_get(sb, replicas);
997 	replicas_v0 = bch2_sb_field_get(sb, replicas_v0);
998 
999 	if (replicas) {
1000 		struct bch_replicas_entry_v1 *r;
1001 
1002 		for_each_replicas_entry(replicas, r)
1003 			for (i = 0; i < r->nr_devs; i++)
1004 				if (r->devs[i] == dev)
1005 					data_has |= 1 << r->data_type;
1006 	} else if (replicas_v0) {
1007 		struct bch_replicas_entry_v0 *r;
1008 
1009 		for_each_replicas_entry_v0(replicas_v0, r)
1010 			for (i = 0; i < r->nr_devs; i++)
1011 				if (r->devs[i] == dev)
1012 					data_has |= 1 << r->data_type;
1013 	}
1014 
1015 
1016 	return data_has;
1017 }
1018 
1019 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1020 {
1021 	unsigned ret;
1022 
1023 	mutex_lock(&c->sb_lock);
1024 	ret = bch2_sb_dev_has_data(c->disk_sb.sb, ca->dev_idx);
1025 	mutex_unlock(&c->sb_lock);
1026 
1027 	return ret;
1028 }
1029 
1030 void bch2_fs_replicas_exit(struct bch_fs *c)
1031 {
1032 	unsigned i;
1033 
1034 	kfree(c->usage_scratch);
1035 	for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1036 		free_percpu(c->usage[i]);
1037 	kfree(c->usage_base);
1038 	kfree(c->replicas.entries);
1039 	kfree(c->replicas_gc.entries);
1040 
1041 	mempool_exit(&c->replicas_delta_pool);
1042 }
1043 
1044 int bch2_fs_replicas_init(struct bch_fs *c)
1045 {
1046 	bch2_journal_entry_res_resize(&c->journal,
1047 			&c->replicas_journal_res,
1048 			reserve_journal_replicas(c, &c->replicas));
1049 
1050 	return mempool_init_kmalloc_pool(&c->replicas_delta_pool, 1,
1051 					 REPLICAS_DELTA_LIST_MAX) ?:
1052 		replicas_table_update(c, &c->replicas);
1053 }
1054