1 // SPDX-License-Identifier: GPL-2.0
9 #include <linux/sort.h>
11 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *,
12 struct bch_replicas_cpu *);
14 /* Some (buggy!) compilers don't allow memcmp to be passed as a pointer */
15 static int bch2_memcmp(const void *l, const void *r, const void *priv)
17 size_t size = (size_t) priv;
18 return memcmp(l, r, size);
21 /* Replicas tracking - in memory: */
23 static void verify_replicas_entry(struct bch_replicas_entry_v1 *e)
25 #ifdef CONFIG_BCACHEFS_DEBUG
28 BUG_ON(e->data_type >= BCH_DATA_NR);
30 BUG_ON(e->nr_required > 1 &&
31 e->nr_required >= e->nr_devs);
33 for (i = 0; i + 1 < e->nr_devs; i++)
34 BUG_ON(e->devs[i] >= e->devs[i + 1]);
38 void bch2_replicas_entry_sort(struct bch_replicas_entry_v1 *e)
40 bubble_sort(e->devs, e->nr_devs, u8_cmp);
43 static void bch2_cpu_replicas_sort(struct bch_replicas_cpu *r)
45 eytzinger0_sort_r(r->entries, r->nr, r->entry_size,
46 bch2_memcmp, NULL, (void *)(size_t)r->entry_size);
49 static void bch2_replicas_entry_v0_to_text(struct printbuf *out,
50 struct bch_replicas_entry_v0 *e)
52 bch2_prt_data_type(out, e->data_type);
54 prt_printf(out, ": %u [", e->nr_devs);
55 for (unsigned i = 0; i < e->nr_devs; i++)
56 prt_printf(out, i ? " %u" : "%u", e->devs[i]);
60 void bch2_replicas_entry_to_text(struct printbuf *out,
61 struct bch_replicas_entry_v1 *e)
63 bch2_prt_data_type(out, e->data_type);
65 prt_printf(out, ": %u/%u [", e->nr_required, e->nr_devs);
66 for (unsigned i = 0; i < e->nr_devs; i++)
67 prt_printf(out, i ? " %u" : "%u", e->devs[i]);
71 int bch2_replicas_entry_validate(struct bch_replicas_entry_v1 *r,
76 prt_printf(err, "no devices in entry ");
80 if (r->nr_required > 1 &&
81 r->nr_required >= r->nr_devs) {
82 prt_printf(err, "bad nr_required in entry ");
86 for (unsigned i = 0; i < r->nr_devs; i++)
87 if (!bch2_dev_exists(sb, r->devs[i])) {
88 prt_printf(err, "invalid device %u in entry ", r->devs[i]);
94 bch2_replicas_entry_to_text(err, r);
95 return -BCH_ERR_invalid_replicas_entry;
98 void bch2_cpu_replicas_to_text(struct printbuf *out,
99 struct bch_replicas_cpu *r)
101 struct bch_replicas_entry_v1 *e;
104 for_each_cpu_replicas_entry(r, e) {
106 prt_printf(out, " ");
109 bch2_replicas_entry_to_text(out, e);
113 static void extent_to_replicas(struct bkey_s_c k,
114 struct bch_replicas_entry_v1 *r)
116 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
117 const union bch_extent_entry *entry;
118 struct extent_ptr_decoded p;
122 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
127 r->devs[r->nr_devs++] = p.ptr.dev;
133 static void stripe_to_replicas(struct bkey_s_c k,
134 struct bch_replicas_entry_v1 *r)
136 struct bkey_s_c_stripe s = bkey_s_c_to_stripe(k);
137 const struct bch_extent_ptr *ptr;
139 r->nr_required = s.v->nr_blocks - s.v->nr_redundant;
141 for (ptr = s.v->ptrs;
142 ptr < s.v->ptrs + s.v->nr_blocks;
144 r->devs[r->nr_devs++] = ptr->dev;
147 void bch2_bkey_to_replicas(struct bch_replicas_entry_v1 *e,
153 case KEY_TYPE_btree_ptr:
154 case KEY_TYPE_btree_ptr_v2:
155 e->data_type = BCH_DATA_btree;
156 extent_to_replicas(k, e);
158 case KEY_TYPE_extent:
159 case KEY_TYPE_reflink_v:
160 e->data_type = BCH_DATA_user;
161 extent_to_replicas(k, e);
163 case KEY_TYPE_stripe:
164 e->data_type = BCH_DATA_parity;
165 stripe_to_replicas(k, e);
169 bch2_replicas_entry_sort(e);
172 void bch2_devlist_to_replicas(struct bch_replicas_entry_v1 *e,
173 enum bch_data_type data_type,
174 struct bch_devs_list devs)
177 data_type == BCH_DATA_sb ||
178 data_type >= BCH_DATA_NR);
180 e->data_type = data_type;
184 darray_for_each(devs, i)
185 e->devs[e->nr_devs++] = *i;
187 bch2_replicas_entry_sort(e);
190 static struct bch_replicas_cpu
191 cpu_replicas_add_entry(struct bch_fs *c,
192 struct bch_replicas_cpu *old,
193 struct bch_replicas_entry_v1 *new_entry)
196 struct bch_replicas_cpu new = {
198 .entry_size = max_t(unsigned, old->entry_size,
199 replicas_entry_bytes(new_entry)),
202 for (i = 0; i < new_entry->nr_devs; i++)
203 BUG_ON(!bch2_dev_exists2(c, new_entry->devs[i]));
205 BUG_ON(!new_entry->data_type);
206 verify_replicas_entry(new_entry);
208 new.entries = kcalloc(new.nr, new.entry_size, GFP_KERNEL);
212 for (i = 0; i < old->nr; i++)
213 memcpy(cpu_replicas_entry(&new, i),
214 cpu_replicas_entry(old, i),
217 memcpy(cpu_replicas_entry(&new, old->nr),
219 replicas_entry_bytes(new_entry));
221 bch2_cpu_replicas_sort(&new);
225 static inline int __replicas_entry_idx(struct bch_replicas_cpu *r,
226 struct bch_replicas_entry_v1 *search)
228 int idx, entry_size = replicas_entry_bytes(search);
230 if (unlikely(entry_size > r->entry_size))
233 verify_replicas_entry(search);
235 #define entry_cmp(_l, _r) memcmp(_l, _r, entry_size)
236 idx = eytzinger0_find(r->entries, r->nr, r->entry_size,
240 return idx < r->nr ? idx : -1;
243 int bch2_replicas_entry_idx(struct bch_fs *c,
244 struct bch_replicas_entry_v1 *search)
246 bch2_replicas_entry_sort(search);
248 return __replicas_entry_idx(&c->replicas, search);
251 static bool __replicas_has_entry(struct bch_replicas_cpu *r,
252 struct bch_replicas_entry_v1 *search)
254 return __replicas_entry_idx(r, search) >= 0;
257 bool bch2_replicas_marked(struct bch_fs *c,
258 struct bch_replicas_entry_v1 *search)
262 if (!search->nr_devs)
265 verify_replicas_entry(search);
267 percpu_down_read(&c->mark_lock);
268 marked = __replicas_has_entry(&c->replicas, search) &&
269 (likely((!c->replicas_gc.entries)) ||
270 __replicas_has_entry(&c->replicas_gc, search));
271 percpu_up_read(&c->mark_lock);
276 static void __replicas_table_update(struct bch_fs_usage *dst,
277 struct bch_replicas_cpu *dst_r,
278 struct bch_fs_usage *src,
279 struct bch_replicas_cpu *src_r)
281 int src_idx, dst_idx;
285 for (src_idx = 0; src_idx < src_r->nr; src_idx++) {
286 if (!src->replicas[src_idx])
289 dst_idx = __replicas_entry_idx(dst_r,
290 cpu_replicas_entry(src_r, src_idx));
293 dst->replicas[dst_idx] = src->replicas[src_idx];
297 static void __replicas_table_update_pcpu(struct bch_fs_usage __percpu *dst_p,
298 struct bch_replicas_cpu *dst_r,
299 struct bch_fs_usage __percpu *src_p,
300 struct bch_replicas_cpu *src_r)
302 unsigned src_nr = sizeof(struct bch_fs_usage) / sizeof(u64) + src_r->nr;
303 struct bch_fs_usage *dst, *src = (void *)
304 bch2_acc_percpu_u64s((u64 __percpu *) src_p, src_nr);
307 dst = this_cpu_ptr(dst_p);
310 __replicas_table_update(dst, dst_r, src, src_r);
314 * Resize filesystem accounting:
316 static int replicas_table_update(struct bch_fs *c,
317 struct bch_replicas_cpu *new_r)
319 struct bch_fs_usage __percpu *new_usage[JOURNAL_BUF_NR];
320 struct bch_fs_usage_online *new_scratch = NULL;
321 struct bch_fs_usage __percpu *new_gc = NULL;
322 struct bch_fs_usage *new_base = NULL;
323 unsigned i, bytes = sizeof(struct bch_fs_usage) +
324 sizeof(u64) * new_r->nr;
325 unsigned scratch_bytes = sizeof(struct bch_fs_usage_online) +
326 sizeof(u64) * new_r->nr;
329 memset(new_usage, 0, sizeof(new_usage));
331 for (i = 0; i < ARRAY_SIZE(new_usage); i++)
332 if (!(new_usage[i] = __alloc_percpu_gfp(bytes,
333 sizeof(u64), GFP_KERNEL)))
336 if (!(new_base = kzalloc(bytes, GFP_KERNEL)) ||
337 !(new_scratch = kmalloc(scratch_bytes, GFP_KERNEL)) ||
339 !(new_gc = __alloc_percpu_gfp(bytes, sizeof(u64), GFP_KERNEL))))
342 for (i = 0; i < ARRAY_SIZE(new_usage); i++)
344 __replicas_table_update_pcpu(new_usage[i], new_r,
345 c->usage[i], &c->replicas);
347 __replicas_table_update(new_base, new_r,
348 c->usage_base, &c->replicas);
350 __replicas_table_update_pcpu(new_gc, new_r,
351 c->usage_gc, &c->replicas);
353 for (i = 0; i < ARRAY_SIZE(new_usage); i++)
354 swap(c->usage[i], new_usage[i]);
355 swap(c->usage_base, new_base);
356 swap(c->usage_scratch, new_scratch);
357 swap(c->usage_gc, new_gc);
358 swap(c->replicas, *new_r);
362 for (i = 0; i < ARRAY_SIZE(new_usage); i++)
363 free_percpu(new_usage[i]);
367 bch_err(c, "error updating replicas table: memory allocation failure");
368 ret = -BCH_ERR_ENOMEM_replicas_table;
372 static unsigned reserve_journal_replicas(struct bch_fs *c,
373 struct bch_replicas_cpu *r)
375 struct bch_replicas_entry_v1 *e;
376 unsigned journal_res_u64s = 0;
380 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
384 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64));
386 /* persistent_reserved: */
388 DIV_ROUND_UP(sizeof(struct jset_entry_usage), sizeof(u64)) *
391 for_each_cpu_replicas_entry(r, e)
393 DIV_ROUND_UP(sizeof(struct jset_entry_data_usage) +
394 e->nr_devs, sizeof(u64));
395 return journal_res_u64s;
399 static int bch2_mark_replicas_slowpath(struct bch_fs *c,
400 struct bch_replicas_entry_v1 *new_entry)
402 struct bch_replicas_cpu new_r, new_gc;
405 verify_replicas_entry(new_entry);
407 memset(&new_r, 0, sizeof(new_r));
408 memset(&new_gc, 0, sizeof(new_gc));
410 mutex_lock(&c->sb_lock);
412 if (c->replicas_gc.entries &&
413 !__replicas_has_entry(&c->replicas_gc, new_entry)) {
414 new_gc = cpu_replicas_add_entry(c, &c->replicas_gc, new_entry);
415 if (!new_gc.entries) {
416 ret = -BCH_ERR_ENOMEM_cpu_replicas;
421 if (!__replicas_has_entry(&c->replicas, new_entry)) {
422 new_r = cpu_replicas_add_entry(c, &c->replicas, new_entry);
423 if (!new_r.entries) {
424 ret = -BCH_ERR_ENOMEM_cpu_replicas;
428 ret = bch2_cpu_replicas_to_sb_replicas(c, &new_r);
432 bch2_journal_entry_res_resize(&c->journal,
433 &c->replicas_journal_res,
434 reserve_journal_replicas(c, &new_r));
437 if (!new_r.entries &&
441 /* allocations done, now commit: */
446 /* don't update in memory replicas until changes are persistent */
447 percpu_down_write(&c->mark_lock);
449 ret = replicas_table_update(c, &new_r);
451 swap(new_gc, c->replicas_gc);
452 percpu_up_write(&c->mark_lock);
454 mutex_unlock(&c->sb_lock);
456 kfree(new_r.entries);
457 kfree(new_gc.entries);
461 bch_err_msg(c, ret, "adding replicas entry");
465 int bch2_mark_replicas(struct bch_fs *c, struct bch_replicas_entry_v1 *r)
467 return likely(bch2_replicas_marked(c, r))
468 ? 0 : bch2_mark_replicas_slowpath(c, r);
471 /* replicas delta list: */
473 int bch2_replicas_delta_list_mark(struct bch_fs *c,
474 struct replicas_delta_list *r)
476 struct replicas_delta *d = r->d;
477 struct replicas_delta *top = (void *) r->d + r->used;
480 for (d = r->d; !ret && d != top; d = replicas_delta_next(d))
481 ret = bch2_mark_replicas(c, &d->r);
486 * Old replicas_gc mechanism: only used for journal replicas entries now, should
490 int bch2_replicas_gc_end(struct bch_fs *c, int ret)
492 lockdep_assert_held(&c->replicas_gc_lock);
494 mutex_lock(&c->sb_lock);
495 percpu_down_write(&c->mark_lock);
498 bch2_cpu_replicas_to_sb_replicas(c, &c->replicas_gc) ?:
499 replicas_table_update(c, &c->replicas_gc);
501 kfree(c->replicas_gc.entries);
502 c->replicas_gc.entries = NULL;
504 percpu_up_write(&c->mark_lock);
509 mutex_unlock(&c->sb_lock);
514 int bch2_replicas_gc_start(struct bch_fs *c, unsigned typemask)
516 struct bch_replicas_entry_v1 *e;
519 lockdep_assert_held(&c->replicas_gc_lock);
521 mutex_lock(&c->sb_lock);
522 BUG_ON(c->replicas_gc.entries);
524 c->replicas_gc.nr = 0;
525 c->replicas_gc.entry_size = 0;
527 for_each_cpu_replicas_entry(&c->replicas, e)
528 if (!((1 << e->data_type) & typemask)) {
530 c->replicas_gc.entry_size =
531 max_t(unsigned, c->replicas_gc.entry_size,
532 replicas_entry_bytes(e));
535 c->replicas_gc.entries = kcalloc(c->replicas_gc.nr,
536 c->replicas_gc.entry_size,
538 if (!c->replicas_gc.entries) {
539 mutex_unlock(&c->sb_lock);
540 bch_err(c, "error allocating c->replicas_gc");
541 return -BCH_ERR_ENOMEM_replicas_gc;
544 for_each_cpu_replicas_entry(&c->replicas, e)
545 if (!((1 << e->data_type) & typemask))
546 memcpy(cpu_replicas_entry(&c->replicas_gc, i++),
547 e, c->replicas_gc.entry_size);
549 bch2_cpu_replicas_sort(&c->replicas_gc);
550 mutex_unlock(&c->sb_lock);
556 * New much simpler mechanism for clearing out unneeded replicas entries - drop
557 * replicas entries that have 0 sectors used.
559 * However, we don't track sector counts for journal usage, so this doesn't drop
560 * any BCH_DATA_journal entries; the old bch2_replicas_gc_(start|end) mechanism
561 * is retained for that.
563 int bch2_replicas_gc2(struct bch_fs *c)
565 struct bch_replicas_cpu new = { 0 };
569 bch2_journal_meta(&c->journal);
571 nr = READ_ONCE(c->replicas.nr);
572 new.entry_size = READ_ONCE(c->replicas.entry_size);
573 new.entries = kcalloc(nr, new.entry_size, GFP_KERNEL);
575 bch_err(c, "error allocating c->replicas_gc");
576 return -BCH_ERR_ENOMEM_replicas_gc;
579 mutex_lock(&c->sb_lock);
580 percpu_down_write(&c->mark_lock);
582 if (nr != c->replicas.nr ||
583 new.entry_size != c->replicas.entry_size) {
584 percpu_up_write(&c->mark_lock);
585 mutex_unlock(&c->sb_lock);
590 for (i = 0; i < c->replicas.nr; i++) {
591 struct bch_replicas_entry_v1 *e =
592 cpu_replicas_entry(&c->replicas, i);
594 if (e->data_type == BCH_DATA_journal ||
595 c->usage_base->replicas[i] ||
596 percpu_u64_get(&c->usage[0]->replicas[i]) ||
597 percpu_u64_get(&c->usage[1]->replicas[i]) ||
598 percpu_u64_get(&c->usage[2]->replicas[i]) ||
599 percpu_u64_get(&c->usage[3]->replicas[i]))
600 memcpy(cpu_replicas_entry(&new, new.nr++),
604 bch2_cpu_replicas_sort(&new);
606 ret = bch2_cpu_replicas_to_sb_replicas(c, &new) ?:
607 replicas_table_update(c, &new);
611 percpu_up_write(&c->mark_lock);
616 mutex_unlock(&c->sb_lock);
621 int bch2_replicas_set_usage(struct bch_fs *c,
622 struct bch_replicas_entry_v1 *r,
625 int ret, idx = bch2_replicas_entry_idx(c, r);
628 struct bch_replicas_cpu n;
630 n = cpu_replicas_add_entry(c, &c->replicas, r);
632 return -BCH_ERR_ENOMEM_cpu_replicas;
634 ret = replicas_table_update(c, &n);
640 idx = bch2_replicas_entry_idx(c, r);
644 c->usage_base->replicas[idx] = sectors;
649 /* Replicas tracking - superblock: */
652 __bch2_sb_replicas_to_cpu_replicas(struct bch_sb_field_replicas *sb_r,
653 struct bch_replicas_cpu *cpu_r)
655 struct bch_replicas_entry_v1 *e, *dst;
656 unsigned nr = 0, entry_size = 0, idx = 0;
658 for_each_replicas_entry(sb_r, e) {
659 entry_size = max_t(unsigned, entry_size,
660 replicas_entry_bytes(e));
664 cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
666 return -BCH_ERR_ENOMEM_cpu_replicas;
669 cpu_r->entry_size = entry_size;
671 for_each_replicas_entry(sb_r, e) {
672 dst = cpu_replicas_entry(cpu_r, idx++);
673 memcpy(dst, e, replicas_entry_bytes(e));
674 bch2_replicas_entry_sort(dst);
681 __bch2_sb_replicas_v0_to_cpu_replicas(struct bch_sb_field_replicas_v0 *sb_r,
682 struct bch_replicas_cpu *cpu_r)
684 struct bch_replicas_entry_v0 *e;
685 unsigned nr = 0, entry_size = 0, idx = 0;
687 for_each_replicas_entry(sb_r, e) {
688 entry_size = max_t(unsigned, entry_size,
689 replicas_entry_bytes(e));
693 entry_size += sizeof(struct bch_replicas_entry_v1) -
694 sizeof(struct bch_replicas_entry_v0);
696 cpu_r->entries = kcalloc(nr, entry_size, GFP_KERNEL);
698 return -BCH_ERR_ENOMEM_cpu_replicas;
701 cpu_r->entry_size = entry_size;
703 for_each_replicas_entry(sb_r, e) {
704 struct bch_replicas_entry_v1 *dst =
705 cpu_replicas_entry(cpu_r, idx++);
707 dst->data_type = e->data_type;
708 dst->nr_devs = e->nr_devs;
709 dst->nr_required = 1;
710 memcpy(dst->devs, e->devs, e->nr_devs);
711 bch2_replicas_entry_sort(dst);
717 int bch2_sb_replicas_to_cpu_replicas(struct bch_fs *c)
719 struct bch_sb_field_replicas *sb_v1;
720 struct bch_sb_field_replicas_v0 *sb_v0;
721 struct bch_replicas_cpu new_r = { 0, 0, NULL };
724 if ((sb_v1 = bch2_sb_field_get(c->disk_sb.sb, replicas)))
725 ret = __bch2_sb_replicas_to_cpu_replicas(sb_v1, &new_r);
726 else if ((sb_v0 = bch2_sb_field_get(c->disk_sb.sb, replicas_v0)))
727 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_v0, &new_r);
731 bch2_cpu_replicas_sort(&new_r);
733 percpu_down_write(&c->mark_lock);
735 ret = replicas_table_update(c, &new_r);
736 percpu_up_write(&c->mark_lock);
738 kfree(new_r.entries);
743 static int bch2_cpu_replicas_to_sb_replicas_v0(struct bch_fs *c,
744 struct bch_replicas_cpu *r)
746 struct bch_sb_field_replicas_v0 *sb_r;
747 struct bch_replicas_entry_v0 *dst;
748 struct bch_replicas_entry_v1 *src;
751 bytes = sizeof(struct bch_sb_field_replicas);
753 for_each_cpu_replicas_entry(r, src)
754 bytes += replicas_entry_bytes(src) - 1;
756 sb_r = bch2_sb_field_resize(&c->disk_sb, replicas_v0,
757 DIV_ROUND_UP(bytes, sizeof(u64)));
759 return -BCH_ERR_ENOSPC_sb_replicas;
761 bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas);
762 sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas_v0);
764 memset(&sb_r->entries, 0,
765 vstruct_end(&sb_r->field) -
766 (void *) &sb_r->entries);
769 for_each_cpu_replicas_entry(r, src) {
770 dst->data_type = src->data_type;
771 dst->nr_devs = src->nr_devs;
772 memcpy(dst->devs, src->devs, src->nr_devs);
774 dst = replicas_entry_next(dst);
776 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
782 static int bch2_cpu_replicas_to_sb_replicas(struct bch_fs *c,
783 struct bch_replicas_cpu *r)
785 struct bch_sb_field_replicas *sb_r;
786 struct bch_replicas_entry_v1 *dst, *src;
787 bool need_v1 = false;
790 bytes = sizeof(struct bch_sb_field_replicas);
792 for_each_cpu_replicas_entry(r, src) {
793 bytes += replicas_entry_bytes(src);
794 if (src->nr_required != 1)
799 return bch2_cpu_replicas_to_sb_replicas_v0(c, r);
801 sb_r = bch2_sb_field_resize(&c->disk_sb, replicas,
802 DIV_ROUND_UP(bytes, sizeof(u64)));
804 return -BCH_ERR_ENOSPC_sb_replicas;
806 bch2_sb_field_delete(&c->disk_sb, BCH_SB_FIELD_replicas_v0);
807 sb_r = bch2_sb_field_get(c->disk_sb.sb, replicas);
809 memset(&sb_r->entries, 0,
810 vstruct_end(&sb_r->field) -
811 (void *) &sb_r->entries);
814 for_each_cpu_replicas_entry(r, src) {
815 memcpy(dst, src, replicas_entry_bytes(src));
817 dst = replicas_entry_next(dst);
819 BUG_ON((void *) dst > vstruct_end(&sb_r->field));
825 static int bch2_cpu_replicas_validate(struct bch_replicas_cpu *cpu_r,
827 struct printbuf *err)
831 sort_r(cpu_r->entries,
835 (void *)(size_t)cpu_r->entry_size);
837 for (i = 0; i < cpu_r->nr; i++) {
838 struct bch_replicas_entry_v1 *e =
839 cpu_replicas_entry(cpu_r, i);
841 int ret = bch2_replicas_entry_validate(e, sb, err);
845 if (i + 1 < cpu_r->nr) {
846 struct bch_replicas_entry_v1 *n =
847 cpu_replicas_entry(cpu_r, i + 1);
849 BUG_ON(memcmp(e, n, cpu_r->entry_size) > 0);
851 if (!memcmp(e, n, cpu_r->entry_size)) {
852 prt_printf(err, "duplicate replicas entry ");
853 bch2_replicas_entry_to_text(err, e);
854 return -BCH_ERR_invalid_sb_replicas;
862 static int bch2_sb_replicas_validate(struct bch_sb *sb, struct bch_sb_field *f,
863 struct printbuf *err)
865 struct bch_sb_field_replicas *sb_r = field_to_type(f, replicas);
866 struct bch_replicas_cpu cpu_r;
869 ret = __bch2_sb_replicas_to_cpu_replicas(sb_r, &cpu_r);
873 ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
874 kfree(cpu_r.entries);
878 static void bch2_sb_replicas_to_text(struct printbuf *out,
880 struct bch_sb_field *f)
882 struct bch_sb_field_replicas *r = field_to_type(f, replicas);
883 struct bch_replicas_entry_v1 *e;
886 for_each_replicas_entry(r, e) {
888 prt_printf(out, " ");
891 bch2_replicas_entry_to_text(out, e);
896 const struct bch_sb_field_ops bch_sb_field_ops_replicas = {
897 .validate = bch2_sb_replicas_validate,
898 .to_text = bch2_sb_replicas_to_text,
901 static int bch2_sb_replicas_v0_validate(struct bch_sb *sb, struct bch_sb_field *f,
902 struct printbuf *err)
904 struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
905 struct bch_replicas_cpu cpu_r;
908 ret = __bch2_sb_replicas_v0_to_cpu_replicas(sb_r, &cpu_r);
912 ret = bch2_cpu_replicas_validate(&cpu_r, sb, err);
913 kfree(cpu_r.entries);
917 static void bch2_sb_replicas_v0_to_text(struct printbuf *out,
919 struct bch_sb_field *f)
921 struct bch_sb_field_replicas_v0 *sb_r = field_to_type(f, replicas_v0);
922 struct bch_replicas_entry_v0 *e;
925 for_each_replicas_entry(sb_r, e) {
927 prt_printf(out, " ");
930 bch2_replicas_entry_v0_to_text(out, e);
935 const struct bch_sb_field_ops bch_sb_field_ops_replicas_v0 = {
936 .validate = bch2_sb_replicas_v0_validate,
937 .to_text = bch2_sb_replicas_v0_to_text,
940 /* Query replicas: */
942 bool bch2_have_enough_devs(struct bch_fs *c, struct bch_devs_mask devs,
943 unsigned flags, bool print)
945 struct bch_replicas_entry_v1 *e;
948 percpu_down_read(&c->mark_lock);
949 for_each_cpu_replicas_entry(&c->replicas, e) {
950 unsigned i, nr_online = 0, nr_failed = 0, dflags = 0;
951 bool metadata = e->data_type < BCH_DATA_user;
953 if (e->data_type == BCH_DATA_cached)
956 for (i = 0; i < e->nr_devs; i++) {
957 struct bch_dev *ca = bch_dev_bkey_exists(c, e->devs[i]);
959 nr_online += test_bit(e->devs[i], devs.d);
960 nr_failed += ca->mi.state == BCH_MEMBER_STATE_failed;
963 if (nr_failed == e->nr_devs)
966 if (nr_online < e->nr_required)
968 ? BCH_FORCE_IF_METADATA_LOST
969 : BCH_FORCE_IF_DATA_LOST;
971 if (nr_online < e->nr_devs)
973 ? BCH_FORCE_IF_METADATA_DEGRADED
974 : BCH_FORCE_IF_DATA_DEGRADED;
976 if (dflags & ~flags) {
978 struct printbuf buf = PRINTBUF;
980 bch2_replicas_entry_to_text(&buf, e);
981 bch_err(c, "insufficient devices online (%u) for replicas entry %s",
990 percpu_up_read(&c->mark_lock);
995 unsigned bch2_sb_dev_has_data(struct bch_sb *sb, unsigned dev)
997 struct bch_sb_field_replicas *replicas;
998 struct bch_sb_field_replicas_v0 *replicas_v0;
999 unsigned i, data_has = 0;
1001 replicas = bch2_sb_field_get(sb, replicas);
1002 replicas_v0 = bch2_sb_field_get(sb, replicas_v0);
1005 struct bch_replicas_entry_v1 *r;
1007 for_each_replicas_entry(replicas, r)
1008 for (i = 0; i < r->nr_devs; i++)
1009 if (r->devs[i] == dev)
1010 data_has |= 1 << r->data_type;
1011 } else if (replicas_v0) {
1012 struct bch_replicas_entry_v0 *r;
1014 for_each_replicas_entry_v0(replicas_v0, r)
1015 for (i = 0; i < r->nr_devs; i++)
1016 if (r->devs[i] == dev)
1017 data_has |= 1 << r->data_type;
1024 unsigned bch2_dev_has_data(struct bch_fs *c, struct bch_dev *ca)
1028 mutex_lock(&c->sb_lock);
1029 ret = bch2_sb_dev_has_data(c->disk_sb.sb, ca->dev_idx);
1030 mutex_unlock(&c->sb_lock);
1035 void bch2_fs_replicas_exit(struct bch_fs *c)
1039 kfree(c->usage_scratch);
1040 for (i = 0; i < ARRAY_SIZE(c->usage); i++)
1041 free_percpu(c->usage[i]);
1042 kfree(c->usage_base);
1043 kfree(c->replicas.entries);
1044 kfree(c->replicas_gc.entries);
1046 mempool_exit(&c->replicas_delta_pool);
1049 int bch2_fs_replicas_init(struct bch_fs *c)
1051 bch2_journal_entry_res_resize(&c->journal,
1052 &c->replicas_journal_res,
1053 reserve_journal_replicas(c, &c->replicas));
1055 return mempool_init_kmalloc_pool(&c->replicas_delta_pool, 1,
1056 REPLICAS_DELTA_LIST_MAX) ?:
1057 replicas_table_update(c, &c->replicas);