1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com>
5 * Code for managing the extent btree and dynamically updating the writeback
10 #include "bkey_methods.h"
12 #include "btree_update.h"
13 #include "btree_update_interior.h"
18 #include "disk_groups.h"
30 unsigned bch2_bkey_nr_ptrs(struct bkey_s_c k)
32 struct bkey_ptrs_c p = bch2_bkey_ptrs_c(k);
33 const struct bch_extent_ptr *ptr;
36 bkey_for_each_ptr(p, ptr)
42 unsigned bch2_bkey_nr_dirty_ptrs(struct bkey_s_c k)
47 case KEY_TYPE_btree_ptr:
48 case KEY_TYPE_extent: {
49 struct bkey_ptrs_c p = bch2_bkey_ptrs_c(k);
50 const struct bch_extent_ptr *ptr;
52 bkey_for_each_ptr(p, ptr)
53 nr_ptrs += !ptr->cached;
57 case KEY_TYPE_reservation:
58 nr_ptrs = bkey_s_c_to_reservation(k).v->nr_replicas;
65 static unsigned bch2_extent_ptr_durability(struct bch_fs *c,
66 struct extent_ptr_decoded p)
68 unsigned i, durability = 0;
74 ca = bch_dev_bkey_exists(c, p.ptr.dev);
76 if (ca->mi.state != BCH_MEMBER_STATE_FAILED)
77 durability = max_t(unsigned, durability, ca->mi.durability);
79 for (i = 0; i < p.ec_nr; i++) {
81 genradix_ptr(&c->stripes[0], p.idx);
86 durability = max_t(unsigned, durability, s->nr_redundant);
92 unsigned bch2_bkey_durability(struct bch_fs *c, struct bkey_s_c k)
94 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
95 const union bch_extent_entry *entry;
96 struct extent_ptr_decoded p;
97 unsigned durability = 0;
99 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
100 durability += bch2_extent_ptr_durability(c, p);
105 static struct bch_dev_io_failures *dev_io_failures(struct bch_io_failures *f,
108 struct bch_dev_io_failures *i;
110 for (i = f->devs; i < f->devs + f->nr; i++)
117 void bch2_mark_io_failure(struct bch_io_failures *failed,
118 struct extent_ptr_decoded *p)
120 struct bch_dev_io_failures *f = dev_io_failures(failed, p->ptr.dev);
123 BUG_ON(failed->nr >= ARRAY_SIZE(failed->devs));
125 f = &failed->devs[failed->nr++];
130 } else if (p->idx != f->idx) {
140 * returns true if p1 is better than p2:
142 static inline bool ptr_better(struct bch_fs *c,
143 const struct extent_ptr_decoded p1,
144 const struct extent_ptr_decoded p2)
146 if (likely(!p1.idx && !p2.idx)) {
147 struct bch_dev *dev1 = bch_dev_bkey_exists(c, p1.ptr.dev);
148 struct bch_dev *dev2 = bch_dev_bkey_exists(c, p2.ptr.dev);
150 u64 l1 = atomic64_read(&dev1->cur_latency[READ]);
151 u64 l2 = atomic64_read(&dev2->cur_latency[READ]);
153 /* Pick at random, biased in favor of the faster device: */
155 return bch2_rand_range(l1 + l2) > l1;
158 if (force_reconstruct_read(c))
159 return p1.idx > p2.idx;
161 return p1.idx < p2.idx;
165 * This picks a non-stale pointer, preferably from a device other than @avoid.
166 * Avoid can be NULL, meaning pick any. If there are no non-stale pointers to
167 * other devices, it will still pick a pointer from avoid.
169 int bch2_bkey_pick_read_device(struct bch_fs *c, struct bkey_s_c k,
170 struct bch_io_failures *failed,
171 struct extent_ptr_decoded *pick)
173 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
174 const union bch_extent_entry *entry;
175 struct extent_ptr_decoded p;
176 struct bch_dev_io_failures *f;
180 if (k.k->type == KEY_TYPE_error)
183 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
184 ca = bch_dev_bkey_exists(c, p.ptr.dev);
187 * If there are any dirty pointers it's an error if we can't
190 if (!ret && !p.ptr.cached)
193 if (p.ptr.cached && ptr_stale(ca, &p.ptr))
196 f = failed ? dev_io_failures(failed, p.ptr.dev) : NULL;
198 p.idx = f->nr_failed < f->nr_retries
203 !bch2_dev_is_readable(ca))
206 if (force_reconstruct_read(c) &&
210 if (p.idx >= p.ec_nr + 1)
213 if (ret > 0 && !ptr_better(c, p, *pick))
223 void bch2_bkey_append_ptr(struct bkey_i *k,
224 struct bch_extent_ptr ptr)
226 EBUG_ON(bch2_bkey_has_device(bkey_i_to_s_c(k), ptr.dev));
229 case KEY_TYPE_btree_ptr:
230 case KEY_TYPE_extent:
231 EBUG_ON(bkey_val_u64s(&k->k) >= BKEY_EXTENT_VAL_U64s_MAX);
233 ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
235 memcpy((void *) &k->v + bkey_val_bytes(&k->k),
245 void bch2_bkey_drop_device(struct bkey_s k, unsigned dev)
247 struct bch_extent_ptr *ptr;
249 bch2_bkey_drop_ptrs(k, ptr, ptr->dev == dev);
252 /* extent specific utility code */
254 const struct bch_extent_ptr *
255 bch2_extent_has_device(struct bkey_s_c_extent e, unsigned dev)
257 const struct bch_extent_ptr *ptr;
259 extent_for_each_ptr(e, ptr)
266 const struct bch_extent_ptr *
267 bch2_extent_has_group(struct bch_fs *c, struct bkey_s_c_extent e, unsigned group)
269 const struct bch_extent_ptr *ptr;
271 extent_for_each_ptr(e, ptr) {
272 struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
275 ca->mi.group - 1 == group)
282 const struct bch_extent_ptr *
283 bch2_extent_has_target(struct bch_fs *c, struct bkey_s_c_extent e, unsigned target)
285 const struct bch_extent_ptr *ptr;
287 extent_for_each_ptr(e, ptr)
288 if (bch2_dev_in_target(c, ptr->dev, target) &&
290 !ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr)))
296 unsigned bch2_extent_is_compressed(struct bkey_s_c k)
301 case KEY_TYPE_extent: {
302 struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
303 const union bch_extent_entry *entry;
304 struct extent_ptr_decoded p;
306 extent_for_each_ptr_decode(e, p, entry)
308 p.crc.compression_type != BCH_COMPRESSION_NONE)
309 ret += p.crc.compressed_size;
316 bool bch2_extent_matches_ptr(struct bch_fs *c, struct bkey_s_c_extent e,
317 struct bch_extent_ptr m, u64 offset)
319 const union bch_extent_entry *entry;
320 struct extent_ptr_decoded p;
322 extent_for_each_ptr_decode(e, p, entry)
323 if (p.ptr.dev == m.dev &&
324 p.ptr.gen == m.gen &&
325 (s64) p.ptr.offset + p.crc.offset - bkey_start_offset(e.k) ==
326 (s64) m.offset - offset)
332 static union bch_extent_entry *extent_entry_prev(struct bkey_ptrs ptrs,
333 union bch_extent_entry *entry)
335 union bch_extent_entry *i = ptrs.start;
340 while (extent_entry_next(i) != entry)
341 i = extent_entry_next(i);
345 union bch_extent_entry *bch2_bkey_drop_ptr(struct bkey_s k,
346 struct bch_extent_ptr *ptr)
348 struct bkey_ptrs ptrs = bch2_bkey_ptrs(k);
349 union bch_extent_entry *dst, *src, *prev;
350 bool drop_crc = true;
352 EBUG_ON(ptr < &ptrs.start->ptr ||
353 ptr >= &ptrs.end->ptr);
354 EBUG_ON(ptr->type != 1 << BCH_EXTENT_ENTRY_ptr);
356 src = extent_entry_next(to_entry(ptr));
357 if (src != ptrs.end &&
358 !extent_entry_is_crc(src))
362 while ((prev = extent_entry_prev(ptrs, dst))) {
363 if (extent_entry_is_ptr(prev))
366 if (extent_entry_is_crc(prev)) {
375 memmove_u64s_down(dst, src,
376 (u64 *) ptrs.end - (u64 *) src);
377 k.k->u64s -= (u64 *) src - (u64 *) dst;
382 static inline bool can_narrow_crc(struct bch_extent_crc_unpacked u,
383 struct bch_extent_crc_unpacked n)
385 return !u.compression_type &&
387 u.uncompressed_size > u.live_size &&
388 bch2_csum_type_is_encryption(u.csum_type) ==
389 bch2_csum_type_is_encryption(n.csum_type);
392 bool bch2_can_narrow_extent_crcs(struct bkey_s_c_extent e,
393 struct bch_extent_crc_unpacked n)
395 struct bch_extent_crc_unpacked crc;
396 const union bch_extent_entry *i;
401 extent_for_each_crc(e, crc, i)
402 if (can_narrow_crc(crc, n))
409 * We're writing another replica for this extent, so while we've got the data in
410 * memory we'll be computing a new checksum for the currently live data.
412 * If there are other replicas we aren't moving, and they are checksummed but
413 * not compressed, we can modify them to point to only the data that is
414 * currently live (so that readers won't have to bounce) while we've got the
417 bool bch2_extent_narrow_crcs(struct bkey_i_extent *e,
418 struct bch_extent_crc_unpacked n)
420 struct bch_extent_crc_unpacked u;
421 struct extent_ptr_decoded p;
422 union bch_extent_entry *i;
425 /* Find a checksum entry that covers only live data: */
427 extent_for_each_crc(extent_i_to_s(e), u, i)
428 if (!u.compression_type &&
430 u.live_size == u.uncompressed_size) {
437 BUG_ON(n.compression_type);
439 BUG_ON(n.live_size != e->k.size);
441 restart_narrow_pointers:
442 extent_for_each_ptr_decode(extent_i_to_s(e), p, i)
443 if (can_narrow_crc(p.crc, n)) {
444 bch2_bkey_drop_ptr(extent_i_to_s(e).s, &i->ptr);
445 p.ptr.offset += p.crc.offset;
447 bch2_extent_ptr_decoded_append(e, &p);
449 goto restart_narrow_pointers;
455 /* returns true if not equal */
456 static inline bool bch2_crc_unpacked_cmp(struct bch_extent_crc_unpacked l,
457 struct bch_extent_crc_unpacked r)
459 return (l.csum_type != r.csum_type ||
460 l.compression_type != r.compression_type ||
461 l.compressed_size != r.compressed_size ||
462 l.uncompressed_size != r.uncompressed_size ||
463 l.offset != r.offset ||
464 l.live_size != r.live_size ||
465 l.nonce != r.nonce ||
466 bch2_crc_cmp(l.csum, r.csum));
469 void bch2_ptr_swab(const struct bkey_format *f, struct bkey_packed *k)
471 union bch_extent_entry *entry;
472 u64 *d = (u64 *) bkeyp_val(f, k);
475 for (i = 0; i < bkeyp_val_u64s(f, k); i++)
478 for (entry = (union bch_extent_entry *) d;
479 entry < (union bch_extent_entry *) (d + bkeyp_val_u64s(f, k));
480 entry = extent_entry_next(entry)) {
481 switch (extent_entry_type(entry)) {
482 case BCH_EXTENT_ENTRY_ptr:
484 case BCH_EXTENT_ENTRY_crc32:
485 entry->crc32.csum = swab32(entry->crc32.csum);
487 case BCH_EXTENT_ENTRY_crc64:
488 entry->crc64.csum_hi = swab16(entry->crc64.csum_hi);
489 entry->crc64.csum_lo = swab64(entry->crc64.csum_lo);
491 case BCH_EXTENT_ENTRY_crc128:
492 entry->crc128.csum.hi = (__force __le64)
493 swab64((__force u64) entry->crc128.csum.hi);
494 entry->crc128.csum.lo = (__force __le64)
495 swab64((__force u64) entry->crc128.csum.lo);
497 case BCH_EXTENT_ENTRY_stripe_ptr:
503 static const char *extent_ptr_invalid(const struct bch_fs *c,
505 const struct bch_extent_ptr *ptr,
506 unsigned size_ondisk,
509 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
510 const struct bch_extent_ptr *ptr2;
513 if (ptr->dev >= c->sb.nr_devices ||
515 return "pointer to invalid device";
517 ca = bch_dev_bkey_exists(c, ptr->dev);
519 return "pointer to invalid device";
521 bkey_for_each_ptr(ptrs, ptr2)
522 if (ptr != ptr2 && ptr->dev == ptr2->dev)
523 return "multiple pointers to same device";
525 if (ptr->offset + size_ondisk > bucket_to_sector(ca, ca->mi.nbuckets))
526 return "offset past end of device";
528 if (ptr->offset < bucket_to_sector(ca, ca->mi.first_bucket))
529 return "offset before first bucket";
531 if (bucket_remainder(ca, ptr->offset) +
532 size_ondisk > ca->mi.bucket_size)
533 return "spans multiple buckets";
538 static void bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
541 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
542 const union bch_extent_entry *entry;
543 struct bch_extent_crc_unpacked crc;
544 const struct bch_extent_ptr *ptr;
545 const struct bch_extent_stripe_ptr *ec;
549 bkey_extent_entry_for_each(ptrs, entry) {
553 switch (__extent_entry_type(entry)) {
554 case BCH_EXTENT_ENTRY_ptr:
555 ptr = entry_to_ptr(entry);
556 ca = ptr->dev < c->sb.nr_devices && c->devs[ptr->dev]
557 ? bch_dev_bkey_exists(c, ptr->dev)
560 pr_buf(out, "ptr: %u:%llu gen %u%s%s", ptr->dev,
561 (u64) ptr->offset, ptr->gen,
562 ptr->cached ? " cached" : "",
563 ca && ptr_stale(ca, ptr)
566 case BCH_EXTENT_ENTRY_crc32:
567 case BCH_EXTENT_ENTRY_crc64:
568 case BCH_EXTENT_ENTRY_crc128:
569 crc = bch2_extent_crc_unpack(k.k, entry_to_crc(entry));
571 pr_buf(out, "crc: c_size %u size %u offset %u nonce %u csum %u compress %u",
573 crc.uncompressed_size,
574 crc.offset, crc.nonce,
576 crc.compression_type);
578 case BCH_EXTENT_ENTRY_stripe_ptr:
579 ec = &entry->stripe_ptr;
581 pr_buf(out, "ec: idx %llu block %u",
582 (u64) ec->idx, ec->block);
585 pr_buf(out, "(invalid extent entry %.16llx)", *((u64 *) entry));
595 const char *bch2_btree_ptr_invalid(const struct bch_fs *c, struct bkey_s_c k)
597 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
598 const union bch_extent_entry *entry;
599 const struct bch_extent_ptr *ptr;
602 if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
603 return "value too big";
605 bkey_extent_entry_for_each(ptrs, entry) {
606 if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX)
607 return "invalid extent entry type";
609 if (!extent_entry_is_ptr(entry))
610 return "has non ptr field";
613 bkey_for_each_ptr(ptrs, ptr) {
614 reason = extent_ptr_invalid(c, k, ptr,
615 c->opts.btree_node_size,
624 void bch2_btree_ptr_debugcheck(struct bch_fs *c, struct btree *b,
627 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
628 const struct bch_extent_ptr *ptr;
632 struct bucket_mark mark;
634 unsigned replicas = 0;
637 bkey_for_each_ptr(ptrs, ptr) {
638 ca = bch_dev_bkey_exists(c, ptr->dev);
641 if (!test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags))
645 if (ptr_stale(ca, ptr))
649 seq = read_seqcount_begin(&c->gc_pos_lock);
650 mark = ptr_bucket_mark(ca, ptr);
652 bad = gc_pos_cmp(c->gc_pos, gc_pos_btree_node(b)) > 0 &&
653 (mark.data_type != BCH_DATA_BTREE ||
654 mark.dirty_sectors < c->opts.btree_node_size);
655 } while (read_seqcount_retry(&c->gc_pos_lock, seq));
657 err = "inconsistent";
662 if (!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
663 !bch2_bkey_replicas_marked(c, k, false)) {
664 bch2_bkey_val_to_text(&PBUF(buf), c, k);
666 "btree key bad (replicas not marked in superblock):\n%s",
673 bch2_bkey_val_to_text(&PBUF(buf), c, k);
674 bch2_fs_bug(c, "%s btree pointer %s: bucket %zi gen %i mark %08x",
675 err, buf, PTR_BUCKET_NR(ca, ptr),
676 mark.gen, (unsigned) mark.v.counter);
679 void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
684 bkey_ptrs_to_text(out, c, k);
686 invalid = bch2_btree_ptr_invalid(c, k);
688 pr_buf(out, " invalid: %s", invalid);
693 bool __bch2_cut_front(struct bpos where, struct bkey_s k)
697 if (bkey_cmp(where, bkey_start_pos(k.k)) <= 0)
700 EBUG_ON(bkey_cmp(where, k.k->p) > 0);
702 len = k.k->p.offset - where.offset;
704 BUG_ON(len > k.k->size);
707 * Don't readjust offset if the key size is now 0, because that could
708 * cause offset to point to the next bucket:
711 k.k->type = KEY_TYPE_deleted;
712 else if (bkey_extent_is_data(k.k)) {
713 struct bkey_s_extent e = bkey_s_to_extent(k);
714 union bch_extent_entry *entry;
715 bool seen_crc = false;
717 extent_for_each_entry(e, entry) {
718 switch (extent_entry_type(entry)) {
719 case BCH_EXTENT_ENTRY_ptr:
721 entry->ptr.offset += e.k->size - len;
723 case BCH_EXTENT_ENTRY_crc32:
724 entry->crc32.offset += e.k->size - len;
726 case BCH_EXTENT_ENTRY_crc64:
727 entry->crc64.offset += e.k->size - len;
729 case BCH_EXTENT_ENTRY_crc128:
730 entry->crc128.offset += e.k->size - len;
732 case BCH_EXTENT_ENTRY_stripe_ptr:
736 if (extent_entry_is_crc(entry))
746 bool bch2_cut_back(struct bpos where, struct bkey *k)
750 if (bkey_cmp(where, k->p) >= 0)
753 EBUG_ON(bkey_cmp(where, bkey_start_pos(k)) < 0);
755 len = where.offset - bkey_start_offset(k);
757 BUG_ON(len > k->size);
763 k->type = KEY_TYPE_deleted;
769 * bch_key_resize - adjust size of @k
771 * bkey_start_offset(k) will be preserved, modifies where the extent ends
773 void bch2_key_resize(struct bkey *k,
776 k->p.offset -= k->size;
777 k->p.offset += new_size;
781 static bool extent_i_save(struct btree *b, struct bkey_packed *dst,
784 struct bkey_format *f = &b->format;
785 struct bkey_i *dst_unpacked;
786 struct bkey_packed tmp;
788 if ((dst_unpacked = packed_to_bkey(dst)))
789 dst_unpacked->k = src->k;
790 else if (bch2_bkey_pack_key(&tmp, &src->k, f))
791 memcpy_u64s(dst, &tmp, f->key_u64s);
795 memcpy_u64s(bkeyp_val(f, dst), &src->v, bkey_val_u64s(&src->k));
799 struct extent_insert_state {
800 struct btree_insert *trans;
801 struct btree_insert_entry *insert;
802 struct bpos committed;
805 struct bkey_i whiteout;
811 static bool bch2_extent_merge_inline(struct bch_fs *,
813 struct bkey_packed *,
814 struct bkey_packed *,
817 static void verify_extent_nonoverlapping(struct btree *b,
818 struct btree_node_iter *_iter,
819 struct bkey_i *insert)
821 #ifdef CONFIG_BCACHEFS_DEBUG
822 struct btree_node_iter iter;
823 struct bkey_packed *k;
827 k = bch2_btree_node_iter_prev_filter(&iter, b, KEY_TYPE_discard);
829 (uk = bkey_unpack_key(b, k),
830 bkey_cmp(uk.p, bkey_start_pos(&insert->k)) > 0));
833 k = bch2_btree_node_iter_peek_filter(&iter, b, KEY_TYPE_discard);
836 (uk = bkey_unpack_key(b, k),
837 bkey_cmp(insert->k.p, bkey_start_pos(&uk))) > 0);
840 (uk = bkey_unpack_key(b, k),
841 bkey_cmp(insert->k.p, bkey_start_pos(&uk))) > 0) {
845 bch2_bkey_to_text(&PBUF(buf1), &insert->k);
846 bch2_bkey_to_text(&PBUF(buf2), &uk);
848 bch2_dump_btree_node(b);
849 panic("insert > next :\n"
859 static void verify_modified_extent(struct btree_iter *iter,
860 struct bkey_packed *k)
862 bch2_btree_iter_verify(iter, iter->l[0].b);
863 bch2_verify_insert_pos(iter->l[0].b, k, k, k->u64s);
866 static void extent_bset_insert(struct bch_fs *c, struct btree_iter *iter,
867 struct bkey_i *insert)
869 struct btree_iter_level *l = &iter->l[0];
870 struct btree_node_iter node_iter;
871 struct bkey_packed *k;
873 BUG_ON(insert->k.u64s > bch_btree_keys_u64s_remaining(c, l->b));
875 EBUG_ON(bkey_deleted(&insert->k) || !insert->k.size);
876 verify_extent_nonoverlapping(l->b, &l->iter, insert);
879 k = bch2_btree_node_iter_prev_filter(&node_iter, l->b, KEY_TYPE_discard);
880 if (k && !bkey_written(l->b, k) &&
881 bch2_extent_merge_inline(c, iter, k, bkey_to_packed(insert), true))
885 k = bch2_btree_node_iter_peek_filter(&node_iter, l->b, KEY_TYPE_discard);
886 if (k && !bkey_written(l->b, k) &&
887 bch2_extent_merge_inline(c, iter, bkey_to_packed(insert), k, false))
890 k = bch2_btree_node_iter_bset_pos(&l->iter, l->b, bset_tree_last(l->b));
892 bch2_bset_insert(l->b, &l->iter, k, insert, 0);
893 bch2_btree_node_iter_fix(iter, l->b, &l->iter, k, 0, k->u64s);
894 bch2_btree_iter_verify(iter, l->b);
897 static void extent_insert_committed(struct extent_insert_state *s)
899 struct bch_fs *c = s->trans->c;
900 struct btree_iter *iter = s->insert->iter;
901 struct bkey_i *insert = s->insert->k;
902 BKEY_PADDED(k) split;
904 EBUG_ON(bkey_cmp(insert->k.p, s->committed) < 0);
905 EBUG_ON(bkey_cmp(s->committed, bkey_start_pos(&insert->k)) < 0);
907 bkey_copy(&split.k, insert);
909 split.k.k.type = KEY_TYPE_discard;
911 bch2_cut_back(s->committed, &split.k.k);
913 if (!bkey_cmp(s->committed, iter->pos))
916 bch2_btree_iter_set_pos_same_leaf(iter, s->committed);
918 if (s->update_btree) {
919 if (debug_check_bkeys(c))
920 bch2_bkey_debugcheck(c, iter->l[0].b,
921 bkey_i_to_s_c(&split.k));
923 EBUG_ON(bkey_deleted(&split.k.k) || !split.k.k.size);
925 extent_bset_insert(c, iter, &split.k);
928 if (s->update_journal) {
929 bkey_copy(&split.k, !s->deleting ? insert : &s->whiteout);
931 split.k.k.type = KEY_TYPE_discard;
933 bch2_cut_back(s->committed, &split.k.k);
935 EBUG_ON(bkey_deleted(&split.k.k) || !split.k.k.size);
937 bch2_btree_journal_key(s->trans, iter, &split.k);
940 bch2_cut_front(s->committed, insert);
942 insert->k.needs_whiteout = false;
945 void bch2_extent_trim_atomic(struct bkey_i *k, struct btree_iter *iter)
947 struct btree *b = iter->l[0].b;
949 BUG_ON(iter->uptodate > BTREE_ITER_NEED_PEEK);
951 bch2_cut_back(b->key.k.p, &k->k);
953 BUG_ON(bkey_cmp(bkey_start_pos(&k->k), b->data->min_key) < 0);
956 enum btree_insert_ret
957 bch2_extent_can_insert(struct btree_insert *trans,
958 struct btree_insert_entry *insert,
961 struct btree_iter_level *l = &insert->iter->l[0];
962 struct btree_node_iter node_iter = l->iter;
963 enum bch_extent_overlap overlap;
964 struct bkey_packed *_k;
965 struct bkey unpacked;
969 BUG_ON(trans->flags & BTREE_INSERT_ATOMIC &&
970 !bch2_extent_is_atomic(&insert->k->k, insert->iter));
973 * We avoid creating whiteouts whenever possible when deleting, but
974 * those optimizations mean we may potentially insert two whiteouts
975 * instead of one (when we overlap with the front of one extent and the
978 if (bkey_whiteout(&insert->k->k))
981 _k = bch2_btree_node_iter_peek_filter(&node_iter, l->b,
984 return BTREE_INSERT_OK;
986 k = bkey_disassemble(l->b, _k, &unpacked);
988 overlap = bch2_extent_overlap(&insert->k->k, k.k);
990 /* account for having to split existing extent: */
991 if (overlap == BCH_EXTENT_OVERLAP_MIDDLE)
994 if (overlap == BCH_EXTENT_OVERLAP_MIDDLE &&
995 (sectors = bch2_extent_is_compressed(k))) {
996 int flags = BCH_DISK_RESERVATION_BTREE_LOCKS_HELD;
998 if (trans->flags & BTREE_INSERT_NOFAIL)
999 flags |= BCH_DISK_RESERVATION_NOFAIL;
1001 switch (bch2_disk_reservation_add(trans->c,
1007 return BTREE_INSERT_ENOSPC;
1009 return BTREE_INSERT_NEED_GC_LOCK;
1015 return BTREE_INSERT_OK;
1019 extent_squash(struct extent_insert_state *s, struct bkey_i *insert,
1020 struct bkey_packed *_k, struct bkey_s k,
1021 enum bch_extent_overlap overlap)
1023 struct bch_fs *c = s->trans->c;
1024 struct btree_iter *iter = s->insert->iter;
1025 struct btree_iter_level *l = &iter->l[0];
1028 case BCH_EXTENT_OVERLAP_FRONT:
1029 /* insert overlaps with start of k: */
1030 __bch2_cut_front(insert->k.p, k);
1031 BUG_ON(bkey_deleted(k.k));
1032 extent_save(l->b, _k, k.k);
1033 verify_modified_extent(iter, _k);
1036 case BCH_EXTENT_OVERLAP_BACK:
1037 /* insert overlaps with end of k: */
1038 bch2_cut_back(bkey_start_pos(&insert->k), k.k);
1039 BUG_ON(bkey_deleted(k.k));
1040 extent_save(l->b, _k, k.k);
1043 * As the auxiliary tree is indexed by the end of the
1044 * key and we've just changed the end, update the
1047 bch2_bset_fix_invalidated_key(l->b, _k);
1048 bch2_btree_node_iter_fix(iter, l->b, &l->iter,
1049 _k, _k->u64s, _k->u64s);
1050 verify_modified_extent(iter, _k);
1053 case BCH_EXTENT_OVERLAP_ALL: {
1054 /* The insert key completely covers k, invalidate k */
1055 if (!bkey_whiteout(k.k))
1056 btree_account_key_drop(l->b, _k);
1059 k.k->type = KEY_TYPE_deleted;
1061 if (_k >= btree_bset_last(l->b)->start) {
1062 unsigned u64s = _k->u64s;
1064 bch2_bset_delete(l->b, _k, _k->u64s);
1065 bch2_btree_node_iter_fix(iter, l->b, &l->iter,
1067 bch2_btree_iter_verify(iter, l->b);
1069 extent_save(l->b, _k, k.k);
1070 bch2_btree_node_iter_fix(iter, l->b, &l->iter,
1071 _k, _k->u64s, _k->u64s);
1072 verify_modified_extent(iter, _k);
1077 case BCH_EXTENT_OVERLAP_MIDDLE: {
1078 BKEY_PADDED(k) split;
1080 * The insert key falls 'in the middle' of k
1081 * The insert key splits k in 3:
1082 * - start only in k, preserve
1083 * - middle common section, invalidate in k
1084 * - end only in k, preserve
1086 * We update the old key to preserve the start,
1087 * insert will be the new common section,
1088 * we manually insert the end that we are preserving.
1090 * modify k _before_ doing the insert (which will move
1093 bkey_reassemble(&split.k, k.s_c);
1094 split.k.k.needs_whiteout |= bkey_written(l->b, _k);
1096 bch2_cut_back(bkey_start_pos(&insert->k), &split.k.k);
1097 BUG_ON(bkey_deleted(&split.k.k));
1099 __bch2_cut_front(insert->k.p, k);
1100 BUG_ON(bkey_deleted(k.k));
1101 extent_save(l->b, _k, k.k);
1102 verify_modified_extent(iter, _k);
1104 extent_bset_insert(c, iter, &split.k);
1110 static void __bch2_insert_fixup_extent(struct extent_insert_state *s)
1112 struct btree_iter *iter = s->insert->iter;
1113 struct btree_iter_level *l = &iter->l[0];
1114 struct bkey_packed *_k;
1115 struct bkey unpacked;
1116 struct bkey_i *insert = s->insert->k;
1118 while (bkey_cmp(s->committed, insert->k.p) < 0 &&
1119 (_k = bch2_btree_node_iter_peek_filter(&l->iter, l->b,
1120 KEY_TYPE_discard))) {
1121 struct bkey_s k = __bkey_disassemble(l->b, _k, &unpacked);
1122 enum bch_extent_overlap overlap = bch2_extent_overlap(&insert->k, k.k);
1124 EBUG_ON(bkey_cmp(iter->pos, k.k->p) >= 0);
1126 if (bkey_cmp(bkey_start_pos(k.k), insert->k.p) >= 0)
1129 s->committed = bpos_min(s->insert->k->k.p, k.k->p);
1131 if (!bkey_whiteout(k.k))
1132 s->update_journal = true;
1134 if (!s->update_journal) {
1135 bch2_cut_front(s->committed, insert);
1136 bch2_cut_front(s->committed, &s->whiteout);
1137 bch2_btree_iter_set_pos_same_leaf(iter, s->committed);
1142 * When deleting, if possible just do it by switching the type
1143 * of the key we're deleting, instead of creating and inserting
1148 !bkey_cmp(insert->k.p, k.k->p) &&
1149 !bkey_cmp(bkey_start_pos(&insert->k), bkey_start_pos(k.k))) {
1150 if (!bkey_whiteout(k.k)) {
1151 btree_account_key_drop(l->b, _k);
1152 _k->type = KEY_TYPE_discard;
1153 reserve_whiteout(l->b, _k);
1158 if (k.k->needs_whiteout || bkey_written(l->b, _k)) {
1159 insert->k.needs_whiteout = true;
1160 s->update_btree = true;
1163 if (s->update_btree &&
1164 overlap == BCH_EXTENT_OVERLAP_ALL &&
1165 bkey_whiteout(k.k) &&
1166 k.k->needs_whiteout) {
1167 unreserve_whiteout(l->b, _k);
1168 _k->needs_whiteout = false;
1171 extent_squash(s, insert, _k, k, overlap);
1173 if (!s->update_btree)
1174 bch2_cut_front(s->committed, insert);
1176 if (overlap == BCH_EXTENT_OVERLAP_FRONT ||
1177 overlap == BCH_EXTENT_OVERLAP_MIDDLE)
1181 if (bkey_cmp(s->committed, insert->k.p) < 0)
1182 s->committed = bpos_min(s->insert->k->k.p, l->b->key.k.p);
1185 * may have skipped past some deleted extents greater than the insert
1186 * key, before we got to a non deleted extent and knew we could bail out
1187 * rewind the iterator a bit if necessary:
1190 struct btree_node_iter node_iter = l->iter;
1192 while ((_k = bch2_btree_node_iter_prev_all(&node_iter, l->b)) &&
1193 bkey_cmp_left_packed(l->b, _k, &s->committed) > 0)
1194 l->iter = node_iter;
1199 * bch_extent_insert_fixup - insert a new extent and deal with overlaps
1201 * this may result in not actually doing the insert, or inserting some subset
1202 * of the insert key. For cmpxchg operations this is where that logic lives.
1204 * All subsets of @insert that need to be inserted are inserted using
1205 * bch2_btree_insert_and_journal(). If @b or @res fills up, this function
1206 * returns false, setting @iter->pos for the prefix of @insert that actually got
1209 * BSET INVARIANTS: this function is responsible for maintaining all the
1210 * invariants for bsets of extents in memory. things get really hairy with 0
1215 * bkey_start_pos(bkey_next(k)) >= k
1216 * or bkey_start_offset(bkey_next(k)) >= k->offset
1218 * i.e. strict ordering, no overlapping extents.
1220 * multiple bsets (i.e. full btree node):
1223 * k.size != 0 ∧ j.size != 0 →
1224 * ¬ (k > bkey_start_pos(j) ∧ k < j)
1226 * i.e. no two overlapping keys _of nonzero size_
1228 * We can't realistically maintain this invariant for zero size keys because of
1229 * the key merging done in bch2_btree_insert_key() - for two mergeable keys k, j
1230 * there may be another 0 size key between them in another bset, and it will
1231 * thus overlap with the merged key.
1233 * In addition, the end of iter->pos indicates how much has been processed.
1234 * If the end of iter->pos is not the same as the end of insert, then
1235 * key insertion needs to continue/be retried.
1237 enum btree_insert_ret
1238 bch2_insert_fixup_extent(struct btree_insert *trans,
1239 struct btree_insert_entry *insert)
1241 struct btree_iter *iter = insert->iter;
1242 struct btree *b = iter->l[0].b;
1243 struct extent_insert_state s = {
1246 .committed = iter->pos,
1248 .whiteout = *insert->k,
1249 .update_journal = !bkey_whiteout(&insert->k->k),
1250 .update_btree = !bkey_whiteout(&insert->k->k),
1251 .deleting = bkey_whiteout(&insert->k->k),
1254 EBUG_ON(iter->level);
1255 EBUG_ON(!insert->k->k.size);
1258 * As we process overlapping extents, we advance @iter->pos both to
1259 * signal to our caller (btree_insert_key()) how much of @insert->k has
1260 * been inserted, and also to keep @iter->pos consistent with
1261 * @insert->k and the node iterator that we're advancing:
1263 EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&insert->k->k)));
1265 __bch2_insert_fixup_extent(&s);
1267 extent_insert_committed(&s);
1269 EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&insert->k->k)));
1270 EBUG_ON(bkey_cmp(iter->pos, s.committed));
1272 if (insert->k->k.size) {
1273 /* got to the end of this leaf node */
1274 BUG_ON(bkey_cmp(iter->pos, b->key.k.p));
1275 return BTREE_INSERT_NEED_TRAVERSE;
1278 return BTREE_INSERT_OK;
1281 const char *bch2_extent_invalid(const struct bch_fs *c, struct bkey_s_c k)
1283 struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
1284 const union bch_extent_entry *entry;
1285 struct bch_extent_crc_unpacked crc;
1286 const struct bch_extent_ptr *ptr;
1287 unsigned size_ondisk = e.k->size;
1289 unsigned nonce = UINT_MAX;
1291 if (bkey_val_u64s(e.k) > BKEY_EXTENT_VAL_U64s_MAX)
1292 return "value too big";
1294 extent_for_each_entry(e, entry) {
1295 if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX)
1296 return "invalid extent entry type";
1298 switch (extent_entry_type(entry)) {
1299 case BCH_EXTENT_ENTRY_ptr:
1300 ptr = entry_to_ptr(entry);
1302 reason = extent_ptr_invalid(c, e.s_c, &entry->ptr,
1303 size_ondisk, false);
1307 case BCH_EXTENT_ENTRY_crc32:
1308 case BCH_EXTENT_ENTRY_crc64:
1309 case BCH_EXTENT_ENTRY_crc128:
1310 crc = bch2_extent_crc_unpack(e.k, entry_to_crc(entry));
1312 if (crc.offset + e.k->size >
1313 crc.uncompressed_size)
1314 return "checksum offset + key size > uncompressed size";
1316 size_ondisk = crc.compressed_size;
1318 if (!bch2_checksum_type_valid(c, crc.csum_type))
1319 return "invalid checksum type";
1321 if (crc.compression_type >= BCH_COMPRESSION_NR)
1322 return "invalid compression type";
1324 if (bch2_csum_type_is_encryption(crc.csum_type)) {
1325 if (nonce == UINT_MAX)
1326 nonce = crc.offset + crc.nonce;
1327 else if (nonce != crc.offset + crc.nonce)
1328 return "incorrect nonce";
1331 case BCH_EXTENT_ENTRY_stripe_ptr:
1339 void bch2_extent_debugcheck(struct bch_fs *c, struct btree *b,
1342 struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
1343 const struct bch_extent_ptr *ptr;
1345 struct bucket_mark mark;
1346 unsigned seq, stale;
1349 unsigned replicas = 0;
1352 * XXX: we should be doing most/all of these checks at startup time,
1353 * where we check bch2_bkey_invalid() in btree_node_read_done()
1355 * But note that we can't check for stale pointers or incorrect gc marks
1356 * until after journal replay is done (it might be an extent that's
1357 * going to get overwritten during replay)
1360 extent_for_each_ptr(e, ptr) {
1361 ca = bch_dev_bkey_exists(c, ptr->dev);
1365 * If journal replay hasn't finished, we might be seeing keys
1366 * that will be overwritten by the time journal replay is done:
1368 if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags))
1374 seq = read_seqcount_begin(&c->gc_pos_lock);
1375 mark = ptr_bucket_mark(ca, ptr);
1377 /* between mark and bucket gen */
1380 stale = ptr_stale(ca, ptr);
1382 bch2_fs_bug_on(stale && !ptr->cached, c,
1383 "stale dirty pointer");
1385 bch2_fs_bug_on(stale > 96, c,
1386 "key too stale: %i",
1392 bad = gc_pos_cmp(c->gc_pos, gc_pos_btree_node(b)) > 0 &&
1393 (mark.data_type != BCH_DATA_USER ||
1395 ? mark.cached_sectors
1396 : mark.dirty_sectors));
1397 } while (read_seqcount_retry(&c->gc_pos_lock, seq));
1403 if (replicas > BCH_REPLICAS_MAX) {
1404 bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c);
1406 "extent key bad (too many replicas: %u): %s",
1411 if (!test_bit(BCH_FS_REBUILD_REPLICAS, &c->flags) &&
1412 !bch2_bkey_replicas_marked(c, e.s_c, false)) {
1413 bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c);
1415 "extent key bad (replicas not marked in superblock):\n%s",
1423 bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c);
1424 bch2_fs_bug(c, "extent pointer bad gc mark: %s:\nbucket %zu "
1425 "gen %i type %u", buf,
1426 PTR_BUCKET_NR(ca, ptr), mark.gen, mark.data_type);
1429 void bch2_extent_to_text(struct printbuf *out, struct bch_fs *c,
1432 const char *invalid;
1434 bkey_ptrs_to_text(out, c, k);
1436 invalid = bch2_extent_invalid(c, k);
1438 pr_buf(out, " invalid: %s", invalid);
1441 static void bch2_extent_crc_init(union bch_extent_crc *crc,
1442 struct bch_extent_crc_unpacked new)
1444 #define common_fields(_crc) \
1445 .csum_type = _crc.csum_type, \
1446 .compression_type = _crc.compression_type, \
1447 ._compressed_size = _crc.compressed_size - 1, \
1448 ._uncompressed_size = _crc.uncompressed_size - 1, \
1449 .offset = _crc.offset
1451 if (bch_crc_bytes[new.csum_type] <= 4 &&
1452 new.uncompressed_size <= CRC32_SIZE_MAX &&
1453 new.nonce <= CRC32_NONCE_MAX) {
1454 crc->crc32 = (struct bch_extent_crc32) {
1455 .type = 1 << BCH_EXTENT_ENTRY_crc32,
1457 .csum = *((__le32 *) &new.csum.lo),
1462 if (bch_crc_bytes[new.csum_type] <= 10 &&
1463 new.uncompressed_size <= CRC64_SIZE_MAX &&
1464 new.nonce <= CRC64_NONCE_MAX) {
1465 crc->crc64 = (struct bch_extent_crc64) {
1466 .type = 1 << BCH_EXTENT_ENTRY_crc64,
1469 .csum_lo = new.csum.lo,
1470 .csum_hi = *((__le16 *) &new.csum.hi),
1475 if (bch_crc_bytes[new.csum_type] <= 16 &&
1476 new.uncompressed_size <= CRC128_SIZE_MAX &&
1477 new.nonce <= CRC128_NONCE_MAX) {
1478 crc->crc128 = (struct bch_extent_crc128) {
1479 .type = 1 << BCH_EXTENT_ENTRY_crc128,
1486 #undef common_fields
1490 void bch2_extent_crc_append(struct bkey_i_extent *e,
1491 struct bch_extent_crc_unpacked new)
1493 bch2_extent_crc_init((void *) extent_entry_last(extent_i_to_s(e)), new);
1494 __extent_entry_push(e);
1497 static inline void __extent_entry_insert(struct bkey_i_extent *e,
1498 union bch_extent_entry *dst,
1499 union bch_extent_entry *new)
1501 union bch_extent_entry *end = extent_entry_last(extent_i_to_s(e));
1503 memmove_u64s_up((u64 *) dst + extent_entry_u64s(new),
1504 dst, (u64 *) end - (u64 *) dst);
1505 e->k.u64s += extent_entry_u64s(new);
1506 memcpy_u64s_small(dst, new, extent_entry_u64s(new));
1509 void bch2_extent_ptr_decoded_append(struct bkey_i_extent *e,
1510 struct extent_ptr_decoded *p)
1512 struct bch_extent_crc_unpacked crc = bch2_extent_crc_unpack(&e->k, NULL);
1513 union bch_extent_entry *pos;
1516 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
1521 extent_for_each_crc(extent_i_to_s(e), crc, pos)
1522 if (!bch2_crc_unpacked_cmp(crc, p->crc)) {
1523 pos = extent_entry_next(pos);
1527 bch2_extent_crc_append(e, p->crc);
1528 pos = extent_entry_last(extent_i_to_s(e));
1530 p->ptr.type = 1 << BCH_EXTENT_ENTRY_ptr;
1531 __extent_entry_insert(e, pos, to_entry(&p->ptr));
1533 for (i = 0; i < p->ec_nr; i++) {
1534 p->ec[i].type = 1 << BCH_EXTENT_ENTRY_stripe_ptr;
1535 __extent_entry_insert(e, pos, to_entry(&p->ec[i]));
1540 * bch_extent_normalize - clean up an extent, dropping stale pointers etc.
1542 * Returns true if @k should be dropped entirely
1544 * For existing keys, only called when btree nodes are being rewritten, not when
1545 * they're merely being compacted/resorted in memory.
1547 bool bch2_extent_normalize(struct bch_fs *c, struct bkey_s k)
1549 struct bch_extent_ptr *ptr;
1551 bch2_bkey_drop_ptrs(k, ptr,
1553 ptr_stale(bch_dev_bkey_exists(c, ptr->dev), ptr));
1555 /* will only happen if all pointers were cached: */
1556 if (!bkey_val_u64s(k.k))
1557 k.k->type = KEY_TYPE_deleted;
1562 void bch2_extent_mark_replicas_cached(struct bch_fs *c,
1563 struct bkey_s_extent e,
1565 unsigned nr_desired_replicas)
1567 union bch_extent_entry *entry;
1568 struct extent_ptr_decoded p;
1569 int extra = bch2_bkey_durability(c, e.s_c) - nr_desired_replicas;
1571 if (target && extra > 0)
1572 extent_for_each_ptr_decode(e, p, entry) {
1573 int n = bch2_extent_ptr_durability(c, p);
1575 if (n && n <= extra &&
1576 !bch2_dev_in_target(c, p.ptr.dev, target)) {
1577 entry->ptr.cached = true;
1583 extent_for_each_ptr_decode(e, p, entry) {
1584 int n = bch2_extent_ptr_durability(c, p);
1586 if (n && n <= extra) {
1587 entry->ptr.cached = true;
1593 enum merge_result bch2_extent_merge(struct bch_fs *c,
1594 struct bkey_i *l, struct bkey_i *r)
1596 struct bkey_s_extent el = bkey_i_to_s_extent(l);
1597 struct bkey_s_extent er = bkey_i_to_s_extent(r);
1598 union bch_extent_entry *en_l, *en_r;
1600 if (bkey_val_u64s(&l->k) != bkey_val_u64s(&r->k))
1601 return BCH_MERGE_NOMERGE;
1603 extent_for_each_entry(el, en_l) {
1604 struct bch_extent_ptr *lp, *rp;
1607 en_r = vstruct_idx(er.v, (u64 *) en_l - el.v->_data);
1609 if ((extent_entry_type(en_l) !=
1610 extent_entry_type(en_r)) ||
1611 !extent_entry_is_ptr(en_l))
1612 return BCH_MERGE_NOMERGE;
1617 if (lp->offset + el.k->size != rp->offset ||
1618 lp->dev != rp->dev ||
1620 return BCH_MERGE_NOMERGE;
1622 /* We don't allow extents to straddle buckets: */
1623 ca = bch_dev_bkey_exists(c, lp->dev);
1625 if (PTR_BUCKET_NR(ca, lp) != PTR_BUCKET_NR(ca, rp))
1626 return BCH_MERGE_NOMERGE;
1629 l->k.needs_whiteout |= r->k.needs_whiteout;
1631 /* Keys with no pointers aren't restricted to one bucket and could
1634 if ((u64) l->k.size + r->k.size > KEY_SIZE_MAX) {
1635 bch2_key_resize(&l->k, KEY_SIZE_MAX);
1636 bch2_cut_front(l->k.p, r);
1637 return BCH_MERGE_PARTIAL;
1640 bch2_key_resize(&l->k, l->k.size + r->k.size);
1642 return BCH_MERGE_MERGE;
1646 * When merging an extent that we're inserting into a btree node, the new merged
1647 * extent could overlap with an existing 0 size extent - if we don't fix that,
1648 * it'll break the btree node iterator so this code finds those 0 size extents
1649 * and shifts them out of the way.
1651 * Also unpacks and repacks.
1653 static bool bch2_extent_merge_inline(struct bch_fs *c,
1654 struct btree_iter *iter,
1655 struct bkey_packed *l,
1656 struct bkey_packed *r,
1659 struct btree *b = iter->l[0].b;
1660 struct btree_node_iter *node_iter = &iter->l[0].iter;
1661 BKEY_PADDED(k) li, ri;
1662 struct bkey_packed *m = back_merge ? l : r;
1663 struct bkey_i *mi = back_merge ? &li.k : &ri.k;
1664 struct bset_tree *t = bch2_bkey_to_bset(b, m);
1665 enum merge_result ret;
1667 EBUG_ON(bkey_written(b, m));
1670 * We need to save copies of both l and r, because we might get a
1671 * partial merge (which modifies both) and then fails to repack
1673 bch2_bkey_unpack(b, &li.k, l);
1674 bch2_bkey_unpack(b, &ri.k, r);
1676 ret = bch2_bkey_merge(c, &li.k, &ri.k);
1677 if (ret == BCH_MERGE_NOMERGE)
1681 * check if we overlap with deleted extents - would break the sort
1685 struct bkey_packed *n = bkey_next(m);
1687 if (n != btree_bkey_last(b, t) &&
1688 bkey_cmp_left_packed(b, n, &li.k.k.p) <= 0 &&
1691 } else if (ret == BCH_MERGE_MERGE) {
1692 struct bkey_packed *prev = bch2_bkey_prev_all(b, t, m);
1695 bkey_cmp_left_packed_byval(b, prev,
1696 bkey_start_pos(&li.k.k)) > 0)
1700 if (ret == BCH_MERGE_PARTIAL) {
1701 if (!extent_i_save(b, m, mi))
1705 bkey_copy(packed_to_bkey(l), &li.k);
1707 bkey_copy(packed_to_bkey(r), &ri.k);
1709 if (!extent_i_save(b, m, &li.k))
1713 bch2_bset_fix_invalidated_key(b, m);
1714 bch2_btree_node_iter_fix(iter, b, node_iter,
1715 m, m->u64s, m->u64s);
1716 verify_modified_extent(iter, m);
1718 return ret == BCH_MERGE_MERGE;
1721 int bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size)
1723 struct btree_iter iter;
1724 struct bpos end = pos;
1730 for_each_btree_key(&iter, c, BTREE_ID_EXTENTS, pos,
1731 BTREE_ITER_SLOTS, k) {
1732 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
1735 if (!bch2_extent_is_fully_allocated(k)) {
1740 bch2_btree_iter_unlock(&iter);
1745 /* KEY_TYPE_reservation: */
1747 const char *bch2_reservation_invalid(const struct bch_fs *c, struct bkey_s_c k)
1749 struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
1751 if (bkey_val_bytes(k.k) != sizeof(struct bch_reservation))
1752 return "incorrect value size";
1754 if (!r.v->nr_replicas || r.v->nr_replicas > BCH_REPLICAS_MAX)
1755 return "invalid nr_replicas";
1760 void bch2_reservation_to_text(struct printbuf *out, struct bch_fs *c,
1763 struct bkey_s_c_reservation r = bkey_s_c_to_reservation(k);
1765 pr_buf(out, "generation %u replicas %u",
1766 le32_to_cpu(r.v->generation),
1770 enum merge_result bch2_reservation_merge(struct bch_fs *c,
1771 struct bkey_i *l, struct bkey_i *r)
1773 struct bkey_i_reservation *li = bkey_i_to_reservation(l);
1774 struct bkey_i_reservation *ri = bkey_i_to_reservation(r);
1776 if (li->v.generation != ri->v.generation ||
1777 li->v.nr_replicas != ri->v.nr_replicas)
1778 return BCH_MERGE_NOMERGE;
1780 l->k.needs_whiteout |= r->k.needs_whiteout;
1782 /* Keys with no pointers aren't restricted to one bucket and could
1785 if ((u64) l->k.size + r->k.size > KEY_SIZE_MAX) {
1786 bch2_key_resize(&l->k, KEY_SIZE_MAX);
1787 bch2_cut_front(l->k.p, r);
1788 return BCH_MERGE_PARTIAL;
1791 bch2_key_resize(&l->k, l->k.size + r->k.size);
1793 return BCH_MERGE_MERGE;