return bch2_rand_range(l1 + l2) > l1;
}
- if (force_reconstruct_read(c))
+ if (bch2_force_reconstruct_read)
return p1.idx > p2.idx;
return p1.idx < p2.idx;
!bch2_dev_is_readable(ca))
p.idx++;
- if (force_reconstruct_read(c) &&
+ if (bch2_force_reconstruct_read &&
!p.idx && p.has_ec)
p.idx++;
const char *bch2_btree_ptr_invalid(const struct bch_fs *c, struct bkey_s_c k)
{
- if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
+ if (bkey_val_u64s(k.k) > BCH_REPLICAS_MAX)
return "value too big";
return bch2_bkey_ptrs_invalid(c, k);
}
-void bch2_btree_ptr_debugcheck(struct bch_fs *c, struct bkey_s_c k)
+void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
+ struct bkey_s_c k)
{
- struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
- const struct bch_extent_ptr *ptr;
- const char *err;
- char buf[160];
- struct bucket_mark mark;
- struct bch_dev *ca;
-
- if (!test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
- return;
-
- if (!percpu_down_read_trylock(&c->mark_lock))
- return;
+ bch2_bkey_ptrs_to_text(out, c, k);
+}
- bkey_for_each_ptr(ptrs, ptr) {
- ca = bch_dev_bkey_exists(c, ptr->dev);
+const char *bch2_btree_ptr_v2_invalid(const struct bch_fs *c, struct bkey_s_c k)
+{
+ struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
- mark = ptr_bucket_mark(ca, ptr);
+ if (bkey_val_bytes(k.k) <= sizeof(*bp.v))
+ return "value too small";
- err = "stale";
- if (gen_after(mark.gen, ptr->gen))
- goto err;
+ if (bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
+ return "value too big";
- err = "inconsistent";
- if (mark.data_type != BCH_DATA_btree ||
- mark.dirty_sectors < c->opts.btree_node_size)
- goto err;
- }
-out:
- percpu_up_read(&c->mark_lock);
- return;
-err:
- bch2_fs_inconsistent(c, "%s btree pointer %s: bucket %zi gen %i mark %08x",
- err, (bch2_bkey_val_to_text(&PBUF(buf), c, k), buf),
- PTR_BUCKET_NR(ca, ptr),
- mark.gen, (unsigned) mark.v.counter);
- goto out;
-}
+ if (c->sb.version < bcachefs_metadata_version_snapshot &&
+ bp.v->min_key.snapshot)
+ return "invalid min_key.snapshot";
-void bch2_btree_ptr_to_text(struct printbuf *out, struct bch_fs *c,
- struct bkey_s_c k)
-{
- bch2_bkey_ptrs_to_text(out, c, k);
+ return bch2_bkey_ptrs_invalid(c, k);
}
void bch2_btree_ptr_v2_to_text(struct printbuf *out, struct bch_fs *c,
{
struct bkey_s_c_btree_ptr_v2 bp = bkey_s_c_to_btree_ptr_v2(k);
- pr_buf(out, "seq %llx sectors %u written %u min_key ",
+ pr_buf(out, "seq %llx written %u min_key %s",
le64_to_cpu(bp.v->seq),
- le16_to_cpu(bp.v->sectors),
- le16_to_cpu(bp.v->sectors_written));
+ le16_to_cpu(bp.v->sectors_written),
+ BTREE_PTR_RANGE_UPDATED(bp.v) ? "R " : "");
bch2_bpos_to_text(out, bp.v->min_key);
pr_buf(out, " ");
btree_node_type_is_extents(btree_id) &&
bkey_cmp(bp.v->min_key, POS_MIN))
bp.v->min_key = write
- ? bkey_predecessor(bp.v->min_key)
- : bkey_successor(bp.v->min_key);
+ ? bpos_nosnap_predecessor(bp.v->min_key)
+ : bpos_nosnap_successor(bp.v->min_key);
}
/* KEY_TYPE_extent: */
return bch2_bkey_ptrs_invalid(c, k);
}
-void bch2_extent_debugcheck(struct bch_fs *c, struct bkey_s_c k)
-{
- struct bkey_s_c_extent e = bkey_s_c_to_extent(k);
- const union bch_extent_entry *entry;
- struct extent_ptr_decoded p;
- char buf[160];
-
- if (!test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) ||
- !test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
- return;
-
- if (!percpu_down_read_trylock(&c->mark_lock))
- return;
-
- extent_for_each_ptr_decode(e, p, entry) {
- struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
- struct bucket_mark mark = ptr_bucket_mark(ca, &p.ptr);
- unsigned stale = gen_after(mark.gen, p.ptr.gen);
- unsigned disk_sectors = ptr_disk_sectors(p);
- unsigned mark_sectors = p.ptr.cached
- ? mark.cached_sectors
- : mark.dirty_sectors;
-
- bch2_fs_inconsistent_on(stale && !p.ptr.cached, c,
- "stale dirty pointer (ptr gen %u bucket %u",
- p.ptr.gen, mark.gen);
-
- bch2_fs_inconsistent_on(stale > 96, c,
- "key too stale: %i", stale);
-
- bch2_fs_inconsistent_on(!stale &&
- (mark.data_type != BCH_DATA_user ||
- mark_sectors < disk_sectors), c,
- "extent pointer not marked: %s:\n"
- "type %u sectors %u < %u",
- (bch2_bkey_val_to_text(&PBUF(buf), c, e.s_c), buf),
- mark.data_type,
- mark_sectors, disk_sectors);
- }
-
- percpu_up_read(&c->mark_lock);
-}
-
void bch2_extent_to_text(struct printbuf *out, struct bch_fs *c,
struct bkey_s_c k)
{
bch2_bkey_ptrs_to_text(out, c, k);
}
-enum merge_result bch2_extent_merge(struct bch_fs *c,
- struct bkey_s _l, struct bkey_s _r)
+bool bch2_extent_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
{
- struct bkey_s_extent l = bkey_s_to_extent(_l);
- struct bkey_s_extent r = bkey_s_to_extent(_r);
- union bch_extent_entry *en_l = l.v->start;
- union bch_extent_entry *en_r = r.v->start;
- struct bch_extent_crc_unpacked crc_l, crc_r;
-
- if (bkey_val_u64s(l.k) != bkey_val_u64s(r.k))
- return BCH_MERGE_NOMERGE;
-
- crc_l = bch2_extent_crc_unpack(l.k, NULL);
-
- extent_for_each_entry(l, en_l) {
- en_r = vstruct_idx(r.v, (u64 *) en_l - l.v->_data);
+ struct bkey_ptrs l_ptrs = bch2_bkey_ptrs(l);
+ struct bkey_ptrs_c r_ptrs = bch2_bkey_ptrs_c(r);
+ union bch_extent_entry *en_l;
+ const union bch_extent_entry *en_r;
+ struct extent_ptr_decoded lp, rp;
+ bool use_right_ptr;
+ struct bch_dev *ca;
+ en_l = l_ptrs.start;
+ en_r = r_ptrs.start;
+ while (en_l < l_ptrs.end && en_r < r_ptrs.end) {
if (extent_entry_type(en_l) != extent_entry_type(en_r))
- return BCH_MERGE_NOMERGE;
-
- switch (extent_entry_type(en_l)) {
- case BCH_EXTENT_ENTRY_ptr: {
- const struct bch_extent_ptr *lp = &en_l->ptr;
- const struct bch_extent_ptr *rp = &en_r->ptr;
- struct bch_dev *ca;
-
- if (lp->offset + crc_l.compressed_size != rp->offset ||
- lp->dev != rp->dev ||
- lp->gen != rp->gen)
- return BCH_MERGE_NOMERGE;
-
- /* We don't allow extents to straddle buckets: */
- ca = bch_dev_bkey_exists(c, lp->dev);
-
- if (PTR_BUCKET_NR(ca, lp) != PTR_BUCKET_NR(ca, rp))
- return BCH_MERGE_NOMERGE;
-
- break;
- }
- case BCH_EXTENT_ENTRY_stripe_ptr:
- if (en_l->stripe_ptr.block != en_r->stripe_ptr.block ||
- en_l->stripe_ptr.idx != en_r->stripe_ptr.idx)
- return BCH_MERGE_NOMERGE;
- break;
- case BCH_EXTENT_ENTRY_crc32:
- case BCH_EXTENT_ENTRY_crc64:
- case BCH_EXTENT_ENTRY_crc128:
- crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
- crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
-
- if (crc_l.csum_type != crc_r.csum_type ||
- crc_l.compression_type != crc_r.compression_type ||
- crc_l.nonce != crc_r.nonce)
- return BCH_MERGE_NOMERGE;
+ return false;
- if (crc_l.offset + crc_l.live_size != crc_l.compressed_size ||
- crc_r.offset)
- return BCH_MERGE_NOMERGE;
-
- if (!bch2_checksum_mergeable(crc_l.csum_type))
- return BCH_MERGE_NOMERGE;
-
- if (crc_is_compressed(crc_l))
- return BCH_MERGE_NOMERGE;
+ en_l = extent_entry_next(en_l);
+ en_r = extent_entry_next(en_r);
+ }
- if (crc_l.csum_type &&
- crc_l.uncompressed_size +
- crc_r.uncompressed_size > c->sb.encoded_extent_max)
- return BCH_MERGE_NOMERGE;
+ if (en_l < l_ptrs.end || en_r < r_ptrs.end)
+ return false;
- if (crc_l.uncompressed_size + crc_r.uncompressed_size >
+ en_l = l_ptrs.start;
+ en_r = r_ptrs.start;
+ lp.crc = bch2_extent_crc_unpack(l.k, NULL);
+ rp.crc = bch2_extent_crc_unpack(r.k, NULL);
+
+ while (__bkey_ptr_next_decode(l.k, l_ptrs.end, lp, en_l) &&
+ __bkey_ptr_next_decode(r.k, r_ptrs.end, rp, en_r)) {
+ if (lp.ptr.offset + lp.crc.offset + lp.crc.live_size !=
+ rp.ptr.offset + rp.crc.offset ||
+ lp.ptr.dev != rp.ptr.dev ||
+ lp.ptr.gen != rp.ptr.gen ||
+ lp.has_ec != rp.has_ec)
+ return false;
+
+ /* Extents may not straddle buckets: */
+ ca = bch_dev_bkey_exists(c, lp.ptr.dev);
+ if (PTR_BUCKET_NR(ca, &lp.ptr) != PTR_BUCKET_NR(ca, &rp.ptr))
+ return false;
+
+ if (lp.has_ec != rp.has_ec ||
+ (lp.has_ec &&
+ (lp.ec.block != rp.ec.block ||
+ lp.ec.redundancy != rp.ec.redundancy ||
+ lp.ec.idx != rp.ec.idx)))
+ return false;
+
+ if (lp.crc.compression_type != rp.crc.compression_type ||
+ lp.crc.nonce != rp.crc.nonce)
+ return false;
+
+ if (lp.crc.offset + lp.crc.live_size + rp.crc.live_size <=
+ lp.crc.uncompressed_size) {
+ /* can use left extent's crc entry */
+ } else if (lp.crc.live_size <= rp.crc.offset ) {
+ /* can use right extent's crc entry */
+ } else {
+ /* check if checksums can be merged: */
+ if (lp.crc.csum_type != rp.crc.csum_type ||
+ lp.crc.nonce != rp.crc.nonce ||
+ crc_is_compressed(lp.crc) ||
+ !bch2_checksum_mergeable(lp.crc.csum_type))
+ return false;
+
+ if (lp.crc.offset + lp.crc.live_size != lp.crc.compressed_size ||
+ rp.crc.offset)
+ return false;
+
+ if (lp.crc.csum_type &&
+ lp.crc.uncompressed_size +
+ rp.crc.uncompressed_size > c->sb.encoded_extent_max)
+ return false;
+
+ if (lp.crc.uncompressed_size + rp.crc.uncompressed_size >
bch2_crc_field_size_max[extent_entry_type(en_l)])
- return BCH_MERGE_NOMERGE;
-
- break;
- default:
- return BCH_MERGE_NOMERGE;
+ return false;
}
- }
-
- extent_for_each_entry(l, en_l) {
- struct bch_extent_crc_unpacked crc_l, crc_r;
-
- en_r = vstruct_idx(r.v, (u64 *) en_l - l.v->_data);
-
- if (!extent_entry_is_crc(en_l))
- continue;
- crc_l = bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
- crc_r = bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
-
- crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
- crc_l.csum,
- crc_r.csum,
- crc_r.uncompressed_size << 9);
+ en_l = extent_entry_next(en_l);
+ en_r = extent_entry_next(en_r);
+ }
- crc_l.uncompressed_size += crc_r.uncompressed_size;
- crc_l.compressed_size += crc_r.compressed_size;
+ use_right_ptr = false;
+ en_l = l_ptrs.start;
+ en_r = r_ptrs.start;
+ while (en_l < l_ptrs.end) {
+ if (extent_entry_type(en_l) == BCH_EXTENT_ENTRY_ptr &&
+ use_right_ptr)
+ en_l->ptr = en_r->ptr;
+
+ if (extent_entry_is_crc(en_l)) {
+ struct bch_extent_crc_unpacked crc_l =
+ bch2_extent_crc_unpack(l.k, entry_to_crc(en_l));
+ struct bch_extent_crc_unpacked crc_r =
+ bch2_extent_crc_unpack(r.k, entry_to_crc(en_r));
+
+ use_right_ptr = false;
+
+ if (crc_l.offset + crc_l.live_size + crc_r.live_size <=
+ crc_l.uncompressed_size) {
+ /* can use left extent's crc entry */
+ } else if (crc_l.live_size <= crc_r.offset ) {
+ /* can use right extent's crc entry */
+ crc_r.offset -= crc_l.live_size;
+ bch2_extent_crc_pack(entry_to_crc(en_l), crc_r,
+ extent_entry_type(en_l));
+ use_right_ptr = true;
+ } else {
+ crc_l.csum = bch2_checksum_merge(crc_l.csum_type,
+ crc_l.csum,
+ crc_r.csum,
+ crc_r.uncompressed_size << 9);
+
+ crc_l.uncompressed_size += crc_r.uncompressed_size;
+ crc_l.compressed_size += crc_r.compressed_size;
+ bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
+ extent_entry_type(en_l));
+ }
+ }
- bch2_extent_crc_pack(entry_to_crc(en_l), crc_l,
- extent_entry_type(en_l));
+ en_l = extent_entry_next(en_l);
+ en_r = extent_entry_next(en_r);
}
bch2_key_resize(l.k, l.k->size + r.k->size);
-
- return BCH_MERGE_MERGE;
+ return true;
}
/* KEY_TYPE_reservation: */
r.v->nr_replicas);
}
-enum merge_result bch2_reservation_merge(struct bch_fs *c,
- struct bkey_s _l, struct bkey_s _r)
+bool bch2_reservation_merge(struct bch_fs *c, struct bkey_s _l, struct bkey_s_c _r)
{
struct bkey_s_reservation l = bkey_s_to_reservation(_l);
- struct bkey_s_reservation r = bkey_s_to_reservation(_r);
+ struct bkey_s_c_reservation r = bkey_s_c_to_reservation(_r);
if (l.v->generation != r.v->generation ||
l.v->nr_replicas != r.v->nr_replicas)
- return BCH_MERGE_NOMERGE;
-
- if ((u64) l.k->size + r.k->size > KEY_SIZE_MAX) {
- bch2_key_resize(l.k, KEY_SIZE_MAX);
- bch2_cut_front_s(l.k->p, r.s);
- return BCH_MERGE_PARTIAL;
- }
+ return false;
bch2_key_resize(l.k, l.k->size + r.k->size);
-
- return BCH_MERGE_MERGE;
+ return true;
}
/* Extent checksum entries: */
}
bool bch2_check_range_allocated(struct bch_fs *c, struct bpos pos, u64 size,
- unsigned nr_replicas)
+ unsigned nr_replicas, bool compressed)
{
struct btree_trans trans;
- struct btree_iter *iter;
+ struct btree_iter iter;
struct bpos end = pos;
struct bkey_s_c k;
bool ret = true;
bch2_trans_init(&trans, c, 0, 0);
- for_each_btree_key(&trans, iter, BTREE_ID_EXTENTS, pos,
+ for_each_btree_key(&trans, iter, BTREE_ID_extents, pos,
BTREE_ITER_SLOTS, k, err) {
if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
break;
- if (nr_replicas > bch2_bkey_nr_ptrs_fully_allocated(k)) {
+ if (nr_replicas > bch2_bkey_replicas(c, k) ||
+ (!compressed && bch2_bkey_sectors_compressed(k))) {
ret = false;
break;
}
}
+ bch2_trans_iter_exit(&trans, &iter);
+
bch2_trans_exit(&trans);
return ret;
}
+unsigned bch2_bkey_replicas(struct bch_fs *c, struct bkey_s_c k)
+{
+ struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ const union bch_extent_entry *entry;
+ struct extent_ptr_decoded p = { 0 };
+ unsigned replicas = 0;
+
+ bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+ if (p.ptr.cached)
+ continue;
+
+ if (p.has_ec)
+ replicas += p.ec.redundancy;
+
+ replicas++;
+
+ }
+
+ return replicas;
+}
+
static unsigned bch2_extent_ptr_durability(struct bch_fs *c,
struct extent_ptr_decoded p)
{
ca = bch_dev_bkey_exists(c, p.ptr.dev);
- if (ca->mi.state != BCH_MEMBER_STATE_FAILED)
+ if (ca->mi.state != BCH_MEMBER_STATE_failed)
durability = max_t(unsigned, durability, ca->mi.durability);
- if (p.has_ec) {
- struct stripe *s =
- genradix_ptr(&c->stripes[0], p.ec.idx);
-
- if (WARN_ON(!s))
- goto out;
+ if (p.has_ec)
+ durability += p.ec.redundancy;
- durability += s->nr_redundant;
- }
-out:
return durability;
}
}
}
+void bch2_bkey_extent_entry_drop(struct bkey_i *k, union bch_extent_entry *entry)
+{
+ union bch_extent_entry *end = bkey_val_end(bkey_i_to_s(k));
+ union bch_extent_entry *next = extent_entry_next(entry);
+
+ memmove_u64s(entry, next, (u64 *) end - (u64 *) next);
+ k->k.u64s -= extent_entry_u64s(entry);
+}
+
void bch2_bkey_append_ptr(struct bkey_i *k,
struct bch_extent_ptr ptr)
{
/* will only happen if all pointers were cached: */
if (!bch2_bkey_nr_ptrs(k.s_c))
- k.k->type = KEY_TYPE_discard;
+ k.k->type = KEY_TYPE_deleted;
- return bkey_whiteout(k.k);
+ return bkey_deleted(k.k);
}
void bch2_bkey_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
const char *bch2_bkey_ptrs_invalid(const struct bch_fs *c, struct bkey_s_c k)
{
struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+ struct bch_devs_list devs;
const union bch_extent_entry *entry;
struct bch_extent_crc_unpacked crc;
unsigned size_ondisk = k.k->size;
const char *reason;
unsigned nonce = UINT_MAX;
+ unsigned i;
- if (k.k->type == KEY_TYPE_btree_ptr)
+ if (k.k->type == KEY_TYPE_btree_ptr ||
+ k.k->type == KEY_TYPE_btree_ptr_v2)
size_ondisk = c->opts.btree_node_size;
- if (k.k->type == KEY_TYPE_btree_ptr_v2)
- size_ondisk = le16_to_cpu(bkey_s_c_to_btree_ptr_v2(k).v->sectors);
bkey_extent_entry_for_each(ptrs, entry) {
if (__extent_entry_type(entry) >= BCH_EXTENT_ENTRY_MAX)
}
}
+ devs = bch2_bkey_devs(k);
+ bubble_sort(devs.devs, devs.nr, u8_cmp);
+ for (i = 0; i + 1 < devs.nr; i++)
+ if (devs.devs[i] == devs.devs[i + 1])
+ return "multiple ptrs to same device";
+
return NULL;
}
le64_add_cpu(&p.v->idx, sub);
break;
}
- case KEY_TYPE_inline_data: {
- struct bkey_s_inline_data d = bkey_s_to_inline_data(k);
+ case KEY_TYPE_inline_data:
+ case KEY_TYPE_indirect_inline_data: {
+ void *p = bkey_inline_data_p(k);
+ unsigned bytes = bkey_inline_data_bytes(k.k);
- sub = min_t(u64, sub << 9, bkey_val_bytes(d.k));
+ sub = min_t(u64, sub << 9, bytes);
- memmove(d.v->data,
- d.v->data + sub,
- bkey_val_bytes(d.k) - sub);
+ memmove(p, p + sub, bytes - sub);
new_val_u64s -= sub >> 3;
break;
len = where.offset - bkey_start_offset(k.k);
- k.k->p = where;
+ k.k->p.offset = where.offset;
k.k->size = len;
if (!len) {
switch (k.k->type) {
case KEY_TYPE_inline_data:
- new_val_u64s = min(new_val_u64s, k.k->size << 6);
+ case KEY_TYPE_indirect_inline_data:
+ new_val_u64s = (bkey_inline_data_offset(k.k) +
+ min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
break;
}
projects / bcachefs-tools-debian / blobdiff