thin-provisioning-tools/btree.tcc

781 lines
19 KiB
Plaintext
Raw Normal View History

2011-12-16 00:04:31 +05:30
// Copyright (C) 2011 Red Hat, Inc. All rights reserved.
2011-12-06 19:23:05 +05:30
//
2011-12-06 19:13:56 +05:30
// This file is part of the thin-provisioning-tools source.
//
// thin-provisioning-tools is free software: you can redistribute it
// and/or modify it under the terms of the GNU General Public License
// as published by the Free Software Foundation, either version 3 of
// the License, or (at your option) any later version.
//
// thin-provisioning-tools is distributed in the hope that it will be
// useful, but WITHOUT ANY WARRANTY; without even the implied warranty
// of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License along
// with thin-provisioning-tools. If not, see
// <http://www.gnu.org/licenses/>.
#include "btree.h"
#include "checksum.h"
2011-06-27 15:15:30 +05:30
#include "transaction_manager.h"
2011-07-15 19:51:28 +05:30
#include <iostream>
2011-07-15 19:51:28 +05:30
using namespace btree_detail;
using namespace persistent_data;
2011-07-15 19:51:28 +05:30
using namespace std;
//----------------------------------------------------------------
namespace {
struct btree_node_validator : public block_manager<>::validator {
virtual void check(block_manager<>::const_buffer &b, block_address location) const {
disk_node const *data = reinterpret_cast<disk_node const *>(&b);
node_header const *n = &data->header;
crc32c sum(BTREE_CSUM_XOR);
sum.append(&n->flags, MD_BLOCK_SIZE - sizeof(uint32_t));
if (sum.get_sum() != to_cpu<uint32_t>(n->csum))
throw runtime_error("bad checksum in btree node");
if (to_cpu<uint64_t>(n->blocknr) != location)
throw runtime_error("bad block nr in btree node");
}
virtual void prepare(block_manager<>::buffer &b, block_address location) const {
disk_node *data = reinterpret_cast<disk_node *>(&b);
node_header *n = &data->header;
n->blocknr = to_disk<base::__le64, uint64_t>(location);
crc32c sum(BTREE_CSUM_XOR);
sum.append(&n->flags, MD_BLOCK_SIZE - sizeof(uint32_t));
n->csum = to_disk<base::__le32>(sum.get_sum());
}
};
block_manager<>::validator::ptr
btree_validator() {
return block_manager<>::validator::ptr(new btree_node_validator);
}
}
//----------------------------------------------------------------
inline void
ro_spine::step(block_address b)
{
spine_.push_back(tm_->read_lock(b, btree_validator()));
if (spine_.size() > 2)
spine_.pop_front();
}
inline bool
shadow_spine::step(block_address b)
{
pair<write_ref, bool> p = tm_->shadow(b, btree_validator());
try {
step(p.first);
} catch (...) {
tm_->get_sm()->dec(p.first.get_location());
throw;
}
return p.second;
}
//----------------------------------------------------------------
template <typename ValueTraits>
node_ref<ValueTraits>::node_ref(block_address location, disk_node *raw)
2011-08-22 18:39:37 +05:30
: location_(location),
raw_(raw)
2011-07-15 19:51:28 +05:30
{
}
2011-09-16 14:36:37 +05:30
template <typename ValueTraits>
uint32_t
node_ref<ValueTraits>::get_checksum() const
{
return to_cpu<uint32_t>(raw_->header.csum);
}
template <typename ValueTraits>
2011-08-23 16:25:37 +05:30
block_address
node_ref<ValueTraits>::get_block_nr() const
2011-08-23 16:25:37 +05:30
{
return to_cpu<uint64_t>(raw_->header.blocknr);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
btree_detail::node_type
node_ref<ValueTraits>::get_type() const
2011-07-15 19:51:28 +05:30
{
uint32_t flags = to_cpu<uint32_t>(raw_->header.flags);
2011-11-21 17:47:16 +05:30
if (flags & INTERNAL_NODE) {
if (flags & LEAF_NODE)
throw runtime_error("btree node is both internal and leaf");
2011-07-15 19:51:28 +05:30
return INTERNAL;
2011-11-21 17:47:16 +05:30
} else if (flags & LEAF_NODE)
2011-07-15 19:51:28 +05:30
return LEAF;
else
2011-09-02 19:35:26 +05:30
throw runtime_error("unknown node type");
2011-07-15 19:51:28 +05:30
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::set_type(node_type t)
2011-07-15 19:51:28 +05:30
{
uint32_t flags = to_cpu<uint32_t>(raw_->header.flags);
switch (t) {
case INTERNAL:
2011-11-21 17:47:16 +05:30
flags = INTERNAL_NODE;
2011-07-15 19:51:28 +05:30
break;
case LEAF:
2011-11-21 17:47:16 +05:30
flags = LEAF_NODE;
2011-07-15 19:51:28 +05:30
break;
}
raw_->header.flags = to_disk<__le32>(flags);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
unsigned
node_ref<ValueTraits>::get_nr_entries() const
2011-07-15 19:51:28 +05:30
{
return to_cpu<uint32_t>(raw_->header.nr_entries);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::set_nr_entries(unsigned n)
2011-07-15 19:51:28 +05:30
{
raw_->header.nr_entries = to_disk<__le32>(n);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
unsigned
node_ref<ValueTraits>::get_max_entries() const
2011-07-15 19:51:28 +05:30
{
return to_cpu<uint32_t>(raw_->header.max_entries);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::set_max_entries(unsigned n)
2011-07-15 19:51:28 +05:30
{
raw_->header.max_entries = to_disk<__le32>(n);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::set_max_entries()
2011-07-15 19:51:28 +05:30
{
set_max_entries(calc_max_entries());
}
template <typename ValueTraits>
2011-08-23 16:25:37 +05:30
size_t
node_ref<ValueTraits>::get_value_size() const
2011-08-23 16:25:37 +05:30
{
return to_cpu<uint32_t>(raw_->header.value_size);
}
2011-11-01 17:01:03 +05:30
template <typename ValueTraits>
void
node_ref<ValueTraits>::set_value_size(size_t s)
{
raw_->header.value_size = to_disk<__le32>(static_cast<uint32_t>(s));
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
uint64_t
node_ref<ValueTraits>::key_at(unsigned i) const
2011-07-15 19:51:28 +05:30
{
2011-08-22 18:39:37 +05:30
if (i >= get_nr_entries())
throw runtime_error("key index out of bounds");
2011-07-15 19:51:28 +05:30
return to_cpu<uint64_t>(raw_->keys[i]);
}
2011-06-27 15:15:30 +05:30
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::set_key(unsigned i, uint64_t k)
2011-07-15 19:51:28 +05:30
{
raw_->keys[i] = to_disk<__le64>(k);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
typename ValueTraits::value_type
node_ref<ValueTraits>::value_at(unsigned i) const
2011-07-15 19:51:28 +05:30
{
2011-08-22 18:39:37 +05:30
if (i >= get_nr_entries())
throw runtime_error("value index out of bounds");
2011-07-15 19:51:28 +05:30
// We have to copy because of alignment issues.
typename ValueTraits::disk_type d;
::memcpy(&d, value_ptr(i), sizeof(d));
2011-07-15 19:51:28 +05:30
typename ValueTraits::value_type v;
ValueTraits::unpack(d, v);
return v;
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::set_value(unsigned i,
typename ValueTraits::value_type const &v)
2011-07-15 19:51:28 +05:30
{
typename ValueTraits::disk_type d;
ValueTraits::pack(v, d);
::memcpy(value_ptr(i), &d, sizeof(d));
}
2011-06-27 15:15:30 +05:30
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::insert_at(unsigned i,
uint64_t key,
typename ValueTraits::value_type const &v)
2011-07-15 19:51:28 +05:30
{
unsigned n = get_nr_entries();
if ((n + 1) > get_max_entries())
throw runtime_error("too many entries");
set_nr_entries(n + 1);
2011-08-22 18:39:37 +05:30
::memmove(key_ptr(i + 1), key_ptr(i), sizeof(uint64_t) * (n - i));
::memmove(value_ptr(i + 1), value_ptr(i), sizeof(typename ValueTraits::disk_type) * (n - i));
2011-07-15 19:51:28 +05:30
overwrite_at(i, key, v);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::overwrite_at(unsigned i,
uint64_t key,
typename ValueTraits::value_type const &v)
2011-07-15 19:51:28 +05:30
{
set_key(i, key);
set_value(i, v);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void
node_ref<ValueTraits>::copy_entries(node_ref const &rhs,
unsigned begin,
unsigned end)
2011-07-15 19:51:28 +05:30
{
unsigned count = end - begin;
unsigned n = get_nr_entries();
if ((n + count) > get_max_entries())
throw runtime_error("too many entries");
2011-07-22 20:39:56 +05:30
::memcpy(key_ptr(n), rhs.key_ptr(begin), sizeof(uint64_t) * count);
::memcpy(value_ptr(n), rhs.value_ptr(begin), sizeof(typename ValueTraits::disk_type) * count);
2011-11-01 17:01:03 +05:30
set_nr_entries(n + count);
2011-07-15 19:51:28 +05:30
}
2011-06-27 15:15:30 +05:30
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
int
node_ref<ValueTraits>::bsearch(uint64_t key, int want_hi) const
2011-07-15 19:51:28 +05:30
{
int lo = -1, hi = get_nr_entries();
2011-06-27 15:15:30 +05:30
2011-07-15 19:51:28 +05:30
while(hi - lo > 1) {
int mid = lo + ((hi - lo) / 2);
uint64_t mid_key = key_at(mid);
2011-06-27 15:15:30 +05:30
2011-07-15 19:51:28 +05:30
if (mid_key == key)
return mid;
2011-06-27 15:15:30 +05:30
2011-07-15 19:51:28 +05:30
if (mid_key < key)
lo = mid;
else
hi = mid;
}
2011-06-27 15:15:30 +05:30
2011-07-15 19:51:28 +05:30
return want_hi ? hi : lo;
}
2011-06-27 15:15:30 +05:30
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
optional<unsigned>
node_ref<ValueTraits>::exact_search(uint64_t key) const
2011-07-15 19:51:28 +05:30
{
int i = bsearch(key, 0);
if (i < 0 || static_cast<unsigned>(i) >= get_nr_entries())
return optional<unsigned>();
2011-11-01 17:01:03 +05:30
if (key != key_at(i))
return optional<unsigned>();
2011-07-15 19:51:28 +05:30
return optional<unsigned>(i);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
int
node_ref<ValueTraits>::lower_bound(uint64_t key) const
2011-07-15 19:51:28 +05:30
{
return bsearch(key, 0);
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
unsigned
node_ref<ValueTraits>::calc_max_entries(void)
2011-07-15 19:51:28 +05:30
{
uint32_t total;
2011-07-15 19:51:28 +05:30
// key + value
size_t elt_size = sizeof(uint64_t) + sizeof(typename ValueTraits::disk_type);
total = (MD_BLOCK_SIZE - sizeof(struct node_header)) / elt_size;
2011-07-15 19:51:28 +05:30
return (total / 3) * 3; // rounds down
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void *
node_ref<ValueTraits>::key_ptr(unsigned i) const
2011-07-15 19:51:28 +05:30
{
return raw_->keys + i;
}
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
void *
node_ref<ValueTraits>::value_ptr(unsigned i) const
2011-07-15 19:51:28 +05:30
{
void *value_base = &raw_->keys[to_cpu<uint32_t>(raw_->header.max_entries)];
return static_cast<unsigned char *>(value_base) +
sizeof(typename ValueTraits::disk_type) * i;
}
template <typename ValueTraits>
2011-07-22 20:39:56 +05:30
template <typename RefCounter>
void
node_ref<ValueTraits>::inc_children(RefCounter &rc)
2011-07-22 20:39:56 +05:30
{
unsigned nr_entries = get_nr_entries();
for (unsigned i = 0; i < nr_entries; i++) {
typename ValueTraits::value_type v;
typename ValueTraits::disk_type d;
::memcpy(&d, value_ptr(i), sizeof(d));
ValueTraits::unpack(d, v);
rc.inc(v);
}
}
2011-07-15 19:51:28 +05:30
//----------------------------------------------------------------
template <unsigned Levels, typename ValueTraits>
btree<Levels, ValueTraits>::
btree(typename transaction_manager::ptr tm,
2011-07-22 20:39:56 +05:30
typename ValueTraits::ref_counter rc)
: tm_(tm),
2011-07-22 20:39:56 +05:30
destroy_(false),
rc_(rc)
{
2011-06-27 15:15:30 +05:30
using namespace btree_detail;
write_ref root = tm_->new_block(btree_validator());
leaf_node n = to_node<ValueTraits>(root);
2011-07-15 19:51:28 +05:30
n.set_type(btree_detail::LEAF);
n.set_nr_entries(0);
2011-07-15 19:51:28 +05:30
n.set_max_entries();
2011-11-01 17:01:03 +05:30
n.set_value_size(sizeof(typename ValueTraits::disk_type));
2011-06-27 15:15:30 +05:30
root_ = root.get_location();
}
template <unsigned Levels, typename ValueTraits>
btree<Levels, ValueTraits>::
btree(typename transaction_manager::ptr tm,
2011-07-22 20:39:56 +05:30
block_address root,
typename ValueTraits::ref_counter rc)
: tm_(tm),
destroy_(false),
2011-07-22 20:39:56 +05:30
root_(root),
rc_(rc)
{
}
template <unsigned Levels, typename ValueTraits>
btree<Levels, ValueTraits>::~btree()
{
}
2011-11-01 17:01:03 +05:30
namespace {
template <typename ValueTraits>
struct lower_bound_search {
static optional<unsigned> search(btree_detail::node_ref<ValueTraits> n, uint64_t key) {
return n.lower_bound(key);
}
};
template <typename ValueTraits>
struct exact_search {
static optional<unsigned> search(btree_detail::node_ref<ValueTraits> n, uint64_t key) {
return n.exact_search(key);
}
};
}
template <unsigned Levels, typename ValueTraits>
typename btree<Levels, ValueTraits>::maybe_value
btree<Levels, ValueTraits>::lookup(key const &key) const
{
2011-06-27 15:15:30 +05:30
using namespace btree_detail;
ro_spine spine(tm_);
block_address root = root_;
for (unsigned level = 0; level < Levels - 1; ++level) {
2011-06-27 15:15:30 +05:30
optional<block_address> mroot =
2011-11-01 17:01:03 +05:30
lookup_raw<uint64_traits, lower_bound_search<uint64_traits> >(spine, root, key[level]);
if (!mroot)
return maybe_value();
root = *mroot;
}
2011-11-01 17:01:03 +05:30
return lookup_raw<ValueTraits, exact_search<ValueTraits> >(spine, root, key[Levels - 1]);
}
template <unsigned Levels, typename ValueTraits>
typename btree<Levels, ValueTraits>::maybe_pair
btree<Levels, ValueTraits>::lookup_le(key const &key) const
{
2011-06-27 15:15:30 +05:30
using namespace btree_detail;
return maybe_pair();
}
template <unsigned Levels, typename ValueTraits>
typename btree<Levels, ValueTraits>::maybe_pair
btree<Levels, ValueTraits>::lookup_ge(key const &key) const
{
2011-06-27 15:15:30 +05:30
using namespace btree_detail;
return maybe_pair();
}
template <unsigned Levels, typename ValueTraits>
void
btree<Levels, ValueTraits>::
2011-07-15 19:51:28 +05:30
insert(key const &key,
typename ValueTraits::value_type const &value)
{
2011-06-27 15:15:30 +05:30
using namespace btree_detail;
2011-07-15 19:51:28 +05:30
block_address block = root_;
int index = 0; // FIXME: ???
shadow_spine spine(tm_);
2011-07-15 19:51:28 +05:30
for (unsigned level = 0; level < Levels - 1; ++level) {
bool need_insert = insert_location<uint64_traits>(spine, block, key[level], &index);
internal_node n = spine.template get_node<uint64_traits>();
2011-07-15 19:51:28 +05:30
if (need_insert) {
btree<Levels - 1, ValueTraits> new_tree(tm_, rc_);
2011-07-15 19:51:28 +05:30
n.insert_at(index, key[level], new_tree.get_root());
}
block = n.value_at(index);
}
bool need_insert = insert_location<ValueTraits>(spine, block, key[Levels - 1], &index);
leaf_node n = spine.template get_node<ValueTraits>();
2011-07-15 19:51:28 +05:30
if (need_insert)
n.insert_at(index, key[Levels - 1], value);
else
// FIXME: check if we're overwriting with the same value.
n.set_value(index, value);
}
template <unsigned Levels, typename ValueTraits>
void
btree<Levels, ValueTraits>::remove(key const &key)
{
2011-06-27 15:15:30 +05:30
using namespace btree_detail;
}
template <unsigned Levels, typename ValueTraits>
block_address
btree<Levels, ValueTraits>::get_root() const
{
2011-07-22 20:39:56 +05:30
return root_;
}
template <unsigned Levels, typename ValueTraits>
void
btree<Levels, ValueTraits>::set_root(block_address root)
{
2011-06-27 15:15:30 +05:30
using namespace btree_detail;
2011-07-22 20:39:56 +05:30
root_ = root;
}
template <unsigned Levels, typename ValueTraits>
typename btree<Levels, ValueTraits>::ptr
btree<Levels, ValueTraits>::clone() const
{
2011-08-31 18:10:52 +05:30
tm_->get_sm()->inc(root_);
return ptr(new btree<Levels, ValueTraits>(tm_, root_, rc_));
}
2011-07-22 20:39:56 +05:30
#if 0
template <unsigned Levels, typename ValueTraits>
void
btree<Levels, ValueTraits>::destroy()
{
2011-06-27 15:15:30 +05:30
using namespace btree_detail;
}
#endif
2011-07-15 19:51:28 +05:30
2011-11-01 17:01:03 +05:30
template <unsigned Levels, typename _>
template <typename ValueTraits, typename Search>
optional<typename ValueTraits::value_type>
btree<Levels, _>::
lookup_raw(ro_spine &spine, block_address block, uint64_t key) const
{
using namespace boost;
typedef typename ValueTraits::value_type leaf_type;
for (;;) {
spine.step(block);
node_ref<ValueTraits> leaf = spine.template get_node<ValueTraits>();
optional<unsigned> mi;
if (leaf.get_type() == btree_detail::LEAF) {
mi = Search::search(leaf, key);
if (!mi)
return optional<leaf_type>();
return optional<leaf_type>(leaf.value_at(*mi));
}
mi = leaf.lower_bound(key);
if (!mi || *mi < 0)
return optional<leaf_type>();
node_ref<uint64_traits> internal = spine.template get_node<uint64_traits>();
block = internal.value_at(*mi);
}
}
template <unsigned Levels, typename _>
2011-07-15 19:51:28 +05:30
template <typename ValueTraits>
void
btree<Levels, _>::
split_node(btree_detail::shadow_spine &spine,
2011-07-15 19:51:28 +05:30
block_address parent_index,
uint64_t key,
bool top)
{
node_ref<ValueTraits> n = spine.template get_node<ValueTraits>();
2011-07-15 19:51:28 +05:30
if (n.get_nr_entries() == n.get_max_entries()) {
if (top)
split_beneath<ValueTraits>(spine, key);
else
split_sibling<ValueTraits>(spine, parent_index, key);
}
}
template <unsigned Levels, typename _>
2011-07-15 19:51:28 +05:30
template <typename ValueTraits>
void
btree<Levels, _>::
split_beneath(btree_detail::shadow_spine &spine,
2011-07-15 19:51:28 +05:30
uint64_t key)
{
using namespace btree_detail;
node_type type;
unsigned nr_left, nr_right;
write_ref left = tm_->new_block(btree_validator());
node_ref<ValueTraits> l = to_node<ValueTraits>(left);
2011-07-15 19:51:28 +05:30
l.set_nr_entries(0);
l.set_max_entries();
2011-11-01 17:01:03 +05:30
l.set_value_size(sizeof(typename ValueTraits::disk_type));
2011-07-15 19:51:28 +05:30
write_ref right = tm_->new_block(btree_validator());
node_ref<ValueTraits> r = to_node<ValueTraits>(right);
2011-07-15 19:51:28 +05:30
r.set_nr_entries(0);
r.set_max_entries();
2011-11-01 17:01:03 +05:30
r.set_value_size(sizeof(typename ValueTraits::disk_type));
2011-07-15 19:51:28 +05:30
{
node_ref<ValueTraits> p = spine.template get_node<ValueTraits>();
2011-11-01 17:01:03 +05:30
if (p.get_value_size() != sizeof(typename ValueTraits::disk_type))
throw std::runtime_error("bad value_size");
2011-07-15 19:51:28 +05:30
nr_left = p.get_nr_entries() / 2;
nr_right = p.get_nr_entries() - nr_left;
type = p.get_type();
l.set_type(type);
l.copy_entries(p, 0, nr_left);
r.set_type(type);
r.copy_entries(p, nr_left, nr_left + nr_right);
}
{
// The parent may have changed value type, so we re-get it.
internal_node p = spine.template get_node<uint64_traits>();
2011-07-15 19:51:28 +05:30
p.set_type(btree_detail::INTERNAL);
2011-11-21 17:47:16 +05:30
p.set_max_entries();
2011-07-15 19:51:28 +05:30
p.set_nr_entries(2);
2011-11-01 17:01:03 +05:30
p.set_value_size(sizeof(typename uint64_traits::disk_type));
2011-07-15 19:51:28 +05:30
p.overwrite_at(0, l.key_at(0), left.get_location());
p.overwrite_at(1, r.key_at(0), right.get_location());
}
if (key < r.key_at(0))
spine.step(left);
else
spine.step(right);
}
template <unsigned Levels, typename _>
2011-07-15 19:51:28 +05:30
template <typename ValueTraits>
void
btree<Levels, _>::
split_sibling(btree_detail::shadow_spine &spine,
2011-07-22 20:39:56 +05:30
block_address parent_index,
uint64_t key)
2011-07-15 19:51:28 +05:30
{
using namespace btree_detail;
node_ref<ValueTraits> l = spine.template get_node<ValueTraits>();
block_address left = spine.get_block();
2011-07-15 19:51:28 +05:30
write_ref right = tm_->new_block(btree_validator());
node_ref<ValueTraits> r = to_node<ValueTraits>(right);
2011-07-15 19:51:28 +05:30
unsigned nr_left = l.get_nr_entries() / 2;
unsigned nr_right = l.get_nr_entries() - nr_left;
r.set_nr_entries(0);
2011-11-21 17:47:16 +05:30
r.set_max_entries();
2011-07-15 19:51:28 +05:30
r.set_type(l.get_type());
2011-11-21 17:47:16 +05:30
r.set_value_size(sizeof(typename ValueTraits::disk_type));
2011-07-15 19:51:28 +05:30
r.copy_entries(l, nr_left, nr_left + nr_right);
l.set_nr_entries(nr_left);
internal_node p = spine.get_parent();
2011-07-15 19:51:28 +05:30
p.overwrite_at(parent_index, l.key_at(0), left);
p.insert_at(parent_index + 1, r.key_at(0), right.get_location());
spine.pop();
if (key < r.key_at(0))
spine.step(left);
else
spine.step(right);
}
// Returns true if we need a new insertion, rather than overwrite.
template <unsigned Levels, typename _>
2011-07-15 19:51:28 +05:30
template <typename ValueTraits>
bool
btree<Levels, _>::
insert_location(btree_detail::shadow_spine &spine,
2011-07-15 19:51:28 +05:30
block_address block,
uint64_t key,
int *index)
{
using namespace btree_detail;
bool top = true; // this isn't the same as spine.has_parent()
int i = *index;
bool inc = false;
for (;;) {
inc = spine.step(block);
#if 0
if (inc)
inc_children<ValueTraits>();
#endif
// patch up the parent to point to the new shadow
if (spine.has_parent()) {
internal_node p = spine.get_parent();
2011-07-15 19:51:28 +05:30
p.set_value(i, spine.get_block());
}
internal_node internal = spine.template get_node<uint64_traits>();
2011-07-15 19:51:28 +05:30
// Split the node if we're full
if (internal.get_type() == INTERNAL)
split_node<uint64_traits>(spine, i, key, top);
else
split_node<ValueTraits>(spine, i, key, top);
internal = spine.template get_node<uint64_traits>();
i = internal.lower_bound(key);
if (internal.get_type() == btree_detail::LEAF)
break;
if (i < 0) {
internal.set_key(0, key);
i = 0;
}
block = internal.value_at(i);
top = false;
}
node_ref<ValueTraits> leaf = spine.template get_node<ValueTraits>();
2011-07-15 19:51:28 +05:30
// FIXME: gross
if (i < 0 || leaf.key_at(i) != key)
i++;
// do decrement the old value if it already exists
// FIXME: I'm not sure about this, I don't understand the |inc| reference
2011-08-22 18:39:37 +05:30
if (static_cast<unsigned>(i) < leaf.get_nr_entries() && leaf.key_at(i) == key && inc) {
2011-07-15 19:51:28 +05:30
// dec old entry
}
*index = i;
return ((static_cast<unsigned>(i) >= leaf.get_nr_entries()) ||
(leaf.key_at(i) != key));
}
template <unsigned Levels, typename ValueTraits>
void
btree<Levels, ValueTraits>::visit(typename visitor::ptr visitor) const
{
2011-10-10 18:40:30 +05:30
walk_tree(visitor, 0, true, boost::optional<uint64_t>(), root_);
visitor->visit_complete();
}
template <unsigned Levels, typename ValueTraits>
void
btree<Levels, ValueTraits>::
walk_tree(typename visitor::ptr visitor,
2011-10-10 18:40:30 +05:30
unsigned level, bool sub_root,
boost::optional<uint64_t> key,
2011-08-25 20:57:58 +05:30
block_address b) const
{
using namespace btree_detail;
2011-08-23 16:25:37 +05:30
read_ref blk = tm_->read_lock(b);
internal_node o = to_node<uint64_traits>(blk);
if (o.get_type() == INTERNAL) {
2011-10-10 18:40:30 +05:30
if (visitor->visit_internal(level, sub_root, key, o))
for (unsigned i = 0; i < o.get_nr_entries(); i++)
2011-10-10 18:40:30 +05:30
walk_tree(visitor, level, false, o.key_at(i), o.value_at(i));
} else if (level < Levels - 1) {
2011-10-10 18:40:30 +05:30
if (visitor->visit_internal_leaf(level, sub_root, key, o))
for (unsigned i = 0; i < o.get_nr_entries(); i++)
2011-10-10 18:40:30 +05:30
walk_tree(visitor, level + 1, true, boost::optional<uint64_t>(o.key_at(i)), o.value_at(i));
} else {
leaf_node ov = to_node<ValueTraits>(blk);
2011-10-10 18:40:30 +05:30
visitor->visit_leaf(level, sub_root, key, ov);
}
}
2011-07-15 19:51:28 +05:30
//----------------------------------------------------------------