thin-provisioning-tools/persistent-data/data-structures/btree.h

461 lines
12 KiB
C
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/>.
#ifndef BTREE_H
#define BTREE_H
#include "persistent-data/endian_utils.h"
#include "persistent-data/transaction_manager.h"
2013-03-13 18:28:05 +05:30
#include "persistent-data/data-structures/ref_counter.h"
2011-07-15 19:51:28 +05:30
#include <boost/noncopyable.hpp>
#include <boost/optional.hpp>
#include <list>
2013-05-15 18:07:30 +05:30
#include <deque>
2011-07-15 19:51:28 +05:30
//----------------------------------------------------------------
namespace persistent_data {
class block_ref_counter : public ref_counter<block_address> {
public:
block_ref_counter(space_map::ptr sm);
virtual void set(block_address const &v, uint32_t rc);
virtual void inc(block_address const &v);
virtual void dec(block_address const &v);
private:
space_map::ptr sm_;
};
// FIXME: move to sep file. I don't think it's directly used by
// the btree code.
2011-07-15 19:51:28 +05:30
struct uint64_traits {
2013-03-22 19:12:20 +05:30
typedef base::le64 disk_type;
2011-07-15 19:51:28 +05:30
typedef uint64_t value_type;
typedef no_op_ref_counter<uint64_t> ref_counter;
2011-07-15 19:51:28 +05:30
static void unpack(disk_type const &disk, value_type &value) {
value = base::to_cpu<uint64_t>(disk);
}
static void pack(value_type const &value, disk_type &disk) {
2013-03-22 19:12:20 +05:30
disk = base::to_disk<base::le64>(value);
2011-07-15 19:51:28 +05:30
}
};
struct block_traits {
typedef base::le64 disk_type;
typedef block_address value_type;
typedef block_ref_counter ref_counter;
static void unpack(disk_type const &disk, value_type &value) {
value = base::to_cpu<uint64_t>(disk);
}
static void pack(value_type const &value, disk_type &disk) {
disk = base::to_disk<base::le64>(value);
}
};
2011-07-15 19:51:28 +05:30
namespace btree_detail {
using namespace base;
using namespace std;
2011-09-16 14:36:37 +05:30
uint32_t const BTREE_CSUM_XOR = 121107;
2011-07-15 19:51:28 +05:30
//------------------------------------------------
// On disk data layout for btree nodes
enum node_flags {
INTERNAL_NODE = 1,
LEAF_NODE = 1 << 1
};
struct node_header {
2013-03-22 19:12:20 +05:30
le32 csum;
le32 flags;
le64 blocknr; /* which block this node is supposed to live in */
le32 nr_entries;
le32 max_entries;
le32 value_size;
le32 padding;
2011-07-15 19:51:28 +05:30
} __attribute__((packed));
struct disk_node {
struct node_header header;
2013-03-22 19:12:20 +05:30
le64 keys[0];
2011-07-15 19:51:28 +05:30
} __attribute__((packed));
enum node_type {
INTERNAL,
LEAF
};
//------------------------------------------------
// Class that acts as an interface over the raw little endian btree
// node data.
template <typename ValueTraits>
2011-07-15 19:51:28 +05:30
class node_ref {
public:
2011-08-22 18:44:10 +05:30
explicit node_ref(block_address b, disk_node *raw);
2011-09-16 14:36:37 +05:30
uint32_t get_checksum() const;
2011-08-22 18:44:10 +05:30
block_address get_location() const {
return location_;
}
2011-07-15 19:51:28 +05:30
2011-08-23 16:25:37 +05:30
block_address get_block_nr() const;
2011-07-15 19:51:28 +05:30
node_type get_type() const;
void set_type(node_type t);
unsigned get_nr_entries() const;
void set_nr_entries(unsigned n);
unsigned get_max_entries() const;
void set_max_entries(unsigned n);
2011-07-22 20:39:56 +05:30
// FIXME: remove this, and get the constructor to do it.
2011-07-15 19:51:28 +05:30
void set_max_entries(); // calculates the max for you.
2011-08-23 16:25:37 +05:30
size_t get_value_size() const;
2011-11-01 17:01:03 +05:30
void set_value_size(size_t);
2011-08-23 16:25:37 +05:30
2011-07-15 19:51:28 +05:30
uint64_t key_at(unsigned i) const;
void set_key(unsigned i, uint64_t k);
typename ValueTraits::value_type value_at(unsigned i) const;
void set_value(unsigned i,
typename ValueTraits::value_type const &v);
// Increments the nr_entries field
void insert_at(unsigned i,
uint64_t key,
typename ValueTraits::value_type const &v);
// Does not increment nr_entries
void overwrite_at(unsigned i,
uint64_t key,
typename ValueTraits::value_type const &v);
// Copies entries from another node, appends them
// to the back of this node. Adjusts nr_entries.
void copy_entries(node_ref const &rhs,
unsigned begin,
unsigned end);
// Various searches
int bsearch(uint64_t key, int want_hi) const;
2013-06-25 18:18:02 +05:30
boost::optional<unsigned> exact_search(uint64_t key) const;
2011-07-15 19:51:28 +05:30
int lower_bound(uint64_t key) const;
2011-07-22 20:39:56 +05:30
template <typename RefCounter>
void inc_children(RefCounter &rc);
2011-09-16 14:36:37 +05:30
disk_node *raw() {
return raw_;
}
disk_node const *raw() const {
2011-07-22 20:39:56 +05:30
return raw_;
}
2011-07-15 19:51:28 +05:30
private:
static unsigned calc_max_entries(void);
void *key_ptr(unsigned i) const;
void *value_ptr(unsigned i) const;
2011-08-22 18:44:10 +05:30
block_address location_;
2011-07-15 19:51:28 +05:30
disk_node *raw_;
};
//------------------------------------------------
//
template <typename ValueTraits>
node_ref<ValueTraits>
to_node(typename block_manager<>::read_ref &b)
2011-07-15 19:51:28 +05:30
{
// FIXME: this should return a const read_ref somehow.
return node_ref<ValueTraits>(
2011-08-22 18:44:10 +05:30
b.get_location(),
2011-07-15 19:51:28 +05:30
reinterpret_cast<disk_node *>(
const_cast<unsigned char *>(b.data().raw())));
2011-07-15 19:51:28 +05:30
}
template <typename ValueTraits>
node_ref<ValueTraits>
to_node(typename block_manager<>::write_ref &b)
2011-07-15 19:51:28 +05:30
{
return node_ref<ValueTraits>(
2011-08-22 18:44:10 +05:30
b.get_location(),
2011-07-15 19:51:28 +05:30
reinterpret_cast<disk_node *>(
const_cast<unsigned char *>(b.data().raw())));
2011-07-15 19:51:28 +05:30
}
2013-06-25 18:18:02 +05:30
class ro_spine : private boost::noncopyable {
2011-07-15 19:51:28 +05:30
public:
ro_spine(transaction_manager::ptr tm,
typename block_manager<>::validator::ptr v)
: tm_(tm),
validator_(v) {
2011-07-15 19:51:28 +05:30
}
void step(block_address b);
2011-07-15 19:51:28 +05:30
template <typename ValueTraits>
node_ref<ValueTraits> get_node() {
return to_node<ValueTraits>(spine_.back());
2011-07-15 19:51:28 +05:30
}
private:
2011-09-01 15:12:57 +05:30
transaction_manager::ptr tm_;
typename block_manager<>::validator::ptr validator_;
2011-09-01 15:12:57 +05:30
std::list<block_manager<>::read_ref> spine_;
2011-07-15 19:51:28 +05:30
};
2013-06-25 18:18:02 +05:30
class shadow_spine : private boost::noncopyable {
2011-07-15 19:51:28 +05:30
public:
2011-09-01 15:12:57 +05:30
typedef transaction_manager::read_ref read_ref;
typedef transaction_manager::write_ref write_ref;
shadow_spine(transaction_manager::ptr tm,
typename block_manager<>::validator::ptr v)
: tm_(tm),
validator_(v) {
2011-07-15 19:51:28 +05:30
}
// true if the children of the shadow need incrementing
bool step(block_address b);
2011-09-01 15:12:57 +05:30
void step(transaction_manager::write_ref b) {
2011-07-15 19:51:28 +05:30
spine_.push_back(b);
if (spine_.size() == 1)
root_ = spine_.front().get_location();
else if (spine_.size() > 2)
spine_.pop_front();
}
void pop() {
spine_.pop_back();
}
template <typename ValueTraits>
node_ref<ValueTraits> get_node() {
return to_node<ValueTraits>(spine_.back());
2011-07-15 19:51:28 +05:30
}
block_address get_block() const {
return spine_.back().get_location();
}
bool has_parent() const {
return spine_.size() > 1;
}
node_ref<block_traits> get_parent() {
2011-07-15 19:51:28 +05:30
if (spine_.size() < 2)
throw std::runtime_error("no parent");
return to_node<block_traits>(spine_.front());
2011-07-15 19:51:28 +05:30
}
2011-07-22 20:39:56 +05:30
block_address get_parent_location() const {
return spine_.front().get_location();
}
2011-07-15 19:51:28 +05:30
block_address get_root() const {
return root_;
}
private:
2011-09-01 15:12:57 +05:30
transaction_manager::ptr tm_;
typename block_manager<>::validator::ptr validator_;
2011-09-01 15:12:57 +05:30
std::list<block_manager<>::write_ref> spine_;
2011-07-15 19:51:28 +05:30
block_address root_;
};
2013-05-17 15:28:34 +05:30
// Used to keep a record of a nested btree's position.
typedef std::vector<uint64_t> btree_path;
// Used when visiting the nodes that make up a btree.
struct node_location {
2013-05-15 18:07:30 +05:30
node_location()
: depth(0) {
}
void inc_depth() {
depth++;
}
void push_key(uint64_t k) {
path.push_back(k);
depth = 0;
}
bool is_sub_root() const {
return depth == 0; // && path.size();
}
unsigned level() const {
return path.size();
}
// Keys used to access this sub tree
2013-05-17 15:28:34 +05:30
btree_path path;
2013-05-15 18:07:30 +05:30
// in this sub tree
unsigned depth;
2013-05-15 18:07:30 +05:30
// This is the key from the parent node to this
// node. If this node is a root then there will be
// no parent, and hence no key.
boost::optional<uint64_t> key;
};
2011-07-15 19:51:28 +05:30
}
template <unsigned Levels, typename ValueTraits>
class btree {
public:
typedef boost::shared_ptr<btree<Levels, ValueTraits> > ptr;
2011-07-15 19:51:28 +05:30
typedef uint64_t key[Levels];
typedef typename ValueTraits::value_type value_type;
typedef boost::optional<value_type> maybe_value;
typedef boost::optional<std::pair<unsigned, value_type> > maybe_pair;
typedef typename block_manager<>::read_ref read_ref;
typedef typename block_manager<>::write_ref write_ref;
typedef typename btree_detail::node_ref<ValueTraits> leaf_node;
typedef typename btree_detail::node_ref<block_traits> internal_node;
btree(typename persistent_data::transaction_manager::ptr tm,
2011-07-22 20:39:56 +05:30
typename ValueTraits::ref_counter rc);
btree(typename transaction_manager::ptr tm,
2011-07-22 20:39:56 +05:30
block_address root,
typename ValueTraits::ref_counter rc);
~btree();
maybe_value lookup(key const &key) const;
maybe_pair lookup_le(key const &key) const;
maybe_pair lookup_ge(key const &key) const;
void insert(key const &key, typename ValueTraits::value_type const &value);
void remove(key const &key);
void set_root(block_address root);
block_address get_root() const;
ptr clone() const;
// free the on disk btree when the destructor is called
void destroy();
// Derive a class from this base class if you need to
// inspect the individual nodes that make up a btree.
class visitor {
public:
typedef boost::shared_ptr<visitor> ptr;
typedef btree_detail::node_location node_location;
virtual ~visitor() {}
// The bool return values indicate whether the walk
// should be continued into sub trees of the node (true == continue).
virtual bool visit_internal(node_location const &l,
internal_node const &n) = 0;
virtual bool visit_internal_leaf(node_location const &l,
internal_node const &n) = 0;
virtual bool visit_leaf(node_location const &l,
leaf_node const &n) = 0;
2011-10-10 18:40:30 +05:30
virtual void visit_complete() {}
enum error_outcome {
EXCEPTION_HANDLED,
RETHROW_EXCEPTION
};
virtual error_outcome error_accessing_node(node_location const &l, block_address b,
std::string const &what) {
return RETHROW_EXCEPTION;
}
};
// Walks the tree in depth first order
void visit_depth_first(visitor &visitor) const;
private:
2011-11-01 17:01:03 +05:30
template <typename ValueTraits2, typename Search>
2013-05-28 18:18:10 +05:30
boost::optional<typename ValueTraits2::value_type>
2011-11-01 17:01:03 +05:30
lookup_raw(btree_detail::ro_spine &spine, block_address block, uint64_t key) const;
2011-07-15 19:51:28 +05:30
template <typename ValueTraits2>
void split_node(btree_detail::shadow_spine &spine,
2011-07-15 19:51:28 +05:30
block_address parent_index,
uint64_t key,
bool top);
2011-06-27 15:15:30 +05:30
2011-07-15 19:51:28 +05:30
template <typename ValueTraits2>
void split_beneath(btree_detail::shadow_spine &spine, uint64_t key);
2011-06-27 15:15:30 +05:30
2011-07-15 19:51:28 +05:30
template <typename ValueTraits2>
void split_sibling(btree_detail::shadow_spine &spine,
2011-07-15 19:51:28 +05:30
block_address parent_index,
uint64_t key);
2011-06-27 15:15:30 +05:30
template <typename ValueTraits2, typename RC>
2011-07-15 19:51:28 +05:30
bool
insert_location(btree_detail::shadow_spine &spine,
2011-07-15 19:51:28 +05:30
block_address block,
uint64_t key,
int *index,
RC &leaf_rc);
2011-07-15 19:51:28 +05:30
void walk_tree(visitor &visitor,
btree_detail::node_location const &loc,
2011-08-25 20:57:58 +05:30
block_address b) const;
void walk_tree_internal(visitor &visitor,
btree_detail::node_location const &loc,
block_address b) const;
2013-06-20 16:47:16 +05:30
template <typename ValueTraits2, typename RefCounter>
void inc_children(btree_detail::shadow_spine &spine,
RefCounter &leaf_rc);
typename persistent_data::transaction_manager::ptr tm_;
2011-07-15 19:51:28 +05:30
bool destroy_;
block_address root_;
block_ref_counter internal_rc_;
2011-07-22 20:39:56 +05:30
typename ValueTraits::ref_counter rc_;
typename block_manager<>::validator::ptr validator_;
2011-06-27 15:15:30 +05:30
};
};
2011-06-27 15:15:30 +05:30
#include "btree.tcc"
//----------------------------------------------------------------
#endif