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

510 lines
13 KiB
C
Raw Normal View History

2013-05-08 16:30:24 +05:30
#ifndef PERSISTENT_DATA_DATA_STRUCTURES_DAMAGE_VISITOR_H
#define PERSISTENT_DATA_DATA_STRUCTURES_DAMAGE_VISITOR_H
#include "persistent-data/data-structures/btree.h"
2013-05-28 16:21:44 +05:30
#include "persistent-data/run.h"
2013-05-08 16:30:24 +05:30
//----------------------------------------------------------------
namespace persistent_data {
2013-05-08 21:08:38 +05:30
namespace btree_detail {
struct damage {
typedef boost::shared_ptr<damage> ptr;
2013-05-28 16:50:05 +05:30
damage(run<uint64_t> lost_keys,
2013-05-08 21:08:38 +05:30
std::string const &desc)
: lost_keys_(lost_keys),
2013-05-08 21:08:38 +05:30
desc_(desc) {
}
2013-05-28 16:50:05 +05:30
run<uint64_t> lost_keys_;
2013-05-08 21:08:38 +05:30
std::string desc_;
};
2013-05-13 15:57:38 +05:30
inline std::ostream &operator <<(std::ostream &out, damage const &d) {
out << "btree damage[lost_keys = " << d.lost_keys_
<< ", \"" << d.desc_ << "\"]";
return out;
}
class noop_damage_visitor {
public:
virtual void visit(btree_path const &path, damage const &d) {
}
};
2013-05-15 18:07:30 +05:30
// Tracks damage in a single level btree. Use multiple
// trackers if you have a multilayer tree.
2013-05-13 15:57:38 +05:30
class damage_tracker {
public:
damage_tracker()
: damaged_(false),
damage_begin_(0) {
}
2013-05-28 17:30:30 +05:30
typedef run<uint64_t> run64;
2013-05-28 16:50:05 +05:30
typedef boost::optional<run64> maybe_run64;
2013-05-13 15:57:38 +05:30
void bad_node() {
damaged_ = true;
}
2013-05-28 16:50:05 +05:30
maybe_run64 good_internal(block_address begin) {
maybe_run64 r;
2013-05-13 15:57:38 +05:30
if (damaged_) {
2013-05-28 16:50:05 +05:30
r = maybe_run64(run64(damage_begin_, begin));
2013-05-13 15:57:38 +05:30
damaged_ = false;
}
damage_begin_ = begin;
return r;
}
// remember 'end' is the one-past-the-end value, so
2013-05-13 15:57:38 +05:30
// take the last key in the leaf and add one.
2013-05-28 16:50:05 +05:30
maybe_run64 good_leaf(block_address begin, block_address end) {
maybe_run64 r;
2013-05-13 15:57:38 +05:30
if (damaged_) {
2013-05-28 16:50:05 +05:30
r = maybe_run64(run64(damage_begin_, begin));
2013-05-13 15:57:38 +05:30
damaged_ = false;
}
damage_begin_ = end;
return r;
}
2013-05-28 16:50:05 +05:30
maybe_run64 end() {
maybe_run64 r;
2013-05-13 15:57:38 +05:30
if (damaged_)
r = maybe_run64(damage_begin_);
2013-05-13 15:57:38 +05:30
else
r = maybe_run64();
damaged_ = false;
damage_begin_ = 0;
return r;
2013-05-13 15:57:38 +05:30
}
private:
bool damaged_;
block_address damage_begin_;
};
2013-05-08 21:08:38 +05:30
2013-05-20 17:01:47 +05:30
// As we walk a btree we need to know if we've moved into a
// different sub tree (by looking at the btree_path).
class path_tracker {
public:
path_tracker() {
// We push an empty path, to ensure there
// is always a current_path.
paths_.push_back(btree_path());
}
2013-05-20 17:01:47 +05:30
// returns the old path if the tree has changed.
btree_path const *next_path(btree_path const &p) {
if (p != current_path()) {
if (paths_.size() == 2)
paths_.pop_front();
paths_.push_back(p);
return &paths_.front();
2013-05-20 17:01:47 +05:30
}
return NULL;
2013-05-20 17:01:47 +05:30
}
btree_path const &current_path() const {
return paths_.back();
2013-05-20 17:01:47 +05:30
}
private:
std::list<btree_path> paths_;
2013-05-20 17:01:47 +05:30
};
//----------------------------------------------------------------
// This class implements consistency checking for the btrees. It
// also allows the caller to visit all accessible values.
// Derive from this if you want some additional checks. It's worth
// summarising what is checked:
//
// Implemented
// -----------
//
// - block_nr
// - nr_entries < max_entries
// - max_entries fits in block
// - max_entries is divisible by 3
// - nr_entries > minimum (except for root nodes)
//
// Not implemented
// ---------------
//
// - leaf | internal flags (this can be inferred from siblings)
//----------------------------------------------------------------
template <typename ValueVisitor, typename DamageVisitor, uint32_t Levels, typename ValueTraits>
class btree_damage_visitor : public btree<Levels, ValueTraits>::visitor {
public:
typedef btree_detail::node_location node_location;
2013-05-28 16:50:05 +05:30
typedef run<block_address> run64;
typedef boost::optional<run64> maybe_run64;
2013-05-08 16:30:24 +05:30
2013-12-11 22:58:14 +05:30
btree_damage_visitor(ValueVisitor &value_visitor,
DamageVisitor &damage_visitor,
bool avoid_repeated_visits = true)
: avoid_repeated_visits_(avoid_repeated_visits),
value_visitor_(value_visitor),
damage_visitor_(damage_visitor) {
}
2013-05-08 16:30:24 +05:30
bool visit_internal(node_location const &loc,
btree_detail::node_ref<block_traits> const &n) {
2013-05-20 17:01:47 +05:30
update_path(loc.path);
return check_internal(loc, n);
}
2013-05-08 16:30:24 +05:30
bool visit_internal_leaf(node_location const &loc,
btree_detail::node_ref<block_traits> const &n) {
2013-05-20 17:01:47 +05:30
update_path(loc.path);
return check_leaf(loc, n);
}
2013-05-09 18:01:04 +05:30
bool visit_leaf(node_location const &loc,
btree_detail::node_ref<ValueTraits> const &n) {
2013-05-20 17:01:47 +05:30
update_path(loc.path);
bool r = check_leaf(loc, n);
2013-05-09 18:01:04 +05:30
// If anything goes wrong with the checks, we skip
// the value visiting.
if (!r)
return false;
2013-05-09 18:01:04 +05:30
visit_values(loc.path, n);
2013-05-08 16:30:24 +05:30
return true;
}
void visit_complete() {
end_walk();
}
2013-05-08 21:08:38 +05:30
typedef typename btree<Levels, ValueTraits>::visitor::error_outcome error_outcome;
2013-05-08 21:08:38 +05:30
error_outcome error_accessing_node(node_location const &l, block_address b,
std::string const &what) {
update_path(l.path);
report_damage(what);
return btree<Levels, ValueTraits>::visitor::EXCEPTION_HANDLED;
}
2013-05-09 18:01:04 +05:30
private:
void visit_values(btree_path const &path,
node_ref<ValueTraits> const &n) {
btree_path p2(path);
unsigned nr = n.get_nr_entries();
for (unsigned i = 0; i < nr; i++) {
p2.push_back(n.key_at(i));
value_visitor_.visit(p2, n.value_at(i));
p2.pop_back();
}
2013-05-08 16:30:24 +05:30
}
bool check_internal(node_location const &loc,
btree_detail::node_ref<block_traits> const &n) {
if (!already_visited(n) &&
check_block_nr(n) &&
check_value_size(n) &&
check_max_entries(n) &&
check_nr_entries(n, loc.is_sub_root()) &&
check_ordered_keys(n) &&
2013-06-25 18:18:02 +05:30
check_parent_key(loc.is_sub_root() ? boost::optional<uint64_t>() : loc.key, n)) {
if (loc.is_sub_root())
new_root(loc.level());
good_internal(n.key_at(0));
return true;
}
return false;
}
2013-05-08 16:30:24 +05:30
template <typename ValueTraits2>
bool check_leaf(node_location const &loc,
btree_detail::node_ref<ValueTraits2> const &n) {
if (!already_visited(n) &&
check_block_nr(n) &&
check_value_size(n) &&
check_max_entries(n) &&
check_nr_entries(n, loc.is_sub_root()) &&
check_ordered_keys(n) &&
2013-06-25 18:18:02 +05:30
check_parent_key(loc.is_sub_root() ? boost::optional<uint64_t>() : loc.key, n)) {
if (loc.is_sub_root())
new_root(loc.level());
bool r = check_leaf_key(loc.level(), n);
if (r && n.get_nr_entries() > 0)
good_leaf(n.key_at(0), n.key_at(n.get_nr_entries() - 1) + 1);
return r;
}
2013-05-13 15:57:38 +05:30
return false;
2013-05-08 16:30:24 +05:30
}
template <typename node>
bool already_visited(node const &n) {
block_address b = n.get_location();
2013-05-08 16:30:24 +05:30
if (avoid_repeated_visits_) {
if (seen_.count(b) > 0)
return true;
2013-05-08 16:30:24 +05:30
seen_.insert(b);
}
2013-05-08 16:30:24 +05:30
return false;
2013-05-08 16:30:24 +05:30
}
template <typename node>
bool check_block_nr(node const &n) {
if (n.get_location() != n.get_block_nr()) {
std::ostringstream out;
out << "block number mismatch: actually "
<< n.get_location()
<< ", claims " << n.get_block_nr();
2013-05-08 16:30:24 +05:30
report_damage(out.str());
return false;
}
2013-05-08 21:08:38 +05:30
return true;
2013-05-08 16:30:24 +05:30
}
template <typename node>
bool check_value_size(node const &n) {
if (!n.value_sizes_match()) {
report_damage(n.value_mismatch_string());
return false;
}
return true;
}
template <typename node>
bool check_max_entries(node const &n) {
size_t elt_size = sizeof(uint64_t) + n.get_value_size();
if (elt_size * n.get_max_entries() + sizeof(node_header) > MD_BLOCK_SIZE) {
std::ostringstream out;
2016-02-26 22:58:41 +05:30
out << "max entries too large: " << n.get_max_entries()
<< " (block " << n.get_location() << ")";
report_damage(out.str());
return false;
}
2013-05-08 16:30:24 +05:30
if (n.get_max_entries() % 3) {
std::ostringstream out;
2016-02-26 22:58:41 +05:30
out << "max entries is not divisible by 3: " << n.get_max_entries()
<< " (block " << n.get_location() << ")";
report_damage(out.str());
return false;
}
2013-05-08 16:30:24 +05:30
return true;
2013-05-08 16:30:24 +05:30
}
template <typename node>
bool check_nr_entries(node const &n, bool is_root) {
if (n.get_nr_entries() > n.get_max_entries()) {
std::ostringstream out;
out << "bad nr_entries: "
<< n.get_nr_entries() << " < "
2016-02-26 22:58:41 +05:30
<< n.get_max_entries()
<< " (block " << n.get_location() << ")";
report_damage(out.str());
return false;
}
2013-05-08 16:30:24 +05:30
block_address min = n.get_max_entries() / 3;
if (!is_root && (n.get_nr_entries() < min)) {
ostringstream out;
out << "too few entries in btree_node: "
<< n.get_nr_entries()
<< ", expected at least "
<< min
2016-02-26 22:58:41 +05:30
<< " (block " << n.get_location()
<< ", max_entries = " << n.get_max_entries() << ")";
report_damage(out.str());
return false;
}
2013-05-08 16:30:24 +05:30
return true;
2013-05-08 16:30:24 +05:30
}
template <typename node>
bool check_ordered_keys(node const &n) {
unsigned nr_entries = n.get_nr_entries();
2013-05-08 16:30:24 +05:30
if (nr_entries == 0)
return true; // can only happen if a root node
2013-05-08 16:30:24 +05:30
uint64_t last_key = n.key_at(0);
2013-05-08 16:30:24 +05:30
for (unsigned i = 1; i < nr_entries; i++) {
uint64_t k = n.key_at(i);
if (k <= last_key) {
ostringstream out;
2016-02-26 22:58:41 +05:30
out << "keys are out of order, " << k << " <= " << last_key
<< " (block " << n.get_location() << ")";
report_damage(out.str());
return false;
}
last_key = k;
}
return true;
}
2013-05-08 16:30:24 +05:30
template <typename node>
bool check_parent_key(boost::optional<uint64_t> key, node const &n) {
if (!key)
return true;
if (*key > n.key_at(0)) {
2013-05-08 16:30:24 +05:30
ostringstream out;
out << "parent key mismatch: parent was " << *key
2016-02-26 22:58:41 +05:30
<< ", but lowest in node was " << n.key_at(0)
<< " (block " << n.get_location() << ")";
report_damage(out.str());
2013-05-08 16:30:24 +05:30
return false;
}
return true;
}
template <typename node>
bool check_leaf_key(unsigned level, node const &n) {
if (n.get_nr_entries() == 0)
return true; // can only happen if a root node
2013-05-08 16:30:24 +05:30
if (last_leaf_key_[level] && *last_leaf_key_[level] >= n.key_at(0)) {
ostringstream out;
out << "the last key of the previous leaf was " << *last_leaf_key_[level]
2016-02-26 22:58:41 +05:30
<< " and the first key of this leaf is " << n.key_at(0)
<< " (block " << n.get_location() << ")";
report_damage(out.str());
return false;
}
2013-05-08 16:30:24 +05:30
last_leaf_key_[level] = n.key_at(n.get_nr_entries() - 1);
return true;
2013-05-08 16:30:24 +05:30
}
void new_root(unsigned level) {
// we're starting a new subtree, so should
// reset the last_leaf value.
last_leaf_key_[level] = boost::optional<uint64_t>();
}
2013-05-08 16:30:24 +05:30
//--------------------------------
2013-05-08 16:30:24 +05:30
// damage tracking
void report_damage(std::string const &desc) {
damage_reasons_.push_back(desc);
dt_.bad_node();
}
void good_internal(block_address b) {
2013-05-28 16:50:05 +05:30
maybe_run64 mr = dt_.good_internal(b);
if (mr)
2013-05-20 17:01:47 +05:30
issue_damage(path_tracker_.current_path(), *mr);
}
2013-05-13 15:57:38 +05:30
void good_leaf(block_address b, block_address e) {
2013-05-28 16:50:05 +05:30
maybe_run64 mr = dt_.good_leaf(b, e);
if (mr)
2013-05-20 17:01:47 +05:30
issue_damage(path_tracker_.current_path(), *mr);
}
2013-05-13 15:57:38 +05:30
void end_walk() {
2013-05-20 17:45:51 +05:30
maybe_issue_damage(path_tracker_.current_path());
}
2013-05-28 16:50:05 +05:30
void issue_damage(btree_path const &path, run64 const &r) {
damage d(r, build_damage_desc());
clear_damage_desc();
2013-05-20 17:01:47 +05:30
damage_visitor_.visit(path, d);
}
std::string build_damage_desc() const {
std::string r;
2013-05-08 21:08:38 +05:30
std::list<std::string>::const_iterator it, end = damage_reasons_.end();
for (it = damage_reasons_.begin(); it != end; ++it)
r += *it;
2013-05-13 15:57:38 +05:30
return r;
}
2013-05-13 15:57:38 +05:30
void clear_damage_desc() {
damage_reasons_.clear();
}
2013-05-13 15:57:38 +05:30
2013-05-20 17:45:51 +05:30
void maybe_issue_damage(btree_path const &path) {
2013-05-28 16:50:05 +05:30
maybe_run64 mr = dt_.end();
2013-05-20 17:45:51 +05:30
if (mr)
issue_damage(path, *mr);
}
2013-05-08 21:08:38 +05:30
2013-05-20 17:01:47 +05:30
void update_path(btree_path const &path) {
btree_path const *old_path = path_tracker_.next_path(path);
2013-05-20 17:45:51 +05:30
if (old_path)
2013-05-20 17:01:47 +05:30
// we need to emit any errors that
// were accrued against the old
// path.
2013-05-20 17:45:51 +05:30
maybe_issue_damage(*old_path);
2013-05-20 17:01:47 +05:30
}
//--------------------------------
bool avoid_repeated_visits_;
2013-05-13 15:57:38 +05:30
ValueVisitor &value_visitor_;
DamageVisitor &damage_visitor_;
2013-05-08 21:08:38 +05:30
std::set<block_address> seen_;
boost::optional<uint64_t> last_leaf_key_[Levels];
2013-05-08 16:30:24 +05:30
2013-05-20 17:01:47 +05:30
path_tracker path_tracker_;
damage_tracker dt_;
std::list<std::string> damage_reasons_;
};
}
2013-05-13 15:57:38 +05:30
template <unsigned Levels, typename ValueTraits, typename ValueVisitor, typename DamageVisitor>
void btree_visit_values(btree<Levels, ValueTraits> const &tree,
ValueVisitor &value_visitor,
DamageVisitor &damage_visitor) {
btree_detail::btree_damage_visitor<ValueVisitor, DamageVisitor, Levels, ValueTraits>
2013-12-11 22:58:14 +05:30
v(value_visitor, damage_visitor);
tree.visit_depth_first(v);
}
2013-05-08 16:30:24 +05:30
}
//----------------------------------------------------------------
#endif