1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
|
// Copyright (c) 2012 The Chromium Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
//
// A read-only set implementation for |SBPrefix| items. Prefixes are
// sorted and stored as 16-bit deltas from the previous prefix. An
// index structure provides quick random access, and also handles
// cases where 16 bits cannot encode a delta.
//
// For example, the sequence {20, 25, 41, 65432, 150000, 160000} would
// be stored as:
// A pair {20, 0} in |index_|.
// 5, 16, 65391 in |deltas_|.
// A pair {150000, 3} in |index_|.
// 10000 in |deltas_|.
// |index_.size()| will be 2, |deltas_.size()| will be 4.
//
// This structure is intended for storage of sparse uniform sets of
// prefixes of a certain size. As of this writing, my safe-browsing
// database contains:
// 653132 add prefixes
// 6446 are duplicates (from different chunks)
// 24301 w/in 2^8 of the prior prefix
// 622337 w/in 2^16 of the prior prefix
// 47 further than 2^16 from the prior prefix
// For this input, the memory usage is approximately 2 bytes per
// prefix, a bit over 1.2M. The bloom filter used 25 bits per prefix,
// a bit over 1.9M on this data.
//
// Experimenting with random selections of the above data, storage
// size drops almost linearly as prefix count drops, until the index
// overhead starts to become a problem a bit under 200k prefixes. The
// memory footprint gets worse than storing the raw prefix data around
// 75k prefixes. Fortunately, the actual memory footprint also falls.
// If the prefix count increases the memory footprint should increase
// approximately linearly. The worst-case would be 2^16 items all
// 2^16 apart, which would need 512k (versus 256k to store the raw
// data).
//
// The on-disk format looks like:
// 4 byte magic number
// 4 byte version number
// 4 byte |index_.size()|
// 4 byte |deltas_.size()|
// n * 8 byte |&index_[0]..&index_[n]|
// m * 2 byte |&deltas_[0]..&deltas_[m]|
// 16 byte digest
#ifndef COMPONENTS_SAFE_BROWSING_DB_PREFIX_SET_H_
#define COMPONENTS_SAFE_BROWSING_DB_PREFIX_SET_H_
#include <stddef.h>
#include <stdint.h>
#include <utility>
#include <vector>
#include "base/gtest_prod_util.h"
#include "base/macros.h"
#include "base/memory/scoped_ptr.h"
#include "components/safe_browsing_db/util.h"
namespace base {
class FilePath;
}
namespace safe_browsing {
class PrefixSet {
public:
~PrefixSet();
// |true| if |hash| is in the hashes passed to the set's builder, or if
// |hash.prefix| is one of the prefixes passed to the set's builder.
bool Exists(const SBFullHash& hash) const;
// Persist the set on disk.
static scoped_ptr<const PrefixSet> LoadFile(
const base::FilePath& filter_name);
bool WriteFile(const base::FilePath& filter_name) const;
private:
friend class PrefixSetBuilder;
friend class PrefixSetTest;
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, AllBig);
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, EdgeCases);
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, Empty);
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, FullHashBuild);
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, IntMinMax);
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, OneElement);
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, ReadWrite);
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, ReadWriteSigned);
FRIEND_TEST_ALL_PREFIXES(PrefixSetTest, Version3);
FRIEND_TEST_ALL_PREFIXES(SafeBrowsingStoreFileTest, BasicStore);
FRIEND_TEST_ALL_PREFIXES(SafeBrowsingStoreFileTest, DeleteChunks);
FRIEND_TEST_ALL_PREFIXES(SafeBrowsingStoreFileTest, DetectsCorruption);
FRIEND_TEST_ALL_PREFIXES(SafeBrowsingStoreFileTest, Empty);
FRIEND_TEST_ALL_PREFIXES(SafeBrowsingStoreFileTest, PrefixMinMax);
FRIEND_TEST_ALL_PREFIXES(SafeBrowsingStoreFileTest, SubKnockout);
FRIEND_TEST_ALL_PREFIXES(SafeBrowsingStoreFileTest, Version7);
FRIEND_TEST_ALL_PREFIXES(SafeBrowsingStoreFileTest, Version8);
// Maximum number of consecutive deltas to encode before generating
// a new index entry. This helps keep the worst-case performance
// for |Exists()| under control.
static const size_t kMaxRun = 100;
// Helpers to make |index_| easier to deal with.
typedef std::pair<SBPrefix, uint32_t> IndexPair;
typedef std::vector<IndexPair> IndexVector;
static bool PrefixLess(const IndexPair& a, const IndexPair& b);
// Helper to let |PrefixSetBuilder| add a run of data. |index_prefix| is
// added to |index_|, with the other elements added into |deltas_|.
void AddRun(SBPrefix index_prefix,
const uint16_t* run_begin,
const uint16_t* run_end);
// |true| if |prefix| is one of the prefixes passed to the set's builder.
// Provided for testing purposes.
bool PrefixExists(SBPrefix prefix) const;
// Regenerate the vector of prefixes passed to the constructor into
// |prefixes|. Prefixes will be added in sorted order. Useful for testing.
void GetPrefixes(std::vector<SBPrefix>* prefixes) const;
// Used by |PrefixSetBuilder|.
PrefixSet();
// Helper for |LoadFile()|. Steals vector contents using |swap()|.
PrefixSet(IndexVector* index,
std::vector<uint16_t>* deltas,
std::vector<SBFullHash>* full_hashes);
// Top-level index of prefix to offset in |deltas_|. Each pair
// indicates a base prefix and where the deltas from that prefix
// begin in |deltas_|. The deltas for a pair end at the next pair's
// index into |deltas_|.
IndexVector index_;
// Deltas which are added to the prefix in |index_| to generate
// prefixes. Deltas are only valid between consecutive items from
// |index_|, or the end of |deltas_| for the last |index_| pair.
std::vector<uint16_t> deltas_;
// Full hashes ordered by SBFullHashLess.
std::vector<SBFullHash> full_hashes_;
DISALLOW_COPY_AND_ASSIGN(PrefixSet);
};
// Helper to incrementally build a PrefixSet from a stream of sorted prefixes.
class PrefixSetBuilder {
public:
PrefixSetBuilder();
~PrefixSetBuilder();
// Helper for unit tests and format conversion.
explicit PrefixSetBuilder(const std::vector<SBPrefix>& prefixes);
// Add a prefix to the set. Prefixes must arrive in ascending order.
// Duplicate prefixes are dropped.
void AddPrefix(SBPrefix prefix);
// Flush any buffered prefixes, and return the final PrefixSet instance.
// |hashes| are sorted and stored in |full_hashes_|. Any call other than the
// destructor is illegal after this call.
scoped_ptr<const PrefixSet> GetPrefixSet(
const std::vector<SBFullHash>& hashes);
// Helper for clients which only track prefixes. Calls GetPrefixSet() with
// empty hash vector.
scoped_ptr<const PrefixSet> GetPrefixSetNoHashes();
private:
// Encode a run of deltas for |AddRun()|. The run is broken by a too-large
// delta, or kMaxRun, whichever comes first.
void EmitRun();
// Buffers prefixes until enough are avaliable to emit a run.
std::vector<SBPrefix> buffer_;
// The PrefixSet being built.
scoped_ptr<PrefixSet> prefix_set_;
};
} // namespace safe_browsing
#endif // COMPONENTS_SAFE_BROWSING_DB_PREFIX_SET_H_
|
