Alex Deymo | f006135 | 2015-07-01 14:59:15 -0700 | [diff] [blame] | 1 | // Copyright 2015 The Chromium OS Authors. All rights reserved. |
| 2 | // Use of this source code is governed by a BSD-style license that can be |
| 3 | // found in the LICENSE file. |
| 4 | |
| 5 | #include "update_engine/payload_generator/block_mapping.h" |
| 6 | |
| 7 | #include <fcntl.h> |
| 8 | #include <sys/stat.h> |
| 9 | #include <sys/types.h> |
| 10 | |
| 11 | #include <functional> |
| 12 | #include <string> |
| 13 | #include <vector> |
| 14 | |
| 15 | #include "update_engine/utils.h" |
| 16 | |
| 17 | using std::string; |
| 18 | using std::vector; |
| 19 | |
| 20 | namespace { |
| 21 | |
| 22 | size_t HashValue(const chromeos::Blob& blob) { |
| 23 | std::hash<string> hash_fn; |
| 24 | return hash_fn(string(blob.begin(), blob.end())); |
| 25 | } |
| 26 | |
| 27 | } // namespace |
| 28 | |
| 29 | namespace chromeos_update_engine { |
| 30 | |
| 31 | BlockMapping::BlockId BlockMapping::AddBlock(const chromeos::Blob& block_data) { |
| 32 | return AddBlock(-1, 0, block_data); |
| 33 | } |
| 34 | |
| 35 | BlockMapping::BlockId BlockMapping::AddDiskBlock(int fd, off_t byte_offset) { |
| 36 | chromeos::Blob blob(block_size_); |
| 37 | ssize_t bytes_read = 0; |
| 38 | if (!utils::PReadAll(fd, blob.data(), block_size_, byte_offset, &bytes_read)) |
| 39 | return -1; |
| 40 | if (static_cast<size_t>(bytes_read) != block_size_) |
| 41 | return -1; |
| 42 | return AddBlock(fd, byte_offset, blob); |
| 43 | } |
| 44 | |
| 45 | bool BlockMapping::AddManyDiskBlocks(int fd, |
| 46 | off_t initial_byte_offset, |
| 47 | size_t num_blocks, |
| 48 | std::vector<BlockId>* block_ids) { |
| 49 | bool ret = true; |
| 50 | block_ids->resize(num_blocks); |
| 51 | for (size_t block = 0; block < num_blocks; block++) { |
| 52 | (*block_ids)[block] = AddDiskBlock( |
| 53 | fd, initial_byte_offset + block * block_size_); |
| 54 | ret = ret && (*block_ids)[block] != -1; |
| 55 | } |
| 56 | return ret; |
| 57 | } |
| 58 | |
| 59 | BlockMapping::BlockId BlockMapping::AddBlock(int fd, |
| 60 | off_t byte_offset, |
| 61 | const chromeos::Blob& block_data) { |
| 62 | if (block_data.size() != block_size_) |
| 63 | return -1; |
| 64 | size_t h = HashValue(block_data); |
| 65 | |
| 66 | // We either reuse a UniqueBlock or create a new one. If we need a new |
| 67 | // UniqueBlock it could also be part of a new or existing bucket (if there is |
| 68 | // a hash collision). |
| 69 | vector<UniqueBlock> *bucket = nullptr; |
| 70 | |
| 71 | auto mapping_it = mapping_.find(h); |
| 72 | if (mapping_it == mapping_.end()) { |
| 73 | bucket = &mapping_[h]; |
| 74 | } else { |
| 75 | for (UniqueBlock& existing_block : mapping_it->second) { |
| 76 | bool equals = false; |
| 77 | if (!existing_block.CompareData(block_data, &equals)) |
| 78 | return -1; |
| 79 | if (equals) |
| 80 | return existing_block.block_id; |
| 81 | } |
| 82 | bucket = &mapping_it->second; |
| 83 | } |
| 84 | |
| 85 | // No existing block was found at this point, so we create and fill in a new |
| 86 | // one. |
| 87 | bucket->emplace_back(); |
| 88 | UniqueBlock *new_ublock = &bucket->back(); |
| 89 | |
| 90 | new_ublock->times_read = 1; |
| 91 | new_ublock->fd = fd; |
| 92 | new_ublock->byte_offset = byte_offset; |
| 93 | new_ublock->block_id = used_block_ids++; |
| 94 | // We need to cache blocks that are not referencing any disk location. |
| 95 | if (fd == -1) |
| 96 | new_ublock->block_data = block_data; |
| 97 | |
| 98 | return new_ublock->block_id; |
| 99 | } |
| 100 | |
| 101 | bool BlockMapping::UniqueBlock::CompareData(const chromeos::Blob& other_block, |
| 102 | bool* equals) { |
| 103 | if (!block_data.empty()) { |
| 104 | *equals = block_data == other_block; |
| 105 | return true; |
| 106 | } |
| 107 | const size_t block_size = other_block.size(); |
| 108 | chromeos::Blob blob(block_size); |
| 109 | ssize_t bytes_read = 0; |
| 110 | if (!utils::PReadAll(fd, blob.data(), block_size, byte_offset, &bytes_read)) |
| 111 | return false; |
| 112 | if (static_cast<size_t>(bytes_read) != block_size) |
| 113 | return false; |
| 114 | *equals = blob == other_block; |
| 115 | |
| 116 | // We increase the number of times we had to read this block from disk and |
| 117 | // we cache this block based on that. This caching method is optimized for |
| 118 | // the common use case of having two partitions that share blocks between them |
| 119 | // but have few repeated blocks inside each partition, such as the block |
| 120 | // with all zeros or duplicated files. |
| 121 | times_read++; |
| 122 | if (times_read > 3) |
| 123 | block_data = std::move(blob); |
| 124 | return true; |
| 125 | } |
| 126 | |
| 127 | bool MapPartitionBlocks(const string& old_part, |
| 128 | const string& new_part, |
| 129 | size_t old_size, |
| 130 | size_t new_size, |
| 131 | size_t block_size, |
| 132 | vector<BlockMapping::BlockId>* old_block_ids, |
| 133 | vector<BlockMapping::BlockId>* new_block_ids) { |
| 134 | BlockMapping mapping(block_size); |
| 135 | if (mapping.AddBlock(chromeos::Blob(block_size, '\0')) != 0) |
| 136 | return false; |
| 137 | int old_fd = HANDLE_EINTR(open(old_part.c_str(), O_RDONLY)); |
| 138 | int new_fd = HANDLE_EINTR(open(new_part.c_str(), O_RDONLY)); |
| 139 | ScopedFdCloser old_fd_closer(&old_fd); |
| 140 | ScopedFdCloser new_fd_closer(&new_fd); |
| 141 | |
| 142 | TEST_AND_RETURN_FALSE(mapping.AddManyDiskBlocks( |
| 143 | old_fd, 0, old_size / block_size, old_block_ids)); |
| 144 | TEST_AND_RETURN_FALSE(mapping.AddManyDiskBlocks( |
| 145 | new_fd, 0, new_size / block_size, new_block_ids)); |
| 146 | return true; |
| 147 | } |
| 148 | |
| 149 | } // namespace chromeos_update_engine |