| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 1 | // | 
|  | 2 | // Copyright (C) 2021 The Android Open Source Project | 
|  | 3 | // | 
|  | 4 | // Licensed under the Apache License, Version 2.0 (the "License"); | 
|  | 5 | // you may not use this file except in compliance with the License. | 
|  | 6 | // You may obtain a copy of the License at | 
|  | 7 | // | 
|  | 8 | //      http://www.apache.org/licenses/LICENSE-2.0 | 
|  | 9 | // | 
|  | 10 | // Unless required by applicable law or agreed to in writing, software | 
|  | 11 | // distributed under the License is distributed on an "AS IS" BASIS, | 
|  | 12 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | 
|  | 13 | // See the License for the specific language governing permissions and | 
|  | 14 | // limitations under the License. | 
|  | 15 | // | 
|  | 16 |  | 
|  | 17 | #include "update_engine/payload_generator/erofs_filesystem.h" | 
|  | 18 |  | 
|  | 19 | #include <time.h> | 
|  | 20 |  | 
|  | 21 | #include <string> | 
|  | 22 | #include <mutex> | 
|  | 23 |  | 
|  | 24 | #include <erofs/internal.h> | 
|  | 25 | #include <erofs/dir.h> | 
|  | 26 | #include <erofs/io.h> | 
|  | 27 |  | 
|  | 28 | #include "erofs_iterate.h" | 
|  | 29 | #include "lz4diff/lz4diff.pb.h" | 
| Kelvin Zhang | 8389dfe | 2022-01-13 12:47:11 -0800 | [diff] [blame] | 30 | #include "lz4diff/lz4patch.h" | 
|  | 31 | #include "lz4diff/lz4diff.h" | 
| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 32 | #include "update_engine/common/utils.h" | 
|  | 33 | #include "update_engine/payload_generator/delta_diff_generator.h" | 
|  | 34 | #include "update_engine/payload_generator/extent_ranges.h" | 
|  | 35 | #include "update_engine/payload_generator/extent_utils.h" | 
| Kelvin Zhang | 8389dfe | 2022-01-13 12:47:11 -0800 | [diff] [blame] | 36 | #include "update_engine/payload_generator/filesystem_interface.h" | 
| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 37 |  | 
|  | 38 | namespace chromeos_update_engine { | 
|  | 39 |  | 
|  | 40 | namespace { | 
|  | 41 |  | 
|  | 42 | static constexpr int GetOccupiedSize(const struct erofs_inode* inode, | 
|  | 43 | erofs_off_t* size) { | 
|  | 44 | *size = 0; | 
|  | 45 | switch (inode->datalayout) { | 
|  | 46 | case EROFS_INODE_FLAT_INLINE: | 
|  | 47 | case EROFS_INODE_FLAT_PLAIN: | 
|  | 48 | case EROFS_INODE_CHUNK_BASED: | 
|  | 49 | *size = inode->i_size; | 
|  | 50 | break; | 
|  | 51 | case EROFS_INODE_FLAT_COMPRESSION_LEGACY: | 
|  | 52 | case EROFS_INODE_FLAT_COMPRESSION: | 
|  | 53 | *size = inode->u.i_blocks * EROFS_BLKSIZ; | 
|  | 54 | break; | 
|  | 55 | default: | 
|  | 56 | LOG(ERROR) << "unknown datalayout " << inode->datalayout; | 
|  | 57 | return -1; | 
|  | 58 | } | 
|  | 59 | return 0; | 
|  | 60 | } | 
|  | 61 |  | 
|  | 62 | static int ErofsMapBlocks(struct erofs_inode* inode, | 
|  | 63 | struct erofs_map_blocks* map, | 
|  | 64 | int flags) { | 
|  | 65 | if (erofs_inode_is_data_compressed(inode->datalayout)) { | 
|  | 66 | return z_erofs_map_blocks_iter(inode, map, flags); | 
|  | 67 | } | 
|  | 68 | return erofs_map_blocks(inode, map, flags); | 
|  | 69 | } | 
|  | 70 |  | 
|  | 71 | static constexpr bool IsBlockCompressed(const struct erofs_map_blocks& block) { | 
|  | 72 | // Z_EROFS_COMPRESSION_SHIFTED means data inside this block are merely | 
|  | 73 | // memmove()'ed in place, instead of going through some compression function | 
|  | 74 | // like LZ4 or LZMA | 
|  | 75 | return block.m_flags & EROFS_MAP_ENCODED && | 
|  | 76 | block.m_algorithmformat != Z_EROFS_COMPRESSION_SHIFTED; | 
|  | 77 | } | 
|  | 78 |  | 
|  | 79 | static void FillCompressedBlockInfo(FilesystemInterface::File* p_file, | 
|  | 80 | std::string_view image_filename, | 
|  | 81 | struct erofs_inode* inode) { | 
|  | 82 | auto& file = *p_file; | 
|  | 83 | if (!file.is_compressed) { | 
|  | 84 | return; | 
|  | 85 | } | 
| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 86 |  | 
|  | 87 | struct erofs_map_blocks block {}; | 
|  | 88 | block.m_la = 0; | 
|  | 89 | block.index = UINT_MAX; | 
|  | 90 |  | 
|  | 91 | const erofs_off_t uncompressed_size = file.file_stat.st_size; | 
|  | 92 | auto& compressed_blocks = file.compressed_file_info.blocks; | 
|  | 93 | auto last_pa = block.m_pa; | 
|  | 94 | auto last_plen = 0; | 
|  | 95 | while (block.m_la < uncompressed_size) { | 
|  | 96 | auto error = ErofsMapBlocks(inode, &block, EROFS_GET_BLOCKS_FIEMAP); | 
|  | 97 | if (error) { | 
|  | 98 | LOG(FATAL) << "Failed to map blocks for " << file.name << " in " | 
|  | 99 | << image_filename; | 
|  | 100 | } | 
|  | 101 | // Certain uncompressed blocks have physical size > logical size. Usually | 
|  | 102 | // the physical block contains bunch of trailing zeros. Include thees | 
|  | 103 | // bytes in the logical size as well. | 
|  | 104 | if (!IsBlockCompressed(block)) { | 
|  | 105 | CHECK_LE(block.m_llen, block.m_plen); | 
|  | 106 | block.m_llen = block.m_plen; | 
|  | 107 | } | 
|  | 108 |  | 
|  | 109 | if (last_pa + last_plen != block.m_pa) { | 
|  | 110 | if (last_plen != 0) { | 
|  | 111 | file.extents.push_back(ExtentForRange( | 
|  | 112 | last_pa / kBlockSize, utils::DivRoundUp(last_plen, kBlockSize))); | 
|  | 113 | } | 
|  | 114 | last_pa = block.m_pa; | 
|  | 115 | last_plen = block.m_plen; | 
|  | 116 | } else { | 
|  | 117 | last_plen += block.m_plen; | 
|  | 118 | } | 
|  | 119 | // If logical size and physical size are the same, this block is | 
|  | 120 | // uncompressed. Join consecutive uncompressed blocks to save a bit memory | 
|  | 121 | // storing metadata. | 
|  | 122 | if (block.m_llen == block.m_plen && !compressed_blocks.empty() && | 
|  | 123 | !compressed_blocks.back().IsCompressed()) { | 
|  | 124 | compressed_blocks.back().compressed_length += block.m_llen; | 
|  | 125 | compressed_blocks.back().uncompressed_length += block.m_llen; | 
|  | 126 | } else { | 
|  | 127 | compressed_blocks.push_back( | 
|  | 128 | CompressedBlock(block.m_la, block.m_plen, block.m_llen)); | 
|  | 129 | } | 
|  | 130 |  | 
|  | 131 | block.m_la += block.m_llen; | 
|  | 132 | } | 
|  | 133 | file.extents.push_back(ExtentForRange( | 
|  | 134 | last_pa / kBlockSize, utils::DivRoundUp(last_plen, kBlockSize))); | 
|  | 135 | return; | 
|  | 136 | } | 
|  | 137 |  | 
|  | 138 | }  // namespace | 
|  | 139 |  | 
|  | 140 | static_assert(kBlockSize == EROFS_BLKSIZ); | 
|  | 141 |  | 
|  | 142 | std::unique_ptr<ErofsFilesystem> ErofsFilesystem::CreateFromFile( | 
| Kelvin Zhang | 8389dfe | 2022-01-13 12:47:11 -0800 | [diff] [blame] | 143 | const std::string& filename, const CompressionAlgorithm& algo) { | 
| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 144 | // erofs-utils makes heavy use of global variables. Hence its functions aren't | 
|  | 145 | // thread safe. For example, it stores a global int holding file descriptors | 
|  | 146 | // to the opened EROFS image. It doesn't even support opening more than 1 | 
|  | 147 | // imaeg at a time. | 
|  | 148 | // TODO(b/202784930) Replace erofs-utils with a cleaner and more C++ friendly | 
|  | 149 | // library. (Or turn erofs-utils into one) | 
|  | 150 | static std::mutex m; | 
|  | 151 | std::lock_guard g{m}; | 
|  | 152 |  | 
|  | 153 | if (const auto err = dev_open_ro(filename.c_str()); err) { | 
|  | 154 | PLOG(INFO) << "Failed to open " << filename; | 
|  | 155 | return nullptr; | 
|  | 156 | } | 
|  | 157 | DEFER { dev_close(); }; | 
|  | 158 |  | 
|  | 159 | if (const auto err = erofs_read_superblock(); err) { | 
|  | 160 | PLOG(INFO) << "Failed to parse " << filename << " as EROFS image"; | 
|  | 161 | return nullptr; | 
|  | 162 | } | 
|  | 163 | struct stat st; | 
|  | 164 | if (const auto err = fstat(erofs_devfd, &st); err) { | 
|  | 165 | PLOG(ERROR) << "Failed to stat() " << filename; | 
|  | 166 | return nullptr; | 
|  | 167 | } | 
|  | 168 | const time_t time = sbi.build_time; | 
|  | 169 | LOG(INFO) << "Parsed EROFS image of size " << st.st_size << " built in " | 
|  | 170 | << ctime(&time) << " " << filename; | 
|  | 171 | std::vector<File> files; | 
| Kelvin Zhang | 8389dfe | 2022-01-13 12:47:11 -0800 | [diff] [blame] | 172 | if (!ErofsFilesystem::GetFiles(filename, &files, algo)) { | 
| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 173 | return nullptr; | 
|  | 174 | } | 
| Kelvin Zhang | 8389dfe | 2022-01-13 12:47:11 -0800 | [diff] [blame] | 175 | LOG(INFO) << "Using compression algo " << algo << " for " << filename; | 
| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 176 | // private ctor doesn't work with make_unique | 
|  | 177 | return std::unique_ptr<ErofsFilesystem>( | 
|  | 178 | new ErofsFilesystem(filename, st.st_size, std::move(files))); | 
|  | 179 | } | 
|  | 180 |  | 
|  | 181 | bool ErofsFilesystem::GetFiles(std::vector<File>* files) const { | 
|  | 182 | *files = files_; | 
|  | 183 | return true; | 
|  | 184 | } | 
|  | 185 |  | 
|  | 186 | bool ErofsFilesystem::GetFiles(const std::string& filename, | 
| Kelvin Zhang | 8389dfe | 2022-01-13 12:47:11 -0800 | [diff] [blame] | 187 | std::vector<File>* files, | 
|  | 188 | const CompressionAlgorithm& algo) { | 
| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 189 | erofs_iterate_root_dir(&sbi, [&](struct erofs_iterate_dir_context* p_info) { | 
|  | 190 | const auto& info = *p_info; | 
|  | 191 | if (info.ctx.de_ftype != EROFS_FT_REG_FILE) { | 
|  | 192 | return 0; | 
|  | 193 | } | 
|  | 194 | struct erofs_inode inode; | 
|  | 195 | inode.nid = info.ctx.de_nid; | 
|  | 196 | int err = erofs_read_inode_from_disk(&inode); | 
|  | 197 | if (err) { | 
|  | 198 | LOG(ERROR) << "Failed to read inode " << inode.nid; | 
|  | 199 | return err; | 
|  | 200 | } | 
|  | 201 | const auto uncompressed_size = inode.i_size; | 
|  | 202 | erofs_off_t compressed_size = 0; | 
|  | 203 | if (uncompressed_size == 0) { | 
|  | 204 | return 0; | 
|  | 205 | } | 
|  | 206 | err = GetOccupiedSize(&inode, &compressed_size); | 
|  | 207 | if (err) { | 
|  | 208 | LOG(FATAL) << "Failed to get occupied size for " << filename; | 
|  | 209 | return err; | 
|  | 210 | } | 
|  | 211 | // If data is packed inline, likely this node is stored on block unalighed | 
|  | 212 | // addresses. OTA doesn't work for non-block aligned files. All blocks not | 
|  | 213 | // reported by |GetFiles| will be updated in 1 operation. Ignore inline | 
|  | 214 | // files for now. | 
|  | 215 | // TODO(b/206729162) Support un-aligned files. | 
|  | 216 | if (inode.datalayout == EROFS_INODE_FLAT_INLINE) { | 
|  | 217 | return 0; | 
|  | 218 | } | 
|  | 219 |  | 
|  | 220 | File file; | 
|  | 221 | file.name = info.path; | 
|  | 222 | file.compressed_file_info.zero_padding_enabled = | 
|  | 223 | erofs_sb_has_lz4_0padding(); | 
|  | 224 | file.is_compressed = compressed_size != uncompressed_size; | 
|  | 225 |  | 
|  | 226 | file.file_stat.st_size = uncompressed_size; | 
|  | 227 | file.file_stat.st_ino = inode.nid; | 
|  | 228 | FillCompressedBlockInfo(&file, filename, &inode); | 
| Kelvin Zhang | 8389dfe | 2022-01-13 12:47:11 -0800 | [diff] [blame] | 229 | file.compressed_file_info.algo = algo; | 
| Kelvin Zhang | 446989a | 2021-12-08 13:49:07 -0800 | [diff] [blame] | 230 |  | 
|  | 231 | files->emplace_back(std::move(file)); | 
|  | 232 | return 0; | 
|  | 233 | }); | 
|  | 234 |  | 
|  | 235 | for (auto& file : *files) { | 
|  | 236 | NormalizeExtents(&file.extents); | 
|  | 237 | } | 
|  | 238 | return true; | 
|  | 239 | } | 
|  | 240 |  | 
|  | 241 | }  // namespace chromeos_update_engine |