delta_diff_utils.cc revision 49fdb09d1ff2e078fa44c26540a5b82900df0f9a
1// 2// Copyright (C) 2015 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/delta_diff_utils.h" 18 19#include <endian.h> 20// TODO: Remove these pragmas when b/35721782 is fixed. 21#pragma clang diagnostic push 22#pragma clang diagnostic ignored "-Wmacro-redefined" 23#include <ext2fs/ext2fs.h> 24#pragma clang diagnostic pop 25#include <unistd.h> 26 27#include <algorithm> 28#include <map> 29 30#include <base/files/file_util.h> 31#include <base/format_macros.h> 32#include <base/strings/stringprintf.h> 33#include <base/threading/simple_thread.h> 34#include <brillo/data_encoding.h> 35 36#include "update_engine/common/hash_calculator.h" 37#include "update_engine/common/subprocess.h" 38#include "update_engine/common/utils.h" 39#include "update_engine/payload_generator/block_mapping.h" 40#include "update_engine/payload_generator/bzip.h" 41#include "update_engine/payload_generator/delta_diff_generator.h" 42#include "update_engine/payload_generator/extent_ranges.h" 43#include "update_engine/payload_generator/extent_utils.h" 44#include "update_engine/payload_generator/xz.h" 45 46using std::map; 47using std::string; 48using std::vector; 49 50namespace chromeos_update_engine { 51namespace { 52 53const char* const kBsdiffPath = "bsdiff"; 54 55// The maximum destination size allowed for bsdiff. In general, bsdiff should 56// work for arbitrary big files, but the payload generation and payload 57// application requires a significant amount of RAM. We put a hard-limit of 58// 200 MiB that should not affect any released board, but will limit the 59// Chrome binary in ASan builders. 60const uint64_t kMaxBsdiffDestinationSize = 200 * 1024 * 1024; // bytes 61 62// The maximum destination size allowed for puffdiff. In general, puffdiff 63// should work for arbitrary big files, but the payload application is quite 64// memory intensive, so we limit these operations to 50 MiB. 65const uint64_t kMaxPuffdiffDestinationSize = 50 * 1024 * 1024; // bytes 66 67// Process a range of blocks from |range_start| to |range_end| in the extent at 68// position |*idx_p| of |extents|. If |do_remove| is true, this range will be 69// removed, which may cause the extent to be trimmed, split or removed entirely. 70// The value of |*idx_p| is updated to point to the next extent to be processed. 71// Returns true iff the next extent to process is a new or updated one. 72bool ProcessExtentBlockRange(vector<Extent>* extents, size_t* idx_p, 73 const bool do_remove, uint64_t range_start, 74 uint64_t range_end) { 75 size_t idx = *idx_p; 76 uint64_t start_block = (*extents)[idx].start_block(); 77 uint64_t num_blocks = (*extents)[idx].num_blocks(); 78 uint64_t range_size = range_end - range_start; 79 80 if (do_remove) { 81 if (range_size == num_blocks) { 82 // Remove the entire extent. 83 extents->erase(extents->begin() + idx); 84 } else if (range_end == num_blocks) { 85 // Trim the end of the extent. 86 (*extents)[idx].set_num_blocks(num_blocks - range_size); 87 idx++; 88 } else if (range_start == 0) { 89 // Trim the head of the extent. 90 (*extents)[idx].set_start_block(start_block + range_size); 91 (*extents)[idx].set_num_blocks(num_blocks - range_size); 92 } else { 93 // Trim the middle, splitting the remainder into two parts. 94 (*extents)[idx].set_num_blocks(range_start); 95 Extent e; 96 e.set_start_block(start_block + range_end); 97 e.set_num_blocks(num_blocks - range_end); 98 idx++; 99 extents->insert(extents->begin() + idx, e); 100 } 101 } else if (range_end == num_blocks) { 102 // Done with this extent. 103 idx++; 104 } else { 105 return false; 106 } 107 108 *idx_p = idx; 109 return true; 110} 111 112// Remove identical corresponding block ranges in |src_extents| and 113// |dst_extents|. Used for preventing moving of blocks onto themselves during 114// MOVE operations. The value of |total_bytes| indicates the actual length of 115// content; this may be slightly less than the total size of blocks, in which 116// case the last block is only partly occupied with data. Returns the total 117// number of bytes removed. 118size_t RemoveIdenticalBlockRanges(vector<Extent>* src_extents, 119 vector<Extent>* dst_extents, 120 const size_t total_bytes) { 121 size_t src_idx = 0; 122 size_t dst_idx = 0; 123 uint64_t src_offset = 0, dst_offset = 0; 124 size_t removed_bytes = 0, nonfull_block_bytes; 125 bool do_remove = false; 126 while (src_idx < src_extents->size() && dst_idx < dst_extents->size()) { 127 do_remove = ((*src_extents)[src_idx].start_block() + src_offset == 128 (*dst_extents)[dst_idx].start_block() + dst_offset); 129 130 uint64_t src_num_blocks = (*src_extents)[src_idx].num_blocks(); 131 uint64_t dst_num_blocks = (*dst_extents)[dst_idx].num_blocks(); 132 uint64_t min_num_blocks = std::min(src_num_blocks - src_offset, 133 dst_num_blocks - dst_offset); 134 uint64_t prev_src_offset = src_offset; 135 uint64_t prev_dst_offset = dst_offset; 136 src_offset += min_num_blocks; 137 dst_offset += min_num_blocks; 138 139 bool new_src = ProcessExtentBlockRange(src_extents, &src_idx, do_remove, 140 prev_src_offset, src_offset); 141 bool new_dst = ProcessExtentBlockRange(dst_extents, &dst_idx, do_remove, 142 prev_dst_offset, dst_offset); 143 if (new_src) { 144 src_offset = 0; 145 } 146 if (new_dst) { 147 dst_offset = 0; 148 } 149 150 if (do_remove) 151 removed_bytes += min_num_blocks * kBlockSize; 152 } 153 154 // If we removed the last block and this block is only partly used by file 155 // content, deduct the unused portion from the total removed byte count. 156 if (do_remove && (nonfull_block_bytes = total_bytes % kBlockSize)) 157 removed_bytes -= kBlockSize - nonfull_block_bytes; 158 159 return removed_bytes; 160} 161 162} // namespace 163 164namespace diff_utils { 165 166// This class encapsulates a file delta processing thread work. The 167// processor computes the delta between the source and target files; 168// and write the compressed delta to the blob. 169class FileDeltaProcessor : public base::DelegateSimpleThread::Delegate { 170 public: 171 FileDeltaProcessor(const string& old_part, 172 const string& new_part, 173 const PayloadVersion& version, 174 const vector<Extent>& old_extents, 175 const vector<Extent>& new_extents, 176 const string& name, 177 ssize_t chunk_blocks, 178 BlobFileWriter* blob_file) 179 : old_part_(old_part), 180 new_part_(new_part), 181 version_(version), 182 old_extents_(old_extents), 183 new_extents_(new_extents), 184 name_(name), 185 chunk_blocks_(chunk_blocks), 186 blob_file_(blob_file) {} 187 188 FileDeltaProcessor(FileDeltaProcessor&& processor) = default; 189 190 ~FileDeltaProcessor() override = default; 191 192 // Overrides DelegateSimpleThread::Delegate. 193 // Calculate the list of operations and write their corresponding deltas to 194 // the blob_file. 195 void Run() override; 196 197 // Merge each file processor's ops list to aops. 198 void MergeOperation(vector<AnnotatedOperation>* aops); 199 200 private: 201 const string& old_part_; 202 const string& new_part_; 203 const PayloadVersion& version_; 204 205 // The block ranges of the old/new file within the src/tgt image 206 const vector<Extent> old_extents_; 207 const vector<Extent> new_extents_; 208 const string name_; 209 // Block limit of one aop. 210 ssize_t chunk_blocks_; 211 BlobFileWriter* blob_file_; 212 213 // The list of ops to reach the new file from the old file. 214 vector<AnnotatedOperation> file_aops_; 215 216 DISALLOW_COPY_AND_ASSIGN(FileDeltaProcessor); 217}; 218 219void FileDeltaProcessor::Run() { 220 TEST_AND_RETURN(blob_file_ != nullptr); 221 222 if (!DeltaReadFile(&file_aops_, 223 old_part_, 224 new_part_, 225 old_extents_, 226 new_extents_, 227 name_, 228 chunk_blocks_, 229 version_, 230 blob_file_)) { 231 LOG(ERROR) << "Failed to generate delta for " << name_ << " (" 232 << BlocksInExtents(new_extents_) << " blocks)"; 233 } 234} 235 236void FileDeltaProcessor::MergeOperation(vector<AnnotatedOperation>* aops) { 237 aops->reserve(aops->size() + file_aops_.size()); 238 std::move(file_aops_.begin(), file_aops_.end(), std::back_inserter(*aops)); 239} 240 241bool DeltaReadPartition(vector<AnnotatedOperation>* aops, 242 const PartitionConfig& old_part, 243 const PartitionConfig& new_part, 244 ssize_t hard_chunk_blocks, 245 size_t soft_chunk_blocks, 246 const PayloadVersion& version, 247 BlobFileWriter* blob_file) { 248 ExtentRanges old_visited_blocks; 249 ExtentRanges new_visited_blocks; 250 251 TEST_AND_RETURN_FALSE(DeltaMovedAndZeroBlocks( 252 aops, 253 old_part.path, 254 new_part.path, 255 old_part.size / kBlockSize, 256 new_part.size / kBlockSize, 257 soft_chunk_blocks, 258 version, 259 blob_file, 260 &old_visited_blocks, 261 &new_visited_blocks)); 262 263 map<string, vector<Extent>> old_files_map; 264 if (old_part.fs_interface) { 265 vector<FilesystemInterface::File> old_files; 266 old_part.fs_interface->GetFiles(&old_files); 267 for (const FilesystemInterface::File& file : old_files) 268 old_files_map[file.name] = file.extents; 269 } 270 271 TEST_AND_RETURN_FALSE(new_part.fs_interface); 272 vector<FilesystemInterface::File> new_files; 273 new_part.fs_interface->GetFiles(&new_files); 274 275 vector<FileDeltaProcessor> file_delta_processors; 276 277 // The processing is very straightforward here, we generate operations for 278 // every file (and pseudo-file such as the metadata) in the new filesystem 279 // based on the file with the same name in the old filesystem, if any. 280 // Files with overlapping data blocks (like hardlinks or filesystems with tail 281 // packing or compression where the blocks store more than one file) are only 282 // generated once in the new image, but are also used only once from the old 283 // image due to some simplifications (see below). 284 for (const FilesystemInterface::File& new_file : new_files) { 285 // Ignore the files in the new filesystem without blocks. Symlinks with 286 // data blocks (for example, symlinks bigger than 60 bytes in ext2) are 287 // handled as normal files. We also ignore blocks that were already 288 // processed by a previous file. 289 vector<Extent> new_file_extents = FilterExtentRanges( 290 new_file.extents, new_visited_blocks); 291 new_visited_blocks.AddExtents(new_file_extents); 292 293 if (new_file_extents.empty()) 294 continue; 295 296 LOG(INFO) << "Encoding file " << new_file.name << " (" 297 << BlocksInExtents(new_file_extents) << " blocks)"; 298 299 // We can't visit each dst image inode more than once, as that would 300 // duplicate work. Here, we avoid visiting each source image inode 301 // more than once. Technically, we could have multiple operations 302 // that read the same blocks from the source image for diffing, but 303 // we choose not to avoid complexity. Eventually we will move away 304 // from using a graph/cycle detection/etc to generate diffs, and at that 305 // time, it will be easy (non-complex) to have many operations read 306 // from the same source blocks. At that time, this code can die. -adlr 307 vector<Extent> old_file_extents = FilterExtentRanges( 308 old_files_map[new_file.name], old_visited_blocks); 309 old_visited_blocks.AddExtents(old_file_extents); 310 311 file_delta_processors.emplace_back(old_part.path, 312 new_part.path, 313 version, 314 std::move(old_file_extents), 315 std::move(new_file_extents), 316 new_file.name, // operation name 317 hard_chunk_blocks, 318 blob_file); 319 } 320 321 size_t max_threads = GetMaxThreads(); 322 base::DelegateSimpleThreadPool thread_pool("incremental-update-generator", 323 max_threads); 324 thread_pool.Start(); 325 for (auto& processor : file_delta_processors) { 326 thread_pool.AddWork(&processor); 327 } 328 thread_pool.JoinAll(); 329 330 for (auto& processor : file_delta_processors) { 331 processor.MergeOperation(aops); 332 } 333 334 // Process all the blocks not included in any file. We provided all the unused 335 // blocks in the old partition as available data. 336 vector<Extent> new_unvisited = { 337 ExtentForRange(0, new_part.size / kBlockSize)}; 338 new_unvisited = FilterExtentRanges(new_unvisited, new_visited_blocks); 339 if (new_unvisited.empty()) 340 return true; 341 342 vector<Extent> old_unvisited; 343 if (old_part.fs_interface) { 344 old_unvisited.push_back(ExtentForRange(0, old_part.size / kBlockSize)); 345 old_unvisited = FilterExtentRanges(old_unvisited, old_visited_blocks); 346 } 347 348 LOG(INFO) << "Scanning " << BlocksInExtents(new_unvisited) 349 << " unwritten blocks using chunk size of " 350 << soft_chunk_blocks << " blocks."; 351 // We use the soft_chunk_blocks limit for the <non-file-data> as we don't 352 // really know the structure of this data and we should not expect it to have 353 // redundancy between partitions. 354 TEST_AND_RETURN_FALSE(DeltaReadFile(aops, 355 old_part.path, 356 new_part.path, 357 old_unvisited, 358 new_unvisited, 359 "<non-file-data>", // operation name 360 soft_chunk_blocks, 361 version, 362 blob_file)); 363 364 return true; 365} 366 367bool DeltaMovedAndZeroBlocks(vector<AnnotatedOperation>* aops, 368 const string& old_part, 369 const string& new_part, 370 size_t old_num_blocks, 371 size_t new_num_blocks, 372 ssize_t chunk_blocks, 373 const PayloadVersion& version, 374 BlobFileWriter* blob_file, 375 ExtentRanges* old_visited_blocks, 376 ExtentRanges* new_visited_blocks) { 377 vector<BlockMapping::BlockId> old_block_ids; 378 vector<BlockMapping::BlockId> new_block_ids; 379 TEST_AND_RETURN_FALSE(MapPartitionBlocks(old_part, 380 new_part, 381 old_num_blocks * kBlockSize, 382 new_num_blocks * kBlockSize, 383 kBlockSize, 384 &old_block_ids, 385 &new_block_ids)); 386 387 // If the update is inplace, we map all the blocks that didn't move, 388 // regardless of the contents since they are already copied and no operation 389 // is required. 390 if (version.InplaceUpdate()) { 391 uint64_t num_blocks = std::min(old_num_blocks, new_num_blocks); 392 for (uint64_t block = 0; block < num_blocks; block++) { 393 if (old_block_ids[block] == new_block_ids[block] && 394 !old_visited_blocks->ContainsBlock(block) && 395 !new_visited_blocks->ContainsBlock(block)) { 396 old_visited_blocks->AddBlock(block); 397 new_visited_blocks->AddBlock(block); 398 } 399 } 400 } 401 402 // A mapping from the block_id to the list of block numbers with that block id 403 // in the old partition. This is used to lookup where in the old partition 404 // is a block from the new partition. 405 map<BlockMapping::BlockId, vector<uint64_t>> old_blocks_map; 406 407 for (uint64_t block = old_num_blocks; block-- > 0; ) { 408 if (old_block_ids[block] != 0 && !old_visited_blocks->ContainsBlock(block)) 409 old_blocks_map[old_block_ids[block]].push_back(block); 410 411 // Mark all zeroed blocks in the old image as "used" since it doesn't make 412 // any sense to spend I/O to read zeros from the source partition and more 413 // importantly, these could sometimes be blocks discarded in the SSD which 414 // would read non-zero values. 415 if (old_block_ids[block] == 0) 416 old_visited_blocks->AddBlock(block); 417 } 418 419 // The collection of blocks in the new partition with just zeros. This is a 420 // common case for free-space that's also problematic for bsdiff, so we want 421 // to optimize it using REPLACE_BZ operations. The blob for a REPLACE_BZ of 422 // just zeros is so small that it doesn't make sense to spend the I/O reading 423 // zeros from the old partition. 424 vector<Extent> new_zeros; 425 426 vector<Extent> old_identical_blocks; 427 vector<Extent> new_identical_blocks; 428 429 for (uint64_t block = 0; block < new_num_blocks; block++) { 430 // Only produce operations for blocks that were not yet visited. 431 if (new_visited_blocks->ContainsBlock(block)) 432 continue; 433 if (new_block_ids[block] == 0) { 434 AppendBlockToExtents(&new_zeros, block); 435 continue; 436 } 437 438 auto old_blocks_map_it = old_blocks_map.find(new_block_ids[block]); 439 // Check if the block exists in the old partition at all. 440 if (old_blocks_map_it == old_blocks_map.end() || 441 old_blocks_map_it->second.empty()) 442 continue; 443 444 AppendBlockToExtents(&old_identical_blocks, 445 old_blocks_map_it->second.back()); 446 AppendBlockToExtents(&new_identical_blocks, block); 447 // We can't reuse source blocks in minor version 1 because the cycle 448 // breaking algorithm used in the in-place update doesn't support that. 449 if (version.InplaceUpdate()) 450 old_blocks_map_it->second.pop_back(); 451 } 452 453 // Produce operations for the zero blocks split per output extent. 454 // TODO(deymo): Produce ZERO operations instead of calling DeltaReadFile(). 455 size_t num_ops = aops->size(); 456 new_visited_blocks->AddExtents(new_zeros); 457 for (const Extent& extent : new_zeros) { 458 TEST_AND_RETURN_FALSE(DeltaReadFile(aops, 459 "", 460 new_part, 461 vector<Extent>(), // old_extents 462 vector<Extent>{extent}, // new_extents 463 "<zeros>", 464 chunk_blocks, 465 version, 466 blob_file)); 467 } 468 LOG(INFO) << "Produced " << (aops->size() - num_ops) << " operations for " 469 << BlocksInExtents(new_zeros) << " zeroed blocks"; 470 471 // Produce MOVE/SOURCE_COPY operations for the moved blocks. 472 num_ops = aops->size(); 473 if (chunk_blocks == -1) 474 chunk_blocks = new_num_blocks; 475 uint64_t used_blocks = 0; 476 old_visited_blocks->AddExtents(old_identical_blocks); 477 new_visited_blocks->AddExtents(new_identical_blocks); 478 for (const Extent& extent : new_identical_blocks) { 479 // We split the operation at the extent boundary or when bigger than 480 // chunk_blocks. 481 for (uint64_t op_block_offset = 0; op_block_offset < extent.num_blocks(); 482 op_block_offset += chunk_blocks) { 483 aops->emplace_back(); 484 AnnotatedOperation* aop = &aops->back(); 485 aop->name = "<identical-blocks>"; 486 aop->op.set_type(version.OperationAllowed(InstallOperation::SOURCE_COPY) 487 ? InstallOperation::SOURCE_COPY 488 : InstallOperation::MOVE); 489 490 uint64_t chunk_num_blocks = 491 std::min(static_cast<uint64_t>(extent.num_blocks()) - op_block_offset, 492 static_cast<uint64_t>(chunk_blocks)); 493 494 // The current operation represents the move/copy operation for the 495 // sublist starting at |used_blocks| of length |chunk_num_blocks| where 496 // the src and dst are from |old_identical_blocks| and 497 // |new_identical_blocks| respectively. 498 StoreExtents( 499 ExtentsSublist(old_identical_blocks, used_blocks, chunk_num_blocks), 500 aop->op.mutable_src_extents()); 501 502 Extent* op_dst_extent = aop->op.add_dst_extents(); 503 op_dst_extent->set_start_block(extent.start_block() + op_block_offset); 504 op_dst_extent->set_num_blocks(chunk_num_blocks); 505 CHECK( 506 vector<Extent>{*op_dst_extent} == // NOLINT(whitespace/braces) 507 ExtentsSublist(new_identical_blocks, used_blocks, chunk_num_blocks)); 508 509 used_blocks += chunk_num_blocks; 510 } 511 } 512 LOG(INFO) << "Produced " << (aops->size() - num_ops) << " operations for " 513 << used_blocks << " identical blocks moved"; 514 515 return true; 516} 517 518bool DeltaReadFile(vector<AnnotatedOperation>* aops, 519 const string& old_part, 520 const string& new_part, 521 const vector<Extent>& old_extents, 522 const vector<Extent>& new_extents, 523 const string& name, 524 ssize_t chunk_blocks, 525 const PayloadVersion& version, 526 BlobFileWriter* blob_file) { 527 brillo::Blob data; 528 InstallOperation operation; 529 530 uint64_t total_blocks = BlocksInExtents(new_extents); 531 if (chunk_blocks == -1) 532 chunk_blocks = total_blocks; 533 534 for (uint64_t block_offset = 0; block_offset < total_blocks; 535 block_offset += chunk_blocks) { 536 // Split the old/new file in the same chunks. Note that this could drop 537 // some information from the old file used for the new chunk. If the old 538 // file is smaller (or even empty when there's no old file) the chunk will 539 // also be empty. 540 vector<Extent> old_extents_chunk = ExtentsSublist( 541 old_extents, block_offset, chunk_blocks); 542 vector<Extent> new_extents_chunk = ExtentsSublist( 543 new_extents, block_offset, chunk_blocks); 544 NormalizeExtents(&old_extents_chunk); 545 NormalizeExtents(&new_extents_chunk); 546 547 TEST_AND_RETURN_FALSE(ReadExtentsToDiff(old_part, 548 new_part, 549 old_extents_chunk, 550 new_extents_chunk, 551 version, 552 &data, 553 &operation)); 554 555 // Check if the operation writes nothing. 556 if (operation.dst_extents_size() == 0) { 557 if (operation.type() == InstallOperation::MOVE) { 558 LOG(INFO) << "Empty MOVE operation (" 559 << name << "), skipping"; 560 continue; 561 } else { 562 LOG(ERROR) << "Empty non-MOVE operation"; 563 return false; 564 } 565 } 566 567 // Now, insert into the list of operations. 568 AnnotatedOperation aop; 569 aop.name = name; 570 if (static_cast<uint64_t>(chunk_blocks) < total_blocks) { 571 aop.name = base::StringPrintf("%s:%" PRIu64, 572 name.c_str(), block_offset / chunk_blocks); 573 } 574 aop.op = operation; 575 576 // Write the data 577 TEST_AND_RETURN_FALSE(aop.SetOperationBlob(data, blob_file)); 578 aops->emplace_back(aop); 579 } 580 return true; 581} 582 583bool GenerateBestFullOperation(const brillo::Blob& new_data, 584 const PayloadVersion& version, 585 brillo::Blob* out_blob, 586 InstallOperation_Type* out_type) { 587 if (new_data.empty()) 588 return false; 589 590 if (version.OperationAllowed(InstallOperation::ZERO) && 591 std::all_of( 592 new_data.begin(), new_data.end(), [](uint8_t x) { return x == 0; })) { 593 // The read buffer is all zeros, so produce a ZERO operation. No need to 594 // check other types of operations in this case. 595 *out_blob = brillo::Blob(); 596 *out_type = InstallOperation::ZERO; 597 return true; 598 } 599 600 bool out_blob_set = false; 601 602 // Try compressing |new_data| with xz first. 603 if (version.OperationAllowed(InstallOperation::REPLACE_XZ)) { 604 brillo::Blob new_data_xz; 605 if (XzCompress(new_data, &new_data_xz) && !new_data_xz.empty()) { 606 *out_type = InstallOperation::REPLACE_XZ; 607 *out_blob = std::move(new_data_xz); 608 out_blob_set = true; 609 } 610 } 611 612 // Try compressing it with bzip2. 613 if (version.OperationAllowed(InstallOperation::REPLACE_BZ)) { 614 brillo::Blob new_data_bz; 615 // TODO(deymo): Implement some heuristic to determine if it is worth trying 616 // to compress the blob with bzip2 if we already have a good REPLACE_XZ. 617 if (BzipCompress(new_data, &new_data_bz) && !new_data_bz.empty() && 618 (!out_blob_set || out_blob->size() > new_data_bz.size())) { 619 // A REPLACE_BZ is better or nothing else was set. 620 *out_type = InstallOperation::REPLACE_BZ; 621 *out_blob = std::move(new_data_bz); 622 out_blob_set = true; 623 } 624 } 625 626 // If nothing else worked or it was badly compressed we try a REPLACE. 627 if (!out_blob_set || out_blob->size() >= new_data.size()) { 628 *out_type = InstallOperation::REPLACE; 629 // This needs to make a copy of the data in the case bzip or xz didn't 630 // compress well, which is not the common case so the performance hit is 631 // low. 632 *out_blob = new_data; 633 } 634 return true; 635} 636 637bool ReadExtentsToDiff(const string& old_part, 638 const string& new_part, 639 const vector<Extent>& old_extents, 640 const vector<Extent>& new_extents, 641 const PayloadVersion& version, 642 brillo::Blob* out_data, 643 InstallOperation* out_op) { 644 InstallOperation operation; 645 646 // We read blocks from old_extents and write blocks to new_extents. 647 uint64_t blocks_to_read = BlocksInExtents(old_extents); 648 uint64_t blocks_to_write = BlocksInExtents(new_extents); 649 650 // Disable bsdiff, and puffdiff when the data is too big. 651 bool bsdiff_allowed = 652 version.OperationAllowed(InstallOperation::SOURCE_BSDIFF) || 653 version.OperationAllowed(InstallOperation::BSDIFF); 654 if (bsdiff_allowed && 655 blocks_to_read * kBlockSize > kMaxBsdiffDestinationSize) { 656 LOG(INFO) << "bsdiff blacklisted, data too big: " 657 << blocks_to_read * kBlockSize << " bytes"; 658 bsdiff_allowed = false; 659 } 660 661 bool puffdiff_allowed = version.OperationAllowed(InstallOperation::PUFFDIFF); 662 if (puffdiff_allowed && 663 blocks_to_read * kBlockSize > kMaxPuffdiffDestinationSize) { 664 LOG(INFO) << "puffdiff blacklisted, data too big: " 665 << blocks_to_read * kBlockSize << " bytes"; 666 puffdiff_allowed = false; 667 } 668 669 // Make copies of the extents so we can modify them. 670 vector<Extent> src_extents = old_extents; 671 vector<Extent> dst_extents = new_extents; 672 673 // Read in bytes from new data. 674 brillo::Blob new_data; 675 TEST_AND_RETURN_FALSE(utils::ReadExtents(new_part, 676 new_extents, 677 &new_data, 678 kBlockSize * blocks_to_write, 679 kBlockSize)); 680 TEST_AND_RETURN_FALSE(!new_data.empty()); 681 682 // Data blob that will be written to delta file. 683 brillo::Blob data_blob; 684 685 // Try generating a full operation for the given new data, regardless of the 686 // old_data. 687 InstallOperation_Type op_type; 688 TEST_AND_RETURN_FALSE( 689 GenerateBestFullOperation(new_data, version, &data_blob, &op_type)); 690 operation.set_type(op_type); 691 692 brillo::Blob old_data; 693 if (blocks_to_read > 0) { 694 // Read old data. 695 TEST_AND_RETURN_FALSE( 696 utils::ReadExtents(old_part, src_extents, &old_data, 697 kBlockSize * blocks_to_read, kBlockSize)); 698 if (old_data == new_data) { 699 // No change in data. 700 operation.set_type(version.OperationAllowed(InstallOperation::SOURCE_COPY) 701 ? InstallOperation::SOURCE_COPY 702 : InstallOperation::MOVE); 703 data_blob = brillo::Blob(); 704 } else if (bsdiff_allowed || puffdiff_allowed) { 705 // If the source file is considered bsdiff safe (no bsdiff bugs 706 // triggered), see if BSDIFF encoding is smaller. 707 base::FilePath old_chunk; 708 TEST_AND_RETURN_FALSE(base::CreateTemporaryFile(&old_chunk)); 709 ScopedPathUnlinker old_unlinker(old_chunk.value()); 710 TEST_AND_RETURN_FALSE(utils::WriteFile( 711 old_chunk.value().c_str(), old_data.data(), old_data.size())); 712 base::FilePath new_chunk; 713 TEST_AND_RETURN_FALSE(base::CreateTemporaryFile(&new_chunk)); 714 ScopedPathUnlinker new_unlinker(new_chunk.value()); 715 TEST_AND_RETURN_FALSE(utils::WriteFile( 716 new_chunk.value().c_str(), new_data.data(), new_data.size())); 717 718 if (bsdiff_allowed) { 719 brillo::Blob bsdiff_delta; 720 TEST_AND_RETURN_FALSE(DiffFiles( 721 kBsdiffPath, old_chunk.value(), new_chunk.value(), &bsdiff_delta)); 722 CHECK_GT(bsdiff_delta.size(), static_cast<brillo::Blob::size_type>(0)); 723 if (bsdiff_delta.size() < data_blob.size()) { 724 operation.set_type( 725 version.OperationAllowed(InstallOperation::SOURCE_BSDIFF) 726 ? InstallOperation::SOURCE_BSDIFF 727 : InstallOperation::BSDIFF); 728 data_blob = std::move(bsdiff_delta); 729 } 730 } 731 if (puffdiff_allowed) { 732 LOG(ERROR) << "puffdiff is not supported yet!"; 733 return false; 734 } 735 } 736 } 737 738 size_t removed_bytes = 0; 739 // Remove identical src/dst block ranges in MOVE operations. 740 if (operation.type() == InstallOperation::MOVE) { 741 removed_bytes = RemoveIdenticalBlockRanges( 742 &src_extents, &dst_extents, new_data.size()); 743 } 744 // Set legacy src_length and dst_length fields. 745 operation.set_src_length(old_data.size() - removed_bytes); 746 operation.set_dst_length(new_data.size() - removed_bytes); 747 748 // Embed extents in the operation. 749 StoreExtents(src_extents, operation.mutable_src_extents()); 750 StoreExtents(dst_extents, operation.mutable_dst_extents()); 751 752 // Replace operations should not have source extents. 753 if (IsAReplaceOperation(operation.type())) { 754 operation.clear_src_extents(); 755 operation.clear_src_length(); 756 } 757 758 *out_data = std::move(data_blob); 759 *out_op = operation; 760 return true; 761} 762 763// Runs the bsdiff tool in |diff_path| on two files and returns the 764// resulting delta in |out|. Returns true on success. 765bool DiffFiles(const string& diff_path, 766 const string& old_file, 767 const string& new_file, 768 brillo::Blob* out) { 769 const string kPatchFile = "delta.patchXXXXXX"; 770 string patch_file_path; 771 772 TEST_AND_RETURN_FALSE( 773 utils::MakeTempFile(kPatchFile, &patch_file_path, nullptr)); 774 775 vector<string> cmd; 776 cmd.push_back(diff_path); 777 cmd.push_back(old_file); 778 cmd.push_back(new_file); 779 cmd.push_back(patch_file_path); 780 781 int rc = 1; 782 string stdout; 783 TEST_AND_RETURN_FALSE(Subprocess::SynchronousExec(cmd, &rc, &stdout)); 784 if (rc != 0) { 785 LOG(ERROR) << diff_path << " returned " << rc << std::endl << stdout; 786 return false; 787 } 788 TEST_AND_RETURN_FALSE(utils::ReadFile(patch_file_path, out)); 789 unlink(patch_file_path.c_str()); 790 return true; 791} 792 793bool IsAReplaceOperation(InstallOperation_Type op_type) { 794 return (op_type == InstallOperation::REPLACE || 795 op_type == InstallOperation::REPLACE_BZ || 796 op_type == InstallOperation::REPLACE_XZ); 797} 798 799// Returns true if |op| is a no-op operation that doesn't do any useful work 800// (e.g., a move operation that copies blocks onto themselves). 801bool IsNoopOperation(const InstallOperation& op) { 802 return (op.type() == InstallOperation::MOVE && 803 ExpandExtents(op.src_extents()) == ExpandExtents(op.dst_extents())); 804} 805 806void FilterNoopOperations(vector<AnnotatedOperation>* ops) { 807 ops->erase( 808 std::remove_if( 809 ops->begin(), ops->end(), 810 [](const AnnotatedOperation& aop){return IsNoopOperation(aop.op);}), 811 ops->end()); 812} 813 814bool InitializePartitionInfo(const PartitionConfig& part, PartitionInfo* info) { 815 info->set_size(part.size); 816 HashCalculator hasher; 817 TEST_AND_RETURN_FALSE(hasher.UpdateFile(part.path, part.size) == 818 static_cast<off_t>(part.size)); 819 TEST_AND_RETURN_FALSE(hasher.Finalize()); 820 const brillo::Blob& hash = hasher.raw_hash(); 821 info->set_hash(hash.data(), hash.size()); 822 LOG(INFO) << part.path << ": size=" << part.size 823 << " hash=" << brillo::data_encoding::Base64Encode(hash); 824 return true; 825} 826 827bool CompareAopsByDestination(AnnotatedOperation first_aop, 828 AnnotatedOperation second_aop) { 829 // We want empty operations to be at the end of the payload. 830 if (!first_aop.op.dst_extents().size() || !second_aop.op.dst_extents().size()) 831 return ((!first_aop.op.dst_extents().size()) < 832 (!second_aop.op.dst_extents().size())); 833 uint32_t first_dst_start = first_aop.op.dst_extents(0).start_block(); 834 uint32_t second_dst_start = second_aop.op.dst_extents(0).start_block(); 835 return first_dst_start < second_dst_start; 836} 837 838bool IsExtFilesystem(const string& device) { 839 brillo::Blob header; 840 // See include/linux/ext2_fs.h for more details on the structure. We obtain 841 // ext2 constants from ext2fs/ext2fs.h header but we don't link with the 842 // library. 843 if (!utils::ReadFileChunk( 844 device, 0, SUPERBLOCK_OFFSET + SUPERBLOCK_SIZE, &header) || 845 header.size() < SUPERBLOCK_OFFSET + SUPERBLOCK_SIZE) 846 return false; 847 848 const uint8_t* superblock = header.data() + SUPERBLOCK_OFFSET; 849 850 // ext3_fs.h: ext3_super_block.s_blocks_count 851 uint32_t block_count = 852 *reinterpret_cast<const uint32_t*>(superblock + 1 * sizeof(int32_t)); 853 854 // ext3_fs.h: ext3_super_block.s_log_block_size 855 uint32_t log_block_size = 856 *reinterpret_cast<const uint32_t*>(superblock + 6 * sizeof(int32_t)); 857 858 // ext3_fs.h: ext3_super_block.s_magic 859 uint16_t magic = 860 *reinterpret_cast<const uint16_t*>(superblock + 14 * sizeof(int32_t)); 861 862 block_count = le32toh(block_count); 863 log_block_size = le32toh(log_block_size) + EXT2_MIN_BLOCK_LOG_SIZE; 864 magic = le16toh(magic); 865 866 if (magic != EXT2_SUPER_MAGIC) 867 return false; 868 869 // Sanity check the parameters. 870 TEST_AND_RETURN_FALSE(log_block_size >= EXT2_MIN_BLOCK_LOG_SIZE && 871 log_block_size <= EXT2_MAX_BLOCK_LOG_SIZE); 872 TEST_AND_RETURN_FALSE(block_count > 0); 873 return true; 874} 875 876// Return the number of CPUs on the machine, and 4 threads in minimum. 877size_t GetMaxThreads() { 878 return std::max(sysconf(_SC_NPROCESSORS_ONLN), 4L); 879} 880 881} // namespace diff_utils 882 883} // namespace chromeos_update_engine 884