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 #if defined(__clang__)
21 // TODO(*): Remove these pragmas when b/35721782 is fixed.
22 #pragma clang diagnostic push
23 #pragma clang diagnostic ignored "-Wmacro-redefined"
24 #endif
25 #include <ext2fs/ext2fs.h>
26 #if defined(__clang__)
27 #pragma clang diagnostic pop
28 #endif
29 #include <unistd.h>
30 
31 #include <algorithm>
32 #include <functional>
33 #include <list>
34 #include <map>
35 #include <memory>
36 #include <numeric>
37 #include <utility>
38 
39 #include <base/files/file_util.h>
40 #include <base/format_macros.h>
41 #include <base/strings/string_util.h>
42 #include <base/strings/stringprintf.h>
43 #include <base/threading/simple_thread.h>
44 #include <base/time/time.h>
45 #include <brillo/data_encoding.h>
46 #include <bsdiff/bsdiff.h>
47 #include <bsdiff/patch_writer_factory.h>
48 #include <puffin/utils.h>
49 
50 #include "update_engine/common/hash_calculator.h"
51 #include "update_engine/common/subprocess.h"
52 #include "update_engine/common/utils.h"
53 #include "update_engine/payload_consumer/payload_constants.h"
54 #include "update_engine/payload_generator/ab_generator.h"
55 #include "update_engine/payload_generator/block_mapping.h"
56 #include "update_engine/payload_generator/bzip.h"
57 #include "update_engine/payload_generator/deflate_utils.h"
58 #include "update_engine/payload_generator/delta_diff_generator.h"
59 #include "update_engine/payload_generator/extent_ranges.h"
60 #include "update_engine/payload_generator/extent_utils.h"
61 #include "update_engine/payload_generator/squashfs_filesystem.h"
62 #include "update_engine/payload_generator/xz.h"
63 
64 using std::list;
65 using std::map;
66 using std::string;
67 using std::vector;
68 
69 namespace chromeos_update_engine {
70 namespace {
71 
72 // The maximum destination size allowed for bsdiff. In general, bsdiff should
73 // work for arbitrary big files, but the payload generation and payload
74 // application requires a significant amount of RAM. We put a hard-limit of
75 // 200 MiB that should not affect any released board, but will limit the
76 // Chrome binary in ASan builders.
77 const uint64_t kMaxBsdiffDestinationSize = 200 * 1024 * 1024;  // bytes
78 
79 // The maximum destination size allowed for puffdiff. In general, puffdiff
80 // should work for arbitrary big files, but the payload application is quite
81 // memory intensive, so we limit these operations to 150 MiB.
82 const uint64_t kMaxPuffdiffDestinationSize = 150 * 1024 * 1024;  // bytes
83 
84 const int kBrotliCompressionQuality = 11;
85 
86 // Process a range of blocks from |range_start| to |range_end| in the extent at
87 // position |*idx_p| of |extents|. If |do_remove| is true, this range will be
88 // removed, which may cause the extent to be trimmed, split or removed entirely.
89 // The value of |*idx_p| is updated to point to the next extent to be processed.
90 // Returns true iff the next extent to process is a new or updated one.
91 bool ProcessExtentBlockRange(vector<Extent>* extents,
92                              size_t* idx_p,
93                              const bool do_remove,
94                              uint64_t range_start,
95                              uint64_t range_end) {
96   size_t idx = *idx_p;
97   uint64_t start_block = (*extents)[idx].start_block();
98   uint64_t num_blocks = (*extents)[idx].num_blocks();
99   uint64_t range_size = range_end - range_start;
100 
101   if (do_remove) {
102     if (range_size == num_blocks) {
103       // Remove the entire extent.
104       extents->erase(extents->begin() + idx);
105     } else if (range_end == num_blocks) {
106       // Trim the end of the extent.
107       (*extents)[idx].set_num_blocks(num_blocks - range_size);
108       idx++;
109     } else if (range_start == 0) {
110       // Trim the head of the extent.
111       (*extents)[idx].set_start_block(start_block + range_size);
112       (*extents)[idx].set_num_blocks(num_blocks - range_size);
113     } else {
114       // Trim the middle, splitting the remainder into two parts.
115       (*extents)[idx].set_num_blocks(range_start);
116       Extent e;
117       e.set_start_block(start_block + range_end);
118       e.set_num_blocks(num_blocks - range_end);
119       idx++;
120       extents->insert(extents->begin() + idx, e);
121     }
122   } else if (range_end == num_blocks) {
123     // Done with this extent.
124     idx++;
125   } else {
126     return false;
127   }
128 
129   *idx_p = idx;
130   return true;
131 }
132 
133 // Remove identical corresponding block ranges in |src_extents| and
134 // |dst_extents|. Used for preventing moving of blocks onto themselves during
135 // MOVE operations. The value of |total_bytes| indicates the actual length of
136 // content; this may be slightly less than the total size of blocks, in which
137 // case the last block is only partly occupied with data. Returns the total
138 // number of bytes removed.
139 size_t RemoveIdenticalBlockRanges(vector<Extent>* src_extents,
140                                   vector<Extent>* dst_extents,
141                                   const size_t total_bytes) {
142   size_t src_idx = 0;
143   size_t dst_idx = 0;
144   uint64_t src_offset = 0, dst_offset = 0;
145   size_t removed_bytes = 0, nonfull_block_bytes;
146   bool do_remove = false;
147   while (src_idx < src_extents->size() && dst_idx < dst_extents->size()) {
148     do_remove = ((*src_extents)[src_idx].start_block() + src_offset ==
149                  (*dst_extents)[dst_idx].start_block() + dst_offset);
150 
151     uint64_t src_num_blocks = (*src_extents)[src_idx].num_blocks();
152     uint64_t dst_num_blocks = (*dst_extents)[dst_idx].num_blocks();
153     uint64_t min_num_blocks =
154         std::min(src_num_blocks - src_offset, dst_num_blocks - dst_offset);
155     uint64_t prev_src_offset = src_offset;
156     uint64_t prev_dst_offset = dst_offset;
157     src_offset += min_num_blocks;
158     dst_offset += min_num_blocks;
159 
160     bool new_src = ProcessExtentBlockRange(
161         src_extents, &src_idx, do_remove, prev_src_offset, src_offset);
162     bool new_dst = ProcessExtentBlockRange(
163         dst_extents, &dst_idx, do_remove, prev_dst_offset, dst_offset);
164     if (new_src) {
165       src_offset = 0;
166     }
167     if (new_dst) {
168       dst_offset = 0;
169     }
170 
171     if (do_remove)
172       removed_bytes += min_num_blocks * kBlockSize;
173   }
174 
175   // If we removed the last block and this block is only partly used by file
176   // content, deduct the unused portion from the total removed byte count.
177   if (do_remove && (nonfull_block_bytes = total_bytes % kBlockSize))
178     removed_bytes -= kBlockSize - nonfull_block_bytes;
179 
180   return removed_bytes;
181 }
182 
183 // Storing a diff operation has more overhead over replace operation in the
184 // manifest, we need to store an additional src_sha256_hash which is 32 bytes
185 // and not compressible, and also src_extents which could use anywhere from a
186 // few bytes to hundreds of bytes depending on the number of extents.
187 // This function evaluates the overhead tradeoff and determines if it's worth to
188 // use a diff operation with data blob of |diff_size| and |num_src_extents|
189 // extents over an existing |op| with data blob of |old_blob_size|.
190 bool IsDiffOperationBetter(const InstallOperation& op,
191                            size_t old_blob_size,
192                            size_t diff_size,
193                            size_t num_src_extents) {
194   if (!diff_utils::IsAReplaceOperation(op.type()))
195     return diff_size < old_blob_size;
196 
197   // Reference: https://developers.google.com/protocol-buffers/docs/encoding
198   // For |src_sha256_hash| we need 1 byte field number/type, 1 byte size and 32
199   // bytes data, for |src_extents| we need 1 byte field number/type and 1 byte
200   // size.
201   constexpr size_t kDiffOverhead = 1 + 1 + 32 + 1 + 1;
202   // Each extent has two variable length encoded uint64, here we use a rough
203   // estimate of 6 bytes overhead per extent, since |num_blocks| is usually
204   // very small.
205   constexpr size_t kDiffOverheadPerExtent = 6;
206 
207   return diff_size + kDiffOverhead + num_src_extents * kDiffOverheadPerExtent <
208          old_blob_size;
209 }
210 
211 // Returns the levenshtein distance between string |a| and |b|.
212 // https://en.wikipedia.org/wiki/Levenshtein_distance
213 int LevenshteinDistance(const string& a, const string& b) {
214   vector<int> distances(a.size() + 1);
215   std::iota(distances.begin(), distances.end(), 0);
216 
217   for (size_t i = 1; i <= b.size(); i++) {
218     distances[0] = i;
219     int previous_distance = i - 1;
220     for (size_t j = 1; j <= a.size(); j++) {
221       int new_distance =
222           std::min({distances[j] + 1,
223                     distances[j - 1] + 1,
224                     previous_distance + (a[j - 1] == b[i - 1] ? 0 : 1)});
225       previous_distance = distances[j];
226       distances[j] = new_distance;
227     }
228   }
229   return distances.back();
230 }
231 }  // namespace
232 
233 namespace diff_utils {
234 
235 // This class encapsulates a file delta processing thread work. The
236 // processor computes the delta between the source and target files;
237 // and write the compressed delta to the blob.
238 class FileDeltaProcessor : public base::DelegateSimpleThread::Delegate {
239  public:
240   FileDeltaProcessor(const string& old_part,
241                      const string& new_part,
242                      const PayloadVersion& version,
243                      const vector<Extent>& old_extents,
244                      const vector<Extent>& new_extents,
245                      const vector<puffin::BitExtent>& old_deflates,
246                      const vector<puffin::BitExtent>& new_deflates,
247                      const string& name,
248                      ssize_t chunk_blocks,
249                      BlobFileWriter* blob_file)
250       : old_part_(old_part),
251         new_part_(new_part),
252         version_(version),
253         old_extents_(old_extents),
254         new_extents_(new_extents),
255         new_extents_blocks_(utils::BlocksInExtents(new_extents)),
256         old_deflates_(old_deflates),
257         new_deflates_(new_deflates),
258         name_(name),
259         chunk_blocks_(chunk_blocks),
260         blob_file_(blob_file) {}
261 
262   bool operator>(const FileDeltaProcessor& other) const {
263     return new_extents_blocks_ > other.new_extents_blocks_;
264   }
265 
266   ~FileDeltaProcessor() override = default;
267 
268   // Overrides DelegateSimpleThread::Delegate.
269   // Calculate the list of operations and write their corresponding deltas to
270   // the blob_file.
271   void Run() override;
272 
273   // Merge each file processor's ops list to aops.
274   bool MergeOperation(vector<AnnotatedOperation>* aops);
275 
276  private:
277   const string& old_part_;  // NOLINT(runtime/member_string_references)
278   const string& new_part_;  // NOLINT(runtime/member_string_references)
279   const PayloadVersion& version_;
280 
281   // The block ranges of the old/new file within the src/tgt image
282   const vector<Extent> old_extents_;
283   const vector<Extent> new_extents_;
284   const size_t new_extents_blocks_;
285   const vector<puffin::BitExtent> old_deflates_;
286   const vector<puffin::BitExtent> new_deflates_;
287   const string name_;
288   // Block limit of one aop.
289   const ssize_t chunk_blocks_;
290   BlobFileWriter* blob_file_;
291 
292   // The list of ops to reach the new file from the old file.
293   vector<AnnotatedOperation> file_aops_;
294 
295   bool failed_ = false;
296 
297   DISALLOW_COPY_AND_ASSIGN(FileDeltaProcessor);
298 };
299 
300 void FileDeltaProcessor::Run() {
301   TEST_AND_RETURN(blob_file_ != nullptr);
302   base::TimeTicks start = base::TimeTicks::Now();
303 
304   if (!DeltaReadFile(&file_aops_,
305                      old_part_,
306                      new_part_,
307                      old_extents_,
308                      new_extents_,
309                      old_deflates_,
310                      new_deflates_,
311                      name_,
312                      chunk_blocks_,
313                      version_,
314                      blob_file_)) {
315     LOG(ERROR) << "Failed to generate delta for " << name_ << " ("
316                << new_extents_blocks_ << " blocks)";
317     failed_ = true;
318     return;
319   }
320 
321   if (!version_.InplaceUpdate()) {
322     if (!ABGenerator::FragmentOperations(
323             version_, &file_aops_, new_part_, blob_file_)) {
324       LOG(ERROR) << "Failed to fragment operations for " << name_;
325       failed_ = true;
326       return;
327     }
328   }
329 
330   LOG(INFO) << "Encoded file " << name_ << " (" << new_extents_blocks_
331             << " blocks) in " << (base::TimeTicks::Now() - start);
332 }
333 
334 bool FileDeltaProcessor::MergeOperation(vector<AnnotatedOperation>* aops) {
335   if (failed_)
336     return false;
337   aops->reserve(aops->size() + file_aops_.size());
338   std::move(file_aops_.begin(), file_aops_.end(), std::back_inserter(*aops));
339   return true;
340 }
341 
342 FilesystemInterface::File GetOldFile(
343     const map<string, FilesystemInterface::File>& old_files_map,
344     const string& new_file_name) {
345   if (old_files_map.empty())
346     return {};
347 
348   auto old_file_iter = old_files_map.find(new_file_name);
349   if (old_file_iter != old_files_map.end())
350     return old_file_iter->second;
351 
352   // No old file match for the new file name, use a similar file with the
353   // shortest levenshtein distance.
354   // This works great if the file has version number in it, but even for
355   // a completely new file, using a similar file can still help.
356   int min_distance = new_file_name.size();
357   const FilesystemInterface::File* old_file;
358   for (const auto& pair : old_files_map) {
359     int distance = LevenshteinDistance(new_file_name, pair.first);
360     if (distance < min_distance) {
361       min_distance = distance;
362       old_file = &pair.second;
363     }
364   }
365   LOG(INFO) << "Using " << old_file->name << " as source for " << new_file_name;
366   return *old_file;
367 }
368 
369 bool DeltaReadPartition(vector<AnnotatedOperation>* aops,
370                         const PartitionConfig& old_part,
371                         const PartitionConfig& new_part,
372                         ssize_t hard_chunk_blocks,
373                         size_t soft_chunk_blocks,
374                         const PayloadVersion& version,
375                         BlobFileWriter* blob_file) {
376   ExtentRanges old_visited_blocks;
377   ExtentRanges new_visited_blocks;
378 
379   // If verity is enabled, mark those blocks as visited to skip generating
380   // operations for them.
381   if (version.minor >= kVerityMinorPayloadVersion &&
382       !new_part.verity.IsEmpty()) {
383     LOG(INFO) << "Skipping verity hash tree blocks: "
384               << ExtentsToString({new_part.verity.hash_tree_extent});
385     new_visited_blocks.AddExtent(new_part.verity.hash_tree_extent);
386     LOG(INFO) << "Skipping verity FEC blocks: "
387               << ExtentsToString({new_part.verity.fec_extent});
388     new_visited_blocks.AddExtent(new_part.verity.fec_extent);
389   }
390 
391   ExtentRanges old_zero_blocks;
392   TEST_AND_RETURN_FALSE(DeltaMovedAndZeroBlocks(aops,
393                                                 old_part.path,
394                                                 new_part.path,
395                                                 old_part.size / kBlockSize,
396                                                 new_part.size / kBlockSize,
397                                                 soft_chunk_blocks,
398                                                 version,
399                                                 blob_file,
400                                                 &old_visited_blocks,
401                                                 &new_visited_blocks,
402                                                 &old_zero_blocks));
403 
404   bool puffdiff_allowed = version.OperationAllowed(InstallOperation::PUFFDIFF);
405   map<string, FilesystemInterface::File> old_files_map;
406   if (old_part.fs_interface) {
407     vector<FilesystemInterface::File> old_files;
408     TEST_AND_RETURN_FALSE(deflate_utils::PreprocessPartitionFiles(
409         old_part, &old_files, puffdiff_allowed));
410     for (const FilesystemInterface::File& file : old_files)
411       old_files_map[file.name] = file;
412   }
413 
414   TEST_AND_RETURN_FALSE(new_part.fs_interface);
415   vector<FilesystemInterface::File> new_files;
416   TEST_AND_RETURN_FALSE(deflate_utils::PreprocessPartitionFiles(
417       new_part, &new_files, puffdiff_allowed));
418 
419   list<FileDeltaProcessor> file_delta_processors;
420 
421   // The processing is very straightforward here, we generate operations for
422   // every file (and pseudo-file such as the metadata) in the new filesystem
423   // based on the file with the same name in the old filesystem, if any.
424   // Files with overlapping data blocks (like hardlinks or filesystems with tail
425   // packing or compression where the blocks store more than one file) are only
426   // generated once in the new image, but are also used only once from the old
427   // image due to some simplifications (see below).
428   for (const FilesystemInterface::File& new_file : new_files) {
429     // Ignore the files in the new filesystem without blocks. Symlinks with
430     // data blocks (for example, symlinks bigger than 60 bytes in ext2) are
431     // handled as normal files. We also ignore blocks that were already
432     // processed by a previous file.
433     vector<Extent> new_file_extents =
434         FilterExtentRanges(new_file.extents, new_visited_blocks);
435     new_visited_blocks.AddExtents(new_file_extents);
436 
437     if (new_file_extents.empty())
438       continue;
439 
440     // We can't visit each dst image inode more than once, as that would
441     // duplicate work. Here, we avoid visiting each source image inode
442     // more than once. Technically, we could have multiple operations
443     // that read the same blocks from the source image for diffing, but
444     // we choose not to avoid complexity. Eventually we will move away
445     // from using a graph/cycle detection/etc to generate diffs, and at that
446     // time, it will be easy (non-complex) to have many operations read
447     // from the same source blocks. At that time, this code can die. -adlr
448     FilesystemInterface::File old_file =
449         GetOldFile(old_files_map, new_file.name);
450     vector<Extent> old_file_extents;
451     if (version.InplaceUpdate())
452       old_file_extents =
453           FilterExtentRanges(old_file.extents, old_visited_blocks);
454     else
455       old_file_extents = FilterExtentRanges(old_file.extents, old_zero_blocks);
456     old_visited_blocks.AddExtents(old_file_extents);
457 
458     file_delta_processors.emplace_back(old_part.path,
459                                        new_part.path,
460                                        version,
461                                        std::move(old_file_extents),
462                                        std::move(new_file_extents),
463                                        old_file.deflates,
464                                        new_file.deflates,
465                                        new_file.name,  // operation name
466                                        hard_chunk_blocks,
467                                        blob_file);
468   }
469   // Process all the blocks not included in any file. We provided all the unused
470   // blocks in the old partition as available data.
471   vector<Extent> new_unvisited = {
472       ExtentForRange(0, new_part.size / kBlockSize)};
473   new_unvisited = FilterExtentRanges(new_unvisited, new_visited_blocks);
474   if (!new_unvisited.empty()) {
475     vector<Extent> old_unvisited;
476     if (old_part.fs_interface) {
477       old_unvisited.push_back(ExtentForRange(0, old_part.size / kBlockSize));
478       old_unvisited = FilterExtentRanges(old_unvisited, old_visited_blocks);
479     }
480 
481     LOG(INFO) << "Scanning " << utils::BlocksInExtents(new_unvisited)
482               << " unwritten blocks using chunk size of " << soft_chunk_blocks
483               << " blocks.";
484     // We use the soft_chunk_blocks limit for the <non-file-data> as we don't
485     // really know the structure of this data and we should not expect it to
486     // have redundancy between partitions.
487     file_delta_processors.emplace_back(
488         old_part.path,
489         new_part.path,
490         version,
491         std::move(old_unvisited),
492         std::move(new_unvisited),
493         vector<puffin::BitExtent>{},  // old_deflates,
494         vector<puffin::BitExtent>{},  // new_deflates
495         "<non-file-data>",            // operation name
496         soft_chunk_blocks,
497         blob_file);
498   }
499 
500   size_t max_threads = GetMaxThreads();
501 
502   // Sort the files in descending order based on number of new blocks to make
503   // sure we start the largest ones first.
504   if (file_delta_processors.size() > max_threads) {
505     file_delta_processors.sort(std::greater<FileDeltaProcessor>());
506   }
507 
508   base::DelegateSimpleThreadPool thread_pool("incremental-update-generator",
509                                              max_threads);
510   thread_pool.Start();
511   for (auto& processor : file_delta_processors) {
512     thread_pool.AddWork(&processor);
513   }
514   thread_pool.JoinAll();
515 
516   for (auto& processor : file_delta_processors) {
517     TEST_AND_RETURN_FALSE(processor.MergeOperation(aops));
518   }
519 
520   return true;
521 }
522 
523 bool DeltaMovedAndZeroBlocks(vector<AnnotatedOperation>* aops,
524                              const string& old_part,
525                              const string& new_part,
526                              size_t old_num_blocks,
527                              size_t new_num_blocks,
528                              ssize_t chunk_blocks,
529                              const PayloadVersion& version,
530                              BlobFileWriter* blob_file,
531                              ExtentRanges* old_visited_blocks,
532                              ExtentRanges* new_visited_blocks,
533                              ExtentRanges* old_zero_blocks) {
534   vector<BlockMapping::BlockId> old_block_ids;
535   vector<BlockMapping::BlockId> new_block_ids;
536   TEST_AND_RETURN_FALSE(MapPartitionBlocks(old_part,
537                                            new_part,
538                                            old_num_blocks * kBlockSize,
539                                            new_num_blocks * kBlockSize,
540                                            kBlockSize,
541                                            &old_block_ids,
542                                            &new_block_ids));
543 
544   // If the update is inplace, we map all the blocks that didn't move,
545   // regardless of the contents since they are already copied and no operation
546   // is required.
547   if (version.InplaceUpdate()) {
548     uint64_t num_blocks = std::min(old_num_blocks, new_num_blocks);
549     for (uint64_t block = 0; block < num_blocks; block++) {
550       if (old_block_ids[block] == new_block_ids[block] &&
551           !old_visited_blocks->ContainsBlock(block) &&
552           !new_visited_blocks->ContainsBlock(block)) {
553         old_visited_blocks->AddBlock(block);
554         new_visited_blocks->AddBlock(block);
555       }
556     }
557   }
558 
559   // A mapping from the block_id to the list of block numbers with that block id
560   // in the old partition. This is used to lookup where in the old partition
561   // is a block from the new partition.
562   map<BlockMapping::BlockId, vector<uint64_t>> old_blocks_map;
563 
564   for (uint64_t block = old_num_blocks; block-- > 0;) {
565     if (old_block_ids[block] != 0 && !old_visited_blocks->ContainsBlock(block))
566       old_blocks_map[old_block_ids[block]].push_back(block);
567 
568     // Mark all zeroed blocks in the old image as "used" since it doesn't make
569     // any sense to spend I/O to read zeros from the source partition and more
570     // importantly, these could sometimes be blocks discarded in the SSD which
571     // would read non-zero values.
572     if (old_block_ids[block] == 0)
573       old_zero_blocks->AddBlock(block);
574   }
575   old_visited_blocks->AddRanges(*old_zero_blocks);
576 
577   // The collection of blocks in the new partition with just zeros. This is a
578   // common case for free-space that's also problematic for bsdiff, so we want
579   // to optimize it using REPLACE_BZ operations. The blob for a REPLACE_BZ of
580   // just zeros is so small that it doesn't make sense to spend the I/O reading
581   // zeros from the old partition.
582   vector<Extent> new_zeros;
583 
584   vector<Extent> old_identical_blocks;
585   vector<Extent> new_identical_blocks;
586 
587   for (uint64_t block = 0; block < new_num_blocks; block++) {
588     // Only produce operations for blocks that were not yet visited.
589     if (new_visited_blocks->ContainsBlock(block))
590       continue;
591     if (new_block_ids[block] == 0) {
592       AppendBlockToExtents(&new_zeros, block);
593       continue;
594     }
595 
596     auto old_blocks_map_it = old_blocks_map.find(new_block_ids[block]);
597     // Check if the block exists in the old partition at all.
598     if (old_blocks_map_it == old_blocks_map.end() ||
599         old_blocks_map_it->second.empty())
600       continue;
601 
602     AppendBlockToExtents(&old_identical_blocks,
603                          old_blocks_map_it->second.back());
604     AppendBlockToExtents(&new_identical_blocks, block);
605     // We can't reuse source blocks in minor version 1 because the cycle
606     // breaking algorithm used in the in-place update doesn't support that.
607     if (version.InplaceUpdate())
608       old_blocks_map_it->second.pop_back();
609   }
610 
611   if (chunk_blocks == -1)
612     chunk_blocks = new_num_blocks;
613 
614   // Produce operations for the zero blocks split per output extent.
615   size_t num_ops = aops->size();
616   new_visited_blocks->AddExtents(new_zeros);
617   for (const Extent& extent : new_zeros) {
618     if (version.OperationAllowed(InstallOperation::ZERO)) {
619       for (uint64_t offset = 0; offset < extent.num_blocks();
620            offset += chunk_blocks) {
621         uint64_t num_blocks =
622             std::min(static_cast<uint64_t>(extent.num_blocks()) - offset,
623                      static_cast<uint64_t>(chunk_blocks));
624         InstallOperation operation;
625         operation.set_type(InstallOperation::ZERO);
626         *(operation.add_dst_extents()) =
627             ExtentForRange(extent.start_block() + offset, num_blocks);
628         aops->push_back({.name = "<zeros>", .op = operation});
629       }
630     } else {
631       TEST_AND_RETURN_FALSE(DeltaReadFile(aops,
632                                           "",
633                                           new_part,
634                                           {},        // old_extents
635                                           {extent},  // new_extents
636                                           {},        // old_deflates
637                                           {},        // new_deflates
638                                           "<zeros>",
639                                           chunk_blocks,
640                                           version,
641                                           blob_file));
642     }
643   }
644   LOG(INFO) << "Produced " << (aops->size() - num_ops) << " operations for "
645             << utils::BlocksInExtents(new_zeros) << " zeroed blocks";
646 
647   // Produce MOVE/SOURCE_COPY operations for the moved blocks.
648   num_ops = aops->size();
649   uint64_t used_blocks = 0;
650   old_visited_blocks->AddExtents(old_identical_blocks);
651   new_visited_blocks->AddExtents(new_identical_blocks);
652   for (const Extent& extent : new_identical_blocks) {
653     // We split the operation at the extent boundary or when bigger than
654     // chunk_blocks.
655     for (uint64_t op_block_offset = 0; op_block_offset < extent.num_blocks();
656          op_block_offset += chunk_blocks) {
657       aops->emplace_back();
658       AnnotatedOperation* aop = &aops->back();
659       aop->name = "<identical-blocks>";
660       aop->op.set_type(version.OperationAllowed(InstallOperation::SOURCE_COPY)
661                            ? InstallOperation::SOURCE_COPY
662                            : InstallOperation::MOVE);
663 
664       uint64_t chunk_num_blocks =
665           std::min(static_cast<uint64_t>(extent.num_blocks()) - op_block_offset,
666                    static_cast<uint64_t>(chunk_blocks));
667 
668       // The current operation represents the move/copy operation for the
669       // sublist starting at |used_blocks| of length |chunk_num_blocks| where
670       // the src and dst are from |old_identical_blocks| and
671       // |new_identical_blocks| respectively.
672       StoreExtents(
673           ExtentsSublist(old_identical_blocks, used_blocks, chunk_num_blocks),
674           aop->op.mutable_src_extents());
675 
676       Extent* op_dst_extent = aop->op.add_dst_extents();
677       op_dst_extent->set_start_block(extent.start_block() + op_block_offset);
678       op_dst_extent->set_num_blocks(chunk_num_blocks);
679       CHECK(
680           vector<Extent>{*op_dst_extent} ==  // NOLINT(whitespace/braces)
681           ExtentsSublist(new_identical_blocks, used_blocks, chunk_num_blocks));
682 
683       used_blocks += chunk_num_blocks;
684     }
685   }
686   LOG(INFO) << "Produced " << (aops->size() - num_ops) << " operations for "
687             << used_blocks << " identical blocks moved";
688 
689   return true;
690 }
691 
692 bool DeltaReadFile(vector<AnnotatedOperation>* aops,
693                    const string& old_part,
694                    const string& new_part,
695                    const vector<Extent>& old_extents,
696                    const vector<Extent>& new_extents,
697                    const vector<puffin::BitExtent>& old_deflates,
698                    const vector<puffin::BitExtent>& new_deflates,
699                    const string& name,
700                    ssize_t chunk_blocks,
701                    const PayloadVersion& version,
702                    BlobFileWriter* blob_file) {
703   brillo::Blob data;
704   InstallOperation operation;
705 
706   uint64_t total_blocks = utils::BlocksInExtents(new_extents);
707   if (chunk_blocks == -1)
708     chunk_blocks = total_blocks;
709 
710   for (uint64_t block_offset = 0; block_offset < total_blocks;
711        block_offset += chunk_blocks) {
712     // Split the old/new file in the same chunks. Note that this could drop
713     // some information from the old file used for the new chunk. If the old
714     // file is smaller (or even empty when there's no old file) the chunk will
715     // also be empty.
716     vector<Extent> old_extents_chunk =
717         ExtentsSublist(old_extents, block_offset, chunk_blocks);
718     vector<Extent> new_extents_chunk =
719         ExtentsSublist(new_extents, block_offset, chunk_blocks);
720     NormalizeExtents(&old_extents_chunk);
721     NormalizeExtents(&new_extents_chunk);
722 
723     TEST_AND_RETURN_FALSE(ReadExtentsToDiff(old_part,
724                                             new_part,
725                                             old_extents_chunk,
726                                             new_extents_chunk,
727                                             old_deflates,
728                                             new_deflates,
729                                             version,
730                                             &data,
731                                             &operation));
732 
733     // Check if the operation writes nothing.
734     if (operation.dst_extents_size() == 0) {
735       if (operation.type() == InstallOperation::MOVE) {
736         LOG(INFO) << "Empty MOVE operation (" << name << "), skipping";
737         continue;
738       } else {
739         LOG(ERROR) << "Empty non-MOVE operation";
740         return false;
741       }
742     }
743 
744     // Now, insert into the list of operations.
745     AnnotatedOperation aop;
746     aop.name = name;
747     if (static_cast<uint64_t>(chunk_blocks) < total_blocks) {
748       aop.name = base::StringPrintf(
749           "%s:%" PRIu64, name.c_str(), block_offset / chunk_blocks);
750     }
751     aop.op = operation;
752 
753     // Write the data
754     TEST_AND_RETURN_FALSE(aop.SetOperationBlob(data, blob_file));
755     aops->emplace_back(aop);
756   }
757   return true;
758 }
759 
760 bool GenerateBestFullOperation(const brillo::Blob& new_data,
761                                const PayloadVersion& version,
762                                brillo::Blob* out_blob,
763                                InstallOperation::Type* out_type) {
764   if (new_data.empty())
765     return false;
766 
767   if (version.OperationAllowed(InstallOperation::ZERO) &&
768       std::all_of(
769           new_data.begin(), new_data.end(), [](uint8_t x) { return x == 0; })) {
770     // The read buffer is all zeros, so produce a ZERO operation. No need to
771     // check other types of operations in this case.
772     *out_blob = brillo::Blob();
773     *out_type = InstallOperation::ZERO;
774     return true;
775   }
776 
777   bool out_blob_set = false;
778 
779   // Try compressing |new_data| with xz first.
780   if (version.OperationAllowed(InstallOperation::REPLACE_XZ)) {
781     brillo::Blob new_data_xz;
782     if (XzCompress(new_data, &new_data_xz) && !new_data_xz.empty()) {
783       *out_type = InstallOperation::REPLACE_XZ;
784       *out_blob = std::move(new_data_xz);
785       out_blob_set = true;
786     }
787   }
788 
789   // Try compressing it with bzip2.
790   if (version.OperationAllowed(InstallOperation::REPLACE_BZ)) {
791     brillo::Blob new_data_bz;
792     // TODO(deymo): Implement some heuristic to determine if it is worth trying
793     // to compress the blob with bzip2 if we already have a good REPLACE_XZ.
794     if (BzipCompress(new_data, &new_data_bz) && !new_data_bz.empty() &&
795         (!out_blob_set || out_blob->size() > new_data_bz.size())) {
796       // A REPLACE_BZ is better or nothing else was set.
797       *out_type = InstallOperation::REPLACE_BZ;
798       *out_blob = std::move(new_data_bz);
799       out_blob_set = true;
800     }
801   }
802 
803   // If nothing else worked or it was badly compressed we try a REPLACE.
804   if (!out_blob_set || out_blob->size() >= new_data.size()) {
805     *out_type = InstallOperation::REPLACE;
806     // This needs to make a copy of the data in the case bzip or xz didn't
807     // compress well, which is not the common case so the performance hit is
808     // low.
809     *out_blob = new_data;
810   }
811   return true;
812 }
813 
814 bool ReadExtentsToDiff(const string& old_part,
815                        const string& new_part,
816                        const vector<Extent>& old_extents,
817                        const vector<Extent>& new_extents,
818                        const vector<puffin::BitExtent>& old_deflates,
819                        const vector<puffin::BitExtent>& new_deflates,
820                        const PayloadVersion& version,
821                        brillo::Blob* out_data,
822                        InstallOperation* out_op) {
823   InstallOperation operation;
824 
825   // We read blocks from old_extents and write blocks to new_extents.
826   uint64_t blocks_to_read = utils::BlocksInExtents(old_extents);
827   uint64_t blocks_to_write = utils::BlocksInExtents(new_extents);
828 
829   // Disable bsdiff, and puffdiff when the data is too big.
830   bool bsdiff_allowed =
831       version.OperationAllowed(InstallOperation::SOURCE_BSDIFF) ||
832       version.OperationAllowed(InstallOperation::BSDIFF);
833   if (bsdiff_allowed &&
834       blocks_to_read * kBlockSize > kMaxBsdiffDestinationSize) {
835     LOG(INFO) << "bsdiff blacklisted, data too big: "
836               << blocks_to_read * kBlockSize << " bytes";
837     bsdiff_allowed = false;
838   }
839 
840   bool puffdiff_allowed = version.OperationAllowed(InstallOperation::PUFFDIFF);
841   if (puffdiff_allowed &&
842       blocks_to_read * kBlockSize > kMaxPuffdiffDestinationSize) {
843     LOG(INFO) << "puffdiff blacklisted, data too big: "
844               << blocks_to_read * kBlockSize << " bytes";
845     puffdiff_allowed = false;
846   }
847 
848   // Make copies of the extents so we can modify them.
849   vector<Extent> src_extents = old_extents;
850   vector<Extent> dst_extents = new_extents;
851 
852   // Read in bytes from new data.
853   brillo::Blob new_data;
854   TEST_AND_RETURN_FALSE(utils::ReadExtents(new_part,
855                                            new_extents,
856                                            &new_data,
857                                            kBlockSize * blocks_to_write,
858                                            kBlockSize));
859   TEST_AND_RETURN_FALSE(!new_data.empty());
860 
861   // Data blob that will be written to delta file.
862   brillo::Blob data_blob;
863 
864   // Try generating a full operation for the given new data, regardless of the
865   // old_data.
866   InstallOperation::Type op_type;
867   TEST_AND_RETURN_FALSE(
868       GenerateBestFullOperation(new_data, version, &data_blob, &op_type));
869   operation.set_type(op_type);
870 
871   brillo::Blob old_data;
872   if (blocks_to_read > 0) {
873     // Read old data.
874     TEST_AND_RETURN_FALSE(utils::ReadExtents(old_part,
875                                              src_extents,
876                                              &old_data,
877                                              kBlockSize * blocks_to_read,
878                                              kBlockSize));
879     if (old_data == new_data) {
880       // No change in data.
881       operation.set_type(version.OperationAllowed(InstallOperation::SOURCE_COPY)
882                              ? InstallOperation::SOURCE_COPY
883                              : InstallOperation::MOVE);
884       data_blob = brillo::Blob();
885     } else if (IsDiffOperationBetter(
886                    operation, data_blob.size(), 0, src_extents.size())) {
887       // No point in trying diff if zero blob size diff operation is
888       // still worse than replace.
889       if (bsdiff_allowed) {
890         base::FilePath patch;
891         TEST_AND_RETURN_FALSE(base::CreateTemporaryFile(&patch));
892         ScopedPathUnlinker unlinker(patch.value());
893 
894         std::unique_ptr<bsdiff::PatchWriterInterface> bsdiff_patch_writer;
895         InstallOperation::Type operation_type = InstallOperation::BSDIFF;
896         if (version.OperationAllowed(InstallOperation::BROTLI_BSDIFF)) {
897           bsdiff_patch_writer =
898               bsdiff::CreateBSDF2PatchWriter(patch.value(),
899                                              bsdiff::CompressorType::kBrotli,
900                                              kBrotliCompressionQuality);
901           operation_type = InstallOperation::BROTLI_BSDIFF;
902         } else {
903           bsdiff_patch_writer = bsdiff::CreateBsdiffPatchWriter(patch.value());
904           if (version.OperationAllowed(InstallOperation::SOURCE_BSDIFF)) {
905             operation_type = InstallOperation::SOURCE_BSDIFF;
906           }
907         }
908 
909         brillo::Blob bsdiff_delta;
910         TEST_AND_RETURN_FALSE(0 == bsdiff::bsdiff(old_data.data(),
911                                                   old_data.size(),
912                                                   new_data.data(),
913                                                   new_data.size(),
914                                                   bsdiff_patch_writer.get(),
915                                                   nullptr));
916 
917         TEST_AND_RETURN_FALSE(utils::ReadFile(patch.value(), &bsdiff_delta));
918         CHECK_GT(bsdiff_delta.size(), static_cast<brillo::Blob::size_type>(0));
919         if (IsDiffOperationBetter(operation,
920                                   data_blob.size(),
921                                   bsdiff_delta.size(),
922                                   src_extents.size())) {
923           operation.set_type(operation_type);
924           data_blob = std::move(bsdiff_delta);
925         }
926       }
927       if (puffdiff_allowed) {
928         // Find all deflate positions inside the given extents and then put all
929         // deflates together because we have already read all the extents into
930         // one buffer.
931         vector<puffin::BitExtent> src_deflates;
932         TEST_AND_RETURN_FALSE(deflate_utils::FindAndCompactDeflates(
933             src_extents, old_deflates, &src_deflates));
934 
935         vector<puffin::BitExtent> dst_deflates;
936         TEST_AND_RETURN_FALSE(deflate_utils::FindAndCompactDeflates(
937             dst_extents, new_deflates, &dst_deflates));
938 
939         puffin::RemoveEqualBitExtents(
940             old_data, new_data, &src_deflates, &dst_deflates);
941 
942         // See crbug.com/915559.
943         if (version.minor <= kPuffdiffMinorPayloadVersion) {
944           TEST_AND_RETURN_FALSE(puffin::RemoveDeflatesWithBadDistanceCaches(
945               old_data, &src_deflates));
946 
947           TEST_AND_RETURN_FALSE(puffin::RemoveDeflatesWithBadDistanceCaches(
948               new_data, &dst_deflates));
949         }
950 
951         // Only Puffdiff if both files have at least one deflate left.
952         if (!src_deflates.empty() && !dst_deflates.empty()) {
953           brillo::Blob puffdiff_delta;
954           string temp_file_path;
955           TEST_AND_RETURN_FALSE(utils::MakeTempFile(
956               "puffdiff-delta.XXXXXX", &temp_file_path, nullptr));
957           ScopedPathUnlinker temp_file_unlinker(temp_file_path);
958 
959           // Perform PuffDiff operation.
960           TEST_AND_RETURN_FALSE(puffin::PuffDiff(old_data,
961                                                  new_data,
962                                                  src_deflates,
963                                                  dst_deflates,
964                                                  temp_file_path,
965                                                  &puffdiff_delta));
966           TEST_AND_RETURN_FALSE(puffdiff_delta.size() > 0);
967           if (IsDiffOperationBetter(operation,
968                                     data_blob.size(),
969                                     puffdiff_delta.size(),
970                                     src_extents.size())) {
971             operation.set_type(InstallOperation::PUFFDIFF);
972             data_blob = std::move(puffdiff_delta);
973           }
974         }
975       }
976     }
977   }
978 
979   // Remove identical src/dst block ranges in MOVE operations.
980   if (operation.type() == InstallOperation::MOVE) {
981     auto removed_bytes =
982         RemoveIdenticalBlockRanges(&src_extents, &dst_extents, new_data.size());
983     operation.set_src_length(old_data.size() - removed_bytes);
984     operation.set_dst_length(new_data.size() - removed_bytes);
985   }
986 
987   // WARNING: We always set legacy |src_length| and |dst_length| fields for
988   // BSDIFF. For SOURCE_BSDIFF we only set them for minor version 3 and
989   // lower. This is needed because we used to use these two parameters in the
990   // SOURCE_BSDIFF for minor version 3 and lower, but we do not need them
991   // anymore in higher minor versions. This means if we stop adding these
992   // parameters for those minor versions, the delta payloads will be invalid.
993   if (operation.type() == InstallOperation::BSDIFF ||
994       (operation.type() == InstallOperation::SOURCE_BSDIFF &&
995        version.minor <= kOpSrcHashMinorPayloadVersion)) {
996     operation.set_src_length(old_data.size());
997     operation.set_dst_length(new_data.size());
998   }
999 
1000   // Embed extents in the operation. Replace (all variants), zero and discard
1001   // operations should not have source extents.
1002   if (!IsNoSourceOperation(operation.type())) {
1003     StoreExtents(src_extents, operation.mutable_src_extents());
1004   }
1005   // All operations have dst_extents.
1006   StoreExtents(dst_extents, operation.mutable_dst_extents());
1007 
1008   *out_data = std::move(data_blob);
1009   *out_op = operation;
1010   return true;
1011 }
1012 
1013 bool IsAReplaceOperation(InstallOperation::Type op_type) {
1014   return (op_type == InstallOperation::REPLACE ||
1015           op_type == InstallOperation::REPLACE_BZ ||
1016           op_type == InstallOperation::REPLACE_XZ);
1017 }
1018 
1019 bool IsNoSourceOperation(InstallOperation::Type op_type) {
1020   return (IsAReplaceOperation(op_type) || op_type == InstallOperation::ZERO ||
1021           op_type == InstallOperation::DISCARD);
1022 }
1023 
1024 // Returns true if |op| is a no-op operation that doesn't do any useful work
1025 // (e.g., a move operation that copies blocks onto themselves).
1026 bool IsNoopOperation(const InstallOperation& op) {
1027   return (op.type() == InstallOperation::MOVE &&
1028           ExpandExtents(op.src_extents()) == ExpandExtents(op.dst_extents()));
1029 }
1030 
1031 void FilterNoopOperations(vector<AnnotatedOperation>* ops) {
1032   ops->erase(std::remove_if(ops->begin(),
1033                             ops->end(),
1034                             [](const AnnotatedOperation& aop) {
1035                               return IsNoopOperation(aop.op);
1036                             }),
1037              ops->end());
1038 }
1039 
1040 bool InitializePartitionInfo(const PartitionConfig& part, PartitionInfo* info) {
1041   info->set_size(part.size);
1042   HashCalculator hasher;
1043   TEST_AND_RETURN_FALSE(hasher.UpdateFile(part.path, part.size) ==
1044                         static_cast<off_t>(part.size));
1045   TEST_AND_RETURN_FALSE(hasher.Finalize());
1046   const brillo::Blob& hash = hasher.raw_hash();
1047   info->set_hash(hash.data(), hash.size());
1048   LOG(INFO) << part.path << ": size=" << part.size
1049             << " hash=" << brillo::data_encoding::Base64Encode(hash);
1050   return true;
1051 }
1052 
1053 bool CompareAopsByDestination(AnnotatedOperation first_aop,
1054                               AnnotatedOperation second_aop) {
1055   // We want empty operations to be at the end of the payload.
1056   if (!first_aop.op.dst_extents().size() || !second_aop.op.dst_extents().size())
1057     return ((!first_aop.op.dst_extents().size()) <
1058             (!second_aop.op.dst_extents().size()));
1059   uint32_t first_dst_start = first_aop.op.dst_extents(0).start_block();
1060   uint32_t second_dst_start = second_aop.op.dst_extents(0).start_block();
1061   return first_dst_start < second_dst_start;
1062 }
1063 
1064 bool IsExtFilesystem(const string& device) {
1065   brillo::Blob header;
1066   // See include/linux/ext2_fs.h for more details on the structure. We obtain
1067   // ext2 constants from ext2fs/ext2fs.h header but we don't link with the
1068   // library.
1069   if (!utils::ReadFileChunk(
1070           device, 0, SUPERBLOCK_OFFSET + SUPERBLOCK_SIZE, &header) ||
1071       header.size() < SUPERBLOCK_OFFSET + SUPERBLOCK_SIZE)
1072     return false;
1073 
1074   const uint8_t* superblock = header.data() + SUPERBLOCK_OFFSET;
1075 
1076   // ext3_fs.h: ext3_super_block.s_blocks_count
1077   uint32_t block_count =
1078       *reinterpret_cast<const uint32_t*>(superblock + 1 * sizeof(int32_t));
1079 
1080   // ext3_fs.h: ext3_super_block.s_log_block_size
1081   uint32_t log_block_size =
1082       *reinterpret_cast<const uint32_t*>(superblock + 6 * sizeof(int32_t));
1083 
1084   // ext3_fs.h: ext3_super_block.s_magic
1085   uint16_t magic =
1086       *reinterpret_cast<const uint16_t*>(superblock + 14 * sizeof(int32_t));
1087 
1088   block_count = le32toh(block_count);
1089   log_block_size = le32toh(log_block_size) + EXT2_MIN_BLOCK_LOG_SIZE;
1090   magic = le16toh(magic);
1091 
1092   if (magic != EXT2_SUPER_MAGIC)
1093     return false;
1094 
1095   // Sanity check the parameters.
1096   TEST_AND_RETURN_FALSE(log_block_size >= EXT2_MIN_BLOCK_LOG_SIZE &&
1097                         log_block_size <= EXT2_MAX_BLOCK_LOG_SIZE);
1098   TEST_AND_RETURN_FALSE(block_count > 0);
1099   return true;
1100 }
1101 
1102 // Return the number of CPUs on the machine, and 4 threads in minimum.
1103 size_t GetMaxThreads() {
1104   return std::max(sysconf(_SC_NPROCESSORS_ONLN), 4L);
1105 }
1106 
1107 }  // namespace diff_utils
1108 
1109 }  // namespace chromeos_update_engine
1110