Avoid mapping output to memory while writing to a file.

It's unnecessary, and causes issues when the uncompressed
output is large.

Bug: http://b/21558406
Change-Id: I99cfb3933b094c2826c7e6c6de9aab03478fcc53
(cherry picked from commit f899bd534b2dc51b9db8d27c76394b192fe51155)

2 files changed, 269 insertions, 90 deletions

diff --git a/libziparchive/zip_archive.cc b/libziparchive/zip_archive.cc
index 858234446..4ba91dfb9 100644
--- a/libziparchive/zip_archive.cc
+++ b/libziparchive/zip_archive.cc
@@ -85,7 +85,8 @@ struct EocdRecord {
   // Length of the central directory comment.
   uint16_t comment_length;
  private:
-  DISALLOW_IMPLICIT_CONSTRUCTORS(EocdRecord);
+  EocdRecord() = default;
+  DISALLOW_COPY_AND_ASSIGN(EocdRecord);
 } __attribute__((packed));
 
 // A structure representing the fixed length fields for a single
@@ -138,7 +139,8 @@ struct CentralDirectoryRecord {
   // beginning of this archive.
   uint32_t local_file_header_offset;
  private:
-  DISALLOW_IMPLICIT_CONSTRUCTORS(CentralDirectoryRecord);
+  CentralDirectoryRecord() = default;
+  DISALLOW_COPY_AND_ASSIGN(CentralDirectoryRecord);
 } __attribute__((packed));
 
 // The local file header for a given entry. This duplicates information
@@ -175,7 +177,8 @@ struct LocalFileHeader {
   // will appear immediately after the entry file name.
   uint16_t extra_field_length;
  private:
-  DISALLOW_IMPLICIT_CONSTRUCTORS(LocalFileHeader);
+  LocalFileHeader() = default;
+  DISALLOW_COPY_AND_ASSIGN(LocalFileHeader);
 } __attribute__((packed));
 
 struct DataDescriptor {
@@ -189,10 +192,10 @@ struct DataDescriptor {
   // Uncompressed size of the entry.
   uint32_t uncompressed_size;
  private:
-  DISALLOW_IMPLICIT_CONSTRUCTORS(DataDescriptor);
+  DataDescriptor() = default;
+  DISALLOW_COPY_AND_ASSIGN(DataDescriptor);
 } __attribute__((packed));
 
-#undef DISALLOW_IMPLICIT_CONSTRUCTORS
 
 static const uint32_t kGPBDDFlagMask = 0x0008;         // mask value that signifies that the entry has a DD
 
@@ -324,35 +327,6 @@ struct ZipArchive {
   }
 };
 
-static int32_t CopyFileToFile(int fd, uint8_t* begin, const uint32_t length, uint64_t *crc_out) {
-  static const uint32_t kBufSize = 32768;
-  uint8_t buf[kBufSize];
-
-  uint32_t count = 0;
-  uint64_t crc = 0;
-  while (count < length) {
-    uint32_t remaining = length - count;
-
-    // Safe conversion because kBufSize is narrow enough for a 32 bit signed
-    // value.
-    ssize_t get_size = (remaining > kBufSize) ? kBufSize : remaining;
-    ssize_t actual = TEMP_FAILURE_RETRY(read(fd, buf, get_size));
-
-    if (actual != get_size) {
-      ALOGW("CopyFileToFile: copy read failed (" ZD " vs " ZD ")", actual, get_size);
-      return kIoError;
-    }
-
-    memcpy(begin + count, buf, get_size);
-    crc = crc32(crc, buf, get_size);
-    count += get_size;
-  }
-
-  *crc_out = crc;
-
-  return 0;
-}
-
 /*
  * Round up to the next highest power of 2.
  *
@@ -972,6 +946,128 @@ int32_t Next(void* cookie, ZipEntry* data, ZipEntryName* name) {
   return kIterationEnd;
 }
 
+class Writer {
+ public:
+  virtual bool Append(uint8_t* buf, size_t buf_size) = 0;
+  virtual ~Writer() {}
+ protected:
+  Writer() = default;
+ private:
+  DISALLOW_COPY_AND_ASSIGN(Writer);
+};
+
+// A Writer that writes data to a fixed size memory region.
+// The size of the memory region must be equal to the total size of
+// the data appended to it.
+class MemoryWriter : public Writer {
+ public:
+  MemoryWriter(uint8_t* buf, size_t size) : Writer(),
+      buf_(buf), size_(size), bytes_written_(0) {
+  }
+
+  virtual bool Append(uint8_t* buf, size_t buf_size) override {
+    if (bytes_written_ + buf_size > size_) {
+      ALOGW("Zip: Unexpected size " ZD " (declared) vs " ZD " (actual)",
+            size_, bytes_written_ + buf_size);
+      return false;
+    }
+
+    memcpy(buf_ + bytes_written_, buf, buf_size);
+    bytes_written_ += buf_size;
+    return true;
+  }
+
+ private:
+  uint8_t* const buf_;
+  const size_t size_;
+  size_t bytes_written_;
+};
+
+// A Writer that appends data to a file |fd| at its current position.
+// The file will be truncated to the end of the written data.
+class FileWriter : public Writer {
+ public:
+
+  // Creates a FileWriter for |fd| and prepare to write |entry| to it,
+  // guaranteeing that the file descriptor is valid and that there's enough
+  // space on the volume to write out the entry completely and that the file
+  // is truncated to the correct length.
+  //
+  // Returns a valid FileWriter on success, |nullptr| if an error occurred.
+  static std::unique_ptr<FileWriter> Create(int fd, const ZipEntry* entry) {
+    const uint32_t declared_length = entry->uncompressed_length;
+    const off64_t current_offset = lseek64(fd, 0, SEEK_CUR);
+    if (current_offset == -1) {
+      ALOGW("Zip: unable to seek to current location on fd %d: %s", fd, strerror(errno));
+      return nullptr;
+    }
+
+    int result = 0;
+#if defined(__linux__)
+    if (declared_length > 0) {
+      // Make sure we have enough space on the volume to extract the compressed
+      // entry. Note that the call to ftruncate below will change the file size but
+      // will not allocate space on disk and this call to fallocate will not
+      // change the file size.
+      result = TEMP_FAILURE_RETRY(fallocate(fd, 0, current_offset, declared_length));
+      if (result == -1) {
+        ALOGW("Zip: unable to allocate space for file to %" PRId64 ": %s",
+              static_cast<int64_t>(declared_length + current_offset), strerror(errno));
+        return std::unique_ptr<FileWriter>(nullptr);
+      }
+    }
+#endif  // __linux__
+
+    result = TEMP_FAILURE_RETRY(ftruncate(fd, declared_length + current_offset));
+    if (result == -1) {
+      ALOGW("Zip: unable to truncate file to %" PRId64 ": %s",
+            static_cast<int64_t>(declared_length + current_offset), strerror(errno));
+      return std::unique_ptr<FileWriter>(nullptr);
+    }
+
+    return std::unique_ptr<FileWriter>(new FileWriter(fd, declared_length));
+  }
+
+  virtual bool Append(uint8_t* buf, size_t buf_size) override {
+    if (total_bytes_written_ + buf_size > declared_length_) {
+      ALOGW("Zip: Unexpected size " ZD " (declared) vs " ZD " (actual)",
+            declared_length_, total_bytes_written_ + buf_size);
+      return false;
+    }
+
+    // Keep track of the start position so we can calculate the
+    // total number of bytes written.
+    const uint8_t* const start = buf;
+    size_t bytes_written = 0;
+    while (buf_size > 0) {
+      ssize_t bytes_written = TEMP_FAILURE_RETRY(write(fd_, buf, buf_size));
+      if (bytes_written == -1) {
+        ALOGW("Zip: unable to write " ZD " bytes to file; %s", buf_size, strerror(errno));
+        return false;
+      }
+
+      buf_size -= bytes_written;
+      buf += bytes_written;
+    }
+
+    total_bytes_written_ += static_cast<size_t>(
+        reinterpret_cast<uintptr_t>(buf) - reinterpret_cast<uintptr_t>(start));
+
+    return true;
+  }
+ private:
+  FileWriter(const int fd, const size_t declared_length) :
+      Writer(),
+      fd_(fd),
+      declared_length_(declared_length),
+      total_bytes_written_(0) {
+  }
+
+  const int fd_;
+  const size_t declared_length_;
+  size_t total_bytes_written_;
+};
+
 // This method is using libz macros with old-style-casts
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wold-style-cast"
@@ -980,9 +1076,8 @@ static inline int zlib_inflateInit2(z_stream* stream, int window_bits) {
 }
 #pragma GCC diagnostic pop
 
-static int32_t InflateToFile(int fd, const ZipEntry* entry,
-                             uint8_t* begin, uint32_t length,
-                             uint64_t* crc_out) {
+static int32_t InflateEntryToWriter(int fd, const ZipEntry* entry,
+                                    Writer* writer, uint64_t* crc_out) {
   const size_t kBufSize = 32768;
   std::vector<uint8_t> read_buf(kBufSize);
   std::vector<uint8_t> write_buf(kBufSize);
@@ -1057,12 +1152,10 @@ static int32_t InflateToFile(int fd, const ZipEntry* entry,
     if (zstream.avail_out == 0 ||
       (zerr == Z_STREAM_END && zstream.avail_out != kBufSize)) {
       const size_t write_size = zstream.next_out - &write_buf[0];
-      // The file might have declared a bogus length.
-      if (write_size + write_count > length) {
-        return -1;
+      if (!writer->Append(&write_buf[0], write_size)) {
+        // The file might have declared a bogus length.
+        return kInconsistentInformation;
       }
-      memcpy(begin + write_count, &write_buf[0], write_size);
-      write_count += write_size;
 
       zstream.next_out = &write_buf[0];
       zstream.avail_out = kBufSize;
@@ -1083,8 +1176,41 @@ static int32_t InflateToFile(int fd, const ZipEntry* entry,
   return 0;
 }
 
-int32_t ExtractToMemory(ZipArchiveHandle handle,
-                        ZipEntry* entry, uint8_t* begin, uint32_t size) {
+static int32_t CopyEntryToWriter(int fd, const ZipEntry* entry, Writer* writer,
+                                 uint64_t *crc_out) {
+  static const uint32_t kBufSize = 32768;
+  std::vector<uint8_t> buf(kBufSize);
+
+  const uint32_t length = entry->uncompressed_length;
+  uint32_t count = 0;
+  uint64_t crc = 0;
+  while (count < length) {
+    uint32_t remaining = length - count;
+
+    // Safe conversion because kBufSize is narrow enough for a 32 bit signed
+    // value.
+    const ssize_t block_size = (remaining > kBufSize) ? kBufSize : remaining;
+    const ssize_t actual = TEMP_FAILURE_RETRY(read(fd, &buf[0], block_size));
+
+    if (actual != block_size) {
+      ALOGW("CopyFileToFile: copy read failed (" ZD " vs " ZD ")", actual, block_size);
+      return kIoError;
+    }
+
+    if (!writer->Append(&buf[0], block_size)) {
+      return kIoError;
+    }
+    crc = crc32(crc, &buf[0], block_size);
+    count += block_size;
+  }
+
+  *crc_out = crc;
+
+  return 0;
+}
+
+int32_t ExtractToWriter(ZipArchiveHandle handle,
+                        ZipEntry* entry, Writer* writer) {
   ZipArchive* archive = reinterpret_cast<ZipArchive*>(handle);
   const uint16_t method = entry->method;
   off64_t data_offset = entry->offset;
@@ -1098,9 +1224,9 @@ int32_t ExtractToMemory(ZipArchiveHandle handle,
   int32_t return_value = -1;
   uint64_t crc = 0;
   if (method == kCompressStored) {
-    return_value = CopyFileToFile(archive->fd, begin, size, &crc);
+    return_value = CopyEntryToWriter(archive->fd, entry, writer, &crc);
   } else if (method == kCompressDeflated) {
-    return_value = InflateToFile(archive->fd, entry, begin, size, &crc);
+    return_value = InflateEntryToWriter(archive->fd, entry, writer, &crc);
   }
 
   if (!return_value && entry->has_data_descriptor) {
@@ -1120,55 +1246,20 @@ int32_t ExtractToMemory(ZipArchiveHandle handle,
   return return_value;
 }
 
+int32_t ExtractToMemory(ZipArchiveHandle handle, ZipEntry* entry,
+                        uint8_t* begin, uint32_t size) {
+  std::unique_ptr<Writer> writer(new MemoryWriter(begin, size));
+  return ExtractToWriter(handle, entry, writer.get());
+}
+
 int32_t ExtractEntryToFile(ZipArchiveHandle handle,
                            ZipEntry* entry, int fd) {
-  const uint32_t declared_length = entry->uncompressed_length;
-
-  const off64_t current_offset = lseek64(fd, 0, SEEK_CUR);
-  if (current_offset == -1) {
-    ALOGW("Zip: unable to seek to current location on fd %d: %s", fd,
-          strerror(errno));
+  std::unique_ptr<Writer> writer(FileWriter::Create(fd, entry));
+  if (writer.get() == nullptr) {
     return kIoError;
   }
 
-  int result = 0;
-#if defined(__linux__)
-  // Make sure we have enough space on the volume to extract the compressed
-  // entry. Note that the call to ftruncate below will change the file size but
-  // will not allocate space on disk.
-  if (declared_length > 0) {
-    result = TEMP_FAILURE_RETRY(fallocate(fd, 0, current_offset, declared_length));
-    if (result == -1) {
-      ALOGW("Zip: unable to allocate space for file to %" PRId64 ": %s",
-            static_cast<int64_t>(declared_length + current_offset), strerror(errno));
-      return kIoError;
-    }
-  }
-#endif  // defined(__linux__)
-
-  result = TEMP_FAILURE_RETRY(ftruncate(fd, declared_length + current_offset));
-  if (result == -1) {
-    ALOGW("Zip: unable to truncate file to %" PRId64 ": %s",
-          static_cast<int64_t>(declared_length + current_offset), strerror(errno));
-    return kIoError;
-  }
-
-  // Don't attempt to map a region of length 0. We still need the
-  // ftruncate() though, since the API guarantees that we will truncate
-  // the file to the end of the uncompressed output.
-  if (declared_length == 0) {
-      return 0;
-  }
-
-  android::FileMap map;
-  if (!map.create(kTempMappingFileName, fd, current_offset, declared_length, false)) {
-    return kMmapFailed;
-  }
-
-  const int32_t error = ExtractToMemory(handle, entry,
-                                        reinterpret_cast<uint8_t*>(map.getDataPtr()),
-                                        map.getDataLength());
-  return error;
+  return ExtractToWriter(handle, entry, writer.get());
 }
 
 const char* ErrorCodeString(int32_t error_code) {
diff --git a/libziparchive/zip_archive_test.cc b/libziparchive/zip_archive_test.cc
index 64faa6de2..f8952ce65 100644
--- a/libziparchive/zip_archive_test.cc
+++ b/libziparchive/zip_archive_test.cc
@@ -23,6 +23,7 @@
 #include <unistd.h>
 #include <vector>
 
+#include <base/file.h>
 #include <gtest/gtest.h>
 
 static std::string test_data_dir;
@@ -228,6 +229,44 @@ static const uint32_t kEmptyEntriesZip[] = {
       0x54557478, 0x13030005, 0x7552e25c, 0x01000b78, 0x00428904, 0x13880400,
       0x4b500000, 0x00000605, 0x00010000, 0x004f0001, 0x00430000, 0x00000000 };
 
+// This is a zip file containing a single entry (ab.txt) that contains
+// 90072 repetitions of the string "ab\n" and has an uncompressed length
+// of 270216 bytes.
+static const uint16_t kAbZip[] = {
+  0x4b50, 0x0403, 0x0014, 0x0000, 0x0008, 0x51d2, 0x4698, 0xc4b0,
+  0x2cda, 0x011b, 0x0000, 0x1f88, 0x0004, 0x0006, 0x001c, 0x6261,
+  0x742e, 0x7478, 0x5455, 0x0009, 0x7c03, 0x3a09, 0x7c55, 0x3a09,
+  0x7555, 0x0b78, 0x0100, 0x8904, 0x0042, 0x0400, 0x1388, 0x0000,
+  0xc2ed, 0x0d31, 0x0000, 0x030c, 0x7fa0, 0x3b2e, 0x22ff, 0xa2aa,
+  0x841f, 0x45fc, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555, 0x5555,
+  0x5555, 0x5555, 0x5555, 0x5555, 0xdd55, 0x502c, 0x014b, 0x1e02,
+  0x1403, 0x0000, 0x0800, 0xd200, 0x9851, 0xb046, 0xdac4, 0x1b2c,
+  0x0001, 0x8800, 0x041f, 0x0600, 0x1800, 0x0000, 0x0000, 0x0100,
+  0x0000, 0xa000, 0x0081, 0x0000, 0x6100, 0x2e62, 0x7874, 0x5574,
+  0x0554, 0x0300, 0x097c, 0x553a, 0x7875, 0x000b, 0x0401, 0x4289,
+  0x0000, 0x8804, 0x0013, 0x5000, 0x054b, 0x0006, 0x0000, 0x0100,
+  0x0100, 0x4c00, 0x0000, 0x5b00, 0x0001, 0x0000, 0x0000
+};
+
+static const uint8_t kAbTxtName[] = { 'a', 'b', '.', 't', 'x', 't' };
+static const uint16_t kAbTxtNameLength = sizeof(kAbTxtName);
+static const size_t kAbUncompressedSize = 270216;
+
 static int make_temporary_file(const char* file_name_pattern) {
   char full_path[1024];
   // Account for differences between the host and the target.
@@ -275,6 +314,55 @@ TEST(ziparchive, EmptyEntries) {
   close(output_fd);
 }
 
+TEST(ziparchive, EntryLargerThan32K) {
+  char temp_file_pattern[] = "entry_larger_than_32k_test_XXXXXX";
+  int fd = make_temporary_file(temp_file_pattern);
+  ASSERT_NE(-1, fd);
+  ASSERT_TRUE(android::base::WriteFully(fd, reinterpret_cast<const uint8_t*>(kAbZip),
+                         sizeof(kAbZip) - 1));
+  ZipArchiveHandle handle;
+  ASSERT_EQ(0, OpenArchiveFd(fd, "EntryLargerThan32KTest", &handle));
+
+  ZipEntry entry;
+  ZipEntryName ab_name;
+  ab_name.name = kAbTxtName;
+  ab_name.name_length = kAbTxtNameLength;
+  ASSERT_EQ(0, FindEntry(handle, ab_name, &entry));
+  ASSERT_EQ(kAbUncompressedSize, entry.uncompressed_length);
+
+  // Extract the entry to memory.
+  std::vector<uint8_t> buffer(kAbUncompressedSize);
+  ASSERT_EQ(0, ExtractToMemory(handle, &entry, &buffer[0], buffer.size()));
+
+  // Extract the entry to a file.
+  char output_file_pattern[] = "entry_larger_than_32k_test_output_XXXXXX";
+  int output_fd = make_temporary_file(output_file_pattern);
+  ASSERT_NE(-1, output_fd);
+  ASSERT_EQ(0, ExtractEntryToFile(handle, &entry, output_fd));
+
+  // Make sure the extracted file size is as expected.
+  struct stat stat_buf;
+  ASSERT_EQ(0, fstat(output_fd, &stat_buf));
+  ASSERT_EQ(kAbUncompressedSize, static_cast<size_t>(stat_buf.st_size));
+
+  // Read the file back to a buffer and make sure the contents are
+  // the same as the memory buffer we extracted directly to.
+  std::vector<uint8_t> file_contents(kAbUncompressedSize);
+  ASSERT_EQ(0, lseek64(output_fd, 0, SEEK_SET));
+  ASSERT_TRUE(android::base::ReadFully(output_fd, &file_contents[0], file_contents.size()));
+  ASSERT_EQ(file_contents, buffer);
+
+  for (int i = 0; i < 90072; ++i) {
+    const uint8_t* line = &file_contents[0] + (3 * i);
+    ASSERT_EQ('a', line[0]);
+    ASSERT_EQ('b', line[1]);
+    ASSERT_EQ('\n', line[2]);
+  }
+
+  close(fd);
+  close(output_fd);
+}
+
 TEST(ziparchive, TrailerAfterEOCD) {
   char temp_file_pattern[] = "trailer_after_eocd_test_XXXXXX";
   int fd = make_temporary_file(temp_file_pattern);