Avoid duplicate data buffering in CPDF_SyntaxParser::ReadStream().

Allow sub-streams created from an IFX_SeekableReadStream to provide
stream data without copying memory.
The data will only reside in the top-level stream.

For example:
For file
http://www.major-landrover.ru/upload/attachments/f/9/f96aab07dab04ae89c8a509ec1ef2b31.pdf
(18 Mb)

The memory usage is reduced by ~13 Mb.

Change-Id: I2595c014d0fbe1fdd181cc04965cfd7d901c2d88
Reviewed-on: https://pdfium-review.googlesource.com/35930
Commit-Queue: Art Snake <art-snake@yandex-team.ru>
Reviewed-by: dsinclair <dsinclair@chromium.org>

3 files changed, 61 insertions, 12 deletions

diff --git a/core/fpdfapi/edit/cpdf_flateencoder.cpp b/core/fpdfapi/edit/cpdf_flateencoder.cpp
index a290da6dac..573c141ff2 100644
--- a/core/fpdfapi/edit/cpdf_flateencoder.cpp
+++ b/core/fpdfapi/edit/cpdf_flateencoder.cpp
@@ -15,7 +15,7 @@
 
 CPDF_FlateEncoder::CPDF_FlateEncoder(const CPDF_Stream* pStream,
                                      bool bFlateEncode)
-    : m_dwSize(0), m_pAcc(pdfium::MakeRetain<CPDF_StreamAcc>(pStream)) {
+    : m_pAcc(pdfium::MakeRetain<CPDF_StreamAcc>(pStream)), m_dwSize(0) {
   m_pAcc->LoadAllDataRaw();
 
   bool bHasFilter = pStream && pStream->HasFilter();
diff --git a/core/fpdfapi/edit/cpdf_flateencoder.h b/core/fpdfapi/edit/cpdf_flateencoder.h
index 05633f6814..14ca7cec16 100644
--- a/core/fpdfapi/edit/cpdf_flateencoder.h
+++ b/core/fpdfapi/edit/cpdf_flateencoder.h
@@ -34,14 +34,14 @@ class CPDF_FlateEncoder {
   }
 
  private:
+  RetainPtr<CPDF_StreamAcc> m_pAcc;
+
   uint32_t m_dwSize;
   MaybeOwned<uint8_t, FxFreeDeleter> m_pData;
 
   // Only one of these two pointers is valid at any time.
   UnownedPtr<const CPDF_Dictionary> m_pDict;
   std::unique_ptr<CPDF_Dictionary> m_pClonedDict;
-
-  RetainPtr<CPDF_StreamAcc> m_pAcc;
 };
 
 #endif  // CORE_FPDFAPI_EDIT_CPDF_FLATEENCODER_H_
diff --git a/core/fpdfapi/parser/cpdf_syntax_parser.cpp b/core/fpdfapi/parser/cpdf_syntax_parser.cpp
index 00eed49300..6edfb24ba2 100644
--- a/core/fpdfapi/parser/cpdf_syntax_parser.cpp
+++ b/core/fpdfapi/parser/cpdf_syntax_parser.cpp
@@ -34,6 +34,37 @@ namespace {
 
 enum class ReadStatus { Normal, Backslash, Octal, FinishOctal, CarriageReturn };
 
+class ReadableSubStream : public IFX_SeekableReadStream {
+ public:
+  ReadableSubStream(const RetainPtr<IFX_SeekableReadStream>& pFileRead,
+                    FX_FILESIZE part_offset,
+                    FX_FILESIZE part_size)
+      : m_pFileRead(pFileRead),
+        m_PartOffset(part_offset),
+        m_PartSize(part_size) {}
+
+  ~ReadableSubStream() override = default;
+
+  // IFX_SeekableReadStream overrides:
+  bool ReadBlock(void* buffer, FX_FILESIZE offset, size_t size) override {
+    FX_SAFE_FILESIZE safe_end = offset;
+    safe_end += size;
+    // Check that requested range is valid, to prevent calling of ReadBlock
+    // of original m_pFileRead with incorrect params.
+    if (!safe_end.IsValid() || safe_end.ValueOrDie() > m_PartSize)
+      return false;
+
+    return m_pFileRead->ReadBlock(buffer, m_PartOffset + offset, size);
+  }
+
+  FX_FILESIZE GetSize() override { return m_PartSize; }
+
+ private:
+  RetainPtr<IFX_SeekableReadStream> m_pFileRead;
+  FX_FILESIZE m_PartOffset;
+  FX_FILESIZE m_PartSize;
+};
+
 }  // namespace
 
 // static
@@ -577,7 +608,6 @@ std::unique_ptr<CPDF_Stream> CPDF_SyntaxParser::ReadStream(
   ToNextLine();
   const FX_FILESIZE streamStartPos = GetPos();
 
-  std::unique_ptr<uint8_t, FxFreeDeleter> pData;
   if (len > 0) {
     FX_SAFE_FILESIZE pos = GetPos();
     pos += len;
@@ -585,12 +615,18 @@ std::unique_ptr<CPDF_Stream> CPDF_SyntaxParser::ReadStream(
       len = -1;
   }
 
+  RetainPtr<IFX_SeekableReadStream> data;
   if (len > 0) {
-    pData.reset(FX_Alloc(uint8_t, len));
-    // We should try read data first to allow the Validator to request data
+    // Check data availability first to allow the Validator to request data
     // smoothly, without jumps.
-    if (!ReadBlock(pData.get(), len))
+    if (!GetValidator()->CheckDataRangeAndRequestIfUnavailable(
+            m_HeaderOffset + GetPos(), len)) {
       return nullptr;
+    }
+
+    data = pdfium::MakeRetain<ReadableSubStream>(
+        GetValidator(), m_HeaderOffset + GetPos(), len);
+    SetPos(GetPos() + len);
   }
 
   const ByteStringView kEndStreamStr("endstream");
@@ -611,7 +647,7 @@ std::unique_ptr<CPDF_Stream> CPDF_SyntaxParser::ReadStream(
     // specified length, it signals the end of stream.
     if (memcmp(m_WordBuffer, kEndStreamStr.raw_str(),
                kEndStreamStr.GetLength()) != 0) {
-      pData.reset();
+      data.Reset();
       len = -1;
       SetPos(streamStartPos);
     }
@@ -628,14 +664,27 @@ std::unique_ptr<CPDF_Stream> CPDF_SyntaxParser::ReadStream(
     ASSERT(len >= 0);
     if (len > 0) {
       SetPos(streamStartPos);
-      pData.reset(FX_Alloc(uint8_t, len));
-      if (!ReadBlock(pData.get(), len))
+      // Check data availability first to allow the Validator to request data
+      // smoothly, without jumps.
+      if (!GetValidator()->CheckDataRangeAndRequestIfUnavailable(
+              m_HeaderOffset + GetPos(), len)) {
         return nullptr;
+      }
+
+      data = pdfium::MakeRetain<ReadableSubStream>(
+          GetValidator(), m_HeaderOffset + GetPos(), len);
+      SetPos(GetPos() + len);
     }
   }
 
-  auto pStream =
-      pdfium::MakeUnique<CPDF_Stream>(std::move(pData), len, std::move(pDict));
+  auto pStream = pdfium::MakeUnique<CPDF_Stream>();
+  if (data) {
+    pStream->InitStreamFromFile(data, std::move(pDict));
+  } else {
+    DCHECK(!len);
+    // Empty stream
+    pStream->InitStream(nullptr, 0, std::move(pDict));
+  }
   const FX_FILESIZE end_stream_offset = GetPos();
   memset(m_WordBuffer, 0, kEndObjStr.GetLength() + 1);
   GetNextWordInternal(nullptr);