Add new HTTP/2 and HPACK decoder in net/http2/.

net/http2/decoder/decode_buffer.h

Index: net/http2/decoder/decode_buffer.h
diff --git a/net/http2/decoder/decode_buffer.h b/net/http2/decoder/decode_buffer.h
new file mode 100644
index 0000000000000000000000000000000000000000..56f695f4b74bbeaa3caa7809cf923f185ab06b98
--- /dev/null
+++ b/net/http2/decoder/decode_buffer.h
@@ -0,0 +1,289 @@
+// Copyright 2016 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef NET_HTTP2_DECODER_DECODE_BUFFER_H_
+#define NET_HTTP2_DECODER_DECODE_BUFFER_H_
+
+// DecodeBuffer provides primitives for decoding various integer types found
+// in HTTP/2 frames.
+// DecodeBuffer wraps a byte array from which we can read and decode serialized
+// HTTP/2 frames, or parts thereof. DecodeBuffer is intended only for stack
+// allocation, where the caller is typically going to use the DecodeBuffer
+// instance as part of decoding the entire buffer before returning to its own
+// caller. Only the concrete Slow* methods are defined in the cc file,
+// all other methods are defined in this header file to enable inlining.
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include <algorithm>
+
+#include "base/logging.h"
+#include "base/macros.h"
+#include "base/strings/string_piece.h"
+#include "net/base/net_export.h"
+
+namespace net {
+class DecodeBufferSubset;
+
+class NET_EXPORT_PRIVATE DecodeBuffer {
+ public:
+  DecodeBuffer(const char* buffer, size_t len)
+      : buffer_(buffer), cursor_(buffer), beyond_(buffer + len) {
+    DCHECK_NE(buffer, nullptr);
+    DCHECK_LE(len, MaxDecodeBufferLength());
+  }
+  explicit DecodeBuffer(base::StringPiece s)
+      : DecodeBuffer(s.data(), s.size()) {}
+  // Constructor for character arrays, typically in tests. For example:
+  //    const char input[] = { 0x11 };
+  //    DecodeBuffer b(input);
+  template <size_t N>
+  explicit DecodeBuffer(const char (&buf)[N]) : DecodeBuffer(buf, N) {}
+
+  bool Empty() const { return cursor_ >= beyond_; }
+  bool HasData() const { return cursor_ < beyond_; }
+  size_t Remaining() const {
+    DCHECK_LE(cursor_, beyond_);
+    return beyond_ - cursor_;
+  }
+  size_t Offset() const { return cursor_ - buffer_; }
+  size_t FullSize() const { return beyond_ - buffer_; }
+
+  // Returns the minimum of the number of bytes remaining in this DecodeBuffer
+  // and |length|, in support of determining how much of some structure/payload
+  // is in this DecodeBuffer.
+  size_t MinLengthRemaining(size_t length) const {
+    return std::min(length, Remaining());
+  }
+
+  // For string decoding, returns a pointer to the next byte/char to be decoded.
+  const char* cursor() const { return cursor_; }
+  // Advances the cursor (pointer to the next byte/char to be decoded).
+  void AdvanceCursor(size_t amount) {
+    DCHECK_LE(amount, Remaining());  // Need at least that much remaining.
+    DCHECK_EQ(subset_, nullptr) << "Access via subset only when present.";
+    cursor_ += amount;
+  }
+
+  // Only call methods starting "Decode" when there is enough input remaining.
+  char DecodeChar() {
+    DCHECK_LE(1u, Remaining());  // Need at least one byte remaining.
+    DCHECK_EQ(subset_, nullptr) << "Access via subset only when present.";
+    return *cursor_++;
+  }
+
+  uint8_t DecodeUInt8() { return static_cast<uint8_t>(DecodeChar()); }
+
+  uint16_t DecodeUInt16() {
+    DCHECK_LE(2u, Remaining());
+    const uint8_t b1 = DecodeUInt8();
+    const uint8_t b2 = DecodeUInt8();
+    // Note that chars are automatically promoted to ints during arithmetic,
+    // so the b1 << 8 doesn't end up as zero before being or-ed with b2.
+    // And the left-shift operator has higher precedence than the or operator.
+    return b1 << 8 | b2;
+  }
+
+  uint32_t DecodeUInt24() {
+    DCHECK_LE(3u, Remaining());
+    const uint8_t b1 = DecodeUInt8();
+    const uint8_t b2 = DecodeUInt8();
+    const uint8_t b3 = DecodeUInt8();
+    return b1 << 16 | b2 << 8 | b3;
+  }
+
+  // For 31-bit unsigned integers, where the 32nd bit is reserved for future
+  // use (i.e. the high-bit of the first byte of the encoding); examples:
+  // the Stream Id in a frame header or the Window Size Increment in a
+  // WINDOW_UPDATE frame.
+  uint32_t DecodeUInt31() {
+    DCHECK_LE(4u, Remaining());
+    const uint8_t b1 = DecodeUInt8() & 0x7f;  // Mask out the high order bit.
+    const uint8_t b2 = DecodeUInt8();
+    const uint8_t b3 = DecodeUInt8();
+    const uint8_t b4 = DecodeUInt8();
+    return b1 << 24 | b2 << 16 | b3 << 8 | b4;
+  }
+
+  uint32_t DecodeUInt32() {
+    DCHECK_LE(4u, Remaining());
+    const uint8_t b1 = DecodeUInt8();
+    const uint8_t b2 = DecodeUInt8();
+    const uint8_t b3 = DecodeUInt8();
+    const uint8_t b4 = DecodeUInt8();
+    return b1 << 24 | b2 << 16 | b3 << 8 | b4;
+  }
+
+  // SlowDecode* routines are used for decoding a multi-field structure when
+  // there may not be enough bytes in the buffer to decode the entirety of the
+  // structure.
+
+  // Read as much of an unsigned int field of an encoded structure as possible,
+  // keeping track via decode_offset of our position in the encoded structure.
+  // Returns true if the field has been fully decoded.
+  // |field_size| is the number of bytes of the encoding of the field (usually
+  // a compile time fixed value).
+  // |field_offset| is the offset of the first byte of the encoding of the field
+  // within the encoding of that structure (usually a compile time fixed value).
+  // |*decode_offset| is the offset of the byte to be decoded next.
+  // |*value| is the storage for the decoded value, and is used for storing
+  // partially decoded values; if some, but not all, bytes of the encoding are
+  // available then this method will return having stored the decoded bytes into
+  // *value.
+  bool SlowDecodeUnsignedInt(uint32_t field_size,
+                             uint32_t field_offset,
+                             uint32_t* decode_offset,
+                             uint32_t* value);
+
+  // Like SlowDecodeUnsignedInt, but specifically for 8-bit unsigned integers.
+  // Obviously a byte can't be split (on our byte addressable machines), but
+  // a larger structure containing such a field might be.
+  bool SlowDecodeUInt8(uint32_t field_offset,
+                       uint32_t* decode_offset,
+                       uint8_t* value);
+
+  // Like SlowDecodeUnsignedInt, but specifically for 16-bit unsigned integers.
+  bool SlowDecodeUInt16(uint32_t field_offset,
+                        uint32_t* decode_offset,
+                        uint16_t* value);
+
+  // Like SlowDecodeUnsignedInt, but specifically for 24-bit unsigned integers.
+  bool SlowDecodeUInt24(uint32_t field_offset,
+                        uint32_t* decode_offset,
+                        uint32_t* value);
+
+  // Like SlowDecodeUnsignedInt, but specifically for 31-bit unsigned integers.
+  // (same definition as for DecodeUInt31).
+  bool SlowDecodeUInt31(uint32_t field_offset,
+                        uint32_t* decode_offset,
+                        uint32_t* value);
+
+  // Like SlowDecodeUnsignedInt, but specifically for 31-bit unsigned integers.
+  bool SlowDecodeUInt32(uint32_t field_offset,
+                        uint32_t* decode_offset,
+                        uint32_t* value);
+
+  // Decodes an enum value, where the size (in bytes) of the encoding must be
+  // stated explicitly. It is assumed that under the covers enums are really
+  // just integers, and that we can static_cast them to and from uint32.
+  template <typename E>
+  bool SlowDecodeEnum(uint32_t field_size,
+                      uint32_t field_offset,
+                      uint32_t* decode_offset,
+                      E* value) {
+    uint32_t tmp = static_cast<uint32_t>(*value);
+    const bool done =
+        SlowDecodeUnsignedInt(field_size, field_offset, decode_offset, &tmp);
+    *value = static_cast<E>(tmp);
+    DCHECK_EQ(tmp, static_cast<uint32_t>(*value));
+    return done;
+  }
+
+  // We assume the decode buffers will typically be modest in size (i.e. a few
+  // K).
+  // Let's make sure during testing that we don't go very high, with 32MB
+  // selected rather arbitrarily.
+  static constexpr size_t MaxDecodeBufferLength() { return 1 << 25; }
+
+ protected:
+#ifndef NDEBUG
+  // These are part of validating during tests that there is at most one
+  // DecodeBufferSubset instance at a time for any DecodeBuffer instance.
+  void set_subset_of_base(DecodeBuffer* base,
+                          const DecodeBufferSubset* subset) {
+    DCHECK_EQ(this, subset);
+    base->set_subset(subset);
+  }
+  void clear_subset_of_base(DecodeBuffer* base,
+                            const DecodeBufferSubset* subset) {
+    DCHECK_EQ(this, subset);
+    base->clear_subset(subset);
+  }
+#endif
+
+ private:
+#ifndef NDEBUG
+  void set_subset(const DecodeBufferSubset* subset) {
+    DCHECK(subset != nullptr);
+    DCHECK_EQ(subset_, nullptr) << "There is already a subset";
+    subset_ = subset;
+  }
+  void clear_subset(const DecodeBufferSubset* subset) {
+    DCHECK(subset != nullptr);
+    DCHECK_EQ(subset_, subset);
+    subset_ = nullptr;
+  }
+#endif
+
+  // Prevent heap allocation of DecodeBuffer.
+  static void* operator new(size_t s);
+  static void* operator new[](size_t s);
+  static void operator delete(void* p);
+  static void operator delete[](void* p);
+
+  const char* const buffer_;
+  const char* cursor_;
+  const char* const beyond_;
+  const DecodeBufferSubset* subset_ = nullptr;  // Used for DCHECKs.
+
+  DISALLOW_COPY_AND_ASSIGN(DecodeBuffer);
+};
+
+// DecodeBufferSubset is used when decoding a known sized chunk of data, which
+// starts at base->cursor(), and continues for subset_len, which may be
+// entirely in |base|, or may extend beyond it (hence the MinLengthRemaining
+// in the constructor).
+// There are two benefits to using DecodeBufferSubset: it ensures that the
+// cursor of |base| is advanced when the subset's destructor runs, and it
+// ensures that the consumer of the subset can't go beyond the subset which
+// it is intended to decode.
+// There must be only a single DecodeBufferSubset at a time for a base
+// DecodeBuffer, though they can be nested (i.e. a DecodeBufferSubset's
+// base may itself be a DecodeBufferSubset). This avoids the AdvanceCursor
+// being called erroneously.
+class DecodeBufferSubset : public DecodeBuffer {
+ public:
+  DecodeBufferSubset(DecodeBuffer* base, size_t subset_len)
+      : DecodeBuffer(base->cursor(), base->MinLengthRemaining(subset_len)),
+#ifndef NDEBUG
+        start_base_offset_(base->Offset()),
+        max_base_offset_(start_base_offset_ + FullSize()),
+#endif
+        base_buffer_(base) {
+#ifndef NDEBUG
+    DCHECK_LE(max_base_offset_, base->FullSize());
+    set_subset_of_base(base_buffer_, this);
+#endif
+  }
+
+  ~DecodeBufferSubset() {
+    size_t offset = Offset();
+#ifndef NDEBUG
+    clear_subset_of_base(base_buffer_, this);
+    DCHECK_LE(Offset(), FullSize());
+    DCHECK_EQ(start_base_offset_, base_buffer_->Offset())
+        << "The base buffer was modified";
+    DCHECK_LE(offset, FullSize());
+    DCHECK_LE(start_base_offset_ + offset, base_buffer_->FullSize());
+#endif
+    base_buffer_->AdvanceCursor(offset);
+#ifndef NDEBUG
+    DCHECK_GE(max_base_offset_, base_buffer_->Offset());
+#endif
+  }
+
+ private:
+#ifndef NDEBUG
+  const size_t start_base_offset_;  // Used for DCHECKs.
+  const size_t max_base_offset_;    // Used for DCHECKs.
+#endif
+  DecodeBuffer* const base_buffer_;
+
+  DISALLOW_COPY_AND_ASSIGN(DecodeBufferSubset);
+};
+
+}  // namespace net
+
+#endif  // NET_HTTP2_DECODER_DECODE_BUFFER_H_