| 1 |
// Protocol Buffers - Google's data interchange format |
| 2 |
// Copyright 2008 Google Inc. All rights reserved. |
| 3 |
// https://developers.google.com/protocol-buffers/ |
| 4 |
// |
| 5 |
// Redistribution and use in source and binary forms, with or without |
| 6 |
// modification, are permitted provided that the following conditions are |
| 7 |
// met: |
| 8 |
// |
| 9 |
// * Redistributions of source code must retain the above copyright |
| 10 |
// notice, this list of conditions and the following disclaimer. |
| 11 |
// * Redistributions in binary form must reproduce the above |
| 12 |
// copyright notice, this list of conditions and the following disclaimer |
| 13 |
// in the documentation and/or other materials provided with the |
| 14 |
// distribution. |
| 15 |
// * Neither the name of Google Inc. nor the names of its |
| 16 |
// contributors may be used to endorse or promote products derived from |
| 17 |
// this software without specific prior written permission. |
| 18 |
// |
| 19 |
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 20 |
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 21 |
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 22 |
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 23 |
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 24 |
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 25 |
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 26 |
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 27 |
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 28 |
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 29 |
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 30 |
|
| 31 |
#ifndef GOOGLE_PROTOBUF_PARSE_CONTEXT_H__ |
| 32 |
#define GOOGLE_PROTOBUF_PARSE_CONTEXT_H__ |
| 33 |
|
| 34 |
#include <cstdint> |
| 35 |
#include <cstring> |
| 36 |
#include <string> |
| 37 |
|
| 38 |
#include <google/protobuf/io/coded_stream.h> |
| 39 |
#include <google/protobuf/io/zero_copy_stream.h> |
| 40 |
#include <google/protobuf/arena.h> |
| 41 |
#include <google/protobuf/arenastring.h> |
| 42 |
#include <google/protobuf/implicit_weak_message.h> |
| 43 |
#include <google/protobuf/metadata_lite.h> |
| 44 |
#include <google/protobuf/port.h> |
| 45 |
#include <google/protobuf/repeated_field.h> |
| 46 |
#include <google/protobuf/wire_format_lite.h> |
| 47 |
#include <google/protobuf/stubs/strutil.h> |
| 48 |
|
| 49 |
#include <google/protobuf/port_def.inc> |
| 50 |
|
| 51 |
|
| 52 |
namespace google { |
| 53 |
namespace protobuf { |
| 54 |
|
| 55 |
class UnknownFieldSet; |
| 56 |
class DescriptorPool; |
| 57 |
class MessageFactory; |
| 58 |
|
| 59 |
namespace internal { |
| 60 |
|
| 61 |
// Template code below needs to know about the existence of these functions. |
| 62 |
PROTOBUF_EXPORT void WriteVarint(uint32 num, uint64 val, std::string* s); |
| 63 |
PROTOBUF_EXPORT void WriteLengthDelimited(uint32 num, StringPiece val, |
| 64 |
std::string* s); |
| 65 |
// Inline because it is just forwarding to s->WriteVarint |
| 66 |
inline void WriteVarint(uint32 num, uint64 val, UnknownFieldSet* s); |
| 67 |
inline void WriteLengthDelimited(uint32 num, StringPiece val, |
| 68 |
UnknownFieldSet* s); |
| 69 |
|
| 70 |
|
| 71 |
// The basic abstraction the parser is designed for is a slight modification |
| 72 |
// of the ZeroCopyInputStream (ZCIS) abstraction. A ZCIS presents a serialized |
| 73 |
// stream as a series of buffers that concatenate to the full stream. |
| 74 |
// Pictorially a ZCIS presents a stream in chunks like so |
| 75 |
// [---------------------------------------------------------------] |
| 76 |
// [---------------------] chunk 1 |
| 77 |
// [----------------------------] chunk 2 |
| 78 |
// chunk 3 [--------------] |
| 79 |
// |
| 80 |
// Where the '-' represent the bytes which are vertically lined up with the |
| 81 |
// bytes of the stream. The proto parser requires its input to be presented |
| 82 |
// similarly with the extra |
| 83 |
// property that each chunk has kSlopBytes past its end that overlaps with the |
| 84 |
// first kSlopBytes of the next chunk, or if there is no next chunk at least its |
| 85 |
// still valid to read those bytes. Again, pictorially, we now have |
| 86 |
// |
| 87 |
// [---------------------------------------------------------------] |
| 88 |
// [-------------------....] chunk 1 |
| 89 |
// [------------------------....] chunk 2 |
| 90 |
// chunk 3 [------------------..**] |
| 91 |
// chunk 4 [--****] |
| 92 |
// Here '-' mean the bytes of the stream or chunk and '.' means bytes past the |
| 93 |
// chunk that match up with the start of the next chunk. Above each chunk has |
| 94 |
// 4 '.' after the chunk. In the case these 'overflow' bytes represents bytes |
| 95 |
// past the stream, indicated by '*' above, their values are unspecified. It is |
| 96 |
// still legal to read them (ie. should not segfault). Reading past the |
| 97 |
// end should be detected by the user and indicated as an error. |
| 98 |
// |
| 99 |
// The reason for this, admittedly, unconventional invariant is to ruthlessly |
| 100 |
// optimize the protobuf parser. Having an overlap helps in two important ways. |
| 101 |
// Firstly it alleviates having to performing bounds checks if a piece of code |
| 102 |
// is guaranteed to not read more than kSlopBytes. Secondly, and more |
| 103 |
// importantly, the protobuf wireformat is such that reading a key/value pair is |
| 104 |
// always less than 16 bytes. This removes the need to change to next buffer in |
| 105 |
// the middle of reading primitive values. Hence there is no need to store and |
| 106 |
// load the current position. |
| 107 |
|
| 108 |
class PROTOBUF_EXPORT EpsCopyInputStream { |
| 109 |
public: |
| 110 |
enum { kSlopBytes = 16, kMaxCordBytesToCopy = 512 }; |
| 111 |
|
| 112 |
explicit EpsCopyInputStream(bool enable_aliasing) |
| 113 |
: aliasing_(enable_aliasing ? kOnPatch : kNoAliasing) {} |
| 114 |
|
| 115 |
void BackUp(const char* ptr) { |
| 116 |
GOOGLE_DCHECK(ptr <= buffer_end_ + kSlopBytes); |
| 117 |
int count; |
| 118 |
if (next_chunk_ == buffer_) { |
| 119 |
count = static_cast<int>(buffer_end_ + kSlopBytes - ptr); |
| 120 |
} else { |
| 121 |
count = size_ + static_cast<int>(buffer_end_ - ptr); |
| 122 |
} |
| 123 |
if (count > 0) StreamBackUp(count); |
| 124 |
} |
| 125 |
|
| 126 |
// If return value is negative it's an error |
| 127 |
PROTOBUF_MUST_USE_RESULT int PushLimit(const char* ptr, int limit) { |
| 128 |
GOOGLE_DCHECK(limit >= 0 && limit <= INT_MAX - kSlopBytes); |
| 129 |
// This add is safe due to the invariant above, because |
| 130 |
// ptr - buffer_end_ <= kSlopBytes. |
| 131 |
limit += static_cast<int>(ptr - buffer_end_); |
| 132 |
limit_end_ = buffer_end_ + (std::min)(0, limit); |
| 133 |
auto old_limit = limit_; |
| 134 |
limit_ = limit; |
| 135 |
return old_limit - limit; |
| 136 |
} |
| 137 |
|
| 138 |
PROTOBUF_MUST_USE_RESULT bool PopLimit(int delta) { |
| 139 |
if (PROTOBUF_PREDICT_FALSE(!EndedAtLimit())) return false; |
| 140 |
limit_ = limit_ + delta; |
| 141 |
// TODO(gerbens) We could remove this line and hoist the code to |
| 142 |
// DoneFallback. Study the perf/bin-size effects. |
| 143 |
limit_end_ = buffer_end_ + (std::min)(0, limit_); |
| 144 |
return true; |
| 145 |
} |
| 146 |
|
| 147 |
PROTOBUF_MUST_USE_RESULT const char* Skip(const char* ptr, int size) { |
| 148 |
if (size <= buffer_end_ + kSlopBytes - ptr) { |
| 149 |
return ptr + size; |
| 150 |
} |
| 151 |
return SkipFallback(ptr, size); |
| 152 |
} |
| 153 |
PROTOBUF_MUST_USE_RESULT const char* ReadString(const char* ptr, int size, |
| 154 |
std::string* s) { |
| 155 |
if (size <= buffer_end_ + kSlopBytes - ptr) { |
| 156 |
s->assign(ptr, size); |
| 157 |
return ptr + size; |
| 158 |
} |
| 159 |
return ReadStringFallback(ptr, size, s); |
| 160 |
} |
| 161 |
PROTOBUF_MUST_USE_RESULT const char* AppendString(const char* ptr, int size, |
| 162 |
std::string* s) { |
| 163 |
if (size <= buffer_end_ + kSlopBytes - ptr) { |
| 164 |
s->append(ptr, size); |
| 165 |
return ptr + size; |
| 166 |
} |
| 167 |
return AppendStringFallback(ptr, size, s); |
| 168 |
} |
| 169 |
|
| 170 |
template <typename Tag, typename T> |
| 171 |
PROTOBUF_MUST_USE_RESULT const char* ReadRepeatedFixed(const char* ptr, |
| 172 |
Tag expected_tag, |
| 173 |
RepeatedField<T>* out); |
| 174 |
|
| 175 |
template <typename T> |
| 176 |
PROTOBUF_MUST_USE_RESULT const char* ReadPackedFixed(const char* ptr, |
| 177 |
int size, |
| 178 |
RepeatedField<T>* out); |
| 179 |
template <typename Add> |
| 180 |
PROTOBUF_MUST_USE_RESULT const char* ReadPackedVarint(const char* ptr, |
| 181 |
Add add); |
| 182 |
|
| 183 |
uint32 LastTag() const { return last_tag_minus_1_ + 1; } |
| 184 |
bool ConsumeEndGroup(uint32 start_tag) { |
| 185 |
bool res = last_tag_minus_1_ == start_tag; |
| 186 |
last_tag_minus_1_ = 0; |
| 187 |
return res; |
| 188 |
} |
| 189 |
bool EndedAtLimit() const { return last_tag_minus_1_ == 0; } |
| 190 |
bool EndedAtEndOfStream() const { return last_tag_minus_1_ == 1; } |
| 191 |
void SetLastTag(uint32 tag) { last_tag_minus_1_ = tag - 1; } |
| 192 |
void SetEndOfStream() { last_tag_minus_1_ = 1; } |
| 193 |
bool IsExceedingLimit(const char* ptr) { |
| 194 |
return ptr > limit_end_ && |
| 195 |
(next_chunk_ == nullptr || ptr - buffer_end_ > limit_); |
| 196 |
} |
| 197 |
int BytesUntilLimit(const char* ptr) const { |
| 198 |
return limit_ + static_cast<int>(buffer_end_ - ptr); |
| 199 |
} |
| 200 |
// Returns true if more data is available, if false is returned one has to |
| 201 |
// call Done for further checks. |
| 202 |
bool DataAvailable(const char* ptr) { return ptr < limit_end_; } |
| 203 |
|
| 204 |
protected: |
| 205 |
// Returns true is limit (either an explicit limit or end of stream) is |
| 206 |
// reached. It aligns *ptr across buffer seams. |
| 207 |
// If limit is exceeded it returns true and ptr is set to null. |
| 208 |
bool DoneWithCheck(const char** ptr, int d) { |
| 209 |
GOOGLE_DCHECK(*ptr); |
| 210 |
if (PROTOBUF_PREDICT_TRUE(*ptr < limit_end_)) return false; |
| 211 |
int overrun = *ptr - buffer_end_; |
| 212 |
GOOGLE_DCHECK_LE(overrun, kSlopBytes); // Guaranteed by parse loop. |
| 213 |
if (overrun == |
| 214 |
limit_) { // No need to flip buffers if we ended on a limit. |
| 215 |
// If we actually overrun the buffer and next_chunk_ is null. It means |
| 216 |
// the stream ended and we passed the stream end. |
| 217 |
if (overrun > 0 && next_chunk_ == nullptr) *ptr = nullptr; |
| 218 |
return true; |
| 219 |
} |
| 220 |
auto res = DoneFallback(overrun, d); |
| 221 |
*ptr = res.first; |
| 222 |
return res.second; |
| 223 |
} |
| 224 |
|
| 225 |
const char* InitFrom(StringPiece flat) { |
| 226 |
overall_limit_ = 0; |
| 227 |
if (flat.size() > kSlopBytes) { |
| 228 |
limit_ = kSlopBytes; |
| 229 |
limit_end_ = buffer_end_ = flat.data() + flat.size() - kSlopBytes; |
| 230 |
next_chunk_ = buffer_; |
| 231 |
if (aliasing_ == kOnPatch) aliasing_ = kNoDelta; |
| 232 |
return flat.data(); |
| 233 |
} else { |
| 234 |
std::memcpy(buffer_, flat.data(), flat.size()); |
| 235 |
limit_ = 0; |
| 236 |
limit_end_ = buffer_end_ = buffer_ + flat.size(); |
| 237 |
next_chunk_ = nullptr; |
| 238 |
if (aliasing_ == kOnPatch) { |
| 239 |
aliasing_ = reinterpret_cast<std::uintptr_t>(flat.data()) - |
| 240 |
reinterpret_cast<std::uintptr_t>(buffer_); |
| 241 |
} |
| 242 |
return buffer_; |
| 243 |
} |
| 244 |
} |
| 245 |
|
| 246 |
const char* InitFrom(io::ZeroCopyInputStream* zcis); |
| 247 |
|
| 248 |
const char* InitFrom(io::ZeroCopyInputStream* zcis, int limit) { |
| 249 |
if (limit == -1) return InitFrom(zcis); |
| 250 |
overall_limit_ = limit; |
| 251 |
auto res = InitFrom(zcis); |
| 252 |
limit_ = limit - static_cast<int>(buffer_end_ - res); |
| 253 |
limit_end_ = buffer_end_ + (std::min)(0, limit_); |
| 254 |
return res; |
| 255 |
} |
| 256 |
|
| 257 |
private: |
| 258 |
const char* limit_end_; // buffer_end_ + min(limit_, 0) |
| 259 |
const char* buffer_end_; |
| 260 |
const char* next_chunk_; |
| 261 |
int size_; |
| 262 |
int limit_; // relative to buffer_end_; |
| 263 |
io::ZeroCopyInputStream* zcis_ = nullptr; |
| 264 |
char buffer_[2 * kSlopBytes] = {}; |
| 265 |
enum { kNoAliasing = 0, kOnPatch = 1, kNoDelta = 2 }; |
| 266 |
std::uintptr_t aliasing_ = kNoAliasing; |
| 267 |
// This variable is used to communicate how the parse ended, in order to |
| 268 |
// completely verify the parsed data. A wire-format parse can end because of |
| 269 |
// one of the following conditions: |
| 270 |
// 1) A parse can end on a pushed limit. |
| 271 |
// 2) A parse can end on End Of Stream (EOS). |
| 272 |
// 3) A parse can end on 0 tag (only valid for toplevel message). |
| 273 |
// 4) A parse can end on an end-group tag. |
| 274 |
// This variable should always be set to 0, which indicates case 1. If the |
| 275 |
// parse terminated due to EOS (case 2), it's set to 1. In case the parse |
| 276 |
// ended due to a terminating tag (case 3 and 4) it's set to (tag - 1). |
| 277 |
// This var doesn't really belong in EpsCopyInputStream and should be part of |
| 278 |
// the ParseContext, but case 2 is most easily and optimally implemented in |
| 279 |
// DoneFallback. |
| 280 |
uint32 last_tag_minus_1_ = 0; |
| 281 |
int overall_limit_ = INT_MAX; // Overall limit independent of pushed limits. |
| 282 |
// Pretty random large number that seems like a safe allocation on most |
| 283 |
// systems. TODO(gerbens) do we need to set this as build flag? |
| 284 |
enum { kSafeStringSize = 50000000 }; |
| 285 |
|
| 286 |
// Advances to next buffer chunk returns a pointer to the same logical place |
| 287 |
// in the stream as set by overrun. Overrun indicates the position in the slop |
| 288 |
// region the parse was left (0 <= overrun <= kSlopBytes). Returns true if at |
| 289 |
// limit, at which point the returned pointer maybe null if there was an |
| 290 |
// error. The invariant of this function is that it's guaranteed that |
| 291 |
// kSlopBytes bytes can be accessed from the returned ptr. This function might |
| 292 |
// advance more buffers than one in the underlying ZeroCopyInputStream. |
| 293 |
std::pair<const char*, bool> DoneFallback(int overrun, int depth); |
| 294 |
// Advances to the next buffer, at most one call to Next() on the underlying |
| 295 |
// ZeroCopyInputStream is made. This function DOES NOT match the returned |
| 296 |
// pointer to where in the slop region the parse ends, hence no overrun |
| 297 |
// parameter. This is useful for string operations where you always copy |
| 298 |
// to the end of the buffer (including the slop region). |
| 299 |
const char* Next(); |
| 300 |
// overrun is the location in the slop region the stream currently is |
| 301 |
// (0 <= overrun <= kSlopBytes). To prevent flipping to the next buffer of |
| 302 |
// the ZeroCopyInputStream in the case the parse will end in the last |
| 303 |
// kSlopBytes of the current buffer. depth is the current depth of nested |
| 304 |
// groups (or negative if the use case does not need careful tracking). |
| 305 |
inline const char* NextBuffer(int overrun, int depth); |
| 306 |
const char* SkipFallback(const char* ptr, int size); |
| 307 |
const char* AppendStringFallback(const char* ptr, int size, std::string* str); |
| 308 |
const char* ReadStringFallback(const char* ptr, int size, std::string* str); |
| 309 |
bool StreamNext(const void** data) { |
| 310 |
bool res = zcis_->Next(data, &size_); |
| 311 |
if (res) overall_limit_ -= size_; |
| 312 |
return res; |
| 313 |
} |
| 314 |
void StreamBackUp(int count) { |
| 315 |
zcis_->BackUp(count); |
| 316 |
overall_limit_ += count; |
| 317 |
} |
| 318 |
|
| 319 |
template <typename A> |
| 320 |
const char* AppendSize(const char* ptr, int size, const A& append) { |
| 321 |
int chunk_size = buffer_end_ + kSlopBytes - ptr; |
| 322 |
do { |
| 323 |
GOOGLE_DCHECK(size > chunk_size); |
| 324 |
if (next_chunk_ == nullptr) return nullptr; |
| 325 |
append(ptr, chunk_size); |
| 326 |
ptr += chunk_size; |
| 327 |
size -= chunk_size; |
| 328 |
// TODO(gerbens) Next calls NextBuffer which generates buffers with |
| 329 |
// overlap and thus incurs cost of copying the slop regions. This is not |
| 330 |
// necessary for reading strings. We should just call Next buffers. |
| 331 |
if (limit_ <= kSlopBytes) return nullptr; |
| 332 |
ptr = Next(); |
| 333 |
if (ptr == nullptr) return nullptr; // passed the limit |
| 334 |
ptr += kSlopBytes; |
| 335 |
chunk_size = buffer_end_ + kSlopBytes - ptr; |
| 336 |
} while (size > chunk_size); |
| 337 |
append(ptr, size); |
| 338 |
return ptr + size; |
| 339 |
} |
| 340 |
|
| 341 |
// AppendUntilEnd appends data until a limit (either a PushLimit or end of |
| 342 |
// stream. Normal payloads are from length delimited fields which have an |
| 343 |
// explicit size. Reading until limit only comes when the string takes |
| 344 |
// the place of a protobuf, ie RawMessage/StringRawMessage, lazy fields and |
| 345 |
// implicit weak messages. We keep these methods private and friend them. |
| 346 |
template <typename A> |
| 347 |
const char* AppendUntilEnd(const char* ptr, const A& append) { |
| 348 |
if (ptr - buffer_end_ > limit_) return nullptr; |
| 349 |
while (limit_ > kSlopBytes) { |
| 350 |
int chunk_size = buffer_end_ + kSlopBytes - ptr; |
| 351 |
GOOGLE_DCHECK_GE(chunk_size, 0); |
| 352 |
append(ptr, chunk_size); |
| 353 |
ptr = Next(); |
| 354 |
if (ptr == nullptr) return limit_end_; |
| 355 |
ptr += kSlopBytes; |
| 356 |
} |
| 357 |
auto end = buffer_end_ + limit_; |
| 358 |
GOOGLE_DCHECK(end >= ptr); |
| 359 |
append(ptr, end - ptr); |
| 360 |
return end; |
| 361 |
} |
| 362 |
|
| 363 |
PROTOBUF_MUST_USE_RESULT const char* AppendString(const char* ptr, |
| 364 |
std::string* str) { |
| 365 |
return AppendUntilEnd( |
| 366 |
ptr, [str](const char* p, ptrdiff_t s) { str->append(p, s); }); |
| 367 |
} |
| 368 |
friend class ImplicitWeakMessage; |
| 369 |
}; |
| 370 |
|
| 371 |
// ParseContext holds all data that is global to the entire parse. Most |
| 372 |
// importantly it contains the input stream, but also recursion depth and also |
| 373 |
// stores the end group tag, in case a parser ended on a endgroup, to verify |
| 374 |
// matching start/end group tags. |
| 375 |
class PROTOBUF_EXPORT ParseContext : public EpsCopyInputStream { |
| 376 |
public: |
| 377 |
struct Data { |
| 378 |
const DescriptorPool* pool = nullptr; |
| 379 |
MessageFactory* factory = nullptr; |
| 380 |
}; |
| 381 |
|
| 382 |
template <typename... T> |
| 383 |
ParseContext(int depth, bool aliasing, const char** start, T&&... args) |
| 384 |
: EpsCopyInputStream(aliasing), depth_(depth) { |
| 385 |
*start = InitFrom(std::forward<T>(args)...); |
| 386 |
} |
| 387 |
|
| 388 |
void TrackCorrectEnding() { group_depth_ = 0; } |
| 389 |
|
| 390 |
bool Done(const char** ptr) { return DoneWithCheck(ptr, group_depth_); } |
| 391 |
|
| 392 |
int depth() const { return depth_; } |
| 393 |
|
| 394 |
Data& data() { return data_; } |
| 395 |
const Data& data() const { return data_; } |
| 396 |
|
| 397 |
template <typename T> |
| 398 |
PROTOBUF_MUST_USE_RESULT const char* ParseMessage(T* msg, const char* ptr); |
| 399 |
// We outline when the type is generic and we go through a virtual |
| 400 |
const char* ParseMessage(MessageLite* msg, const char* ptr); |
| 401 |
const char* ParseMessage(Message* msg, const char* ptr); |
| 402 |
|
| 403 |
template <typename T> |
| 404 |
PROTOBUF_MUST_USE_RESULT PROTOBUF_ALWAYS_INLINE const char* ParseGroup( |
| 405 |
T* msg, const char* ptr, uint32 tag) { |
| 406 |
if (--depth_ < 0) return nullptr; |
| 407 |
group_depth_++; |
| 408 |
ptr = msg->_InternalParse(ptr, this); |
| 409 |
group_depth_--; |
| 410 |
depth_++; |
| 411 |
if (PROTOBUF_PREDICT_FALSE(!ConsumeEndGroup(tag))) return nullptr; |
| 412 |
return ptr; |
| 413 |
} |
| 414 |
|
| 415 |
private: |
| 416 |
// The context keeps an internal stack to keep track of the recursive |
| 417 |
// part of the parse state. |
| 418 |
// Current depth of the active parser, depth counts down. |
| 419 |
// This is used to limit recursion depth (to prevent overflow on malicious |
| 420 |
// data), but is also used to index in stack_ to store the current state. |
| 421 |
int depth_; |
| 422 |
// Unfortunately necessary for the fringe case of ending on 0 or end-group tag |
| 423 |
// in the last kSlopBytes of a ZeroCopyInputStream chunk. |
| 424 |
int group_depth_ = INT_MIN; |
| 425 |
Data data_; |
| 426 |
}; |
| 427 |
|
| 428 |
template <uint32 tag> |
| 429 |
bool ExpectTag(const char* ptr) { |
| 430 |
if (tag < 128) { |
| 431 |
return *ptr == tag; |
| 432 |
} else { |
| 433 |
static_assert(tag < 128 * 128, "We only expect tags for 1 or 2 bytes"); |
| 434 |
char buf[2] = {static_cast<char>(tag | 0x80), static_cast<char>(tag >> 7)}; |
| 435 |
return std::memcmp(ptr, buf, 2) == 0; |
| 436 |
} |
| 437 |
} |
| 438 |
|
| 439 |
template <int> |
| 440 |
struct EndianHelper; |
| 441 |
|
| 442 |
template <> |
| 443 |
struct EndianHelper<1> { |
| 444 |
static uint8 Load(const void* p) { return *static_cast<const uint8*>(p); } |
| 445 |
}; |
| 446 |
|
| 447 |
template <> |
| 448 |
struct EndianHelper<2> { |
| 449 |
static uint16 Load(const void* p) { |
| 450 |
uint16 tmp; |
| 451 |
std::memcpy(&tmp, p, 2); |
| 452 |
#ifndef PROTOBUF_LITTLE_ENDIAN |
| 453 |
tmp = bswap_16(tmp); |
| 454 |
#endif |
| 455 |
return tmp; |
| 456 |
} |
| 457 |
}; |
| 458 |
|
| 459 |
template <> |
| 460 |
struct EndianHelper<4> { |
| 461 |
static uint32 Load(const void* p) { |
| 462 |
uint32 tmp; |
| 463 |
std::memcpy(&tmp, p, 4); |
| 464 |
#ifndef PROTOBUF_LITTLE_ENDIAN |
| 465 |
tmp = bswap_32(tmp); |
| 466 |
#endif |
| 467 |
return tmp; |
| 468 |
} |
| 469 |
}; |
| 470 |
|
| 471 |
template <> |
| 472 |
struct EndianHelper<8> { |
| 473 |
static uint64 Load(const void* p) { |
| 474 |
uint64 tmp; |
| 475 |
std::memcpy(&tmp, p, 8); |
| 476 |
#ifndef PROTOBUF_LITTLE_ENDIAN |
| 477 |
tmp = bswap_64(tmp); |
| 478 |
#endif |
| 479 |
return tmp; |
| 480 |
} |
| 481 |
}; |
| 482 |
|
| 483 |
template <typename T> |
| 484 |
T UnalignedLoad(const char* p) { |
| 485 |
auto tmp = EndianHelper<sizeof(T)>::Load(p); |
| 486 |
T res; |
| 487 |
memcpy(&res, &tmp, sizeof(T)); |
| 488 |
return res; |
| 489 |
} |
| 490 |
|
| 491 |
PROTOBUF_EXPORT |
| 492 |
std::pair<const char*, uint32> VarintParseSlow32(const char* p, uint32 res); |
| 493 |
PROTOBUF_EXPORT |
| 494 |
std::pair<const char*, uint64> VarintParseSlow64(const char* p, uint32 res); |
| 495 |
|
| 496 |
inline const char* VarintParseSlow(const char* p, uint32 res, uint32* out) { |
| 497 |
auto tmp = VarintParseSlow32(p, res); |
| 498 |
*out = tmp.second; |
| 499 |
return tmp.first; |
| 500 |
} |
| 501 |
|
| 502 |
inline const char* VarintParseSlow(const char* p, uint32 res, uint64* out) { |
| 503 |
auto tmp = VarintParseSlow64(p, res); |
| 504 |
*out = tmp.second; |
| 505 |
return tmp.first; |
| 506 |
} |
| 507 |
|
| 508 |
template <typename T> |
| 509 |
PROTOBUF_MUST_USE_RESULT const char* VarintParse(const char* p, T* out) { |
| 510 |
auto ptr = reinterpret_cast<const uint8*>(p); |
| 511 |
uint32 res = ptr[0]; |
| 512 |
if (!(res & 0x80)) { |
| 513 |
*out = res; |
| 514 |
return p + 1; |
| 515 |
} |
| 516 |
uint32 byte = ptr[1]; |
| 517 |
res += (byte - 1) << 7; |
| 518 |
if (!(byte & 0x80)) { |
| 519 |
*out = res; |
| 520 |
return p + 2; |
| 521 |
} |
| 522 |
return VarintParseSlow(p, res, out); |
| 523 |
} |
| 524 |
|
| 525 |
// Used for tags, could read up to 5 bytes which must be available. |
| 526 |
// Caller must ensure its safe to call. |
| 527 |
|
| 528 |
PROTOBUF_EXPORT |
| 529 |
std::pair<const char*, uint32> ReadTagFallback(const char* p, uint32 res); |
| 530 |
|
| 531 |
// Same as ParseVarint but only accept 5 bytes at most. |
| 532 |
inline const char* ReadTag(const char* p, uint32* out, uint32 /*max_tag*/ = 0) { |
| 533 |
uint32 res = static_cast<uint8>(p[0]); |
| 534 |
if (res < 128) { |
| 535 |
*out = res; |
| 536 |
return p + 1; |
| 537 |
} |
| 538 |
uint32 second = static_cast<uint8>(p[1]); |
| 539 |
res += (second - 1) << 7; |
| 540 |
if (second < 128) { |
| 541 |
*out = res; |
| 542 |
return p + 2; |
| 543 |
} |
| 544 |
auto tmp = ReadTagFallback(p, res); |
| 545 |
*out = tmp.second; |
| 546 |
return tmp.first; |
| 547 |
} |
| 548 |
|
| 549 |
// Decode 2 consecutive bytes of a varint and returns the value, shifted left |
| 550 |
// by 1. It simultaneous updates *ptr to *ptr + 1 or *ptr + 2 depending if the |
| 551 |
// first byte's continuation bit is set. |
| 552 |
// If bit 15 of return value is set (equivalent to the continuation bits of both |
| 553 |
// bytes being set) the varint continues, otherwise the parse is done. On x86 |
| 554 |
// movsx eax, dil |
| 555 |
// add edi, eax |
| 556 |
// adc [rsi], 1 |
| 557 |
// add eax, eax |
| 558 |
// and eax, edi |
| 559 |
inline uint32 DecodeTwoBytes(const char** ptr) { |
| 560 |
uint32 value = UnalignedLoad<uint16>(*ptr); |
| 561 |
// Sign extend the low byte continuation bit |
| 562 |
uint32_t x = static_cast<int8_t>(value); |
| 563 |
// This add is an amazing operation, it cancels the low byte continuation bit |
| 564 |
// from y transferring it to the carry. Simultaneously it also shifts the 7 |
| 565 |
// LSB left by one tightly against high byte varint bits. Hence value now |
| 566 |
// contains the unpacked value shifted left by 1. |
| 567 |
value += x; |
| 568 |
// Use the carry to update the ptr appropriately. |
| 569 |
*ptr += value < x ? 2 : 1; |
| 570 |
return value & (x + x); // Mask out the high byte iff no continuation |
| 571 |
} |
| 572 |
|
| 573 |
// More efficient varint parsing for big varints |
| 574 |
inline const char* ParseBigVarint(const char* p, uint64* out) { |
| 575 |
auto pnew = p; |
| 576 |
auto tmp = DecodeTwoBytes(&pnew); |
| 577 |
uint64 res = tmp >> 1; |
| 578 |
if (PROTOBUF_PREDICT_TRUE(std::int16_t(tmp) >= 0)) { |
| 579 |
*out = res; |
| 580 |
return pnew; |
| 581 |
} |
| 582 |
for (std::uint32_t i = 1; i < 5; i++) { |
| 583 |
pnew = p + 2 * i; |
| 584 |
tmp = DecodeTwoBytes(&pnew); |
| 585 |
res += (static_cast<std::uint64_t>(tmp) - 2) << (14 * i - 1); |
| 586 |
if (PROTOBUF_PREDICT_TRUE(std::int16_t(tmp) >= 0)) { |
| 587 |
*out = res; |
| 588 |
return pnew; |
| 589 |
} |
| 590 |
} |
| 591 |
return nullptr; |
| 592 |
} |
| 593 |
|
| 594 |
PROTOBUF_EXPORT |
| 595 |
std::pair<const char*, int32> ReadSizeFallback(const char* p, uint32 first); |
| 596 |
// Used for tags, could read up to 5 bytes which must be available. Additionally |
| 597 |
// it makes sure the unsigned value fits a int32, otherwise returns nullptr. |
| 598 |
// Caller must ensure its safe to call. |
| 599 |
inline uint32 ReadSize(const char** pp) { |
| 600 |
auto p = *pp; |
| 601 |
uint32 res = static_cast<uint8>(p[0]); |
| 602 |
if (res < 128) { |
| 603 |
*pp = p + 1; |
| 604 |
return res; |
| 605 |
} |
| 606 |
auto x = ReadSizeFallback(p, res); |
| 607 |
*pp = x.first; |
| 608 |
return x.second; |
| 609 |
} |
| 610 |
|
| 611 |
// Some convenience functions to simplify the generated parse loop code. |
| 612 |
// Returning the value and updating the buffer pointer allows for nicer |
| 613 |
// function composition. We rely on the compiler to inline this. |
| 614 |
// Also in debug compiles having local scoped variables tend to generated |
| 615 |
// stack frames that scale as O(num fields). |
| 616 |
inline uint64 ReadVarint64(const char** p) { |
| 617 |
uint64 tmp; |
| 618 |
*p = VarintParse(*p, &tmp); |
| 619 |
return tmp; |
| 620 |
} |
| 621 |
|
| 622 |
inline uint32 ReadVarint32(const char** p) { |
| 623 |
uint32 tmp; |
| 624 |
*p = VarintParse(*p, &tmp); |
| 625 |
return tmp; |
| 626 |
} |
| 627 |
|
| 628 |
inline int64 ReadVarintZigZag64(const char** p) { |
| 629 |
uint64 tmp; |
| 630 |
*p = VarintParse(*p, &tmp); |
| 631 |
return WireFormatLite::ZigZagDecode64(tmp); |
| 632 |
} |
| 633 |
|
| 634 |
inline int32 ReadVarintZigZag32(const char** p) { |
| 635 |
uint64 tmp; |
| 636 |
*p = VarintParse(*p, &tmp); |
| 637 |
return WireFormatLite::ZigZagDecode32(static_cast<uint32>(tmp)); |
| 638 |
} |
| 639 |
|
| 640 |
template <typename T> |
| 641 |
PROTOBUF_MUST_USE_RESULT const char* ParseContext::ParseMessage( |
| 642 |
T* msg, const char* ptr) { |
| 643 |
int size = ReadSize(&ptr); |
| 644 |
if (!ptr) return nullptr; |
| 645 |
auto old = PushLimit(ptr, size); |
| 646 |
if (--depth_ < 0) return nullptr; |
| 647 |
ptr = msg->_InternalParse(ptr, this); |
| 648 |
if (PROTOBUF_PREDICT_FALSE(ptr == nullptr)) return nullptr; |
| 649 |
depth_++; |
| 650 |
if (!PopLimit(old)) return nullptr; |
| 651 |
return ptr; |
| 652 |
} |
| 653 |
|
| 654 |
template <typename Add> |
| 655 |
const char* ReadPackedVarintArray(const char* ptr, const char* end, Add add) { |
| 656 |
while (ptr < end) { |
| 657 |
uint64 varint; |
| 658 |
ptr = VarintParse(ptr, &varint); |
| 659 |
if (ptr == nullptr) return nullptr; |
| 660 |
add(varint); |
| 661 |
} |
| 662 |
return ptr; |
| 663 |
} |
| 664 |
|
| 665 |
template <typename Add> |
| 666 |
const char* EpsCopyInputStream::ReadPackedVarint(const char* ptr, Add add) { |
| 667 |
int size = ReadSize(&ptr); |
| 668 |
if (ptr == nullptr) return nullptr; |
| 669 |
int chunk_size = buffer_end_ - ptr; |
| 670 |
while (size > chunk_size) { |
| 671 |
ptr = ReadPackedVarintArray(ptr, buffer_end_, add); |
| 672 |
if (ptr == nullptr) return nullptr; |
| 673 |
int overrun = ptr - buffer_end_; |
| 674 |
GOOGLE_DCHECK(overrun >= 0 && overrun <= kSlopBytes); |
| 675 |
if (size - chunk_size <= kSlopBytes) { |
| 676 |
// The current buffer contains all the information needed, we don't need |
| 677 |
// to flip buffers. However we must parse from a buffer with enough space |
| 678 |
// so we are not prone to a buffer overflow. |
| 679 |
char buf[kSlopBytes + 10] = {}; |
| 680 |
std::memcpy(buf, buffer_end_, kSlopBytes); |
| 681 |
GOOGLE_CHECK_LE(size - chunk_size, kSlopBytes); |
| 682 |
auto end = buf + (size - chunk_size); |
| 683 |
auto res = ReadPackedVarintArray(buf + overrun, end, add); |
| 684 |
if (res == nullptr || res != end) return nullptr; |
| 685 |
return buffer_end_ + (res - buf); |
| 686 |
} |
| 687 |
size -= overrun + chunk_size; |
| 688 |
GOOGLE_DCHECK_GT(size, 0); |
| 689 |
// We must flip buffers |
| 690 |
if (limit_ <= kSlopBytes) return nullptr; |
| 691 |
ptr = Next(); |
| 692 |
if (ptr == nullptr) return nullptr; |
| 693 |
ptr += overrun; |
| 694 |
chunk_size = buffer_end_ - ptr; |
| 695 |
} |
| 696 |
auto end = ptr + size; |
| 697 |
ptr = ReadPackedVarintArray(ptr, end, add); |
| 698 |
return end == ptr ? ptr : nullptr; |
| 699 |
} |
| 700 |
|
| 701 |
// Helper for verification of utf8 |
| 702 |
PROTOBUF_EXPORT |
| 703 |
bool VerifyUTF8(StringPiece s, const char* field_name); |
| 704 |
|
| 705 |
inline bool VerifyUTF8(const std::string* s, const char* field_name) { |
| 706 |
return VerifyUTF8(*s, field_name); |
| 707 |
} |
| 708 |
|
| 709 |
// All the string parsers with or without UTF checking and for all CTypes. |
| 710 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* InlineGreedyStringParser( |
| 711 |
std::string* s, const char* ptr, ParseContext* ctx); |
| 712 |
|
| 713 |
|
| 714 |
// Add any of the following lines to debug which parse function is failing. |
| 715 |
|
| 716 |
#define GOOGLE_PROTOBUF_ASSERT_RETURN(predicate, ret) \ |
| 717 |
if (!(predicate)) { \ |
| 718 |
/* ::raise(SIGINT); */ \ |
| 719 |
/* GOOGLE_LOG(ERROR) << "Parse failure"; */ \ |
| 720 |
return ret; \ |
| 721 |
} |
| 722 |
|
| 723 |
#define GOOGLE_PROTOBUF_PARSER_ASSERT(predicate) \ |
| 724 |
GOOGLE_PROTOBUF_ASSERT_RETURN(predicate, nullptr) |
| 725 |
|
| 726 |
template <typename T> |
| 727 |
PROTOBUF_MUST_USE_RESULT const char* FieldParser(uint64 tag, T& field_parser, |
| 728 |
const char* ptr, |
| 729 |
ParseContext* ctx) { |
| 730 |
uint32 number = tag >> 3; |
| 731 |
GOOGLE_PROTOBUF_PARSER_ASSERT(number != 0); |
| 732 |
using WireType = internal::WireFormatLite::WireType; |
| 733 |
switch (tag & 7) { |
| 734 |
case WireType::WIRETYPE_VARINT: { |
| 735 |
uint64 value; |
| 736 |
ptr = VarintParse(ptr, &value); |
| 737 |
GOOGLE_PROTOBUF_PARSER_ASSERT(ptr); |
| 738 |
field_parser.AddVarint(number, value); |
| 739 |
break; |
| 740 |
} |
| 741 |
case WireType::WIRETYPE_FIXED64: { |
| 742 |
uint64 value = UnalignedLoad<uint64>(ptr); |
| 743 |
ptr += 8; |
| 744 |
field_parser.AddFixed64(number, value); |
| 745 |
break; |
| 746 |
} |
| 747 |
case WireType::WIRETYPE_LENGTH_DELIMITED: { |
| 748 |
ptr = field_parser.ParseLengthDelimited(number, ptr, ctx); |
| 749 |
GOOGLE_PROTOBUF_PARSER_ASSERT(ptr); |
| 750 |
break; |
| 751 |
} |
| 752 |
case WireType::WIRETYPE_START_GROUP: { |
| 753 |
ptr = field_parser.ParseGroup(number, ptr, ctx); |
| 754 |
GOOGLE_PROTOBUF_PARSER_ASSERT(ptr); |
| 755 |
break; |
| 756 |
} |
| 757 |
case WireType::WIRETYPE_END_GROUP: { |
| 758 |
GOOGLE_LOG(FATAL) << "Can't happen"; |
| 759 |
break; |
| 760 |
} |
| 761 |
case WireType::WIRETYPE_FIXED32: { |
| 762 |
uint32 value = UnalignedLoad<uint32>(ptr); |
| 763 |
ptr += 4; |
| 764 |
field_parser.AddFixed32(number, value); |
| 765 |
break; |
| 766 |
} |
| 767 |
default: |
| 768 |
return nullptr; |
| 769 |
} |
| 770 |
return ptr; |
| 771 |
} |
| 772 |
|
| 773 |
template <typename T> |
| 774 |
PROTOBUF_MUST_USE_RESULT const char* WireFormatParser(T& field_parser, |
| 775 |
const char* ptr, |
| 776 |
ParseContext* ctx) { |
| 777 |
while (!ctx->Done(&ptr)) { |
| 778 |
uint32 tag; |
| 779 |
ptr = ReadTag(ptr, &tag); |
| 780 |
GOOGLE_PROTOBUF_PARSER_ASSERT(ptr != nullptr); |
| 781 |
if (tag == 0 || (tag & 7) == 4) { |
| 782 |
ctx->SetLastTag(tag); |
| 783 |
return ptr; |
| 784 |
} |
| 785 |
ptr = FieldParser(tag, field_parser, ptr, ctx); |
| 786 |
GOOGLE_PROTOBUF_PARSER_ASSERT(ptr != nullptr); |
| 787 |
} |
| 788 |
return ptr; |
| 789 |
} |
| 790 |
|
| 791 |
// The packed parsers parse repeated numeric primitives directly into the |
| 792 |
// corresponding field |
| 793 |
|
| 794 |
// These are packed varints |
| 795 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedInt32Parser( |
| 796 |
void* object, const char* ptr, ParseContext* ctx); |
| 797 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedUInt32Parser( |
| 798 |
void* object, const char* ptr, ParseContext* ctx); |
| 799 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedInt64Parser( |
| 800 |
void* object, const char* ptr, ParseContext* ctx); |
| 801 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedUInt64Parser( |
| 802 |
void* object, const char* ptr, ParseContext* ctx); |
| 803 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedSInt32Parser( |
| 804 |
void* object, const char* ptr, ParseContext* ctx); |
| 805 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedSInt64Parser( |
| 806 |
void* object, const char* ptr, ParseContext* ctx); |
| 807 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedEnumParser( |
| 808 |
void* object, const char* ptr, ParseContext* ctx); |
| 809 |
|
| 810 |
template <typename T> |
| 811 |
PROTOBUF_MUST_USE_RESULT const char* PackedEnumParser( |
| 812 |
void* object, const char* ptr, ParseContext* ctx, bool (*is_valid)(int), |
| 813 |
InternalMetadata* metadata, int field_num) { |
| 814 |
return ctx->ReadPackedVarint( |
| 815 |
ptr, [object, is_valid, metadata, field_num](uint64 val) { |
| 816 |
if (is_valid(val)) { |
| 817 |
static_cast<RepeatedField<int>*>(object)->Add(val); |
| 818 |
} else { |
| 819 |
WriteVarint(field_num, val, metadata->mutable_unknown_fields<T>()); |
| 820 |
} |
| 821 |
}); |
| 822 |
} |
| 823 |
|
| 824 |
template <typename T> |
| 825 |
PROTOBUF_MUST_USE_RESULT const char* PackedEnumParserArg( |
| 826 |
void* object, const char* ptr, ParseContext* ctx, |
| 827 |
bool (*is_valid)(const void*, int), const void* data, |
| 828 |
InternalMetadata* metadata, int field_num) { |
| 829 |
return ctx->ReadPackedVarint( |
| 830 |
ptr, [object, is_valid, data, metadata, field_num](uint64 val) { |
| 831 |
if (is_valid(data, val)) { |
| 832 |
static_cast<RepeatedField<int>*>(object)->Add(val); |
| 833 |
} else { |
| 834 |
WriteVarint(field_num, val, metadata->mutable_unknown_fields<T>()); |
| 835 |
} |
| 836 |
}); |
| 837 |
} |
| 838 |
|
| 839 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedBoolParser( |
| 840 |
void* object, const char* ptr, ParseContext* ctx); |
| 841 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedFixed32Parser( |
| 842 |
void* object, const char* ptr, ParseContext* ctx); |
| 843 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedSFixed32Parser( |
| 844 |
void* object, const char* ptr, ParseContext* ctx); |
| 845 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedFixed64Parser( |
| 846 |
void* object, const char* ptr, ParseContext* ctx); |
| 847 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedSFixed64Parser( |
| 848 |
void* object, const char* ptr, ParseContext* ctx); |
| 849 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedFloatParser( |
| 850 |
void* object, const char* ptr, ParseContext* ctx); |
| 851 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* PackedDoubleParser( |
| 852 |
void* object, const char* ptr, ParseContext* ctx); |
| 853 |
|
| 854 |
// This is the only recursive parser. |
| 855 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* UnknownGroupLiteParse( |
| 856 |
std::string* unknown, const char* ptr, ParseContext* ctx); |
| 857 |
// This is a helper to for the UnknownGroupLiteParse but is actually also |
| 858 |
// useful in the generated code. It uses overload on std::string* vs |
| 859 |
// UnknownFieldSet* to make the generated code isomorphic between full and lite. |
| 860 |
PROTOBUF_EXPORT PROTOBUF_MUST_USE_RESULT const char* UnknownFieldParse( |
| 861 |
uint32 tag, std::string* unknown, const char* ptr, ParseContext* ctx); |
| 862 |
|
| 863 |
} // namespace internal |
| 864 |
} // namespace protobuf |
| 865 |
} // namespace google |
| 866 |
|
| 867 |
#include <google/protobuf/port_undef.inc> |
| 868 |
|
| 869 |
#endif // GOOGLE_PROTOBUF_PARSE_CONTEXT_H__ |