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 // Author: kenton@google.com (Kenton Varda)
32 // Based on original Protocol Buffers design by
33 // Sanjay Ghemawat, Jeff Dean, and others.
34 //
35 // This file contains the CodedInputStream and CodedOutputStream classes,
36 // which wrap a ZeroCopyInputStream or ZeroCopyOutputStream, respectively,
37 // and allow you to read or write individual pieces of data in various
38 // formats. In particular, these implement the varint encoding for
39 // integers, a simple variable-length encoding in which smaller numbers
40 // take fewer bytes.
41 //
42 // Typically these classes will only be used internally by the protocol
43 // buffer library in order to encode and decode protocol buffers. Clients
44 // of the library only need to know about this class if they wish to write
45 // custom message parsing or serialization procedures.
46 //
47 // CodedOutputStream example:
48 // // Write some data to "myfile". First we write a 4-byte "magic number"
49 // // to identify the file type, then write a length-delimited string. The
50 // // string is composed of a varint giving the length followed by the raw
51 // // bytes.
52 // int fd = open("myfile", O_CREAT | O_WRONLY);
53 // ZeroCopyOutputStream* raw_output = new FileOutputStream(fd);
54 // CodedOutputStream* coded_output = new CodedOutputStream(raw_output);
55 //
56 // int magic_number = 1234;
57 // char text[] = "Hello world!";
58 // coded_output->WriteLittleEndian32(magic_number);
59 // coded_output->WriteVarint32(strlen(text));
60 // coded_output->WriteRaw(text, strlen(text));
61 //
62 // delete coded_output;
63 // delete raw_output;
64 // close(fd);
65 //
66 // CodedInputStream example:
67 // // Read a file created by the above code.
68 // int fd = open("myfile", O_RDONLY);
69 // ZeroCopyInputStream* raw_input = new FileInputStream(fd);
70 // CodedInputStream* coded_input = new CodedInputStream(raw_input);
71 //
72 // coded_input->ReadLittleEndian32(&magic_number);
73 // if (magic_number != 1234) {
74 // cerr << "File not in expected format." << endl;
75 // return;
76 // }
77 //
78 // uint32_t size;
79 // coded_input->ReadVarint32(&size);
80 //
81 // char* text = new char[size + 1];
82 // coded_input->ReadRaw(buffer, size);
83 // text[size] = '\0';
84 //
85 // delete coded_input;
86 // delete raw_input;
87 // close(fd);
88 //
89 // cout << "Text is: " << text << endl;
90 // delete [] text;
91 //
92 // For those who are interested, varint encoding is defined as follows:
93 //
94 // The encoding operates on unsigned integers of up to 64 bits in length.
95 // Each byte of the encoded value has the format:
96 // * bits 0-6: Seven bits of the number being encoded.
97 // * bit 7: Zero if this is the last byte in the encoding (in which
98 // case all remaining bits of the number are zero) or 1 if
99 // more bytes follow.
100 // The first byte contains the least-significant 7 bits of the number, the
101 // second byte (if present) contains the next-least-significant 7 bits,
102 // and so on. So, the binary number 1011000101011 would be encoded in two
103 // bytes as "10101011 00101100".
104 //
105 // In theory, varint could be used to encode integers of any length.
106 // However, for practicality we set a limit at 64 bits. The maximum encoded
107 // length of a number is thus 10 bytes.
108
109 #ifndef GOOGLE_PROTOBUF_IO_CODED_STREAM_H__
110 #define GOOGLE_PROTOBUF_IO_CODED_STREAM_H__
111
112
113 #include <assert.h>
114
115 #include <atomic>
116 #include <climits>
117 #include <cstddef>
118 #include <cstring>
119 #include <limits>
120 #include <string>
121 #include <type_traits>
122 #include <utility>
123
124 #ifdef _WIN32
125 // Assuming windows is always little-endian.
126 #if !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST)
127 #define PROTOBUF_LITTLE_ENDIAN 1
128 #endif
129 #if defined(_MSC_VER) && _MSC_VER >= 1300 && !defined(__INTEL_COMPILER)
130 // If MSVC has "/RTCc" set, it will complain about truncating casts at
131 // runtime. This file contains some intentional truncating casts.
132 #pragma runtime_checks("c", off)
133 #endif
134 #else
135 #ifdef __APPLE__
136 #include <machine/endian.h> // __BYTE_ORDER
137 #elif defined(__FreeBSD__)
138 #include <sys/endian.h> // __BYTE_ORDER
139 #elif (defined(sun) || defined(__sun)) && (defined(__SVR4) || defined(__svr4__))
140 #include <sys/isa_defs.h> // __BYTE_ORDER
141 #elif defined(_AIX) || defined(__TOS_AIX__)
142 #include <sys/machine.h> // BYTE_ORDER
143 #else
144 #if !defined(__QNX__)
145 #include <endian.h> // __BYTE_ORDER
146 #endif
147 #endif
148 #if ((defined(__LITTLE_ENDIAN__) && !defined(__BIG_ENDIAN__)) || \
149 (defined(__BYTE_ORDER) && __BYTE_ORDER == __LITTLE_ENDIAN)) && \
150 !defined(PROTOBUF_DISABLE_LITTLE_ENDIAN_OPT_FOR_TEST)
151 #define PROTOBUF_LITTLE_ENDIAN 1
152 #endif
153 #endif
154 #include <google/protobuf/stubs/common.h>
155 #include <google/protobuf/stubs/logging.h>
156 #include <google/protobuf/stubs/strutil.h>
157 #include <google/protobuf/port.h>
158 #include <google/protobuf/stubs/port.h>
159
160
161 // Must be included last.
162 #include <google/protobuf/port_def.inc>
163
164 namespace google {
165 namespace protobuf {
166
167 class DescriptorPool;
168 class MessageFactory;
169 class ZeroCopyCodedInputStream;
170
171 namespace internal {
172 void MapTestForceDeterministic();
173 class EpsCopyByteStream;
174 } // namespace internal
175
176 namespace io {
177
178 // Defined in this file.
179 class CodedInputStream;
180 class CodedOutputStream;
181
182 // Defined in other files.
183 class ZeroCopyInputStream; // zero_copy_stream.h
184 class ZeroCopyOutputStream; // zero_copy_stream.h
185
186 // Class which reads and decodes binary data which is composed of varint-
187 // encoded integers and fixed-width pieces. Wraps a ZeroCopyInputStream.
188 // Most users will not need to deal with CodedInputStream.
189 //
190 // Most methods of CodedInputStream that return a bool return false if an
191 // underlying I/O error occurs or if the data is malformed. Once such a
192 // failure occurs, the CodedInputStream is broken and is no longer useful.
193 // After a failure, callers also should assume writes to "out" args may have
194 // occurred, though nothing useful can be determined from those writes.
195 class PROTOBUF_EXPORT CodedInputStream {
196 public:
197 // Create a CodedInputStream that reads from the given ZeroCopyInputStream.
198 explicit CodedInputStream(ZeroCopyInputStream* input);
199
200 // Create a CodedInputStream that reads from the given flat array. This is
201 // faster than using an ArrayInputStream. PushLimit(size) is implied by
202 // this constructor.
203 explicit CodedInputStream(const uint8_t* buffer, int size);
204
205 // Destroy the CodedInputStream and position the underlying
206 // ZeroCopyInputStream at the first unread byte. If an error occurred while
207 // reading (causing a method to return false), then the exact position of
208 // the input stream may be anywhere between the last value that was read
209 // successfully and the stream's byte limit.
210 ~CodedInputStream();
211
212 // Return true if this CodedInputStream reads from a flat array instead of
213 // a ZeroCopyInputStream.
214 inline bool IsFlat() const;
215
216 // Skips a number of bytes. Returns false if an underlying read error
217 // occurs.
218 inline bool Skip(int count);
219
220 // Sets *data to point directly at the unread part of the CodedInputStream's
221 // underlying buffer, and *size to the size of that buffer, but does not
222 // advance the stream's current position. This will always either produce
223 // a non-empty buffer or return false. If the caller consumes any of
224 // this data, it should then call Skip() to skip over the consumed bytes.
225 // This may be useful for implementing external fast parsing routines for
226 // types of data not covered by the CodedInputStream interface.
227 bool GetDirectBufferPointer(const void** data, int* size);
228
229 // Like GetDirectBufferPointer, but this method is inlined, and does not
230 // attempt to Refresh() if the buffer is currently empty.
231 PROTOBUF_ALWAYS_INLINE
232 void GetDirectBufferPointerInline(const void** data, int* size);
233
234 // Read raw bytes, copying them into the given buffer.
235 bool ReadRaw(void* buffer, int size);
236
237 // Like ReadRaw, but reads into a string.
238 bool ReadString(std::string* buffer, int size);
239
240
241 // Read a 32-bit little-endian integer.
242 bool ReadLittleEndian32(uint32_t* value);
243 // Read a 64-bit little-endian integer.
244 bool ReadLittleEndian64(uint64_t* value);
245
246 // These methods read from an externally provided buffer. The caller is
247 // responsible for ensuring that the buffer has sufficient space.
248 // Read a 32-bit little-endian integer.
249 static const uint8_t* ReadLittleEndian32FromArray(const uint8_t* buffer,
250 uint32_t* value);
251 // Read a 64-bit little-endian integer.
252 static const uint8_t* ReadLittleEndian64FromArray(const uint8_t* buffer,
253 uint64_t* value);
254
255 // Read an unsigned integer with Varint encoding, truncating to 32 bits.
256 // Reading a 32-bit value is equivalent to reading a 64-bit one and casting
257 // it to uint32_t, but may be more efficient.
258 bool ReadVarint32(uint32_t* value);
259 // Read an unsigned integer with Varint encoding.
260 bool ReadVarint64(uint64_t* value);
261
262 // Reads a varint off the wire into an "int". This should be used for reading
263 // sizes off the wire (sizes of strings, submessages, bytes fields, etc).
264 //
265 // The value from the wire is interpreted as unsigned. If its value exceeds
266 // the representable value of an integer on this platform, instead of
267 // truncating we return false. Truncating (as performed by ReadVarint32()
268 // above) is an acceptable approach for fields representing an integer, but
269 // when we are parsing a size from the wire, truncating the value would result
270 // in us misparsing the payload.
271 bool ReadVarintSizeAsInt(int* value);
272
273 // Read a tag. This calls ReadVarint32() and returns the result, or returns
274 // zero (which is not a valid tag) if ReadVarint32() fails. Also, ReadTag
275 // (but not ReadTagNoLastTag) updates the last tag value, which can be checked
276 // with LastTagWas().
277 //
278 // Always inline because this is only called in one place per parse loop
279 // but it is called for every iteration of said loop, so it should be fast.
280 // GCC doesn't want to inline this by default.
ReadTag()281 PROTOBUF_ALWAYS_INLINE uint32_t ReadTag() {
282 return last_tag_ = ReadTagNoLastTag();
283 }
284
285 PROTOBUF_ALWAYS_INLINE uint32_t ReadTagNoLastTag();
286
287 // This usually a faster alternative to ReadTag() when cutoff is a manifest
288 // constant. It does particularly well for cutoff >= 127. The first part
289 // of the return value is the tag that was read, though it can also be 0 in
290 // the cases where ReadTag() would return 0. If the second part is true
291 // then the tag is known to be in [0, cutoff]. If not, the tag either is
292 // above cutoff or is 0. (There's intentional wiggle room when tag is 0,
293 // because that can arise in several ways, and for best performance we want
294 // to avoid an extra "is tag == 0?" check here.)
295 PROTOBUF_ALWAYS_INLINE
ReadTagWithCutoff(uint32_t cutoff)296 std::pair<uint32_t, bool> ReadTagWithCutoff(uint32_t cutoff) {
297 std::pair<uint32_t, bool> result = ReadTagWithCutoffNoLastTag(cutoff);
298 last_tag_ = result.first;
299 return result;
300 }
301
302 PROTOBUF_ALWAYS_INLINE
303 std::pair<uint32_t, bool> ReadTagWithCutoffNoLastTag(uint32_t cutoff);
304
305 // Usually returns true if calling ReadVarint32() now would produce the given
306 // value. Will always return false if ReadVarint32() would not return the
307 // given value. If ExpectTag() returns true, it also advances past
308 // the varint. For best performance, use a compile-time constant as the
309 // parameter.
310 // Always inline because this collapses to a small number of instructions
311 // when given a constant parameter, but GCC doesn't want to inline by default.
312 PROTOBUF_ALWAYS_INLINE bool ExpectTag(uint32_t expected);
313
314 // Like above, except this reads from the specified buffer. The caller is
315 // responsible for ensuring that the buffer is large enough to read a varint
316 // of the expected size. For best performance, use a compile-time constant as
317 // the expected tag parameter.
318 //
319 // Returns a pointer beyond the expected tag if it was found, or NULL if it
320 // was not.
321 PROTOBUF_ALWAYS_INLINE
322 static const uint8_t* ExpectTagFromArray(const uint8_t* buffer,
323 uint32_t expected);
324
325 // Usually returns true if no more bytes can be read. Always returns false
326 // if more bytes can be read. If ExpectAtEnd() returns true, a subsequent
327 // call to LastTagWas() will act as if ReadTag() had been called and returned
328 // zero, and ConsumedEntireMessage() will return true.
329 bool ExpectAtEnd();
330
331 // If the last call to ReadTag() or ReadTagWithCutoff() returned the given
332 // value, returns true. Otherwise, returns false.
333 // ReadTagNoLastTag/ReadTagWithCutoffNoLastTag do not preserve the last
334 // returned value.
335 //
336 // This is needed because parsers for some types of embedded messages
337 // (with field type TYPE_GROUP) don't actually know that they've reached the
338 // end of a message until they see an ENDGROUP tag, which was actually part
339 // of the enclosing message. The enclosing message would like to check that
340 // tag to make sure it had the right number, so it calls LastTagWas() on
341 // return from the embedded parser to check.
342 bool LastTagWas(uint32_t expected);
SetLastTag(uint32_t tag)343 void SetLastTag(uint32_t tag) { last_tag_ = tag; }
344
345 // When parsing message (but NOT a group), this method must be called
346 // immediately after MergeFromCodedStream() returns (if it returns true)
347 // to further verify that the message ended in a legitimate way. For
348 // example, this verifies that parsing did not end on an end-group tag.
349 // It also checks for some cases where, due to optimizations,
350 // MergeFromCodedStream() can incorrectly return true.
351 bool ConsumedEntireMessage();
SetConsumed()352 void SetConsumed() { legitimate_message_end_ = true; }
353
354 // Limits ----------------------------------------------------------
355 // Limits are used when parsing length-delimited embedded messages.
356 // After the message's length is read, PushLimit() is used to prevent
357 // the CodedInputStream from reading beyond that length. Once the
358 // embedded message has been parsed, PopLimit() is called to undo the
359 // limit.
360
361 // Opaque type used with PushLimit() and PopLimit(). Do not modify
362 // values of this type yourself. The only reason that this isn't a
363 // struct with private internals is for efficiency.
364 typedef int Limit;
365
366 // Places a limit on the number of bytes that the stream may read,
367 // starting from the current position. Once the stream hits this limit,
368 // it will act like the end of the input has been reached until PopLimit()
369 // is called.
370 //
371 // As the names imply, the stream conceptually has a stack of limits. The
372 // shortest limit on the stack is always enforced, even if it is not the
373 // top limit.
374 //
375 // The value returned by PushLimit() is opaque to the caller, and must
376 // be passed unchanged to the corresponding call to PopLimit().
377 Limit PushLimit(int byte_limit);
378
379 // Pops the last limit pushed by PushLimit(). The input must be the value
380 // returned by that call to PushLimit().
381 void PopLimit(Limit limit);
382
383 // Returns the number of bytes left until the nearest limit on the
384 // stack is hit, or -1 if no limits are in place.
385 int BytesUntilLimit() const;
386
387 // Returns current position relative to the beginning of the input stream.
388 int CurrentPosition() const;
389
390 // Total Bytes Limit -----------------------------------------------
391 // To prevent malicious users from sending excessively large messages
392 // and causing memory exhaustion, CodedInputStream imposes a hard limit on
393 // the total number of bytes it will read.
394
395 // Sets the maximum number of bytes that this CodedInputStream will read
396 // before refusing to continue. To prevent servers from allocating enormous
397 // amounts of memory to hold parsed messages, the maximum message length
398 // should be limited to the shortest length that will not harm usability.
399 // The default limit is INT_MAX (~2GB) and apps should set shorter limits
400 // if possible. An error will always be printed to stderr if the limit is
401 // reached.
402 //
403 // Note: setting a limit less than the current read position is interpreted
404 // as a limit on the current position.
405 //
406 // This is unrelated to PushLimit()/PopLimit().
407 void SetTotalBytesLimit(int total_bytes_limit);
408
409 // The Total Bytes Limit minus the Current Position, or -1 if the total bytes
410 // limit is INT_MAX.
411 int BytesUntilTotalBytesLimit() const;
412
413 // Recursion Limit -------------------------------------------------
414 // To prevent corrupt or malicious messages from causing stack overflows,
415 // we must keep track of the depth of recursion when parsing embedded
416 // messages and groups. CodedInputStream keeps track of this because it
417 // is the only object that is passed down the stack during parsing.
418
419 // Sets the maximum recursion depth. The default is 100.
420 void SetRecursionLimit(int limit);
RecursionBudget()421 int RecursionBudget() { return recursion_budget_; }
422
GetDefaultRecursionLimit()423 static int GetDefaultRecursionLimit() { return default_recursion_limit_; }
424
425 // Increments the current recursion depth. Returns true if the depth is
426 // under the limit, false if it has gone over.
427 bool IncrementRecursionDepth();
428
429 // Decrements the recursion depth if possible.
430 void DecrementRecursionDepth();
431
432 // Decrements the recursion depth blindly. This is faster than
433 // DecrementRecursionDepth(). It should be used only if all previous
434 // increments to recursion depth were successful.
435 void UnsafeDecrementRecursionDepth();
436
437 // Shorthand for make_pair(PushLimit(byte_limit), --recursion_budget_).
438 // Using this can reduce code size and complexity in some cases. The caller
439 // is expected to check that the second part of the result is non-negative (to
440 // bail out if the depth of recursion is too high) and, if all is well, to
441 // later pass the first part of the result to PopLimit() or similar.
442 std::pair<CodedInputStream::Limit, int> IncrementRecursionDepthAndPushLimit(
443 int byte_limit);
444
445 // Shorthand for PushLimit(ReadVarint32(&length) ? length : 0).
446 Limit ReadLengthAndPushLimit();
447
448 // Helper that is equivalent to: {
449 // bool result = ConsumedEntireMessage();
450 // PopLimit(limit);
451 // UnsafeDecrementRecursionDepth();
452 // return result; }
453 // Using this can reduce code size and complexity in some cases.
454 // Do not use unless the current recursion depth is greater than zero.
455 bool DecrementRecursionDepthAndPopLimit(Limit limit);
456
457 // Helper that is equivalent to: {
458 // bool result = ConsumedEntireMessage();
459 // PopLimit(limit);
460 // return result; }
461 // Using this can reduce code size and complexity in some cases.
462 bool CheckEntireMessageConsumedAndPopLimit(Limit limit);
463
464 // Extension Registry ----------------------------------------------
465 // ADVANCED USAGE: 99.9% of people can ignore this section.
466 //
467 // By default, when parsing extensions, the parser looks for extension
468 // definitions in the pool which owns the outer message's Descriptor.
469 // However, you may call SetExtensionRegistry() to provide an alternative
470 // pool instead. This makes it possible, for example, to parse a message
471 // using a generated class, but represent some extensions using
472 // DynamicMessage.
473
474 // Set the pool used to look up extensions. Most users do not need to call
475 // this as the correct pool will be chosen automatically.
476 //
477 // WARNING: It is very easy to misuse this. Carefully read the requirements
478 // below. Do not use this unless you are sure you need it. Almost no one
479 // does.
480 //
481 // Let's say you are parsing a message into message object m, and you want
482 // to take advantage of SetExtensionRegistry(). You must follow these
483 // requirements:
484 //
485 // The given DescriptorPool must contain m->GetDescriptor(). It is not
486 // sufficient for it to simply contain a descriptor that has the same name
487 // and content -- it must be the *exact object*. In other words:
488 // assert(pool->FindMessageTypeByName(m->GetDescriptor()->full_name()) ==
489 // m->GetDescriptor());
490 // There are two ways to satisfy this requirement:
491 // 1) Use m->GetDescriptor()->pool() as the pool. This is generally useless
492 // because this is the pool that would be used anyway if you didn't call
493 // SetExtensionRegistry() at all.
494 // 2) Use a DescriptorPool which has m->GetDescriptor()->pool() as an
495 // "underlay". Read the documentation for DescriptorPool for more
496 // information about underlays.
497 //
498 // You must also provide a MessageFactory. This factory will be used to
499 // construct Message objects representing extensions. The factory's
500 // GetPrototype() MUST return non-NULL for any Descriptor which can be found
501 // through the provided pool.
502 //
503 // If the provided factory might return instances of protocol-compiler-
504 // generated (i.e. compiled-in) types, or if the outer message object m is
505 // a generated type, then the given factory MUST have this property: If
506 // GetPrototype() is given a Descriptor which resides in
507 // DescriptorPool::generated_pool(), the factory MUST return the same
508 // prototype which MessageFactory::generated_factory() would return. That
509 // is, given a descriptor for a generated type, the factory must return an
510 // instance of the generated class (NOT DynamicMessage). However, when
511 // given a descriptor for a type that is NOT in generated_pool, the factory
512 // is free to return any implementation.
513 //
514 // The reason for this requirement is that generated sub-objects may be
515 // accessed via the standard (non-reflection) extension accessor methods,
516 // and these methods will down-cast the object to the generated class type.
517 // If the object is not actually of that type, the results would be undefined.
518 // On the other hand, if an extension is not compiled in, then there is no
519 // way the code could end up accessing it via the standard accessors -- the
520 // only way to access the extension is via reflection. When using reflection,
521 // DynamicMessage and generated messages are indistinguishable, so it's fine
522 // if these objects are represented using DynamicMessage.
523 //
524 // Using DynamicMessageFactory on which you have called
525 // SetDelegateToGeneratedFactory(true) should be sufficient to satisfy the
526 // above requirement.
527 //
528 // If either pool or factory is NULL, both must be NULL.
529 //
530 // Note that this feature is ignored when parsing "lite" messages as they do
531 // not have descriptors.
532 void SetExtensionRegistry(const DescriptorPool* pool,
533 MessageFactory* factory);
534
535 // Get the DescriptorPool set via SetExtensionRegistry(), or NULL if no pool
536 // has been provided.
537 const DescriptorPool* GetExtensionPool();
538
539 // Get the MessageFactory set via SetExtensionRegistry(), or NULL if no
540 // factory has been provided.
541 MessageFactory* GetExtensionFactory();
542
543 private:
544 GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(CodedInputStream);
545
546 const uint8_t* buffer_;
547 const uint8_t* buffer_end_; // pointer to the end of the buffer.
548 ZeroCopyInputStream* input_;
549 int total_bytes_read_; // total bytes read from input_, including
550 // the current buffer
551
552 // If total_bytes_read_ surpasses INT_MAX, we record the extra bytes here
553 // so that we can BackUp() on destruction.
554 int overflow_bytes_;
555
556 // LastTagWas() stuff.
557 uint32_t last_tag_; // result of last ReadTag() or ReadTagWithCutoff().
558
559 // This is set true by ReadTag{Fallback/Slow}() if it is called when exactly
560 // at EOF, or by ExpectAtEnd() when it returns true. This happens when we
561 // reach the end of a message and attempt to read another tag.
562 bool legitimate_message_end_;
563
564 // See EnableAliasing().
565 bool aliasing_enabled_;
566
567 // Limits
568 Limit current_limit_; // if position = -1, no limit is applied
569
570 // For simplicity, if the current buffer crosses a limit (either a normal
571 // limit created by PushLimit() or the total bytes limit), buffer_size_
572 // only tracks the number of bytes before that limit. This field
573 // contains the number of bytes after it. Note that this implies that if
574 // buffer_size_ == 0 and buffer_size_after_limit_ > 0, we know we've
575 // hit a limit. However, if both are zero, it doesn't necessarily mean
576 // we aren't at a limit -- the buffer may have ended exactly at the limit.
577 int buffer_size_after_limit_;
578
579 // Maximum number of bytes to read, period. This is unrelated to
580 // current_limit_. Set using SetTotalBytesLimit().
581 int total_bytes_limit_;
582
583 // Current recursion budget, controlled by IncrementRecursionDepth() and
584 // similar. Starts at recursion_limit_ and goes down: if this reaches
585 // -1 we are over budget.
586 int recursion_budget_;
587 // Recursion depth limit, set by SetRecursionLimit().
588 int recursion_limit_;
589
590 // See SetExtensionRegistry().
591 const DescriptorPool* extension_pool_;
592 MessageFactory* extension_factory_;
593
594 // Private member functions.
595
596 // Fallback when Skip() goes past the end of the current buffer.
597 bool SkipFallback(int count, int original_buffer_size);
598
599 // Advance the buffer by a given number of bytes.
600 void Advance(int amount);
601
602 // Back up input_ to the current buffer position.
603 void BackUpInputToCurrentPosition();
604
605 // Recomputes the value of buffer_size_after_limit_. Must be called after
606 // current_limit_ or total_bytes_limit_ changes.
607 void RecomputeBufferLimits();
608
609 // Writes an error message saying that we hit total_bytes_limit_.
610 void PrintTotalBytesLimitError();
611
612 // Called when the buffer runs out to request more data. Implies an
613 // Advance(BufferSize()).
614 bool Refresh();
615
616 // When parsing varints, we optimize for the common case of small values, and
617 // then optimize for the case when the varint fits within the current buffer
618 // piece. The Fallback method is used when we can't use the one-byte
619 // optimization. The Slow method is yet another fallback when the buffer is
620 // not large enough. Making the slow path out-of-line speeds up the common
621 // case by 10-15%. The slow path is fairly uncommon: it only triggers when a
622 // message crosses multiple buffers. Note: ReadVarint32Fallback() and
623 // ReadVarint64Fallback() are called frequently and generally not inlined, so
624 // they have been optimized to avoid "out" parameters. The former returns -1
625 // if it fails and the uint32_t it read otherwise. The latter has a bool
626 // indicating success or failure as part of its return type.
627 int64_t ReadVarint32Fallback(uint32_t first_byte_or_zero);
628 int ReadVarintSizeAsIntFallback();
629 std::pair<uint64_t, bool> ReadVarint64Fallback();
630 bool ReadVarint32Slow(uint32_t* value);
631 bool ReadVarint64Slow(uint64_t* value);
632 int ReadVarintSizeAsIntSlow();
633 bool ReadLittleEndian32Fallback(uint32_t* value);
634 bool ReadLittleEndian64Fallback(uint64_t* value);
635
636 // Fallback/slow methods for reading tags. These do not update last_tag_,
637 // but will set legitimate_message_end_ if we are at the end of the input
638 // stream.
639 uint32_t ReadTagFallback(uint32_t first_byte_or_zero);
640 uint32_t ReadTagSlow();
641 bool ReadStringFallback(std::string* buffer, int size);
642
643 // Return the size of the buffer.
644 int BufferSize() const;
645
646 static const int kDefaultTotalBytesLimit = INT_MAX;
647
648 static int default_recursion_limit_; // 100 by default.
649
650 friend class google::protobuf::ZeroCopyCodedInputStream;
651 friend class google::protobuf::internal::EpsCopyByteStream;
652 };
653
654 // EpsCopyOutputStream wraps a ZeroCopyOutputStream and exposes a new stream,
655 // which has the property you can write kSlopBytes (16 bytes) from the current
656 // position without bounds checks. The cursor into the stream is managed by
657 // the user of the class and is an explicit parameter in the methods. Careful
658 // use of this class, ie. keep ptr a local variable, eliminates the need to
659 // for the compiler to sync the ptr value between register and memory.
660 class PROTOBUF_EXPORT EpsCopyOutputStream {
661 public:
662 enum { kSlopBytes = 16 };
663
664 // Initialize from a stream.
EpsCopyOutputStream(ZeroCopyOutputStream * stream,bool deterministic,uint8_t ** pp)665 EpsCopyOutputStream(ZeroCopyOutputStream* stream, bool deterministic,
666 uint8_t** pp)
667 : end_(buffer_),
668 stream_(stream),
669 is_serialization_deterministic_(deterministic) {
670 *pp = buffer_;
671 }
672
673 // Only for array serialization. No overflow protection, end_ will be the
674 // pointed to the end of the array. When using this the total size is already
675 // known, so no need to maintain the slop region.
EpsCopyOutputStream(void * data,int size,bool deterministic)676 EpsCopyOutputStream(void* data, int size, bool deterministic)
677 : end_(static_cast<uint8_t*>(data) + size),
678 buffer_end_(nullptr),
679 stream_(nullptr),
680 is_serialization_deterministic_(deterministic) {}
681
682 // Initialize from stream but with the first buffer already given (eager).
EpsCopyOutputStream(void * data,int size,ZeroCopyOutputStream * stream,bool deterministic,uint8_t ** pp)683 EpsCopyOutputStream(void* data, int size, ZeroCopyOutputStream* stream,
684 bool deterministic, uint8_t** pp)
685 : stream_(stream), is_serialization_deterministic_(deterministic) {
686 *pp = SetInitialBuffer(data, size);
687 }
688
689 // Flush everything that's written into the underlying ZeroCopyOutputStream
690 // and trims the underlying stream to the location of ptr.
691 uint8_t* Trim(uint8_t* ptr);
692
693 // After this it's guaranteed you can safely write kSlopBytes to ptr. This
694 // will never fail! The underlying stream can produce an error. Use HadError
695 // to check for errors.
EnsureSpace(uint8_t * ptr)696 PROTOBUF_NODISCARD uint8_t* EnsureSpace(uint8_t* ptr) {
697 if (PROTOBUF_PREDICT_FALSE(ptr >= end_)) {
698 return EnsureSpaceFallback(ptr);
699 }
700 return ptr;
701 }
702
WriteRaw(const void * data,int size,uint8_t * ptr)703 uint8_t* WriteRaw(const void* data, int size, uint8_t* ptr) {
704 if (PROTOBUF_PREDICT_FALSE(end_ - ptr < size)) {
705 return WriteRawFallback(data, size, ptr);
706 }
707 std::memcpy(ptr, data, size);
708 return ptr + size;
709 }
710 // Writes the buffer specified by data, size to the stream. Possibly by
711 // aliasing the buffer (ie. not copying the data). The caller is responsible
712 // to make sure the buffer is alive for the duration of the
713 // ZeroCopyOutputStream.
714 #ifndef NDEBUG
715 PROTOBUF_NOINLINE
716 #endif
WriteRawMaybeAliased(const void * data,int size,uint8_t * ptr)717 uint8_t* WriteRawMaybeAliased(const void* data, int size, uint8_t* ptr) {
718 if (aliasing_enabled_) {
719 return WriteAliasedRaw(data, size, ptr);
720 } else {
721 return WriteRaw(data, size, ptr);
722 }
723 }
724
725
726 #ifndef NDEBUG
727 PROTOBUF_NOINLINE
728 #endif
WriteStringMaybeAliased(uint32_t num,const std::string & s,uint8_t * ptr)729 uint8_t* WriteStringMaybeAliased(uint32_t num, const std::string& s,
730 uint8_t* ptr) {
731 std::ptrdiff_t size = s.size();
732 if (PROTOBUF_PREDICT_FALSE(
733 size >= 128 || end_ - ptr + 16 - TagSize(num << 3) - 1 < size)) {
734 return WriteStringMaybeAliasedOutline(num, s, ptr);
735 }
736 ptr = UnsafeVarint((num << 3) | 2, ptr);
737 *ptr++ = static_cast<uint8_t>(size);
738 std::memcpy(ptr, s.data(), size);
739 return ptr + size;
740 }
WriteBytesMaybeAliased(uint32_t num,const std::string & s,uint8_t * ptr)741 uint8_t* WriteBytesMaybeAliased(uint32_t num, const std::string& s,
742 uint8_t* ptr) {
743 return WriteStringMaybeAliased(num, s, ptr);
744 }
745
746 template <typename T>
WriteString(uint32_t num,const T & s,uint8_t * ptr)747 PROTOBUF_ALWAYS_INLINE uint8_t* WriteString(uint32_t num, const T& s,
748 uint8_t* ptr) {
749 std::ptrdiff_t size = s.size();
750 if (PROTOBUF_PREDICT_FALSE(
751 size >= 128 || end_ - ptr + 16 - TagSize(num << 3) - 1 < size)) {
752 return WriteStringOutline(num, s, ptr);
753 }
754 ptr = UnsafeVarint((num << 3) | 2, ptr);
755 *ptr++ = static_cast<uint8_t>(size);
756 std::memcpy(ptr, s.data(), size);
757 return ptr + size;
758 }
759 template <typename T>
760 #ifndef NDEBUG
761 PROTOBUF_NOINLINE
762 #endif
WriteBytes(uint32_t num,const T & s,uint8_t * ptr)763 uint8_t* WriteBytes(uint32_t num, const T& s, uint8_t* ptr) {
764 return WriteString(num, s, ptr);
765 }
766
767 template <typename T>
WriteInt32Packed(int num,const T & r,int size,uint8_t * ptr)768 PROTOBUF_ALWAYS_INLINE uint8_t* WriteInt32Packed(int num, const T& r,
769 int size, uint8_t* ptr) {
770 return WriteVarintPacked(num, r, size, ptr, Encode64);
771 }
772 template <typename T>
WriteUInt32Packed(int num,const T & r,int size,uint8_t * ptr)773 PROTOBUF_ALWAYS_INLINE uint8_t* WriteUInt32Packed(int num, const T& r,
774 int size, uint8_t* ptr) {
775 return WriteVarintPacked(num, r, size, ptr, Encode32);
776 }
777 template <typename T>
WriteSInt32Packed(int num,const T & r,int size,uint8_t * ptr)778 PROTOBUF_ALWAYS_INLINE uint8_t* WriteSInt32Packed(int num, const T& r,
779 int size, uint8_t* ptr) {
780 return WriteVarintPacked(num, r, size, ptr, ZigZagEncode32);
781 }
782 template <typename T>
WriteInt64Packed(int num,const T & r,int size,uint8_t * ptr)783 PROTOBUF_ALWAYS_INLINE uint8_t* WriteInt64Packed(int num, const T& r,
784 int size, uint8_t* ptr) {
785 return WriteVarintPacked(num, r, size, ptr, Encode64);
786 }
787 template <typename T>
WriteUInt64Packed(int num,const T & r,int size,uint8_t * ptr)788 PROTOBUF_ALWAYS_INLINE uint8_t* WriteUInt64Packed(int num, const T& r,
789 int size, uint8_t* ptr) {
790 return WriteVarintPacked(num, r, size, ptr, Encode64);
791 }
792 template <typename T>
WriteSInt64Packed(int num,const T & r,int size,uint8_t * ptr)793 PROTOBUF_ALWAYS_INLINE uint8_t* WriteSInt64Packed(int num, const T& r,
794 int size, uint8_t* ptr) {
795 return WriteVarintPacked(num, r, size, ptr, ZigZagEncode64);
796 }
797 template <typename T>
WriteEnumPacked(int num,const T & r,int size,uint8_t * ptr)798 PROTOBUF_ALWAYS_INLINE uint8_t* WriteEnumPacked(int num, const T& r, int size,
799 uint8_t* ptr) {
800 return WriteVarintPacked(num, r, size, ptr, Encode64);
801 }
802
803 template <typename T>
WriteFixedPacked(int num,const T & r,uint8_t * ptr)804 PROTOBUF_ALWAYS_INLINE uint8_t* WriteFixedPacked(int num, const T& r,
805 uint8_t* ptr) {
806 ptr = EnsureSpace(ptr);
807 constexpr auto element_size = sizeof(typename T::value_type);
808 auto size = r.size() * element_size;
809 ptr = WriteLengthDelim(num, size, ptr);
810 return WriteRawLittleEndian<element_size>(r.data(), static_cast<int>(size),
811 ptr);
812 }
813
814 // Returns true if there was an underlying I/O error since this object was
815 // created.
HadError()816 bool HadError() const { return had_error_; }
817
818 // Instructs the EpsCopyOutputStream to allow the underlying
819 // ZeroCopyOutputStream to hold pointers to the original structure instead of
820 // copying, if it supports it (i.e. output->AllowsAliasing() is true). If the
821 // underlying stream does not support aliasing, then enabling it has no
822 // affect. For now, this only affects the behavior of
823 // WriteRawMaybeAliased().
824 //
825 // NOTE: It is caller's responsibility to ensure that the chunk of memory
826 // remains live until all of the data has been consumed from the stream.
827 void EnableAliasing(bool enabled);
828
829 // See documentation on CodedOutputStream::SetSerializationDeterministic.
SetSerializationDeterministic(bool value)830 void SetSerializationDeterministic(bool value) {
831 is_serialization_deterministic_ = value;
832 }
833
834 // See documentation on CodedOutputStream::IsSerializationDeterministic.
IsSerializationDeterministic()835 bool IsSerializationDeterministic() const {
836 return is_serialization_deterministic_;
837 }
838
839 // The number of bytes written to the stream at position ptr, relative to the
840 // stream's overall position.
841 int64_t ByteCount(uint8_t* ptr) const;
842
843
844 private:
845 uint8_t* end_;
846 uint8_t* buffer_end_ = buffer_;
847 uint8_t buffer_[2 * kSlopBytes];
848 ZeroCopyOutputStream* stream_;
849 bool had_error_ = false;
850 bool aliasing_enabled_ = false; // See EnableAliasing().
851 bool is_serialization_deterministic_;
852 bool skip_check_consistency = false;
853
854 uint8_t* EnsureSpaceFallback(uint8_t* ptr);
855 inline uint8_t* Next();
856 int Flush(uint8_t* ptr);
GetSize(uint8_t * ptr)857 std::ptrdiff_t GetSize(uint8_t* ptr) const {
858 GOOGLE_DCHECK(ptr <= end_ + kSlopBytes); // NOLINT
859 return end_ + kSlopBytes - ptr;
860 }
861
Error()862 uint8_t* Error() {
863 had_error_ = true;
864 // We use the patch buffer to always guarantee space to write to.
865 end_ = buffer_ + kSlopBytes;
866 return buffer_;
867 }
868
TagSize(uint32_t tag)869 static constexpr int TagSize(uint32_t tag) {
870 return (tag < (1 << 7)) ? 1
871 : (tag < (1 << 14)) ? 2
872 : (tag < (1 << 21)) ? 3
873 : (tag < (1 << 28)) ? 4
874 : 5;
875 }
876
WriteTag(uint32_t num,uint32_t wt,uint8_t * ptr)877 PROTOBUF_ALWAYS_INLINE uint8_t* WriteTag(uint32_t num, uint32_t wt,
878 uint8_t* ptr) {
879 GOOGLE_DCHECK(ptr < end_); // NOLINT
880 return UnsafeVarint((num << 3) | wt, ptr);
881 }
882
WriteLengthDelim(int num,uint32_t size,uint8_t * ptr)883 PROTOBUF_ALWAYS_INLINE uint8_t* WriteLengthDelim(int num, uint32_t size,
884 uint8_t* ptr) {
885 ptr = WriteTag(num, 2, ptr);
886 return UnsafeWriteSize(size, ptr);
887 }
888
889 uint8_t* WriteRawFallback(const void* data, int size, uint8_t* ptr);
890
891 uint8_t* WriteAliasedRaw(const void* data, int size, uint8_t* ptr);
892
893 uint8_t* WriteStringMaybeAliasedOutline(uint32_t num, const std::string& s,
894 uint8_t* ptr);
895 uint8_t* WriteStringOutline(uint32_t num, const std::string& s, uint8_t* ptr);
896
897 template <typename T, typename E>
WriteVarintPacked(int num,const T & r,int size,uint8_t * ptr,const E & encode)898 PROTOBUF_ALWAYS_INLINE uint8_t* WriteVarintPacked(int num, const T& r,
899 int size, uint8_t* ptr,
900 const E& encode) {
901 ptr = EnsureSpace(ptr);
902 ptr = WriteLengthDelim(num, size, ptr);
903 auto it = r.data();
904 auto end = it + r.size();
905 do {
906 ptr = EnsureSpace(ptr);
907 ptr = UnsafeVarint(encode(*it++), ptr);
908 } while (it < end);
909 return ptr;
910 }
911
Encode32(uint32_t v)912 static uint32_t Encode32(uint32_t v) { return v; }
Encode64(uint64_t v)913 static uint64_t Encode64(uint64_t v) { return v; }
ZigZagEncode32(int32_t v)914 static uint32_t ZigZagEncode32(int32_t v) {
915 return (static_cast<uint32_t>(v) << 1) ^ static_cast<uint32_t>(v >> 31);
916 }
ZigZagEncode64(int64_t v)917 static uint64_t ZigZagEncode64(int64_t v) {
918 return (static_cast<uint64_t>(v) << 1) ^ static_cast<uint64_t>(v >> 63);
919 }
920
921 template <typename T>
UnsafeVarint(T value,uint8_t * ptr)922 PROTOBUF_ALWAYS_INLINE static uint8_t* UnsafeVarint(T value, uint8_t* ptr) {
923 static_assert(std::is_unsigned<T>::value,
924 "Varint serialization must be unsigned");
925 ptr[0] = static_cast<uint8_t>(value);
926 if (value < 0x80) {
927 return ptr + 1;
928 }
929 // Turn on continuation bit in the byte we just wrote.
930 ptr[0] |= static_cast<uint8_t>(0x80);
931 value >>= 7;
932 ptr[1] = static_cast<uint8_t>(value);
933 if (value < 0x80) {
934 return ptr + 2;
935 }
936 ptr += 2;
937 do {
938 // Turn on continuation bit in the byte we just wrote.
939 ptr[-1] |= static_cast<uint8_t>(0x80);
940 value >>= 7;
941 *ptr = static_cast<uint8_t>(value);
942 ++ptr;
943 } while (value >= 0x80);
944 return ptr;
945 }
946
UnsafeWriteSize(uint32_t value,uint8_t * ptr)947 PROTOBUF_ALWAYS_INLINE static uint8_t* UnsafeWriteSize(uint32_t value,
948 uint8_t* ptr) {
949 while (PROTOBUF_PREDICT_FALSE(value >= 0x80)) {
950 *ptr = static_cast<uint8_t>(value | 0x80);
951 value >>= 7;
952 ++ptr;
953 }
954 *ptr++ = static_cast<uint8_t>(value);
955 return ptr;
956 }
957
958 template <int S>
959 uint8_t* WriteRawLittleEndian(const void* data, int size, uint8_t* ptr);
960 #ifndef PROTOBUF_LITTLE_ENDIAN
961 uint8_t* WriteRawLittleEndian32(const void* data, int size, uint8_t* ptr);
962 uint8_t* WriteRawLittleEndian64(const void* data, int size, uint8_t* ptr);
963 #endif
964
965 // These methods are for CodedOutputStream. Ideally they should be private
966 // but to match current behavior of CodedOutputStream as close as possible
967 // we allow it some functionality.
968 public:
SetInitialBuffer(void * data,int size)969 uint8_t* SetInitialBuffer(void* data, int size) {
970 auto ptr = static_cast<uint8_t*>(data);
971 if (size > kSlopBytes) {
972 end_ = ptr + size - kSlopBytes;
973 buffer_end_ = nullptr;
974 return ptr;
975 } else {
976 end_ = buffer_ + size;
977 buffer_end_ = ptr;
978 return buffer_;
979 }
980 }
981
982 private:
983 // Needed by CodedOutputStream HadError. HadError needs to flush the patch
984 // buffers to ensure there is no error as of yet.
985 uint8_t* FlushAndResetBuffer(uint8_t*);
986
987 // The following functions mimic the old CodedOutputStream behavior as close
988 // as possible. They flush the current state to the stream, behave as
989 // the old CodedOutputStream and then return to normal operation.
990 bool Skip(int count, uint8_t** pp);
991 bool GetDirectBufferPointer(void** data, int* size, uint8_t** pp);
992 uint8_t* GetDirectBufferForNBytesAndAdvance(int size, uint8_t** pp);
993
994 friend class CodedOutputStream;
995 };
996
997 template <>
998 inline uint8_t* EpsCopyOutputStream::WriteRawLittleEndian<1>(const void* data,
999 int size,
1000 uint8_t* ptr) {
1001 return WriteRaw(data, size, ptr);
1002 }
1003 template <>
1004 inline uint8_t* EpsCopyOutputStream::WriteRawLittleEndian<4>(const void* data,
1005 int size,
1006 uint8_t* ptr) {
1007 #ifdef PROTOBUF_LITTLE_ENDIAN
1008 return WriteRaw(data, size, ptr);
1009 #else
1010 return WriteRawLittleEndian32(data, size, ptr);
1011 #endif
1012 }
1013 template <>
1014 inline uint8_t* EpsCopyOutputStream::WriteRawLittleEndian<8>(const void* data,
1015 int size,
1016 uint8_t* ptr) {
1017 #ifdef PROTOBUF_LITTLE_ENDIAN
1018 return WriteRaw(data, size, ptr);
1019 #else
1020 return WriteRawLittleEndian64(data, size, ptr);
1021 #endif
1022 }
1023
1024 // Class which encodes and writes binary data which is composed of varint-
1025 // encoded integers and fixed-width pieces. Wraps a ZeroCopyOutputStream.
1026 // Most users will not need to deal with CodedOutputStream.
1027 //
1028 // Most methods of CodedOutputStream which return a bool return false if an
1029 // underlying I/O error occurs. Once such a failure occurs, the
1030 // CodedOutputStream is broken and is no longer useful. The Write* methods do
1031 // not return the stream status, but will invalidate the stream if an error
1032 // occurs. The client can probe HadError() to determine the status.
1033 //
1034 // Note that every method of CodedOutputStream which writes some data has
1035 // a corresponding static "ToArray" version. These versions write directly
1036 // to the provided buffer, returning a pointer past the last written byte.
1037 // They require that the buffer has sufficient capacity for the encoded data.
1038 // This allows an optimization where we check if an output stream has enough
1039 // space for an entire message before we start writing and, if there is, we
1040 // call only the ToArray methods to avoid doing bound checks for each
1041 // individual value.
1042 // i.e., in the example above:
1043 //
1044 // CodedOutputStream* coded_output = new CodedOutputStream(raw_output);
1045 // int magic_number = 1234;
1046 // char text[] = "Hello world!";
1047 //
1048 // int coded_size = sizeof(magic_number) +
1049 // CodedOutputStream::VarintSize32(strlen(text)) +
1050 // strlen(text);
1051 //
1052 // uint8_t* buffer =
1053 // coded_output->GetDirectBufferForNBytesAndAdvance(coded_size);
1054 // if (buffer != nullptr) {
1055 // // The output stream has enough space in the buffer: write directly to
1056 // // the array.
1057 // buffer = CodedOutputStream::WriteLittleEndian32ToArray(magic_number,
1058 // buffer);
1059 // buffer = CodedOutputStream::WriteVarint32ToArray(strlen(text), buffer);
1060 // buffer = CodedOutputStream::WriteRawToArray(text, strlen(text), buffer);
1061 // } else {
1062 // // Make bound-checked writes, which will ask the underlying stream for
1063 // // more space as needed.
1064 // coded_output->WriteLittleEndian32(magic_number);
1065 // coded_output->WriteVarint32(strlen(text));
1066 // coded_output->WriteRaw(text, strlen(text));
1067 // }
1068 //
1069 // delete coded_output;
1070 class PROTOBUF_EXPORT CodedOutputStream {
1071 public:
1072 // Creates a CodedOutputStream that writes to the given `stream`.
1073 // The provided stream must publicly derive from `ZeroCopyOutputStream`.
1074 template <class Stream, class = typename std::enable_if<std::is_base_of<
1075 ZeroCopyOutputStream, Stream>::value>::type>
1076 explicit CodedOutputStream(Stream* stream);
1077
1078 // Creates a CodedOutputStream that writes to the given `stream`, and does
1079 // an 'eager initialization' of the internal state if `eager_init` is true.
1080 // The provided stream must publicly derive from `ZeroCopyOutputStream`.
1081 template <class Stream, class = typename std::enable_if<std::is_base_of<
1082 ZeroCopyOutputStream, Stream>::value>::type>
1083 CodedOutputStream(Stream* stream, bool eager_init);
1084
1085 // Destroy the CodedOutputStream and position the underlying
1086 // ZeroCopyOutputStream immediately after the last byte written.
1087 ~CodedOutputStream();
1088
1089 // Returns true if there was an underlying I/O error since this object was
1090 // created. On should call Trim before this function in order to catch all
1091 // errors.
HadError()1092 bool HadError() {
1093 cur_ = impl_.FlushAndResetBuffer(cur_);
1094 GOOGLE_DCHECK(cur_);
1095 return impl_.HadError();
1096 }
1097
1098 // Trims any unused space in the underlying buffer so that its size matches
1099 // the number of bytes written by this stream. The underlying buffer will
1100 // automatically be trimmed when this stream is destroyed; this call is only
1101 // necessary if the underlying buffer is accessed *before* the stream is
1102 // destroyed.
Trim()1103 void Trim() { cur_ = impl_.Trim(cur_); }
1104
1105 // Skips a number of bytes, leaving the bytes unmodified in the underlying
1106 // buffer. Returns false if an underlying write error occurs. This is
1107 // mainly useful with GetDirectBufferPointer().
1108 // Note of caution, the skipped bytes may contain uninitialized data. The
1109 // caller must make sure that the skipped bytes are properly initialized,
1110 // otherwise you might leak bytes from your heap.
Skip(int count)1111 bool Skip(int count) { return impl_.Skip(count, &cur_); }
1112
1113 // Sets *data to point directly at the unwritten part of the
1114 // CodedOutputStream's underlying buffer, and *size to the size of that
1115 // buffer, but does not advance the stream's current position. This will
1116 // always either produce a non-empty buffer or return false. If the caller
1117 // writes any data to this buffer, it should then call Skip() to skip over
1118 // the consumed bytes. This may be useful for implementing external fast
1119 // serialization routines for types of data not covered by the
1120 // CodedOutputStream interface.
GetDirectBufferPointer(void ** data,int * size)1121 bool GetDirectBufferPointer(void** data, int* size) {
1122 return impl_.GetDirectBufferPointer(data, size, &cur_);
1123 }
1124
1125 // If there are at least "size" bytes available in the current buffer,
1126 // returns a pointer directly into the buffer and advances over these bytes.
1127 // The caller may then write directly into this buffer (e.g. using the
1128 // *ToArray static methods) rather than go through CodedOutputStream. If
1129 // there are not enough bytes available, returns NULL. The return pointer is
1130 // invalidated as soon as any other non-const method of CodedOutputStream
1131 // is called.
GetDirectBufferForNBytesAndAdvance(int size)1132 inline uint8_t* GetDirectBufferForNBytesAndAdvance(int size) {
1133 return impl_.GetDirectBufferForNBytesAndAdvance(size, &cur_);
1134 }
1135
1136 // Write raw bytes, copying them from the given buffer.
WriteRaw(const void * buffer,int size)1137 void WriteRaw(const void* buffer, int size) {
1138 cur_ = impl_.WriteRaw(buffer, size, cur_);
1139 }
1140 // Like WriteRaw() but will try to write aliased data if aliasing is
1141 // turned on.
1142 void WriteRawMaybeAliased(const void* data, int size);
1143 // Like WriteRaw() but writing directly to the target array.
1144 // This is _not_ inlined, as the compiler often optimizes memcpy into inline
1145 // copy loops. Since this gets called by every field with string or bytes
1146 // type, inlining may lead to a significant amount of code bloat, with only a
1147 // minor performance gain.
1148 static uint8_t* WriteRawToArray(const void* buffer, int size,
1149 uint8_t* target);
1150
1151 // Equivalent to WriteRaw(str.data(), str.size()).
1152 void WriteString(const std::string& str);
1153 // Like WriteString() but writing directly to the target array.
1154 static uint8_t* WriteStringToArray(const std::string& str, uint8_t* target);
1155 // Write the varint-encoded size of str followed by str.
1156 static uint8_t* WriteStringWithSizeToArray(const std::string& str,
1157 uint8_t* target);
1158
1159
1160 // Write a 32-bit little-endian integer.
WriteLittleEndian32(uint32_t value)1161 void WriteLittleEndian32(uint32_t value) {
1162 cur_ = impl_.EnsureSpace(cur_);
1163 SetCur(WriteLittleEndian32ToArray(value, Cur()));
1164 }
1165 // Like WriteLittleEndian32() but writing directly to the target array.
1166 static uint8_t* WriteLittleEndian32ToArray(uint32_t value, uint8_t* target);
1167 // Write a 64-bit little-endian integer.
WriteLittleEndian64(uint64_t value)1168 void WriteLittleEndian64(uint64_t value) {
1169 cur_ = impl_.EnsureSpace(cur_);
1170 SetCur(WriteLittleEndian64ToArray(value, Cur()));
1171 }
1172 // Like WriteLittleEndian64() but writing directly to the target array.
1173 static uint8_t* WriteLittleEndian64ToArray(uint64_t value, uint8_t* target);
1174
1175 // Write an unsigned integer with Varint encoding. Writing a 32-bit value
1176 // is equivalent to casting it to uint64_t and writing it as a 64-bit value,
1177 // but may be more efficient.
1178 void WriteVarint32(uint32_t value);
1179 // Like WriteVarint32() but writing directly to the target array.
1180 static uint8_t* WriteVarint32ToArray(uint32_t value, uint8_t* target);
1181 // Like WriteVarint32() but writing directly to the target array, and with
1182 // the less common-case paths being out of line rather than inlined.
1183 static uint8_t* WriteVarint32ToArrayOutOfLine(uint32_t value,
1184 uint8_t* target);
1185 // Write an unsigned integer with Varint encoding.
1186 void WriteVarint64(uint64_t value);
1187 // Like WriteVarint64() but writing directly to the target array.
1188 static uint8_t* WriteVarint64ToArray(uint64_t value, uint8_t* target);
1189
1190 // Equivalent to WriteVarint32() except when the value is negative,
1191 // in which case it must be sign-extended to a full 10 bytes.
1192 void WriteVarint32SignExtended(int32_t value);
1193 // Like WriteVarint32SignExtended() but writing directly to the target array.
1194 static uint8_t* WriteVarint32SignExtendedToArray(int32_t value,
1195 uint8_t* target);
1196
1197 // This is identical to WriteVarint32(), but optimized for writing tags.
1198 // In particular, if the input is a compile-time constant, this method
1199 // compiles down to a couple instructions.
1200 // Always inline because otherwise the aforementioned optimization can't work,
1201 // but GCC by default doesn't want to inline this.
1202 void WriteTag(uint32_t value);
1203 // Like WriteTag() but writing directly to the target array.
1204 PROTOBUF_ALWAYS_INLINE
1205 static uint8_t* WriteTagToArray(uint32_t value, uint8_t* target);
1206
1207 // Returns the number of bytes needed to encode the given value as a varint.
1208 static size_t VarintSize32(uint32_t value);
1209 // Returns the number of bytes needed to encode the given value as a varint.
1210 static size_t VarintSize64(uint64_t value);
1211
1212 // If negative, 10 bytes. Otherwise, same as VarintSize32().
1213 static size_t VarintSize32SignExtended(int32_t value);
1214
1215 // Same as above, plus one. The additional one comes at no compute cost.
1216 static size_t VarintSize32PlusOne(uint32_t value);
1217 static size_t VarintSize64PlusOne(uint64_t value);
1218 static size_t VarintSize32SignExtendedPlusOne(int32_t value);
1219
1220 // Compile-time equivalent of VarintSize32().
1221 template <uint32_t Value>
1222 struct StaticVarintSize32 {
1223 static const size_t value = (Value < (1 << 7)) ? 1
1224 : (Value < (1 << 14)) ? 2
1225 : (Value < (1 << 21)) ? 3
1226 : (Value < (1 << 28)) ? 4
1227 : 5;
1228 };
1229
1230 // Returns the total number of bytes written since this object was created.
ByteCount()1231 int ByteCount() const {
1232 return static_cast<int>(impl_.ByteCount(cur_) - start_count_);
1233 }
1234
1235 // Instructs the CodedOutputStream to allow the underlying
1236 // ZeroCopyOutputStream to hold pointers to the original structure instead of
1237 // copying, if it supports it (i.e. output->AllowsAliasing() is true). If the
1238 // underlying stream does not support aliasing, then enabling it has no
1239 // affect. For now, this only affects the behavior of
1240 // WriteRawMaybeAliased().
1241 //
1242 // NOTE: It is caller's responsibility to ensure that the chunk of memory
1243 // remains live until all of the data has been consumed from the stream.
EnableAliasing(bool enabled)1244 void EnableAliasing(bool enabled) { impl_.EnableAliasing(enabled); }
1245
1246 // Indicate to the serializer whether the user wants deterministic
1247 // serialization. The default when this is not called comes from the global
1248 // default, controlled by SetDefaultSerializationDeterministic.
1249 //
1250 // What deterministic serialization means is entirely up to the driver of the
1251 // serialization process (i.e. the caller of methods like WriteVarint32). In
1252 // the case of serializing a proto buffer message using one of the methods of
1253 // MessageLite, this means that for a given binary equal messages will always
1254 // be serialized to the same bytes. This implies:
1255 //
1256 // * Repeated serialization of a message will return the same bytes.
1257 //
1258 // * Different processes running the same binary (including on different
1259 // machines) will serialize equal messages to the same bytes.
1260 //
1261 // Note that this is *not* canonical across languages. It is also unstable
1262 // across different builds with intervening message definition changes, due to
1263 // unknown fields. Users who need canonical serialization (e.g. persistent
1264 // storage in a canonical form, fingerprinting) should define their own
1265 // canonicalization specification and implement the serializer using
1266 // reflection APIs rather than relying on this API.
SetSerializationDeterministic(bool value)1267 void SetSerializationDeterministic(bool value) {
1268 impl_.SetSerializationDeterministic(value);
1269 }
1270
1271 // Return whether the user wants deterministic serialization. See above.
IsSerializationDeterministic()1272 bool IsSerializationDeterministic() const {
1273 return impl_.IsSerializationDeterministic();
1274 }
1275
IsDefaultSerializationDeterministic()1276 static bool IsDefaultSerializationDeterministic() {
1277 return default_serialization_deterministic_.load(
1278 std::memory_order_relaxed) != 0;
1279 }
1280
1281 template <typename Func>
1282 void Serialize(const Func& func);
1283
Cur()1284 uint8_t* Cur() const { return cur_; }
SetCur(uint8_t * ptr)1285 void SetCur(uint8_t* ptr) { cur_ = ptr; }
EpsCopy()1286 EpsCopyOutputStream* EpsCopy() { return &impl_; }
1287
1288 private:
1289 template <class Stream>
1290 void InitEagerly(Stream* stream);
1291
1292 EpsCopyOutputStream impl_;
1293 uint8_t* cur_;
1294 int64_t start_count_;
1295 static std::atomic<bool> default_serialization_deterministic_;
1296
1297 // See above. Other projects may use "friend" to allow them to call this.
1298 // After SetDefaultSerializationDeterministic() completes, all protocol
1299 // buffer serializations will be deterministic by default. Thread safe.
1300 // However, the meaning of "after" is subtle here: to be safe, each thread
1301 // that wants deterministic serialization by default needs to call
1302 // SetDefaultSerializationDeterministic() or ensure on its own that another
1303 // thread has done so.
1304 friend void internal::MapTestForceDeterministic();
SetDefaultSerializationDeterministic()1305 static void SetDefaultSerializationDeterministic() {
1306 default_serialization_deterministic_.store(true, std::memory_order_relaxed);
1307 }
1308 // REQUIRES: value >= 0x80, and that (value & 7f) has been written to *target.
1309 static uint8_t* WriteVarint32ToArrayOutOfLineHelper(uint32_t value,
1310 uint8_t* target);
1311 GOOGLE_DISALLOW_EVIL_CONSTRUCTORS(CodedOutputStream);
1312 };
1313
1314 // inline methods ====================================================
1315 // The vast majority of varints are only one byte. These inline
1316 // methods optimize for that case.
1317
ReadVarint32(uint32_t * value)1318 inline bool CodedInputStream::ReadVarint32(uint32_t* value) {
1319 uint32_t v = 0;
1320 if (PROTOBUF_PREDICT_TRUE(buffer_ < buffer_end_)) {
1321 v = *buffer_;
1322 if (v < 0x80) {
1323 *value = v;
1324 Advance(1);
1325 return true;
1326 }
1327 }
1328 int64_t result = ReadVarint32Fallback(v);
1329 *value = static_cast<uint32_t>(result);
1330 return result >= 0;
1331 }
1332
ReadVarint64(uint64_t * value)1333 inline bool CodedInputStream::ReadVarint64(uint64_t* value) {
1334 if (PROTOBUF_PREDICT_TRUE(buffer_ < buffer_end_) && *buffer_ < 0x80) {
1335 *value = *buffer_;
1336 Advance(1);
1337 return true;
1338 }
1339 std::pair<uint64_t, bool> p = ReadVarint64Fallback();
1340 *value = p.first;
1341 return p.second;
1342 }
1343
ReadVarintSizeAsInt(int * value)1344 inline bool CodedInputStream::ReadVarintSizeAsInt(int* value) {
1345 if (PROTOBUF_PREDICT_TRUE(buffer_ < buffer_end_)) {
1346 int v = *buffer_;
1347 if (v < 0x80) {
1348 *value = v;
1349 Advance(1);
1350 return true;
1351 }
1352 }
1353 *value = ReadVarintSizeAsIntFallback();
1354 return *value >= 0;
1355 }
1356
1357 // static
ReadLittleEndian32FromArray(const uint8_t * buffer,uint32_t * value)1358 inline const uint8_t* CodedInputStream::ReadLittleEndian32FromArray(
1359 const uint8_t* buffer, uint32_t* value) {
1360 #if defined(PROTOBUF_LITTLE_ENDIAN)
1361 memcpy(value, buffer, sizeof(*value));
1362 return buffer + sizeof(*value);
1363 #else
1364 *value = (static_cast<uint32_t>(buffer[0])) |
1365 (static_cast<uint32_t>(buffer[1]) << 8) |
1366 (static_cast<uint32_t>(buffer[2]) << 16) |
1367 (static_cast<uint32_t>(buffer[3]) << 24);
1368 return buffer + sizeof(*value);
1369 #endif
1370 }
1371 // static
ReadLittleEndian64FromArray(const uint8_t * buffer,uint64_t * value)1372 inline const uint8_t* CodedInputStream::ReadLittleEndian64FromArray(
1373 const uint8_t* buffer, uint64_t* value) {
1374 #if defined(PROTOBUF_LITTLE_ENDIAN)
1375 memcpy(value, buffer, sizeof(*value));
1376 return buffer + sizeof(*value);
1377 #else
1378 uint32_t part0 = (static_cast<uint32_t>(buffer[0])) |
1379 (static_cast<uint32_t>(buffer[1]) << 8) |
1380 (static_cast<uint32_t>(buffer[2]) << 16) |
1381 (static_cast<uint32_t>(buffer[3]) << 24);
1382 uint32_t part1 = (static_cast<uint32_t>(buffer[4])) |
1383 (static_cast<uint32_t>(buffer[5]) << 8) |
1384 (static_cast<uint32_t>(buffer[6]) << 16) |
1385 (static_cast<uint32_t>(buffer[7]) << 24);
1386 *value = static_cast<uint64_t>(part0) | (static_cast<uint64_t>(part1) << 32);
1387 return buffer + sizeof(*value);
1388 #endif
1389 }
1390
ReadLittleEndian32(uint32_t * value)1391 inline bool CodedInputStream::ReadLittleEndian32(uint32_t* value) {
1392 #if defined(PROTOBUF_LITTLE_ENDIAN)
1393 if (PROTOBUF_PREDICT_TRUE(BufferSize() >= static_cast<int>(sizeof(*value)))) {
1394 buffer_ = ReadLittleEndian32FromArray(buffer_, value);
1395 return true;
1396 } else {
1397 return ReadLittleEndian32Fallback(value);
1398 }
1399 #else
1400 return ReadLittleEndian32Fallback(value);
1401 #endif
1402 }
1403
ReadLittleEndian64(uint64_t * value)1404 inline bool CodedInputStream::ReadLittleEndian64(uint64_t* value) {
1405 #if defined(PROTOBUF_LITTLE_ENDIAN)
1406 if (PROTOBUF_PREDICT_TRUE(BufferSize() >= static_cast<int>(sizeof(*value)))) {
1407 buffer_ = ReadLittleEndian64FromArray(buffer_, value);
1408 return true;
1409 } else {
1410 return ReadLittleEndian64Fallback(value);
1411 }
1412 #else
1413 return ReadLittleEndian64Fallback(value);
1414 #endif
1415 }
1416
ReadTagNoLastTag()1417 inline uint32_t CodedInputStream::ReadTagNoLastTag() {
1418 uint32_t v = 0;
1419 if (PROTOBUF_PREDICT_TRUE(buffer_ < buffer_end_)) {
1420 v = *buffer_;
1421 if (v < 0x80) {
1422 Advance(1);
1423 return v;
1424 }
1425 }
1426 v = ReadTagFallback(v);
1427 return v;
1428 }
1429
ReadTagWithCutoffNoLastTag(uint32_t cutoff)1430 inline std::pair<uint32_t, bool> CodedInputStream::ReadTagWithCutoffNoLastTag(
1431 uint32_t cutoff) {
1432 // In performance-sensitive code we can expect cutoff to be a compile-time
1433 // constant, and things like "cutoff >= kMax1ByteVarint" to be evaluated at
1434 // compile time.
1435 uint32_t first_byte_or_zero = 0;
1436 if (PROTOBUF_PREDICT_TRUE(buffer_ < buffer_end_)) {
1437 // Hot case: buffer_ non_empty, buffer_[0] in [1, 128).
1438 // TODO(gpike): Is it worth rearranging this? E.g., if the number of fields
1439 // is large enough then is it better to check for the two-byte case first?
1440 first_byte_or_zero = buffer_[0];
1441 if (static_cast<int8_t>(buffer_[0]) > 0) {
1442 const uint32_t kMax1ByteVarint = 0x7f;
1443 uint32_t tag = buffer_[0];
1444 Advance(1);
1445 return std::make_pair(tag, cutoff >= kMax1ByteVarint || tag <= cutoff);
1446 }
1447 // Other hot case: cutoff >= 0x80, buffer_ has at least two bytes available,
1448 // and tag is two bytes. The latter is tested by bitwise-and-not of the
1449 // first byte and the second byte.
1450 if (cutoff >= 0x80 && PROTOBUF_PREDICT_TRUE(buffer_ + 1 < buffer_end_) &&
1451 PROTOBUF_PREDICT_TRUE((buffer_[0] & ~buffer_[1]) >= 0x80)) {
1452 const uint32_t kMax2ByteVarint = (0x7f << 7) + 0x7f;
1453 uint32_t tag = (1u << 7) * buffer_[1] + (buffer_[0] - 0x80);
1454 Advance(2);
1455 // It might make sense to test for tag == 0 now, but it is so rare that
1456 // that we don't bother. A varint-encoded 0 should be one byte unless
1457 // the encoder lost its mind. The second part of the return value of
1458 // this function is allowed to be either true or false if the tag is 0,
1459 // so we don't have to check for tag == 0. We may need to check whether
1460 // it exceeds cutoff.
1461 bool at_or_below_cutoff = cutoff >= kMax2ByteVarint || tag <= cutoff;
1462 return std::make_pair(tag, at_or_below_cutoff);
1463 }
1464 }
1465 // Slow path
1466 const uint32_t tag = ReadTagFallback(first_byte_or_zero);
1467 return std::make_pair(tag, static_cast<uint32_t>(tag - 1) < cutoff);
1468 }
1469
LastTagWas(uint32_t expected)1470 inline bool CodedInputStream::LastTagWas(uint32_t expected) {
1471 return last_tag_ == expected;
1472 }
1473
ConsumedEntireMessage()1474 inline bool CodedInputStream::ConsumedEntireMessage() {
1475 return legitimate_message_end_;
1476 }
1477
ExpectTag(uint32_t expected)1478 inline bool CodedInputStream::ExpectTag(uint32_t expected) {
1479 if (expected < (1 << 7)) {
1480 if (PROTOBUF_PREDICT_TRUE(buffer_ < buffer_end_) &&
1481 buffer_[0] == expected) {
1482 Advance(1);
1483 return true;
1484 } else {
1485 return false;
1486 }
1487 } else if (expected < (1 << 14)) {
1488 if (PROTOBUF_PREDICT_TRUE(BufferSize() >= 2) &&
1489 buffer_[0] == static_cast<uint8_t>(expected | 0x80) &&
1490 buffer_[1] == static_cast<uint8_t>(expected >> 7)) {
1491 Advance(2);
1492 return true;
1493 } else {
1494 return false;
1495 }
1496 } else {
1497 // Don't bother optimizing for larger values.
1498 return false;
1499 }
1500 }
1501
ExpectTagFromArray(const uint8_t * buffer,uint32_t expected)1502 inline const uint8_t* CodedInputStream::ExpectTagFromArray(
1503 const uint8_t* buffer, uint32_t expected) {
1504 if (expected < (1 << 7)) {
1505 if (buffer[0] == expected) {
1506 return buffer + 1;
1507 }
1508 } else if (expected < (1 << 14)) {
1509 if (buffer[0] == static_cast<uint8_t>(expected | 0x80) &&
1510 buffer[1] == static_cast<uint8_t>(expected >> 7)) {
1511 return buffer + 2;
1512 }
1513 }
1514 return nullptr;
1515 }
1516
GetDirectBufferPointerInline(const void ** data,int * size)1517 inline void CodedInputStream::GetDirectBufferPointerInline(const void** data,
1518 int* size) {
1519 *data = buffer_;
1520 *size = static_cast<int>(buffer_end_ - buffer_);
1521 }
1522
ExpectAtEnd()1523 inline bool CodedInputStream::ExpectAtEnd() {
1524 // If we are at a limit we know no more bytes can be read. Otherwise, it's
1525 // hard to say without calling Refresh(), and we'd rather not do that.
1526
1527 if (buffer_ == buffer_end_ && ((buffer_size_after_limit_ != 0) ||
1528 (total_bytes_read_ == current_limit_))) {
1529 last_tag_ = 0; // Pretend we called ReadTag()...
1530 legitimate_message_end_ = true; // ... and it hit EOF.
1531 return true;
1532 } else {
1533 return false;
1534 }
1535 }
1536
CurrentPosition()1537 inline int CodedInputStream::CurrentPosition() const {
1538 return total_bytes_read_ - (BufferSize() + buffer_size_after_limit_);
1539 }
1540
Advance(int amount)1541 inline void CodedInputStream::Advance(int amount) { buffer_ += amount; }
1542
SetRecursionLimit(int limit)1543 inline void CodedInputStream::SetRecursionLimit(int limit) {
1544 recursion_budget_ += limit - recursion_limit_;
1545 recursion_limit_ = limit;
1546 }
1547
IncrementRecursionDepth()1548 inline bool CodedInputStream::IncrementRecursionDepth() {
1549 --recursion_budget_;
1550 return recursion_budget_ >= 0;
1551 }
1552
DecrementRecursionDepth()1553 inline void CodedInputStream::DecrementRecursionDepth() {
1554 if (recursion_budget_ < recursion_limit_) ++recursion_budget_;
1555 }
1556
UnsafeDecrementRecursionDepth()1557 inline void CodedInputStream::UnsafeDecrementRecursionDepth() {
1558 assert(recursion_budget_ < recursion_limit_);
1559 ++recursion_budget_;
1560 }
1561
SetExtensionRegistry(const DescriptorPool * pool,MessageFactory * factory)1562 inline void CodedInputStream::SetExtensionRegistry(const DescriptorPool* pool,
1563 MessageFactory* factory) {
1564 extension_pool_ = pool;
1565 extension_factory_ = factory;
1566 }
1567
GetExtensionPool()1568 inline const DescriptorPool* CodedInputStream::GetExtensionPool() {
1569 return extension_pool_;
1570 }
1571
GetExtensionFactory()1572 inline MessageFactory* CodedInputStream::GetExtensionFactory() {
1573 return extension_factory_;
1574 }
1575
BufferSize()1576 inline int CodedInputStream::BufferSize() const {
1577 return static_cast<int>(buffer_end_ - buffer_);
1578 }
1579
CodedInputStream(ZeroCopyInputStream * input)1580 inline CodedInputStream::CodedInputStream(ZeroCopyInputStream* input)
1581 : buffer_(nullptr),
1582 buffer_end_(nullptr),
1583 input_(input),
1584 total_bytes_read_(0),
1585 overflow_bytes_(0),
1586 last_tag_(0),
1587 legitimate_message_end_(false),
1588 aliasing_enabled_(false),
1589 current_limit_(std::numeric_limits<int32_t>::max()),
1590 buffer_size_after_limit_(0),
1591 total_bytes_limit_(kDefaultTotalBytesLimit),
1592 recursion_budget_(default_recursion_limit_),
1593 recursion_limit_(default_recursion_limit_),
1594 extension_pool_(nullptr),
1595 extension_factory_(nullptr) {
1596 // Eagerly Refresh() so buffer space is immediately available.
1597 Refresh();
1598 }
1599
CodedInputStream(const uint8_t * buffer,int size)1600 inline CodedInputStream::CodedInputStream(const uint8_t* buffer, int size)
1601 : buffer_(buffer),
1602 buffer_end_(buffer + size),
1603 input_(nullptr),
1604 total_bytes_read_(size),
1605 overflow_bytes_(0),
1606 last_tag_(0),
1607 legitimate_message_end_(false),
1608 aliasing_enabled_(false),
1609 current_limit_(size),
1610 buffer_size_after_limit_(0),
1611 total_bytes_limit_(kDefaultTotalBytesLimit),
1612 recursion_budget_(default_recursion_limit_),
1613 recursion_limit_(default_recursion_limit_),
1614 extension_pool_(nullptr),
1615 extension_factory_(nullptr) {
1616 // Note that setting current_limit_ == size is important to prevent some
1617 // code paths from trying to access input_ and segfaulting.
1618 }
1619
IsFlat()1620 inline bool CodedInputStream::IsFlat() const { return input_ == nullptr; }
1621
Skip(int count)1622 inline bool CodedInputStream::Skip(int count) {
1623 if (count < 0) return false; // security: count is often user-supplied
1624
1625 const int original_buffer_size = BufferSize();
1626
1627 if (count <= original_buffer_size) {
1628 // Just skipping within the current buffer. Easy.
1629 Advance(count);
1630 return true;
1631 }
1632
1633 return SkipFallback(count, original_buffer_size);
1634 }
1635
1636 template <class Stream, class>
CodedOutputStream(Stream * stream)1637 inline CodedOutputStream::CodedOutputStream(Stream* stream)
1638 : impl_(stream, IsDefaultSerializationDeterministic(), &cur_),
1639 start_count_(stream->ByteCount()) {
1640 InitEagerly(stream);
1641 }
1642
1643 template <class Stream, class>
CodedOutputStream(Stream * stream,bool eager_init)1644 inline CodedOutputStream::CodedOutputStream(Stream* stream, bool eager_init)
1645 : impl_(stream, IsDefaultSerializationDeterministic(), &cur_),
1646 start_count_(stream->ByteCount()) {
1647 if (eager_init) {
1648 InitEagerly(stream);
1649 }
1650 }
1651
1652 template <class Stream>
InitEagerly(Stream * stream)1653 inline void CodedOutputStream::InitEagerly(Stream* stream) {
1654 void* data;
1655 int size;
1656 if (PROTOBUF_PREDICT_TRUE(stream->Next(&data, &size) && size > 0)) {
1657 cur_ = impl_.SetInitialBuffer(data, size);
1658 }
1659 }
1660
WriteVarint32ToArray(uint32_t value,uint8_t * target)1661 inline uint8_t* CodedOutputStream::WriteVarint32ToArray(uint32_t value,
1662 uint8_t* target) {
1663 return EpsCopyOutputStream::UnsafeVarint(value, target);
1664 }
1665
WriteVarint32ToArrayOutOfLine(uint32_t value,uint8_t * target)1666 inline uint8_t* CodedOutputStream::WriteVarint32ToArrayOutOfLine(
1667 uint32_t value, uint8_t* target) {
1668 target[0] = static_cast<uint8_t>(value);
1669 if (value < 0x80) {
1670 return target + 1;
1671 } else {
1672 return WriteVarint32ToArrayOutOfLineHelper(value, target);
1673 }
1674 }
1675
WriteVarint64ToArray(uint64_t value,uint8_t * target)1676 inline uint8_t* CodedOutputStream::WriteVarint64ToArray(uint64_t value,
1677 uint8_t* target) {
1678 return EpsCopyOutputStream::UnsafeVarint(value, target);
1679 }
1680
WriteVarint32SignExtended(int32_t value)1681 inline void CodedOutputStream::WriteVarint32SignExtended(int32_t value) {
1682 WriteVarint64(static_cast<uint64_t>(value));
1683 }
1684
WriteVarint32SignExtendedToArray(int32_t value,uint8_t * target)1685 inline uint8_t* CodedOutputStream::WriteVarint32SignExtendedToArray(
1686 int32_t value, uint8_t* target) {
1687 return WriteVarint64ToArray(static_cast<uint64_t>(value), target);
1688 }
1689
WriteLittleEndian32ToArray(uint32_t value,uint8_t * target)1690 inline uint8_t* CodedOutputStream::WriteLittleEndian32ToArray(uint32_t value,
1691 uint8_t* target) {
1692 #if defined(PROTOBUF_LITTLE_ENDIAN)
1693 memcpy(target, &value, sizeof(value));
1694 #else
1695 target[0] = static_cast<uint8_t>(value);
1696 target[1] = static_cast<uint8_t>(value >> 8);
1697 target[2] = static_cast<uint8_t>(value >> 16);
1698 target[3] = static_cast<uint8_t>(value >> 24);
1699 #endif
1700 return target + sizeof(value);
1701 }
1702
WriteLittleEndian64ToArray(uint64_t value,uint8_t * target)1703 inline uint8_t* CodedOutputStream::WriteLittleEndian64ToArray(uint64_t value,
1704 uint8_t* target) {
1705 #if defined(PROTOBUF_LITTLE_ENDIAN)
1706 memcpy(target, &value, sizeof(value));
1707 #else
1708 uint32_t part0 = static_cast<uint32_t>(value);
1709 uint32_t part1 = static_cast<uint32_t>(value >> 32);
1710
1711 target[0] = static_cast<uint8_t>(part0);
1712 target[1] = static_cast<uint8_t>(part0 >> 8);
1713 target[2] = static_cast<uint8_t>(part0 >> 16);
1714 target[3] = static_cast<uint8_t>(part0 >> 24);
1715 target[4] = static_cast<uint8_t>(part1);
1716 target[5] = static_cast<uint8_t>(part1 >> 8);
1717 target[6] = static_cast<uint8_t>(part1 >> 16);
1718 target[7] = static_cast<uint8_t>(part1 >> 24);
1719 #endif
1720 return target + sizeof(value);
1721 }
1722
WriteVarint32(uint32_t value)1723 inline void CodedOutputStream::WriteVarint32(uint32_t value) {
1724 cur_ = impl_.EnsureSpace(cur_);
1725 SetCur(WriteVarint32ToArray(value, Cur()));
1726 }
1727
WriteVarint64(uint64_t value)1728 inline void CodedOutputStream::WriteVarint64(uint64_t value) {
1729 cur_ = impl_.EnsureSpace(cur_);
1730 SetCur(WriteVarint64ToArray(value, Cur()));
1731 }
1732
WriteTag(uint32_t value)1733 inline void CodedOutputStream::WriteTag(uint32_t value) {
1734 WriteVarint32(value);
1735 }
1736
WriteTagToArray(uint32_t value,uint8_t * target)1737 inline uint8_t* CodedOutputStream::WriteTagToArray(uint32_t value,
1738 uint8_t* target) {
1739 return WriteVarint32ToArray(value, target);
1740 }
1741
VarintSize32(uint32_t value)1742 inline size_t CodedOutputStream::VarintSize32(uint32_t value) {
1743 // This computes value == 0 ? 1 : floor(log2(value)) / 7 + 1
1744 // Use an explicit multiplication to implement the divide of
1745 // a number in the 1..31 range.
1746 // Explicit OR 0x1 to avoid calling Bits::Log2FloorNonZero(0), which is
1747 // undefined.
1748 uint32_t log2value = Bits::Log2FloorNonZero(value | 0x1);
1749 return static_cast<size_t>((log2value * 9 + 73) / 64);
1750 }
1751
VarintSize32PlusOne(uint32_t value)1752 inline size_t CodedOutputStream::VarintSize32PlusOne(uint32_t value) {
1753 // Same as above, but one more.
1754 uint32_t log2value = Bits::Log2FloorNonZero(value | 0x1);
1755 return static_cast<size_t>((log2value * 9 + 73 + 64) / 64);
1756 }
1757
VarintSize64(uint64_t value)1758 inline size_t CodedOutputStream::VarintSize64(uint64_t value) {
1759 // This computes value == 0 ? 1 : floor(log2(value)) / 7 + 1
1760 // Use an explicit multiplication to implement the divide of
1761 // a number in the 1..63 range.
1762 // Explicit OR 0x1 to avoid calling Bits::Log2FloorNonZero(0), which is
1763 // undefined.
1764 uint32_t log2value = Bits::Log2FloorNonZero64(value | 0x1);
1765 return static_cast<size_t>((log2value * 9 + 73) / 64);
1766 }
1767
VarintSize64PlusOne(uint64_t value)1768 inline size_t CodedOutputStream::VarintSize64PlusOne(uint64_t value) {
1769 // Same as above, but one more.
1770 uint32_t log2value = Bits::Log2FloorNonZero64(value | 0x1);
1771 return static_cast<size_t>((log2value * 9 + 73 + 64) / 64);
1772 }
1773
VarintSize32SignExtended(int32_t value)1774 inline size_t CodedOutputStream::VarintSize32SignExtended(int32_t value) {
1775 return VarintSize64(static_cast<uint64_t>(int64_t{value}));
1776 }
1777
VarintSize32SignExtendedPlusOne(int32_t value)1778 inline size_t CodedOutputStream::VarintSize32SignExtendedPlusOne(
1779 int32_t value) {
1780 return VarintSize64PlusOne(static_cast<uint64_t>(int64_t{value}));
1781 }
1782
WriteString(const std::string & str)1783 inline void CodedOutputStream::WriteString(const std::string& str) {
1784 WriteRaw(str.data(), static_cast<int>(str.size()));
1785 }
1786
WriteRawMaybeAliased(const void * data,int size)1787 inline void CodedOutputStream::WriteRawMaybeAliased(const void* data,
1788 int size) {
1789 cur_ = impl_.WriteRawMaybeAliased(data, size, cur_);
1790 }
1791
WriteRawToArray(const void * data,int size,uint8_t * target)1792 inline uint8_t* CodedOutputStream::WriteRawToArray(const void* data, int size,
1793 uint8_t* target) {
1794 memcpy(target, data, size);
1795 return target + size;
1796 }
1797
WriteStringToArray(const std::string & str,uint8_t * target)1798 inline uint8_t* CodedOutputStream::WriteStringToArray(const std::string& str,
1799 uint8_t* target) {
1800 return WriteRawToArray(str.data(), static_cast<int>(str.size()), target);
1801 }
1802
1803 } // namespace io
1804 } // namespace protobuf
1805 } // namespace google
1806
1807 #if defined(_MSC_VER) && _MSC_VER >= 1300 && !defined(__INTEL_COMPILER)
1808 #pragma runtime_checks("c", restore)
1809 #endif // _MSC_VER && !defined(__INTEL_COMPILER)
1810
1811 #include <google/protobuf/port_undef.inc>
1812
1813 #endif // GOOGLE_PROTOBUF_IO_CODED_STREAM_H__
1814