tor-browser

decode_rust_punycode.cc (10283B)

---

// Copyright 2024 The Abseil Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     https://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include "absl/debugging/internal/decode_rust_punycode.h"

#include <cstddef>
#include <cstdint>
#include <cstring>

#include "absl/base/config.h"
#include "absl/base/nullability.h"
#include "absl/debugging/internal/bounded_utf8_length_sequence.h"
#include "absl/debugging/internal/utf8_for_code_point.h"

namespace absl {
ABSL_NAMESPACE_BEGIN
namespace debugging_internal {

namespace {

// Decoding Punycode requires repeated random-access insertion into a stream of
// variable-length UTF-8 code-point encodings.  We need this to be tolerably
// fast (no N^2 slowdown for unfortunate inputs), and we can't allocate any data
// structures on the heap (async-signal-safety).
//
// It is pragmatic to impose a moderately low limit on the identifier length and
// bail out if we ever hit it.  Then BoundedUtf8LengthSequence efficiently
// determines where to insert the next code point, and memmove efficiently makes
// room for it.
//
// The chosen limit is a round number several times larger than identifiers
// expected in practice, yet still small enough that a memmove of this many
// UTF-8 characters is not much more expensive than the division and modulus
// operations that Punycode decoding requires.
constexpr uint32_t kMaxChars = 256;

// Constants from RFC 3492 section 5.
constexpr uint32_t kBase = 36, kTMin = 1, kTMax = 26, kSkew = 38, kDamp = 700;

constexpr uint32_t kMaxCodePoint = 0x10ffff;

// Overflow threshold in DecodeRustPunycode's inner loop; see comments there.
constexpr uint32_t kMaxI = 1 << 30;

// If punycode_begin .. punycode_end begins with a prefix matching the regular
// expression [0-9a-zA-Z_]+_, removes that prefix, copies all but the final
// underscore into out_begin .. out_end, sets num_ascii_chars to the number of
// bytes copied, and returns true.  (A prefix of this sort represents the
// nonempty subsequence of ASCII characters in the corresponding plaintext.)
//
// If punycode_begin .. punycode_end does not contain an underscore, sets
// num_ascii_chars to zero and returns true.  (The encoding of a plaintext
// without any ASCII characters does not carry such a prefix.)
//
// Returns false and zeroes num_ascii_chars on failure (either parse error or
// not enough space in the output buffer).
bool ConsumeOptionalAsciiPrefix(const char*& punycode_begin,
                               const char* const punycode_end,
                               char* const out_begin,
                               char* const out_end,
                               uint32_t& num_ascii_chars) {
 num_ascii_chars = 0;

 // Remember the last underscore if any.  Also use the same string scan to
 // reject any ASCII bytes that do not belong in an identifier, including NUL,
 // as well as non-ASCII bytes, which should have been delta-encoded instead.
 int last_underscore = -1;
 for (int i = 0; i < punycode_end - punycode_begin; ++i) {
   const char c = punycode_begin[i];
   if (c == '_') {
     last_underscore = i;
     continue;
   }
   // We write out the meaning of absl::ascii_isalnum rather than call that
   // function because its documentation does not promise it will remain
   // async-signal-safe under future development.
   if ('a' <= c && c <= 'z') continue;
   if ('A' <= c && c <= 'Z') continue;
   if ('0' <= c && c <= '9') continue;
   return false;
 }

 // If there was no underscore, that means there were no ASCII characters in
 // the plaintext, so there is no prefix to consume.  Our work is done.
 if (last_underscore < 0) return true;

 // Otherwise there will be an underscore delimiter somewhere.  It can't be
 // initial because then there would be no ASCII characters to its left, and no
 // delimiter would have been added in that case.
 if (last_underscore == 0) return false;

 // Any other position is reasonable.  Make sure there's room in the buffer.
 if (last_underscore + 1 > out_end - out_begin) return false;

 // Consume and write out the ASCII characters.
 num_ascii_chars = static_cast<uint32_t>(last_underscore);
 std::memcpy(out_begin, punycode_begin, num_ascii_chars);
 out_begin[num_ascii_chars] = '\0';
 punycode_begin += num_ascii_chars + 1;
 return true;
}

// Returns the value of `c` as a base-36 digit according to RFC 3492 section 5,
// or -1 if `c` is not such a digit.
int DigitValue(char c) {
 if ('0' <= c && c <= '9') return c - '0' + 26;
 if ('a' <= c && c <= 'z') return c - 'a';
 if ('A' <= c && c <= 'Z') return c - 'A';
 return -1;
}

// Consumes the next delta encoding from punycode_begin .. punycode_end,
// updating i accordingly.  Returns true on success.  Returns false on parse
// failure or arithmetic overflow.
bool ScanNextDelta(const char*& punycode_begin, const char* const punycode_end,
                  uint32_t bias, uint32_t& i) {
 uint64_t w = 1;  // 64 bits to prevent overflow in w *= kBase - t

 // "for k = base to infinity in steps of base do begin ... end" in RFC 3492
 // section 6.2.  Each loop iteration scans one digit of the delta.
 for (uint32_t k = kBase; punycode_begin != punycode_end; k += kBase) {
   const int digit_value = DigitValue(*punycode_begin++);
   if (digit_value < 0) return false;

   // Compute this in 64-bit arithmetic so we can check for overflow afterward.
   const uint64_t new_i = i + static_cast<uint64_t>(digit_value) * w;

   // Valid deltas are bounded by (#chars already emitted) * kMaxCodePoint, but
   // invalid input could encode an arbitrarily large delta.  Nip that in the
   // bud here.
   static_assert(
       kMaxI >= kMaxChars * kMaxCodePoint,
       "kMaxI is too small to prevent spurious failures on good input");
   if (new_i > kMaxI) return false;

   static_assert(
       kMaxI < (uint64_t{1} << 32),
       "Make kMaxI smaller or i 64 bits wide to prevent silent wraparound");
   i = static_cast<uint32_t>(new_i);

   // Compute the threshold that determines whether this is the last digit and
   // (if not) what the next digit's place value will be.  This logic from RFC
   // 3492 section 6.2 is explained in section 3.3.
   uint32_t t;
   if (k <= bias + kTMin) {
     t = kTMin;
   } else if (k >= bias + kTMax) {
     t = kTMax;
   } else {
     t = k - bias;
   }
   if (static_cast<uint32_t>(digit_value) < t) return true;

   // If this gets too large, the range check on new_i in the next iteration
   // will catch it.  We know this multiplication will not overwrap because w
   // is 64 bits wide.
   w *= kBase - t;
 }
 return false;
}

}  // namespace

absl::Nullable<char*> DecodeRustPunycode(DecodeRustPunycodeOptions options) {
 const char* punycode_begin = options.punycode_begin;
 const char* const punycode_end = options.punycode_end;
 char* const out_begin = options.out_begin;
 char* const out_end = options.out_end;

 // Write a NUL terminator first.  Later memcpy calls will keep bumping it
 // along to its new right place.
 const size_t out_size = static_cast<size_t>(out_end - out_begin);
 if (out_size == 0) return nullptr;
 *out_begin = '\0';

 // RFC 3492 section 6.2 begins here.  We retain the names of integer variables
 // appearing in that text.
 uint32_t n = 128, i = 0, bias = 72, num_chars = 0;

 // If there are any ASCII characters, consume them and their trailing
 // underscore delimiter.
 if (!ConsumeOptionalAsciiPrefix(punycode_begin, punycode_end,
                                 out_begin, out_end, num_chars)) {
   return nullptr;
 }
 uint32_t total_utf8_bytes = num_chars;

 BoundedUtf8LengthSequence<kMaxChars> utf8_lengths;

 // "while the input is not exhausted do begin ... end"
 while (punycode_begin != punycode_end) {
   if (num_chars >= kMaxChars) return nullptr;

   const uint32_t old_i = i;

   if (!ScanNextDelta(punycode_begin, punycode_end, bias, i)) return nullptr;

   // Update bias as in RFC 3492 section 6.1.  (We have inlined adapt.)
   uint32_t delta = i - old_i;
   delta /= (old_i == 0 ? kDamp : 2);
   delta += delta/(num_chars + 1);
   bias = 0;
   while (delta > ((kBase - kTMin) * kTMax)/2) {
     delta /= kBase - kTMin;
     bias += kBase;
   }
   bias += ((kBase - kTMin + 1) * delta)/(delta + kSkew);

   // Back in section 6.2, compute the new code point and insertion index.
   static_assert(
       kMaxI + kMaxCodePoint < (uint64_t{1} << 32),
       "Make kMaxI smaller or n 64 bits wide to prevent silent wraparound");
   n += i/(num_chars + 1);
   i %= num_chars + 1;

   // To actually insert, we need to convert the code point n to UTF-8 and the
   // character index i to an index into the byte stream emitted so far.  First
   // prepare the UTF-8 encoding for n, rejecting surrogates, overlarge values,
   // and anything that won't fit into the remaining output storage.
   Utf8ForCodePoint utf8_for_code_point(n);
   if (!utf8_for_code_point.ok()) return nullptr;
   if (total_utf8_bytes + utf8_for_code_point.length + 1 > out_size) {
     return nullptr;
   }

   // Now insert the new character into both our length map and the output.
   uint32_t n_index =
       utf8_lengths.InsertAndReturnSumOfPredecessors(
           i, utf8_for_code_point.length);
   std::memmove(
       out_begin + n_index + utf8_for_code_point.length, out_begin + n_index,
       total_utf8_bytes + 1 - n_index);
   std::memcpy(out_begin + n_index, utf8_for_code_point.bytes,
               utf8_for_code_point.length);
   total_utf8_bytes += utf8_for_code_point.length;
   ++num_chars;

   // Finally, advance to the next state before continuing.
   ++i;
 }

 return out_begin + total_utf8_bytes;
}

}  // namespace debugging_internal
ABSL_NAMESPACE_END
}  // namespace absl