Add FuzzedDataProvider utility for structured fuzzing

fuzzeddataprovider.py

@@ -0,0 +1,244 @@
+"""Pure-Python FuzzedDataProvider matching the atheris API.
+
+This is a drop-in replacement for atheris.FuzzedDataProvider that requires
+no native compilation. It matches atheris's consumption semantics:
+  - ConsumeBytes/ConsumeInt/ConsumeFloat/ConsumeUnicode consume from the FRONT
+  - ConsumeIntInRange/ConsumeBool/PickValueInList consume from the BACK
+
+Reference: https://github.com/google/atheris
+"""
+
+import struct
+
+
+class FuzzedDataProvider:
+    def __init__(self, data):
+        if not isinstance(data, (bytes, bytearray)):
+            raise TypeError("data must be bytes or bytearray")
+        self._data = bytes(data)
+        self._front = 0
+        self._back = len(self._data)
+
+    def remaining_bytes(self):
+        return max(0, self._back - self._front)
+
+    def buffer(self):
+        return self._data[self._front:self._back]
+
+    # -- Front-consuming methods (ConsumeBytes, ConsumeInt, etc.) --
+
+    def _consume_front(self, n):
+        n = min(n, self.remaining_bytes())
+        result = self._data[self._front:self._front + n]
+        self._front += n
+        return result
+
+    def ConsumeBytes(self, count):
+        count = max(0, int(count))
+        return self._consume_front(count)
+
+    def ConsumeInt(self, byte_count):
+        byte_count = max(0, int(byte_count))
+        raw = self._consume_front(byte_count)
+        if not raw:
+            return 0
+        val = int.from_bytes(raw, 'little')
+        bits = len(raw) * 8
+        if val >= (1 << (bits - 1)):
+            val -= (1 << bits)
+        return val
+
+    def ConsumeUInt(self, byte_count):
+        byte_count = max(0, int(byte_count))
+        raw = self._consume_front(byte_count)
+        if not raw:
+            return 0
+        return int.from_bytes(raw, 'little')
+
+    def ConsumeFloat(self):
+        raw = self._consume_front(8)
+        if len(raw) < 8:
+            raw = raw + b'\x00' * (8 - len(raw))
+        return struct.unpack('<d', raw)[0]
+
+    def ConsumeRegularFloat(self):
+        val = self.ConsumeFloat()
+        if val != val or val == float('inf') or val == float('-inf'):
+            return 0.0
+        return val
+
+    def ConsumeUnicode(self, count):
+        count = max(0, int(count))
+        if count == 0 or self.remaining_bytes() == 0:
+            return ''
+        # First byte selects encoding mode (matching atheris behavior)
+        mode_byte = self._consume_front(1)
+        mode = mode_byte[0] if mode_byte else 0
+        if mode == 1:
+            # ASCII mode: one byte per character, masked to 0-127
+            raw = self._consume_front(count)
+            return ''.join(chr(b & 0x7f) for b in raw)
+        elif mode == 2:
+            # UTF-16 mode: two bytes per character
+            raw = self._consume_front(count * 2)
+            chars = []
+            for i in range(0, len(raw) - 1, 2):
+                cp = int.from_bytes(raw[i:i + 2], 'little')
+                chars.append(chr(cp))
+            return ''.join(chars[:count])
+        else:
+            # UTF-32 mode: four bytes per character, clamped to valid range
+            raw = self._consume_front(count * 4)
+            chars = []
+            for i in range(0, len(raw) - 3, 4):
+                cp = int.from_bytes(raw[i:i + 4], 'little') & 0x10ffff
+                try:
+                    chars.append(chr(cp))
+                except (ValueError, OverflowError):
+                    chars.append(' ')
+            return ''.join(chars[:count])
+
+    def ConsumeUnicodeNoSurrogates(self, count):
+        count = max(0, int(count))
+        if count == 0 or self.remaining_bytes() == 0:
+            return ''
+        mode_byte = self._consume_front(1)
+        mode = mode_byte[0] if mode_byte else 0
+        if mode == 1:
+            raw = self._consume_front(count)
+            return ''.join(chr(b & 0x7f) for b in raw)
+        elif mode == 2:
+            raw = self._consume_front(count * 2)
+            chars = []
+            for i in range(0, len(raw) - 1, 2):
+                cp = int.from_bytes(raw[i:i + 2], 'little')
+                if 0xD800 <= cp <= 0xDFFF:
+                    cp -= 0xD800
+                chars.append(chr(cp))
+            return ''.join(chars[:count])
+        else:
+            raw = self._consume_front(count * 4)
+            chars = []
+            for i in range(0, len(raw) - 3, 4):
+                cp = int.from_bytes(raw[i:i + 4], 'little') & 0x10ffff
+                if 0xD800 <= cp <= 0xDFFF:
+                    cp -= 0xD800
+                try:
+                    chars.append(chr(cp))
+                except (ValueError, OverflowError):
+                    chars.append(' ')
+            return ''.join(chars[:count])
+
+    def ConsumeString(self, count):
+        return self.ConsumeUnicode(count)
+
+    # -- Back-consuming methods (ConsumeIntInRange, ConsumeBool, etc.) --
+
+    def _consume_back(self, n):
+        n = min(n, self.remaining_bytes())
+        result = self._data[self._back - n:self._back]
+        self._back -= n
+        return result
+
+    def ConsumeIntInRange(self, lo, hi):
+        lo, hi = int(lo), int(hi)
+        if lo > hi:
+            lo, hi = hi, lo
+        if lo == hi:
+            return lo
+        rng = hi - lo
+        # Match LLVM: consume ceil(bits_needed/8) bytes from back
+        # LLVM loops while offset < sizeof(T)*8 && (range >> offset) > 0
+        nbytes = (rng.bit_length() + 7) // 8
+        raw = self._consume_back(nbytes)
+        if not raw:
+            return lo
+        # LLVM reads bytes from back as big-endian accumulation:
+        #   result = (result << 8) | next_byte_from_back
+        # which equals int.from_bytes(reversed_bytes, 'big')
+        # But since _consume_back returns bytes in memory order and
+        # int.from_bytes(raw, 'little') produces the same value, we use that.
+        val = int.from_bytes(raw, 'little')
+        return lo + (val % (rng + 1))
+
+    # Alias for LLVM naming compatibility
+    ConsumeIntegralInRange = ConsumeIntInRange
+
+    def ConsumeBool(self):
+        # Matches LLVM: 1 & ConsumeIntegral<uint8_t>()
+        # ConsumeIntegral<uint8_t>() = ConsumeIntegralInRange(0, 255)
+        return (self.ConsumeIntInRange(0, 255) & 1) == 1
+
+    def ConsumeProbability(self):
+        # Matches LLVM: ConsumeIntegral<uint64_t>() / UINT64_MAX
+        # ConsumeIntegral<uint64_t>() = ConsumeIntegralInRange(0, 2^64-1)
+        # When range == UINT64_MAX, no modulo is applied (special case)
+        raw = self._consume_back(8)
+        if not raw:
+            return 0.0
+        val = int.from_bytes(raw, 'little')
+        return val / float((1 << 64) - 1)
+
+    def ConsumeFloatInRange(self, lo, hi):
+        lo, hi = float(lo), float(hi)
+        if lo > hi:
+            lo, hi = hi, lo
+        p = self.ConsumeProbability()
+        return lo + (hi - lo) * p
+
+    def PickValueInList(self, lst):
+        if not lst:
+            raise ValueError("list must not be empty")
+        idx = self.ConsumeIntInRange(0, len(lst) - 1)
+        return lst[idx]
+
+    # -- List methods --
+
+    def ConsumeIntList(self, count, byte_count):
+        count = max(0, int(count))
+        return [self.ConsumeInt(byte_count) for _ in range(count)]
+
+    def ConsumeIntListInRange(self, count, lo, hi):
+        count = max(0, int(count))
+        return [self.ConsumeIntInRange(lo, hi) for _ in range(count)]
+
+    def ConsumeFloatList(self, count):
+        count = max(0, int(count))
+        return [self.ConsumeFloat() for _ in range(count)]
+
+    def ConsumeFloatListInRange(self, count, lo, hi):
+        count = max(0, int(count))
+        return [self.ConsumeFloatInRange(lo, hi) for _ in range(count)]
+
+    def ConsumeProbabilityList(self, count):
+        count = max(0, int(count))
+        return [self.ConsumeProbability() for _ in range(count)]
+
+    def ConsumeRegularFloatList(self, count):
+        count = max(0, int(count))
+        return [self.ConsumeRegularFloat() for _ in range(count)]
+
+    # -- Arbitrary value --
+
+    _ANY_TYPE_INT = 0
+    _ANY_TYPE_FLOAT = 1
+    _ANY_TYPE_BOOL = 2
+    _ANY_TYPE_BYTES = 3
+    _ANY_TYPE_STRING = 4
+    _ANY_TYPE_NONE = 5
+
+    def ConsumeRandomValue(self):
+        """Return a value of a randomly chosen primitive type."""
+        t = self.ConsumeIntInRange(self._ANY_TYPE_INT, self._ANY_TYPE_NONE)
+        if t == self._ANY_TYPE_INT:
+            return self.ConsumeInt(4)
+        elif t == self._ANY_TYPE_FLOAT:
+            return self.ConsumeFloat()
+        elif t == self._ANY_TYPE_BOOL:
+            return self.ConsumeBool()
+        elif t == self._ANY_TYPE_BYTES:
+            return self.ConsumeBytes(self.ConsumeIntInRange(0, 64))
+        elif t == self._ANY_TYPE_STRING:
+            return self.ConsumeUnicode(self.ConsumeIntInRange(0, 64))
+        else:
+            return None