GH-50707: [C++][Gandiva] fix out-of-bounds read in mask utf8proc length

cpp/src/gandiva/gdv_function_stubs.cc

@@ -484,7 +484,7 @@ const char* gdv_mask_first_n_utf8_int32(int64_t context, const char* data,
   while ((chars_masked < n_to_mask) && (bytes_masked < data_len)) {
     auto char_len =
         utf8proc_iterate(reinterpret_cast<const utf8proc_uint8_t*>(data + bytes_masked),
-                         data_len, &utf8_char);
+                         data_len - bytes_masked, &utf8_char);
 
     if (char_len < 0) {
       gdv_fn_context_set_error_msg(context, utf8proc_errmsg(char_len));
@@ -592,7 +592,7 @@ const char* gdv_mask_last_n_utf8_int32(int64_t context, const char* data,
   while ((bytes_read < data_len) && (chars_counter < (num_of_chars - n_to_mask))) {
     auto char_len =
         utf8proc_iterate(reinterpret_cast<const utf8proc_uint8_t*>(data + bytes_read),
-                         data_len, &utf8_char);
+                         data_len - bytes_read, &utf8_char);
     chars_counter++;
     bytes_read += static_cast<int>(char_len);
   }
@@ -606,7 +606,7 @@ const char* gdv_mask_last_n_utf8_int32(int64_t context, const char* data,
   while (bytes_read < data_len) {
     auto char_len =
         utf8proc_iterate(reinterpret_cast<const utf8proc_uint8_t*>(data + bytes_read),
-                         data_len, &utf8_char);
+                         data_len - bytes_read, &utf8_char);
     switch (utf8proc_category(utf8_char)) {
       case 1:
         out[out_idx] = 'X';
@@ -695,7 +695,12 @@ const char* mask_utf8_utf8_utf8_utf8(int64_t context, const char* data, int32_t
   while (bytes_read < data_len) {
     auto char_len =
         utf8proc_iterate(reinterpret_cast<const utf8proc_uint8_t*>(data + bytes_read),
-                         data_len, &utf8_char);
+                         data_len - bytes_read, &utf8_char);
+    if (char_len < 0) {
+      gdv_fn_context_set_error_msg(context, utf8proc_errmsg(char_len));
+      *out_len = 0;
+      return nullptr;
+    }
     switch (utf8proc_category(utf8_char)) {
       case UTF8PROC_CATEGORY_LU:
         memcpy(out + out_index, upper, upper_length);

cpp/src/gandiva/gdv_function_stubs_test.cc

@@ -1151,6 +1151,34 @@ TEST(TestGdvFnStubs, TestMaskLastN) {
   EXPECT_EQ(expected, std::string(result, out_len));
 }
 
+TEST(TestGdvFnStubs, TestMaskTruncatedUtf8NoOverread) {
+  gandiva::ExecutionContext ctx;
+  int64_t ctx_ptr = reinterpret_cast<int64_t>(&ctx);
+  int32_t out_len = -1;
+
+  // A byte > 127 routes the mask functions through the utf8proc multi-byte path.
+  // The buffer holds a complete euro sign (0xE2 0x82 0xAC) but the reported
+  // length stops one byte short, so the trailing glyph is truncated. The
+  // functions must not read the byte past data_len: utf8proc_iterate has to be
+  // told only data_len - offset bytes remain, otherwise it consumes the
+  // out-of-range continuation byte and decodes a full glyph instead of
+  // reporting the truncated input.
+  const char buf[] = {'a', static_cast<char>(0xE2), static_cast<char>(0x82),
+                      static_cast<char>(0xAC)};
+  const int32_t truncated_len = 3;  // 'a' + first two bytes of the euro sign
+
+  ctx.Reset();
+  gdv_mask_first_n_utf8_int32(ctx_ptr, buf, truncated_len, 4, &out_len);
+  EXPECT_EQ(out_len, 0);
+  EXPECT_TRUE(ctx.has_error());
+
+  out_len = -1;
+  ctx.Reset();
+  mask_utf8(ctx_ptr, buf, truncated_len, &out_len);
+  EXPECT_EQ(out_len, 0);
+  EXPECT_TRUE(ctx.has_error());
+}
+
 TEST(TestGdvFnStubs, TestTranslate) {
   gandiva::ExecutionContext ctx;
   int64_t ctx_ptr = reinterpret_cast<int64_t>(&ctx);