cdesouza-chromium
275bd15340
[ubsan] Fix number overflow inside base10 funcs This PR fixes `int256_t`/`uint256_t` overflow inside `base10` conversion functions. The checks in place for overflow were not covering the arithmetic operations correctly. Due to being a large number, this was also causing a crash on ubsan, which was not producing a report. This change also does an upstream change to make it possible to use `base/numerics` for the necessary overflow checks over `int256_t`. A CL[1] has been submitted upstream, to have this issue correct in chromium. This is an incidental fix, and large numbers are not supported by clang's builtins. Nonetheless this upstream change allows us to use the slower path. [1] https://crrev.com/c/5914985
14 lines
663 B
Diff
14 lines
663 B
Diff
diff --git a/base/numerics/checked_math_impl.h b/base/numerics/checked_math_impl.h
|
|
index e619956a16e0b5ed2b1ecadeeb97deb6d4df4b94..027c46ed08fac68e0cf715a172ba736e3bc9101f 100644
|
|
--- a/base/numerics/checked_math_impl.h
|
|
+++ b/base/numerics/checked_math_impl.h
|
|
@@ -189,7 +189,7 @@ struct CheckedMulOp<T, U> {
|
|
|
|
Promotion presult = {};
|
|
bool is_valid = true;
|
|
- if (CheckedMulFastOp<Promotion, Promotion>::is_supported) {
|
|
+ if constexpr (CheckedMulFastOp<Promotion, Promotion>::is_supported) {
|
|
// The fast op may be available with the promoted type.
|
|
// The casts here are safe because of the "value in range" conditional
|
|
// above.
|