From 6e389ce2c34355d36009a8fb1666bed29fa2d4f4 Mon Sep 17 00:00:00 2001 From: "Todd C. Miller" Date: Mon, 6 Jun 2022 13:57:38 -0600 Subject: [PATCH] Only shift unsigned values to avoid implementation-specific behavior. This converts the arithmetic shifts to logical shifts. It is based in part on a stackoverflow answer by John Schultz, https://stackoverflow.com/questions/4533076/google-protocol-buffers-zigzag-encoding --- protobuf-c/protobuf-c.c | 23 +++++++++++------------ 1 file changed, 11 insertions(+), 12 deletions(-) diff --git a/protobuf-c/protobuf-c.c b/protobuf-c/protobuf-c.c index ad1bdb1..98052cd 100644 --- a/protobuf-c/protobuf-c.c +++ b/protobuf-c/protobuf-c.c @@ -316,9 +316,8 @@ int32_size(int32_t v) static inline uint32_t zigzag32(int32_t v) { - // Note: the right-shift must be arithmetic - // Note: left shift must be unsigned because of overflow - return ((uint32_t)(v) << 1) ^ (uint32_t)(v >> 31); + // Note: Using unsigned types prevents undefined behavior + return ((uint32_t)v << 1) ^ -((uint32_t)v >> 31); } /** @@ -380,9 +379,8 @@ uint64_size(uint64_t v) static inline uint64_t zigzag64(int64_t v) { - // Note: the right-shift must be arithmetic - // Note: left shift must be unsigned because of overflow - return ((uint64_t)(v) << 1) ^ (uint64_t)(v >> 63); + // Note: Using unsigned types prevents undefined behavior + return ((uint64_t)v << 1) ^ -((uint64_t)v >> 63); } /** @@ -802,7 +800,8 @@ uint32_pack(uint32_t value, uint8_t *out) } /** - * Pack a signed 32-bit integer and return the number of bytes written. + * Pack a signed 32-bit integer and return the number of bytes written, + * passed as unsigned to avoid implementation-specific behavior. * Negative numbers are encoded as two's complement 64-bit integers. * * \param value @@ -813,14 +812,14 @@ uint32_pack(uint32_t value, uint8_t *out) * Number of bytes written to `out`. */ static inline size_t -int32_pack(int32_t value, uint8_t *out) +int32_pack(uint32_t value, uint8_t *out) { - if (value < 0) { + if ((int32_t)value < 0) { out[0] = value | 0x80; out[1] = (value >> 7) | 0x80; out[2] = (value >> 14) | 0x80; out[3] = (value >> 21) | 0x80; - out[4] = (value >> 28) | 0x80; + out[4] = (value >> 28) | 0xf0; out[5] = out[6] = out[7] = out[8] = 0xff; out[9] = 0x01; return 10; @@ -2425,7 +2424,7 @@ static inline int32_t unzigzag32(uint32_t v) { // Note: Using unsigned types prevents undefined behavior - return (int32_t)((v >> 1) ^ (~(v & 1) + 1)); + return (int32_t)((v >> 1) ^ -(v & 1)); } static inline uint32_t @@ -2467,7 +2466,7 @@ static inline int64_t unzigzag64(uint64_t v) { // Note: Using unsigned types prevents undefined behavior - return (int64_t)((v >> 1) ^ (~(v & 1) + 1)); + return (int64_t)((v >> 1) ^ -(v & 1)); } static inline uint64_t