fix memmove direction

src/CMakeLists.txt

@@ -8,6 +8,7 @@ add_library(bbs_sources OBJECT
 	compat-string.c
 )
 add_dependencies(bbs_sources blst)
+set_source_files_properties(compat-string.c PROPERTIES COMPILE_OPTIONS "-fno-builtin")
 
 
 # Main library

src/compat-string.c

@@ -2,9 +2,16 @@
 #include "compat-string.h"
 #include <stdint.h>
 
-// In general, functions which set memory have a memory barrier if the compiler
-// supports them, and functions reading from memory should have timing
-// independent of the respective memory contents.
+// Constant-time memory operations for cryptographic use.
+//
+// Functions that write memory include a compiler barrier to prevent
+// dead-store elimination. Functions that read memory execute in time
+// independent of the memory contents.
+//
+// These are intentionally NOT named memset/memcpy/etc. — overriding libc
+// symbols affects the entire linked program, breaking performance and
+// correctness of unrelated code. The -fno-builtin flag (set in CMakeLists)
+// prevents the compiler from replacing these byte loops with calls to libc.
 
 static void *barrier(void *ret) {
 #if defined(__GNUC__) || defined(__clang__)
@@ -13,11 +20,9 @@ static void *barrier(void *ret) {
 	return ret;
 }
 
-// GCC likes to replace loops with calls to these four functions. That is a) why
-// they still have their original symbol names here, and b) why we must use
-// function attributes to prevent GCC from implementing e.g. memcpy as a
-// recursive call to memcpy *facepalm-emoji*
-
+// GCC's -ftree-loop-distribute-patterns pass recognizes byte-by-byte loops
+// and replaces them with calls to memset/memcpy. This attribute disables
+// that pass per-function. Clang relies on -fno-builtin instead.
 #if defined (__GNUC__) && !defined (__clang__)
 # define NO_OPTIMIZE __attribute__((optimize("no-tree-loop-distribute-patterns")))
 #else
@@ -26,16 +31,16 @@ static void *barrier(void *ret) {
 
 
 NO_OPTIMIZE
-void *memcpy(void *dest, const void *src, size_t n) {
+void *bbs_memcpy(void *dest, const void *src, size_t n) {
 	char *sc = (char*)src, *dc = (char*)dest;
 	while(n--) *dc++ = *sc++;
 	return barrier(dest);
 }
 
 NO_OPTIMIZE
-void *memmove(void *dest, const void *src, size_t n) {
+void *bbs_memmove(void *dest, const void *src, size_t n) {
 	char *sc = (char*)src + n, *dc = (char*)dest + n;
-	if(src < dest) return memcpy(dest, src, n);
+	if(dest <= src) return bbs_memcpy(dest, src, n);
 	while(n--) *--dc = *--sc;
 	return barrier(dest);
 }
@@ -45,39 +50,19 @@ void *memmove(void *dest, const void *src, size_t n) {
 // a public volatile all-zero value.
 volatile int _bbs_zero = 0;
 NO_OPTIMIZE
-int   memcmp(const void *s1, const void *s2, size_t n) {
+int bbs_memcmp(const void *s1, const void *s2, size_t n) {
 	char *s1c = (char*)s1 + n, *s2c = (char*)s2 + n;
 	volatile int res = 0, diff;
 	while(n--) {
 		diff = (*--s1c - *--s2c) ^ _bbs_zero;
 		res = (res & (((diff-1) & ~diff) >> 8)) | diff; // cselect
 	}
-	return res; // right shift on signed values was undefined
+	return res;
 }
 
 NO_OPTIMIZE
-void *memset(void *s, int c, size_t n) {
+void *bbs_memset(void *s, int c, size_t n) {
 	char *sc = (char*)s;
 	while(n--) *sc++ = c;
 	return barrier(s);
 }
-
-// gcc likes to generate calls to the four library functions above, unless we
-// Since our implementations have some guardrails, so we do not want to let the
-// compiler assume we are explicitly using library functions outside this
-// compilation unit. However, GCC will replace e.g. memset with
-// __builtin_memset, which may be implemented and potentially removed inline.
-// Hence the different names given here. That said, do not enable LTO in case
-// compilers get smarter.
-// An alternative, suggested online, is to use -fno-builtins, which is
-// gcc-specific and may in some cases degrade performance.
-
-void *bbs_memset(void *s, int c, size_t n)
-	{ return memset(s, c, n); }
-int   bbs_memcmp(const void *s1, const void *s2, size_t n)
-	{ return memcmp(s1,s2,n); }
-void *bbs_memmove(void *dest, const void *src, size_t n)
-	{ return memmove(dest,src,n); }
-void *bbs_memcpy(void *dest, const void *src, size_t n)
-	{ return memcpy(dest,src,n); }
-