https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105396
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, the bug is clearly in asan_redzone_buffer::emit_redzone_byte.
The off == offset case is handled correctly, but the other case is valid only
if the gap is bigger such that we need to flush in between.
--- gcc/asan.cc.jj 2022-02-19 09:03:50.000000000 +0100
+++ gcc/asan.cc 2022-04-26 16:36:55.717551793 +0200
@@ -1502,6 +1502,15 @@ asan_redzone_buffer::emit_redzone_byte (
m_shadow_bytes.safe_push (value);
flush_if_full ();
}
+ else if (offset < m_prev_offset + ASAN_SHADOW_GRANULARITY * RZ_BUFFER_SIZE
+ && !m_shadow_bytes.is_empty ())
+ {
+ /* Shadow memory byte with a small gap. */
+ for (; off < offset; off += ASAN_SHADOW_GRANULARITY)
+ m_shadow_bytes.safe_push (0);
+ m_shadow_bytes.safe_push (value);
+ flush_if_full ();
+ }
else
{
if (!m_shadow_bytes.is_empty ())
fixes this.