https://bugs.exim.org/show_bug.cgi?id=2173
Bug ID: 2173
Summary: stack frame size detection is broken
Product: PCRE
Version: 8.41
Hardware: x86
OS: Linux
Status: NEW
Severity: bug
Priority: medium
Component: Code
Assignee: ph10@???
Reporter: vuvova@???
CC: pcre-dev@???
The documented way to check the stack frame size is
-pcre_exec(NULL, NULL, NULL, -999, -999, 0, NULL, 0)
This doesn't work for a while, at least since gcc 5.4.0, probably earlier. The
code looks like that:
=====================================================
int
pcre_exec(const pcre *argument_re, const pcre_extra *extra_data,
PCRE_SPTR subject, int length, int start_offset, int options, int *offsets,
int offsetcount)
{
if (re == NULL && extra_data == NULL && subject == NULL && length == -999 &&
start_offset == -999)
return match(NULL, NULL, NULL, 0, NULL, NULL, 0);
...
=====================================================
static int
match(PCRE_PUCHAR eptr, const pcre_uchar *ecode,
PCRE_PUCHAR mstart, int offset_top, match_data *md, eptrblock *eptrb,
unsigned int rdepth)
{
if (ecode == NULL)
{
if (rdepth == 0)
return match((PCRE_PUCHAR)&rdepth, NULL, NULL, 0, NULL, NULL, 1);
else
{
int len = (int)((char *)&rdepth - (char *)eptr);
return (len > 0)? -len : len;
}
}
...
=====================================================
while match() itself is rather big and complex, the code path for ecode==NULL
is simple.
Compiler clones match(), extracting that code path into a separate very small
function and inlines it directly into pcre_exec(). The result is:
$ pcretest -m -C
PCRE version 8.41 2017-07-05
Compiled with
8-bit support
UTF-8 support
Unicode properties support
Just-in-time compiler support: x86 64bit (little endian + unaligned)
Newline sequence is LF
\R matches all Unicode newlines
Internal link size = 2
POSIX malloc threshold = 10
Parentheses nest limit = 250
Default match limit = 10000000
Default recursion depth limit = 8192
Match recursion uses stack: approximate frame size = 4 bytes
As a fix, I've declared match() with __attribute__((noinline,noclone)). This
helped.
--
You are receiving this mail because:
You are on the CC list for the bug.
