cpp token_buffer not being NULL terminated causes fixproto bugs
Zack Weinberg
zack@wolery.cumb.org
Fri Apr 28 14:56:00 GMT 2000
On Fri, Apr 28, 2000 at 04:47:47PM -0400, Kaveh R. Ghazi wrote:
>
> My concern is that I didn't find all the places that make a NULL
> termination assumption... Another set of eyeballs on the code would
> be good.
>
> If you want a test input file, here's errno.h from sunos4. When you
> recode using cpp_idcmp, make sure fix-header does not insert an extra
> "extern int errno;" into this file.
Thanks.
I couldn't find any places where you missed an assumed NUL-
termination. Would you care to give this patch a whirl on sunos4? It
works for me on your errno.h example. As a fringe bonus, we no longer
mess up glibc's getopt.h.
zw
* cpplex.c (cpp_idcmp): New function.
* cpplib.h: Prototype it.
* fix-header.c (read_scan_file): Use it to inspect token
names.
* scan_decls.c (scan_decls): Likewise.
* fix-header.c: If we are scanning getopt.h, define
__GNU_LIBRARY__. Set system_header_p on the file being run
through the preprocessor.
* fixproto: Don't redefine __STDC__ to 0.
* Makefile.in: Remove stale warning message.
===================================================================
Index: Makefile.in
--- Makefile.in 2000/04/28 06:18:14 1.428
+++ Makefile.in 2000/04/28 21:51:42
@@ -2296,7 +2296,6 @@ fixhdr.ready: fix-header
# The if statement is so that we don't run fixproto a second time
# if it has already been run on the files in `include'.
stmp-fixproto: fixhdr.ready fixproto stmp-int-hdrs
- @echo "Various warnings and error messages from fixproto are normal"
if [ -f include/fixed ] ; then true; \
else \
: This line works around a 'make' bug in BSDI 1.1.; \
===================================================================
Index: cpplex.c
--- cpplex.c 2000/04/28 18:17:54 1.30
+++ cpplex.c 2000/04/28 21:51:43
@@ -2048,6 +2048,37 @@ _cpp_init_input_buffer (pfile)
pfile->input_buffer_len = 8192;
}
+/* Utility routine:
+ Compares, in the manner of strcmp(3), the token beginning at TOKEN
+ and extending for LEN characters to the NUL-terminated string
+ STRING. Typical usage:
+
+ if (! cpp_idcmp (pfile->token_buffer + here, CPP_WRITTEN (pfile) - here,
+ "inline"))
+ { ... }
+ */
+
+int
+cpp_idcmp (token, len, string)
+ const U_CHAR *token;
+ size_t len;
+ const char *string;
+{
+ size_t len2 = strlen (string);
+ int r;
+
+ if ((r = memcmp (token, string, MIN (len, len2))))
+ return r;
+
+ /* The longer of the two strings sorts after the shorter. */
+ if (len == len2)
+ return 0;
+ else if (len < len2)
+ return -1;
+ else
+ return 1;
+}
+
#if 0
/* Lexing algorithm.
===================================================================
Index: cpplib.h
--- cpplib.h 2000/04/27 05:49:33 1.86
+++ cpplib.h 2000/04/28 21:51:44
@@ -655,8 +655,8 @@ extern cpp_buffer *cpp_push_buffer PARAM
extern cpp_buffer *cpp_pop_buffer PARAMS ((cpp_reader *));
extern void cpp_scan_buffer PARAMS ((cpp_reader *, cpp_printer *));
extern void cpp_scan_buffer_nooutput PARAMS ((cpp_reader *));
-
-
+extern int cpp_idcmp PARAMS ((const unsigned char *,
+ size_t, const char *));
/* In cpphash.c */
extern int cpp_defined PARAMS ((cpp_reader *,
===================================================================
Index: fix-header.c
--- fix-header.c 2000/04/14 23:29:45 1.41
+++ fix-header.c 2000/04/28 21:51:44
@@ -135,7 +135,11 @@ enum special_file
#ifdef sys_stat_h
#undef sys_stat_h
#endif
- sys_stat_h
+ sys_stat_h,
+#ifdef getopt_h
+#undef getopt_h
+#endif
+ getopt_h
};
/* A NAMELIST is a sequence of names, separated by '\0', and terminated
@@ -637,6 +641,15 @@ read_scan_file (in_fname, argc, argv)
if (! cpp_start_read (&scan_in, 0, in_fname))
exit (FATAL_EXIT_CODE);
+ /* The getopt.h from glibc does not prototype getopt() unless
+ __GNU_LIBRARY__ is defined, and it does not include <features.h>
+ to get it defined. You don't want to know why this is. */
+ if (special_file_handling == getopt_h)
+ cpp_define (&scan_in, "__GNU_LIBRARY__");
+
+ /* We are scanning a system header, so mark it as such. */
+ CPP_BUFFER (&scan_in)->system_header_p = 1;
+
scan_decls (&scan_in, argc, argv);
for (cur_symbols = &symbol_table[0]; cur_symbols->names; cur_symbols++)
check_macro_names (&scan_in, cur_symbols->names);
@@ -663,6 +676,8 @@ read_scan_file (in_fname, argc, argv)
{
enum cpp_ttype token = cpp_get_token (&scan_in);
int length = CPP_WRITTEN (&scan_in) - old_written;
+ unsigned char *id = scan_in.token_buffer + old_written;
+
CPP_SET_WRITTEN (&scan_in, old_written);
if (token == CPP_EOF) /* Should not happen ... */
break;
@@ -671,8 +686,7 @@ read_scan_file (in_fname, argc, argv)
cpp_pop_buffer (&scan_in);
break;
}
- if (token == CPP_NAME && length == 7
- && strcmp ("_filbuf", scan_in.token_buffer + old_written) == 0)
+ if (token == CPP_NAME && cpp_idcmp (id, length, "_filbuf") == 0)
seen_filbuf++;
}
if (seen_filbuf)
@@ -690,7 +704,7 @@ read_scan_file (in_fname, argc, argv)
SET_REQUIRED (fn);
if (need_flsbuf)
SET_REQUIRED (flsbuf_fn);
- if (need_flsbuf + need_filbuf == 2)
+ if (need_flsbuf && need_filbuf)
new_list = "_filbuf\0_flsbuf\0";
else if (need_flsbuf)
new_list = "_flsbuf\0";
@@ -1130,6 +1144,8 @@ main (argc, argv)
special_file_handling = stdlib_h, required_other+=2;
else if (strcmp (inc_filename, "stdio.h") == 0)
special_file_handling = stdio_h;
+ else if (strcmp (inc_filename, "getopt.h") == 0)
+ special_file_handling = getopt_h;
include_entry = std_include_table;
while (include_entry->name != NULL
&& ((strcmp (include_entry->name, CONTINUED) == 0)
===================================================================
Index: fixproto
--- fixproto 2000/04/17 14:40:03 1.9
+++ fixproto 2000/04/28 21:51:45
@@ -58,7 +58,7 @@ dirname=`echo "$0" | sed 's,^[^/]*$,.,;s
progname=`echo "$0" | sed 's,.*/,,'`
original_dir=`pwd`
FIX_HEADER=${FIX_HEADER-$original_dir/fix-header}
-DEFINES="-D__STDC__=0 -D__cplusplus ${FIXPROTO_DEFINES}"
+DEFINES="-D__cplusplus ${FIXPROTO_DEFINES}"
if mkdir -p . 2> /dev/null; then
# Great, mkdir accepts -p
===================================================================
Index: scan-decls.c
--- scan-decls.c 2000/04/20 19:33:11 1.16
+++ scan-decls.c 2000/04/28 21:51:45
@@ -191,9 +191,7 @@ scan_decls (pfile, argc, argv)
maybe_handle_comma:
if (token != CPP_COMMA)
goto new_statement;
-#if 0
- handle_comma:
-#endif
+
/* Handle multiple declarators in a single declaration,
as in: extern char *strcpy (), *strcat (), ... ; */
if (declarator_start == 0)
@@ -202,12 +200,14 @@ scan_decls (pfile, argc, argv)
break;
case CPP_NAME:
/* "inline" and "extern" are recognized but skipped */
- if (strcmp (pfile->token_buffer, "inline") == 0)
+ if (!cpp_idcmp (pfile->token_buffer,
+ CPP_WRITTEN (pfile), "inline"))
{
saw_inline = 1;
CPP_SET_WRITTEN (pfile, start_written);
}
- if (strcmp (pfile->token_buffer, "extern") == 0)
+ else if (!cpp_idcmp (pfile->token_buffer,
+ CPP_WRITTEN (pfile), "extern"))
{
saw_extern = 1;
CPP_SET_WRITTEN (pfile, start_written);
More information about the Gcc-patches
mailing list