2023-01-06 17:44:58

by Warner Losh

Subject: [PATCH v2] tools/nolibc: Fix S_ISxxx macros

The mode field has the type encoded as an value in a field, not as a bit
mask. Mask the mode with S_IFMT instead of each type to test. Otherwise,
false positives are possible: eg S_ISDIR will return true for block
devices because S_IFDIR = 0040000 and S_IFBLK = 0060000 since mode is
masked with S_IFDIR instead of S_IFMT. These macros now match the
similar definitions in tools/include/uapi/linux/stat.h.

Signed-off-by: Warner Losh <[email protected]>
---
tools/include/nolibc/types.h | 14 +++++++-------
1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/tools/include/nolibc/types.h b/tools/include/nolibc/types.h
index 959997034e55..7770328dcef7 100644
--- a/tools/include/nolibc/types.h
+++ b/tools/include/nolibc/types.h
@@ -26,13 +26,13 @@
#define S_IFSOCK 0140000
#define S_IFMT 0170000

-#define S_ISDIR(mode) (((mode) & S_IFDIR) == S_IFDIR)
-#define S_ISCHR(mode) (((mode) & S_IFCHR) == S_IFCHR)
-#define S_ISBLK(mode) (((mode) & S_IFBLK) == S_IFBLK)
-#define S_ISREG(mode) (((mode) & S_IFREG) == S_IFREG)
-#define S_ISFIFO(mode) (((mode) & S_IFIFO) == S_IFIFO)
-#define S_ISLNK(mode) (((mode) & S_IFLNK) == S_IFLNK)
-#define S_ISSOCK(mode) (((mode) & S_IFSOCK) == S_IFSOCK)
+#define S_ISDIR(mode) (((mode) & S_IFMT) == S_IFDIR)
+#define S_ISCHR(mode) (((mode) & S_IFMT) == S_IFCHR)
+#define S_ISBLK(mode) (((mode) & S_IFMT) == S_IFBLK)
+#define S_ISREG(mode) (((mode) & S_IFMT) == S_IFREG)
+#define S_ISFIFO(mode) (((mode) & S_IFMT) == S_IFIFO)
+#define S_ISLNK(mode) (((mode) & S_IFMT) == S_IFLNK)
+#define S_ISSOCK(mode) (((mode) & S_IFMT) == S_IFSOCK)

/* dirent types */
#define DT_UNKNOWN 0x0
--
2.33.1

2023-01-08 10:05:41

by Willy Tarreau

Subject: Re: [PATCH v2] tools/nolibc: Fix S_ISxxx macros

Hello Warner,

On Fri, Jan 06, 2023 at 10:18:34AM -0700, Warner Losh wrote:
> The mode field has the type encoded as an value in a field, not as a bit
> mask. Mask the mode with S_IFMT instead of each type to test. Otherwise,
> false positives are possible: eg S_ISDIR will return true for block
> devices because S_IFDIR = 0040000 and S_IFBLK = 0060000 since mode is
> masked with S_IFDIR instead of S_IFMT. These macros now match the
> similar definitions in tools/include/uapi/linux/stat.h.

I'm impressed to see how I managed to make that gross a mistake!
Thanks for catching it and for the fix, its now queued.

Willy